UNIVERSITY OF MINNESOTA. Kenneth Alan Williams. Lanny D. Schmidt Name of Faculty Adviser. Signature of Faculty Adviser. Date GRADUATE SCHOOL

Transcription

1 UNIVERSITY OF MINNESOTA This is to certify that I have examined this copy of a doctoral dissertation by Kenneth Alan Williams and have found that it is complete and satisfactory in all respects, and that any and all revisions required by the final examining committee have been made. Lanny D. Schmidt Name of Faculty Adviser Signature of Faculty Adviser Date GRADUATE SCHOOL

2 TRANSIENT AND STEADY-STATE PHENOMENA IN SHORT-CONTACT-TIME REACTORS FOR SYNGAS AND OLEFIN PRODUCTION: EXPERIMENTS AND NUMERICAL SIMULATIONS A DISSERTATION SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY KENNETH ALAN WILLIAMS IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY LANNY D. SCHMIDT, ADVISER OCTOBER 2006

3 Kenneth Alan Williams 2006

4 Acknowledgments First, I want to thank God for everything He has given me. I want to thank my advisor, Dr. Lanny D. Schmidt, for allowing me to join his group and for his compassionate guidance and practical approach to mentorship. Thanks to my parents, Mary and Dolon, for their support and to my dad for proofreading this thesis. Appreciation goes to Dr. Corey Leclerc for collaboration on the fast lightoff project and to Dr. Kevin West for his overhaul of the reactor control and data acquisition procedures. Collaboration with Dr. Raimund Horn and Nick Degenstein on the spatial experiments has been very fruitful. Thanks to Dr. Birali Runesha and the Minnesota Supercomputer Institute for the generous computational resources. I deeply appreciate the National Science Foundation Graduate Research Fellowship that funded my stipend and tuition from Finally, I want to thank my wife, Chrysanthi, for all of her support, patience, and love, and for proofreading this thesis. She has truly given me a beautiful life. i

5 Dedication To my wife, Chrysanthi, and my parents, Mary and Dolon ii

6 Abstract Catalytic reforming of hydrocarbon fuels (e.g., gasoline, diesel, or jet fuel) to produce a hydrogen-rich reformate has generated great interest for NO x abatement in diesel engines and electricity production in fuel cells. The mechanism responsible for synthesis gas production (carbon monoxide and hydrogen) and olefins (e.g., ethylene) is studied under steady-state and transient conditions during the catalytic partial oxidation (CPO) of alkane fuels (methane, n-octane, i-octane, n-decane, and n-hexadecane) in a short-contact-time reactor. The reactor employs a noble metal coated ceramic foam and can produce high yields of hydrogen or ethylene while offering improved process intensity over traditional manufacturing routes. The absence of information on the surface mechanism by which large hydrocarbons are converted into valuable chemicals serves as a prime motivator for the work in this dissertation. In order to provide a better understanding of the hypothesized complex homogeneous/heterogeneous mechanistic pathway, a dual experimental/simulation approach was employed. The transient behavior of the catalyst during lightoff and upon forced perturbations in fuel composition was measured using fast response mass spectroscopy and gas chromatography. The rapid start-up of methane and larger alkane fuels is demonstrated in less than 10 s, and the interplay between heterogeneous and homogeneous chemistry is explored. In parallel with experiments, numerical modeling was performed to distinguish between reactor models and surface mechanisms used to explain hydrocarbon oxidation and reforming. The ignition behaviors of higher alkane fuels on Rh-coated foams were investigated under CPO conditions to better understand the surface processes governing lightoff. It is demonstrated that steady-state syngas production can be attained within 5 s after admitting large alkanes (i-octane, n-octane, n-decane, or n- hexadecane) and air into a short-contact-time reactor by using an automotive fuel injector and initially preheating the Rh-coated catalyst above the respective catalytic autoignition temperature for each fuel. Minimum catalytic autoignition temperatures on Rh were ~260 C for n-octane and 240 C for i-octane and n-decane. In contrast, catalytic autoignition of n-hexadecane occurred at lower temperatures (> 220 o C) than for the other fuels investigated because of exothermic homogeneous chemistry that preheated the catalyst (30-60 o C) to a temperature (~280 o C) sufficient for surface lightoff. iii

7 The ignition kinetics for the large alkanes were also determined and compared with those of methane. The dominant energetic step for large alkane surface ignition is hypothesized to be oxygen desorption at saturation coverage as has been suggested for methane. The step(s) controlling surface ignition possessed an apparent activation energy of ~78 kj/mol that was not significantly different between fuels (p > 0.05). However, a significant difference was found between the ignition preexponential for methane, O(10 4 s -1 ), and the other large alkanes, O(10 6 s -1 ). This finding implies that the sticking coefficients for higher alkane fuels should be two orders of magnitude larger than for methane under relevant reactor operating conditions. Additionally, the effect of carbon formation on lightoff was examined and found to significantly affect ignition kinetics. The variation of carbon formation with time on-stream was also measured. Vaporization and mixing of higher alkanes (i.e., n-decane and larger alkanes) with air can be difficult because flames can occur at or below temperatures required for fuel vaporization leading to coke formation under fuel-rich conditions. An automotive fuel injector can be used for rapid and controllable fuel admission that yields suitable composition and temperature profiles to allow rapid vaporization and mixing while avoiding homogeneous reaction before the catalyst. Although reactants are continuously fed at a stoichiometry favorable for autoignition, flames upstream of the catalyst can be avoided under carefully chosen conditions. This work also seeks to better define allowable operating regimes and designs for the CPO reactor through the use of flexible computational simulations. Towards this goal, detailed 2D and 3D computational fluid dynamics models using n-decane with air simulated the upstream mixing phenomena and mapped the autoignition topography inside the reactor. Simulations combining temperature-dependent transport treatment and multi-step homogeneous chemistry allowed the determination of regions upstream of the catalyst capable of autoignition but incapable of flame formation due to the high space velocities used in the short-contact-time reactor. These models explained how autoignition and flames can be avoided before the reaction mixture reaches the catalyst by using fuel injection and could also serve as predictive tools to help in the challenging experimental design of vaporizing and mixing liquid fuels with oxygen to make olefins. iv

8 Table of Contents Acknowledgments... i Dedication Abstract... ii... iii List of Tables...xii List of Figures... xiii List of Abbreviations... xix List of Definitions... xx Chapter 1 Introduction Motivation Background Oxidative Coupling of Methane Methane as a Chemical Feedstock Brief Historical Summary and Current Status of OCM Process Quenching as a Process Improvement Alkane CPO Methane on Rh Applications of Syngas from Higher Alkanes Challenges: Vaporization and Mixing Experiments with Fuel-injected CPO Reactor Specific Aims...13 Chapter 2 Modeling of OCM Introduction Kinetic Models with Detailed Homogeneous Chemistry SENKIN Solution Method and Governing Equations Mechanism for OCM Homogeneous Chemistry Effect of Varying Temperature in the Isothermal Case Effect of Varying Feed Composition in the Isothermal Case Effect of Pressure in the Isothermal Case Quenching using Variable Temperature Profiles PFR Model with Homogeneous and Heterogeneous Chemistry...25 v

9 2.3.1 PFR Solution Method and Governing Equations Mechanism for OCM Homogeneous Chemistry Mechanism for OCM Heterogeneous Chemistry Geometry and Inlet Conditions for the Plug-flow Models Isothermal Catalyst Results Isothermal Composite Model Results Variable Temperature Composite Model Conclusions Nomenclature...30 Chapter 3 OCM over Pt-coated Foams: Experiments and Simulations Introduction Experimental Apparatus and Design Reactor Geometry and Configurations Reactor Components and Catalyst Preparation Temperature Acquisition and Thermocouple Placement Catalyst Contact Time Experimental Procedure Startup, Steady-state Operation, and Shutdown Thermal Quenching of Post-catalyst Gases Product Analysis Experimental Results Effect of Quench Gas Flowrate Effect of GHSV and Reactor Configuration Carbon Formation Modeling of Quench Control Experiments Experimental Axial Temperature Distribution Proposed Axial Temperature Distributions Chemkin Plug-flow Modeling Comparison of Experimental and Simulation Results Quenching with Compact Heat Exchanger Apparatus and Methods Results Conclusions Nomenclature...51 vi

10 Chapter 4 Modeling Steady-state Axial for Methane CPO on Rh-coated Foams Introduction Reactor Models and Multi-step Chemistry Previous Experimental/Numerical Spatial Profile Comparisons Motivation Numerical Methods PFR Model D Pore Channel Model D Porous Media Model Surface Mechanisms Step Mechanism Step Mechanism Results PFR Simulations Based on Experimental Temperature Profiles Adiabatic PFR Simulations Effect of Inlet Temperature on Oxidation Section Length Temperature, Species, and Coverage Profiles: C/O = Temperature, Species, and Coverage Profiles: C/O = 0.7 and D Simulations with 38-Step Mechanism Temperature and Species Profiles: C/O = Temperature and Species Profiles: C/O = 0.7 and Species Coverages Discussion Experimental Uncertainty Uncertainty in Model Parameters Pore Diameter Foam Tortuosity Interphase Heat and Mass Transport Catalytic Surface Area and Site Density Comparison of PFR Models and 2D Models Comparison of 2D Channel Model and 2D Porous Model Comparison of 2D Porous Model and Experimental Profiles Comparison of Mechanism 1 and Mechanism Conclusions...93 vii

11 Chapter 5 Rapid Lightoff of Syngas Production from Methane: A Transient Product Analysis Introduction Materials and Methods Experimental Apparatus and Procedure Determination of QMS System Response Time Numerical Simulations Temperature Profiles for Numerical Simulations Results Temperature and Species Profiles for Short Combustion Times Temperature and Species Profiles for Longer Combustion Times Mechanism Behavior at Steady-state Temperature and Species Profiles for Short Combustion Time Simulations Discussion Minimizing Start-up Time by Modifying Combustion Time Hydrogen Response Time in the QMS Catalyst Deactivation Kinetics and Transport Phenomena in Methane CPO Transient Simulations Temperature Profile Parameters Summary Chapter 6 Catalytic Autoignition of Higher Alkane Partial Oxidation on Rhcoated Foams Introduction Materials and Methods Experimental Apparatus Experimental Procedure Analysis of Ignition Parameters Determination of Ignition Kinetics Statistical Analysis Results Effect of Preheat Temperature on Start-up i-octane and n-octane n-decane n-hexadecane viii

12 Effect of Temperature vs. Fuel Ignition Kinetics Methane, n-octane, i-octane, and n-decane on Rh n-hexadecane: Gas-phase vs. Surface Ignition Discussion Controlling Step for Surface Lightoff Comparison with Previous Experimental Ignition Studies Comparison with Previous Numerical Ignition Studies Effect of C/O Ratio and Carbon Burnoff with n-decane Conclusions Nomenclature Chapter 7 Transient Analysis of Integral Carbon Formation for CPO on Rh Foams: Methane and Decane Introduction Materials and Methods Experimental Apparatus Experimental Procedure Titration Analysis Adsorption Model Results Methane on 10 mm 5 wt % Rh Catalyst C/O = C/O = Methane on 5 mm 20 wt % Rh Catalyst C/O = C/O = n-decane on 5 mm 20 wt % Rh Catalyst, C/O = Catalyst Overheating and Metal Migration Discussion Monolayer or Multilayer Carbon? Chemical Nature of the Carbon Chapter 8 Numerical Analysis of Upstream Mixing and Autoignition Criteria in a Fuel-injected CPO Reactor Introduction ix

13 8.2 Computational Methods Reactor Geometry D and 3D Reactor Models Vaporization Model Solution Method and Constitutive Equations Physical Properties of Components Empirical Flammability Limits Homogeneous Chemistry Mechanisms Computational Mesh Generation CFD Computations Results Heat Transfer Model Effect of Inlet Velocity Condition for 2D Axisymmetric Model Effect of Porous Media Zone in 2D Axisymmetric Model Heat Transfer Validation Summary Vaporization Model Creating an Autoignition Map Ignition Delay Time from Chemistry Mechanisms Discussion Conclusions Nomenclature Chapter 9 Conclusions and Future Work Conclusions OCM and Quenching in Short-contact-time Reactors D Reactor Model for Methane CPO Start-up, Transients, and Ignition Kinetics Carbon Formation versus Time On-stream Modeling Upstream Autoignition in CPO Reactor for Higher Alkanes Towards an Integrated Reactor Model Future Work Perturbation Methods to Probe Surface Kinetics for CPO: Deconvolution of Mass Transfer from Apparent Kinetics Microkinetic Surface Mechanism for Higher Alkane Partial Oxidation Appendix A Physical Properties of Quartz Reactor and Ceramic Foam Catalyst Support x

14 A.1 Thermal Conductivity of Fused Silica (SiO 2 ) A.2 Thermal Conductivity of Polycrystalline Alumina (Al 2 O 3 ) A.3 Pressure Drop Correlation for Ceramic Foam Appendix B Estimation of Binary Mass Diffusion Coefficients B.1 n-decane in Air Bibliography xi

15 List of Tables Table 2-1 Summary for isothermal simulations at 1.7 CH 4 /O 2 molar feed ratio Table 2-2 Summary for variable feed ratio simulations at 1300 C Table 2-3 Temperature profile parameters for Figure Table 2-4 Optimized temperature profile parameters for maximal C 2 H 4 yield Table 3-1 Comparison of experimental and simulation results...53 Table 5-1 Table 5-2 Temperature profile parameters for 1 s combustion time simulation Temperature profile parameters for 5 s combustion time simulation Table 6-1 Minimum catalytic autoignition and lightoff temperatures for C 1 -C Table 6-2 Minimum ignition and lightoff temperature for C 16 H Table 6-3 Kinetic parameters for surface ignition of C 1 C 10 alkanes Table 6-4 Kinetic parameters for ignition of C 16 H Table 6-5 Effect of surface burnoff on kinetic parameters for n-decane ignition Table 7-1 Limits of detection and quantification for C from CO and CO Table 7-2 Summary of adsorption model for C from CO Table 7-3 Summary of adsorption model for C from CO Table 7-4 Summary of adsorption model for total C Table 7-5 Total carbon as a percentage of catalyst weight Table 8-1 Table 8-2 Reynolds numbers for 19 mm ID tube Reynolds numbers for 4 mm ID tube Table 9-1 Optimized parameters for Figure Table 9-2 Optimized parameters for Figure Table 9-3 Methane on Rh mechanism used for C 8 surface simulations Table 9-4 Added steps for C 8 CPO surface mechanism on Rh Table A-1 80 ppi foam flow resistance inputs for Fluent porous media model xii

16 List of Figures Figure 1-1 Schematic of short-contact-time reactor...15 Figure 1-2 Desired axial temperature distributions for OCM (blue) and higher alkane CPO (red) compared with standard SCTR temperature distribution (black) Figure 1-3 Hypothesized chemistry during OCM Figure 1-4 Demonstration of the "rule of 100"...17 Figure 1-5 Melting (T m ), boiling (T b ), and autoignition (T ai ) temperatures for linear n- alkanes Figure 1-6 Schematic of fuel-injected short-contact-time reactor...19 Figure 2-1 Calculated methane conversion and C 2 selectivities versus time...33 Figure 2-2 Effect of transient temperature profile on product composition Figure 2-3 Use of millisecond heating and cooling to maximize C 2 H 4 yield Figure 2-4 Diagram of modeled reactor geometry showing the catalytic section and heat shield (top) and a single pore of the catalyst and heat shield (bottom) Figure 2-5 Demonstration of ethylene yield loss...35 Figure 2-6 Isothermal composite model consisting of 0.1 cm of catalytic section and 0.65 cm of noncatalytic section at 1050 C Figure 2-7 Validation of thermal quenching of post-catalytic gases at high catalyst operating temperature to give maximum ethylene yields Figure 3-1 Schematic and photograph of SCTR reactor for OCM Figure 3-2 Reactor configurations...55 Figure 3-3 Experimental apparatus diagram for default and quench control configurations Figure 3-4 Experimental apparatus diagram for quench configuration Figure 3-5 Photographs of the catalyst and blank ceramic foam monoliths...58 Figure 3-6 Effect of quench N 2 flowrate on C 2 selectivities at 5 slpm feed Figure 3-7 CH 4 and O 2 conversions (A) and static mixer and exhaust temperatures (B) as a function of feed flowrate...60 xiii

17 Figure 3-8 C 2 selectivities versus feed slpm for reactor configuration Figure 3-9 Photograph of premixed methane/air flame downstream of catalyst...62 Figure 3-10 Catalyst and quench sections of the OCM reactor during operation at 3 slpm feed in quench control configuration Figure 3-11 Proposed axial temperature distributions for 5 slpm feed in quench control configuration Figure 3-12 Calculated reactor product composition as a function of axial distance Figure 3-13 Schematic of compact heat exchanger and interface...67 Figure 3-14 Photograph of operating reactor (front view) in "staggered quench" configuration with compact heat exchanger Figure 3-15 Methane and oxygen conversions versus space velocity for reactor configurations with heat exchanger Figure 3-16 Total and individual C 2 product selectivities versus GHSV...69 Figure 3-17 Front side of heat exchanger after operation in "staggered quench" configuration Figure 4-1 Schematic of computational domain for 2D porous media model Figure 4-2 PFR species profiles based on experimental T profiles...95 Figure 4-3 PFR coverages based on experimental T profiles...96 Figure 4-4 Effect of inlet temperature on oxidation zone length (99.9% oxygen conversion) for adiabatic PFR simulations with mechanisms 1 and Figure 4-5 Adiabatic PFR species profiles (5 slpm and C/O= Figure 4-6 Adiabatic PFR coverages (5 slpm and C/O=1.0) Figure 4-7 Adiabatic PFR species profiles (5 slpm, C/O = 0.7) Figure 4-8 Adiabatic PFR species coverages (5 slpm, C/O = 0.7 and 1.3) Figure 4-9 Contour plots of temperature (T), velocity magnitude (v), and molar flows for adiabatic 2D porous media model (5 slpm and C/O = 1.0) Figure 4-10 Centerline temperature and species profiles for 2D adiabatic models Figure 4-11 Centerline temperature and species profiles for 2D adiabatic porous model (5 slpm and C/O = 0.7 and 1.3) Figure 4-12 Centerline species coverages for 2D adiabatic porous model (5 slpm and C/O = 0.7, 1.0, and 1.3) Figure 4-13 Effect of pore diameter on thermal entry length for methane CPO xiv

18 Figure 5-1 Fast lightoff reactor and QMS system schematics Figure 5-2 Normalized response of mass flow controller (MFC) and QMS system to step change in methane flowrate Figure 5-3 Effect of short combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst Figure 5-4 Effect of combustion time on CH 4 conversion (X CH4 ) and syngas selectivities (S CO and S H2 ) versus time for 5 slpm air flowrate on a 5 mm Rh catalyst Figure 5-5 Effect of longer combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst Figure 5-6 Steady-state conversions and selectivities versus isothermal reactor temperature for an atmospheric plug-flow simulation with a methane/air feed at φ = Figure 5-7 Temperature profiles for 1 s combustion time simulation Figure 5-8 Transient species profiles for 1 s combustion time simulation Figure 5-9 Temperature profiles for 5 s combustion time simulation Figure 5-10 Transient species profiles for 5 s combustion time simulation Figure 6-1 Temperatures pertinent to successful vaporization and mixing of C 1 to C 16 alkanes with air and alkane CPO ignition Figure 6-2 Schematic of fuel-injected short-contact-time reactor Figure 6-3 Demonstration of measured ignition parameters with methane (C/O = 1.0, 310 o C preheat, and 5 slpm air flowrate) Figure 6-4 Effect of 300 o C preheat on catalyst temperature and effluent species profiles during start-up for i-octane/air feed (C/O=1) at 10 slpm air flowrate Figure 6-5 Effect of varying preheat temperature on start-up time for i-octane fuel (same feed stoichiometry and flowrate as Figure 6-4) Figure 6-6 Effect of 295 o C preheat on catalyst temperature and effluent species profiles during start-up for n-decane/air feed (C/O=1) at 10 slpm air flowrate xv

19 Figure 6-7 Effect of varying preheat temperature on start-up time for n-decane fuel (same feed stoichiometry and flowrate as Figure 6-6) Figure 6-8 Effect of 315 o C preheat on catalyst temperature and effluent species profiles during start-up for n-hexadecane/air feed (C/O=1) at 5 slpm air flowrate Figure 6-9 Effect of varying preheat temperature on start-up time for n-hexadecane fuel (same feed stoichiometry and flowrate as Figure 6-8) Figure 6-10 Surface ignition kinetic parameters for C 1 -C 10 alkanes from Arrhehius plots Figure 6-11 Gas-phase vs. surface chemistry during ignition for n-hexadecane Figure 6-12 Effect of previous surface burnoff on subsequent ignition kinetics with n- decane Figure 7-1 Schematic of experimental apparatus with capillary sampler Figure 7-2 Orientation and response of sampling capillary Figure 7-3 Effect of O 2 on baseline response Figure 7-4 Variation of CO and CO 2 internal flowrates with O 2 partial pressure Figure 7-5 Example of CO 2 response adjustment model used to correct fluctuations in CO 2 baseline from changing partial pressure of O 2 in mass spectrometer Figure 7-6 Demonstration of baseline correction and titration analysis Figure 7-7 Photographs of 10 mm long, 5 wt% Rh catalyst with 5 wt% washcoat Figure 7-8 Temperature, H 2 O, CO 2, and CO flows during a burnoff at 12 min for C/O = 1.0: Figure 7-9 Effect of run time on titration at C/O = 1.0 for a 5 wt % Rh foam catalyst.214 Figure 7-10 Effect of time on stream on carbon amounts for 5 wt % Rh foam, C/O = Figure 7-11 Effect of run time on titration at C/O = 2.0 for a 5 wt % Rh foam catalyst.216 Figure 7-12 Effect of time on stream of carbon amounts for 5 wt % Rh foam, C/O = Figure 7-13 Photographs of 5 mm long, 20 wt% Rh catalyst with 5 wt% washcoat Figure 7-14 Effect of run time on titration at C/O = 1.0 for a 20 wt % Rh foam catalyst xvi

20 Figure 7-15 Effect of time on stream on carbon amounts for 20 wt % Rh foam, C/O = Figure 7-16 Effect of run time on titration at C/O = 2.0 for a 20 wt % Rh foam catalyst Figure 7-17 Effect of time on stream on carbon amounts for 20 wt % Rh foam, C/O = Figure 7-18 H 2 O, CO 2, and CO flows during a burnoff at 3 min for C/O = 1.0 with n- decane as fuel Figure 7-19 Overheating the catalyst during a carbon titration at C/O = Figure 7-20 Photographs of front heat shield, catalyst, and back heat shield after burnoff with temperature history shown in Figure Figure 7-21 Summary of total carbon amount versus run time parameterized by C/O ratio and catalyst loading for methane as fuel Figure 7-22 Adiabatic oxidation diagram demonstrating the reactivity and amount of carbon as a function of back face temperature Figure 8-1 Temperatures pertinent to successful vaporization and mixing of C 1 to C 16 alkanes with air for CPO Figure 8-2 Schematic and photograph of fuel-injected short-contact-time reactor Figure 8-3 Schematic of 2D axisymmetric model geometry Figure 8-4 Schematic of volumetric source/sink cells for vaporization model Figure 8-5 Computational grid for 3D model Figure 8-6 Comparison of the effect of inlet boundary conditions and static mixer on temperature profile Figure 8-7 Effect of porous media model on velocity and temperature Figure 8-8 Heat transfer validation for 5 slpm total air flowrate Figure 8-9 Heat transfer validation for 10 slpm air flowrate Figure 8-10 Heating efficiency for all heat transfer validation experiments Figure 8-11 Evaporation length vs. heat flux Figure 8-12 Validation of vaporization model at 10 slpm total air flowrate, C/O = Figure 8-13 Concentration and temperature profiles from 2D autoignition model for 10 slpm air, C/O=1, and wall flux = 5000 W/m xvii

21 Figure 8-14 Qualitative example of temperature, n-decane, and O 2 radial profiles at a fictitious downstream location in the preheat section Figure 8-15 Quantitative example of temperature, n-decane, and O 2 radial profiles Figure 8-16 Autoignition map for C/O = Figure 8-17 Dependence of ignition delay time on feed stoichiometry and initial temperature in a PFR model Figure 8-18 Pre-mixed n-decane flame with one-step chemistry Figure 8-19 Velocity pathlines (m/s) for (A) 5 slpm total flow at 25 o C Figure 9-1 Spatio-temporal experiment for methane on Rh Figure 9-2 H 2 and CO production rates in a differential catalyst slice Figure 9-3 Arrhenius analysis for rate data from Figure 9-2 set A Figure 9-4 Arrhenius analysis for rate data from Figure 9-2 set B Figure 9-5 Integral comparison between simulations with n-octane mechanism (Tables 9-3 and 9-4) and experiments for n-octane data at 4 slpm total flowrate Figure 9-6 Spatial profiles for adiabatic plug-flow simulation for n-octane at C/O = 1.0 and 4 slpm total flowrate Figure 9-7 Comparison of C 8 mechanism with C 1 mechanism for methane CPO performance at C/O = Figure A-1 Thermal conductivities of (A) fused silica and (B) polycrystalline alumina Figure B-1 Binary diffusion coefficients for (A) C 10 H 22 /O 2 and (B) C 1O H 22 /N 2 systems xviii

22 List of Abbreviations Although abbreviations are defined in the Nomenclature sections of the chapters in which they are used, this section defines common abbreviations encountered in multiple chapters. BET CPO CFD GHSV ID OCM OD ppi QMS RMS SCTR slpm STP Brunauer, Emmett, and Teller Catalytic partial oxidation Computational fluid dynamics Gas hourly space velocity Inside diameter Oxidative coupling of methane Outside diameter Pores per linear inch Quadrupole mass spectrometer Root-mean-square Short-contact-time reactor Standard liters per minute Standard temperature and pressure xix

23 List of Definitions Autoignition Temperature The autoignition temperature of a vapor is the temperature at which the vapor ignites spontaneously from the energy available in the environment. In the literature, it is usually given for a vapor in air at atmospheric pressure. Gas Hourly Space Velocity (GHSV) GHSV is the inverse of catalyst residence time if the reactant gases remain at 25 C and 1 atm (STP). It is defined as GHSV ν ε V o =, where ν o is the volumetric flowrate of reactants at STP, and ε and V are the void fraction and volume of the catalyst monolith, respectively. Conversion Conversion is the fraction of a reactant species that is consumed by reaction: X A NAo, NA =, N A, o where X A is the conversion of reactant A, N A,o and N A of the initial and final number of moles of reactant. Selectivity Selectivity is the fraction of converted species on an atomic basis that forms a particular product species. For example in the simple reaction A B: S B = N NB N Ao, A, where S B is the selectivity for species B, N A,o is the initial number of moles of A, and N A and N B are the moles of A and B after reaction, respectively. Note that in this example, reactants and products are assumed to contain the same number of atoms in question. If one or more species contain differing numbers of the atom in question, then the xx

24 number of moles of each species must be multiplied by the number of atoms of interest contained in that species. Yield Yield is the fraction of a reactant that forms a particular product. Using the reaction given above, then the yield to product B is YB = X A SB. xxi

25 Chapter 1 Introduction 1.1 Motivation Short-contact-time reactors (SCTRs) efficiently convert gaseous fuels (short-chain alkanes) and air (or oxygen) into useful chemicals. Two very successful and wellcharacterized reactions that can be performed in the SCTR are methane catalytic partial oxidation (CPO) and ethane dehydrogenation represented by the global reactions 1-1 and 1-2, respectively. 1 CH 4 + O2 CO + 2H2 (1-1) 2 1 CH O2 CH HO 2 (1-2) 2 The SCTR can produce synthesis gas (CO and H 2, or syngas) by methane catalytic partial oxidation over a single Pt- or Rh-coated α-alumina foam monolith at millisecond (ms) contact times with selectivities above 90% and methane conversion of at least 90% [1-4]. Ethylene production from ethane through oxidative dehydrogenation (ODH) in the SCTR over Pt-coated monoliths is characterized by selectivities of 85% and ethane conversion of 70% [5-7]. These reactions are extremely fast, exothermic, and autothermal. Consequently, the autothermal SCTR (Figure 1-1) offers a promising alternative to conventional energy-intensive processes such as steam reforming and ethane pyrolysis that produce syngas and olefins, respectively. Typical laboratory-scale operating parameters for short-chain alkane CPO in the SCTR are catalyst temperatures of approximately C, operating pressure of 1 atm, and catalyst residence times on the order of 1 ms. However, another important reaction, the oxidative coupling of methane (OCM) offers a unique challenge for the SCTR. OCM is an excellent example of a catalytic process with a complex relationship between homogeneous and heterogeneous chemistry [8], a relationship that creates a significant roadblock to desired product yields. Over the last 20 years, extensive research has been done to find a catalyst capable of converting methane to C 2 products with high yields (see section 1.2.1). Most of the catalytic reactors run at moderate temperature ( C) and with large catalyst residence times (0.2-1 s). Product throughput is low, reactors are not autothermal, and 1

26 the maximum C 2 single-pass yields are limited to approximately 25%, which is lower than the estimated commercially viable yield (30%). In these experiments, C 2 yield is lost to deep C 2 oxidation in the gas-phase after its creation by heterogeneoushomogeneous chemistry. Rather than running at moderate temperatures and large residence times, OCM has recently been studied in the SCTR over a Pt-coated monolith at higher catalyst temperatures ( C) and very small residence times (~1-2 ms) [9]; however, C 2 yields were still below 25%. As mentioned previously, C 2 yield is low because of hypothesized gas phase chemistry downstream of the reaction zone that oxidizes the valuable C 2 intermediates at high temperatures even during operation at millisecond contact times. Previous studies have paid less attention to optimizing the process (specifically, the reactor axial temperature profile) than optimizing the catalyst. The current approach is to search for active low-temperature catalysts that would attenuate unwanted postcatalyst homogeneous chemistry. However, after an exhaustive trek through the periodic table, future progress looks bleak [8]. Rather than search for a highly active low-temperature catalyst, another approach to boosting C 2 yield over 25% is by autothermally operating the SCTR catalyst at higher temperatures followed by thermal quenching of the non-equilibrium product gases leaving the catalyst (Figure 1-2). Chapters 2 and 3 explore the sensitivity of OCM product composition to rapid cooling of the SCTR post-catalyst effluent through experiments and detailed modeling. In contrast, methane to syngas (CPO) and ethane to ethylene (ODH) at short contact times give very high product yields and do not suffer the yield limitations of OCM. In addition to extensive laboratory research, feasibility studies on large-scale fixed-site applications of millisecond reactors for syngas and ethylene production have demonstrated their competitiveness with conventional endothermic reforming technologies [10, 11]. These findings have led chemical companies to begin adopting catalytic partial oxidation as a first step in new gas to liquids ventures [12-14]. However, most of the research performed on CPO has been for fixed (non-mobile) applications at steady-state conditions, and there is still much debate over the reaction mechanism. The remainder of this thesis focuses on the surface chemistry and transient performance of alkane CPO (see section 1.2.2) through the use of detailed numerical simulations of spatially resolved and transient experiments that characterize the start-up 2

27 and governing chemistry during ignition for C 1 -C 16 alkanes. Recent development of high-resolution spatially-resolved measurements of reactant and product compositions inside the catalyst have shed light on the mechanism for syngas production over Rh and Pt foams [15, 16] when combined with numerical simulation. Chapter 4 discusses the ability of state-of-the art surface mechanisms to reproduce steady-state experimental findings on Rh when combined with the correct reactor model. Development of an accurate reactor model and surface chemistry is crucial to understanding how reactor performance varies during scale-up and with increased pressure. Another critical application of CPO technology is for mobile fuel processors such as fuel cells in automobiles. Given the frequent start-ups and process fluctuations automotive reforming systems will face, they will be required to produce hydrogen from room temperature mixtures in seconds and respond almost instantaneously to dramatic shifts in power output. Therefore, design and optimization of a compact reforming system capable of producing hydrogen within several seconds are critical steps towards creating enabling technologies that can rapidly generate energy with lower harmful emissions than current practices. Chapter 5 focuses on the transient performance of a SCTR designed to begin producing steady-state quantities of hydrogen from methane in less than 5 seconds. The reactor is coupled with a fast-response mass spectrometer to determine optimum operating conditions to minimize start-up time. Study of CPO has not been limited to small alkanes. Currently, there is a focused interest on using higher alkane catalytic partial oxidation to produce syngas for many applications ranging from engine pollution abatement to electricity generation in fuel cells. In order to address these demands, research has been performed to modify the SCTR to handle liquid fuels such as n-decane, n-hexadecane, and diesel fuel [17]. Coupling of a fuel injection system with the SCTR has allowed the processing of liquid fuels with air in a robust system for mobile applications. Chapter 6 extends the transient analysis of methane ignition to the catalytic ignition and start-up with higher alkane fuels such as octane, n-decane, and n-hexadecane with mass spectrometer analysis. A comparison of the dominating energetic pathways for heterogeneous and homogeneous chemistry governing small and larger alkanes from C 1 to C 16 is made using this analysis. A major concern during the start-up of mobile fuel processors is carbon formation on the catalyst and its effect on reformer performance. In Chapter 7, a transient analysis of carbon formation on the catalyst is made for methane and decane as a function of 3

28 time on stream for a Rh catalyst. This analysis also aids in the understanding of the surface chemistry of alkane fuels over Rh as it tests a major assumption in the meanfield models used to simulate reactor performance: monolayer species coverages. Besides carbon formation on the catalyst, partial oxidation of higher alkanes with fuel injection exhibits challenging processing issues (the possibility of autoignition, flames, and carbon formation) upstream of the reaction zone. The precarious autoignition and vaporization relationship of higher alkanes (Section ) places severe constraints on reactor operating conditions. In Chapter 8, computational fluid dynamics models are used to better understand how rapid fuel heating and vaporization prevent flames or explosions. These simulations are the first step in the development of a fully predictive model capable of giving an accurate picture of the mixing phenomena in this system. A predictive model can aid in the design of reactor configurations capable of handling much more explosive feed mixtures such as higher alkanes and oxygen to produce olefins. 1.2 Background Oxidative Coupling of Methane Methane as a Chemical Feedstock Methane, the major component of natural gas reserves and the least valuable hydrocarbon, is currently used for home and industrial heating and electrical power generation. It is a natural resource that rivals liquid petroleum in abundance [18]. With the inevitable decline in liquid petroleum reserves and predicted increase in natural gas reserves, methane could become a major chemical feedstock. Unfortunately, methane is currently an underutilized resource for chemicals and liquid fuels. This underuse is partly because much of the methane is found in regions far removed from industrial centers and is often produced offshore. The development of efficient technologies to convert methane into a transportable fuel or value-added, high volume chemical would allow its extensive and cost-effective utilization. Various direct and indirect methane conversion strategies are being explored. These options include: (1) reforming of methane to create syngas, which could be followed by Fischer-Tropsch chemistry; (2) direct methane conversion to methanol and 4

29 formaldehyde; (3) oxidative coupling of methane to C 2 products; (4) direct conversion to aromatics in the absence of hydrogen. Although much research has been expended into all of these strategies, a major limitation is shared by all: the economical separation of hydrogen or hydrocarbons from dilute product streams and the separation of oxygen from air [18]. OCM is the representative method of direct methane conversion to primarily ethane and ethylene (a feedstock for synthesis of liquid fuels and a large number of synthetic materials) Brief Historical Summary and Current Status of OCM Process OCM has been the subject of intense research effort over the last 20 years. In this process, methane and oxygen are reacted over a catalyst at moderate to high temperatures to form C 2 products (ethane and ethylene). The global reaction steps for OCM are the oxygen-assisted coupling of methane and ethane ODH: 1 2CH 4 + O2 C2H6 + H2O 2 o ΔH r = kj/mol (1-3) 1 CH O2 CH HO 2 2 o ΔH r = kj/mol (1-4) Combining these reactions gives o 2CH + O C H + 2H O ΔH r = kj/mol (1-5) Although ethane and ethylene are intermediate products, the reaction often leads to the formation of thermodynamically-favored syngas. After the initial findings by Keller and Bhasin in 1982 [19], the search for a catalyst that could kinetically control the OCM reaction and enhance C 2 yield led researchers on a long journey through the periodic table. There is general agreement that the OCM mechanism is a complex heterogeneous-homogeneous process where the catalyst surface is used to generate methyl radicals that couple in the gas phase to form C 2 species [8]. Modeling efforts directed towards understanding the homogeneous contributions to OCM [20, 21] agree well with experimental findings sans catalyst; however, gas-phase chemistry by itself cannot explain OCM results over a catalytic surface. A significant challenge to OCM is that the intermediate C 2 species can react with oxygen, which is necessary for the OCM process, to form unwanted CO and CO 2 (Figure 1-3). The result is that a high C 2 selectivity is always accompanied by low methane conversion or vice versa. 5

30 Consequently, the maximum single-pass yield obtained from any catalytic surface is limited to 25%. However, the threshold for economic viability with this process is estimated to be 30-35% [22, 23]. The combined use of heterogeneous mechanisms with modeling has been of no use in searching parameter spaces for process conditions capable of increasing C 2 yields above 25% [24-26]. Many active catalytic surfaces for OCM have been reported including rare earth oxides [24, 27], Li/MgO [28-31], supported Na-MnO [32-34], and Pt [9]. The majority of OCM catalyst research has been performed with oxide catalysts operating at moderate temperatures ( C) and residence times of s giving very low gas hourly space velocities (GHSV). In addition, the reactors must be heated since the reaction on these catalysts is not autothermal. C 2 selectivities (mainly ethane and ethylene) of approximately 80% and methane conversions of 20% are common. Recently, OCM has been performed on a Pt-coated monolith in the autothermal SCTR operating at C and catalyst residence times of 2-3 ms giving large GHSV ( 5 x 10 5 hr -1 ) [9]. Reported maximum C 2 selectivity (mainly acetylene) was approximately 20% with methane conversions of 60%. Interestingly, all of the single-pass results with many different catalysts adhere to the informal "rule of 100", where the sum of methane conversion and C 2 selectivity is 100 or less (Figure 1-4). In an effort to break through the "rule of 100" and push C 2 yield above 25%, efforts to develop new reactor configurations over the last decade have shown some promise. Processes that couple a separation unit with the reactor allow the OCM reaction to take place at high selectivities and low conversion [35-39]. These processes gave overall (not per-pass) C 2 yields greater than 30%. Unfortunately, the use of a separation unit makes these processes more complicated and economically less attractive. The key is finding a process or process conditions capable of single-pass C 2 yields greater than 30%. Recently, new reactor designs have been developed. A membrane reactor configuration has given single pass C 2 yields above 30%, but product throughput is extremely small, and operation is not autothermal [40]. The use of a solid oxide electrolyte reactor [41] gives extremely high C 2 selectivity (85%) but only 4% C 2 yield. Of further note, in a potential process for natural gas upgrading through OCM, ethane may have to be fed with methane through the OCM reactor because their separation may not be economical [42]. This scenario may occur both for the natural 6

31 gas feed and the coupling product recycle stream. The effect of ethane co-fed to the OCM reactor with methane has been well studied [43-45]. Ethane addition does not have a beneficial effect on C 2 production of because of competition between the more reactive ethane with methane for catalytically-active sites Quenching as a Process Improvement Although OCM has been the subject of a huge experimental and modeling effort, very little attention has been paid to addressing what seems to be a fundamental process issue: the reactor temperature profile (Figure 1-2). Can the reactor temperature profile be tuned to maximize C 2 yield? The goal is to maintain the catalyst at high temperatures, thereby achieving significant surface chemistry while using rapid gas cooling to decrease the unwanted post-catalyst homogeneous chemistry (C 2 oxidation) and trap the desired reactive intermediates. With experiments operating at moderate temperatures and low GHSV, it would be hard to decouple the homogeneous and heterogeneous chemistry since all of the significant homogeneous chemistry occurs in the catalyst section of the reactor. However, the SCTR operates at high temperature and GHSV, so it may be possible to separate unwanted homogeneous chemistry from surface chemistry. Indeed, a significant portion of homogeneous chemistry has been shown to occur after the catalyst for ethane oxidative dehydrogenation (a process related to OCM) [7]. Based on this result, the SCTR provides an excellent apparatus to explore the use of rapid and precise cooling of post-catalyst gases for increasing C 2 yields. In Chapter 2, preliminary modeling with detailed chemistry demonstrates the theoretical effectiveness of rapid cooling with the appropriate time constants to optimize C 2 yield. Chapter 3 builds upon previous SCTR OCM work by experimentally demonstrating how coupling the SCTR with a simple quenching system can significantly increase C 2 yield over control experiments. In order to determine the accuracy of chemical mechanisms used in the preliminary modeling in Chapter 3, experimental conditions presented in Chapter 3 are also simulated, and comparisons between simulated and experimental results are made. 7

32 1.2.2 Alkane CPO Methane on Rh There is ongoing debate about the mechanism of methane CPO. The literature reviewed below focuses on results with Rh, but similar arguments may hold for other catalysts. Direct (pyrolysis oxidation) and indirect (combustion reforming) mechanisms are discussed in the literature [46]. The pyrolysis oxidation mechanism assumes that H 2 and CO are primary reaction products formed in the oxidation zone at the catalyst entrance. After methane pyrolysis (CH 4 C s + 4H s ), surface carbon reacts with surface oxygen to form CO (C s + O s CO), and surface hydrogen atoms combine to form H 2 (H s + H s H 2 ). In contrast, the combustion-reforming mechanism postulates a two-zone model with a CH 4 combustion zone at the catalyst entrance (CH 4 + 2O 2 CO 2 + 2H 2 O), and H 2 and CO production in a reforming zone downstream (CH 4 + H 2 O CO + 3H 2, CH 4 + CO 2 2CO + 2H 2, respectively). Nonspatially-resolved (integral) investigations into the mechanism are numerous in the literature. Among these some support the direct mechanism [47-52], some the indirect mechanism [53-55], and others a mixed mechanism [56, 57]. In some cases, the mechanistic conclusions contradict each other. However, the literature results indicate a dependence of the mechanism on (i) temperature [50, 58-60], (ii) pressure [53], (iii) Rh oxidation state [48, 51-54, 57, 59, 60], (iv) catalyst loading [51, 58], and (v) nature of the catalyst support [48, 54, 55, 59, 61]. Therefore, contradictions may arise because the mechanistic conclusions cannot be extrapolated beyond the conditions of the particular experiment. Because the proposed mechanisms postulate different reaction zones, spatiallyresolved measurements would help verify the assumptions. Lyubovsky et al. [56] used a stack of Microlith screens and measured species and temperature behind each screen at pressures between MPa. They argued for mainly direct formation of CO and indirect formation of H 2 by steam reforming. This conclusion was drawn by extrapolating CO and H 2 selectivities to zero CH 4 conversion. But because the spatial resolution was only 2 mm, and CH 4 and O 2 conversions at the first data point were already 40% and >80%, respectively, extrapolation over such a large range could be imprecise. High resolution axial profiles taken on Rh foams show that 40% of H 2 and 60% of CO are 8

33 produced in the oxidation zone demonstrating a mixed mechanism for syngas production [15, 16]. Without discussing individual results, an overall picture of the reaction can be extracted from the studies cited above. It seems that reduced Rh sites (Rh metal) are active for syngas production with CH 4 dissociation as the rate-limiting step. Oxidized Rh sites (Rh x O y ) lead to the formation of total oxidation products (H 2 O and CO 2 ). The oxidation state of the Rh surface depends on temperature and gas atmosphere. Many studies report CO 2 and H 2 O formation after the reaction is started on an oxidized Rh catalyst, but the selectivities turn to syngas formation after a short operational time, because reduced Rh sites are formed. At high temperatures, the overall gas atmosphere is sufficiently reducing to restore metallic Rh sites quickly even if gas phase oxygen is present. Chemisorbed oxygen can lead to total oxidation if the surface temperature is low (e.g., 500 C). In this case H 2 and CO are not desorbing quickly enough and become oxidized. The higher the catalyst temperature, the higher the selectivity to H 2 and CO. Most published studies agree that CO 2 reforming is not important in the reaction network. It seems that the support is important, especially at low Rh loadings. Some oxides can serve as an oxygen source for the Rh surface (inverse spillover of OH or H 2 O), even though the intrinsic catalytic activity of the support is low Applications of Syngas from Higher Alkanes CPO of higher alkanes (> C 8 ) has also been studied extensively [17, 62-74]. In particular, CPO of diesel fuel and its major aliphatic components (n-decane and n- hexadecane) has been touted as a process that can potentially benefit the environment through pollution abatement and clean energy generation. Diesel fuel can be reformed into syngas exothermically according to the following reaction x y CH x y + O 2 H 2 + xco (1-6) 2 2 where C x H y represents any hydrocarbon present in diesel fuel. Two possible uses for the syngas created from diesel fuel are the reduction of diesel engine emissions and the production of electricity using fuel cells. During the operation of a diesel engine, combustion produces nitrogen oxides (NO x ), a key culprit in acid rain production. Fortunately for the environment, the current 9

34 limit for NO X emissions from heavy-duty diesel engines of 2 g bhp -1 hr -1 will be decreased to 0.20 g bhp -1 hr -1 by 2007 [75]. This and other regulations geared toward reducing NO x emissions are forcing the development of new NO x abatement technologies. Syngas produced by partially oxidizing diesel fuel could help reduce NO x emissions in a diesel engine. Of particular note, experiments have shown that addition of hydrogen to gasoline improves the overall performance of the spark ignition (SI) engine [76, 77]. This improvement arises from the many favorable combustion properties of hydrogen. The addition of hydrogen to the gasoline combustion process allows operation at ratios below the lean flammability limit of gasoline/air mixtures. The lower temperatures that result from the hydrogen/gasoline/air mixture lead to decreased NO x formation and reduced engine heat losses. Furthermore, the faster burning velocities associated with hydrogen/gasoline mixtures decrease flame-quenching distances and lower hydrocarbon emissions. Finally, mixture homogeneity and combustion efficiency are improved by hydrogen s high molecular diffusivity into air. For example, an enhanced SI engine that runs with 6% hydrogen enrichment can have a 15-20% reduced fuel consumption compared to standard SI engines, along with reduced NO x and CO emissions [76]. The positive impacts of hydrogen addition to the SI engine have motivated further research on both the onboard generation of hydrogen and the logistics of hydrogen delivery to the engine [77, 78]. It has been shown preferable to generate hydrogen onboard for three reasons: (1) to avoid the use of a secondary tank to store hydrogen, (2) to avoid the need of hydrogen refueling stations, and (3) to reduce the risk of explosion of stored hydrogen. If a small amount of diesel fuel is partially oxidized to a hydrogen-rich stream, this stream can be mixed with the remaining diesel fuel and sent to the engine, thereby effecting lower engine temperatures and NO x emissions and higher efficiencies. A second way the use of syngas could benefit diesel engine emissions involves the catalytic converter. The more benign emissions of most SI engines consist of nitrogen, carbon dioxide, and water. However, smaller amounts of harmful emissions are also produced including carbon monoxide, volatile organic compounds, and NO x. Vehicles with SI engines use a catalytic converter to reduce these harmful emissions. The converter uses a reduction catalyst and an oxidation catalyst. Both catalysts consist of a 10

35 ceramic structure coated with a noble metal. The reduction catalyst uses platinum and rhodium to reduce the NO x emissions via the following reactions: 2NO N + 2O (1-7) CO + 2NO N + 2CO (1-8) 2 2 where CO acts as a reducing agent. The energy required to drive oxygen and nitrogen desorption is generated by the combustion reactions that take place on the oxidation catalyst. Unburned hydrocarbons and carbon monoxide undergo combustion on the oxidation catalyst (platinum and palladium). This catalyst effects the oxidation of CO and hydrocarbons with the remaining oxygen in the exhaust gas: 2CO + O 2CO (1-9) 2 2 y y CH x y + x+ O2 xco2 + HO 2 (1-10) 4 2 Unfortunately, current operation of diesel engines does not allow the use of the catalytic converter. Diesel engines operate under fuel-lean conditions in order to reduce the emission of particulate matter. Fuel-lean operation causes the engine to produce minute quantities of CO and unburned hydrocarbons. Therefore, no reducing agent (CO) is available to reduce NO, and very little heat is generated to help N 2 and O 2 desorb. Lack of a reducing agent renders the catalytic converter useless for a standard diesel engine. Syngas produced by partially oxidizing diesel fuel could solve this problem, as it could be mixed with the diesel engine exhaust gases and sent to a catalytic converter. The CO would serve as a reducing agent for NO. The smaller hydrocarbons, which are produced as side products, can be combusted in the catalytic converter to generate the energy required to desorb N 2 and O 2. Of further note, the compact size and high space velocities associated with the SCTR have sparked interest in its use to reform logistic fuels such as diesel and JP-8 (similar to kerosene and used by the military) into light alkanes and especially H 2 [68, 79]. H 2 can be used with fuel cells, which function either exclusively on H 2 (the protonexchange membrane fuel cell) or with H 2 in the fuel (the solid oxide fuel cell). Since a major interest currently lies in fuel cells for transportation vehicles, gasoline and diesel are essential fuels in the next generation of fuel cell vehicles. 11

36 Challenges: Vaporization and Mixing CPO of higher alkanes (such as diesel fuel) offers a distinct process challenge: fuel vaporization without autoignition. A major concern is possible fuel autoignition before the reactants reach the catalyst. The minimum temperature required for a mixture of fuel and oxygen to ignite spontaneously is usually referred to as the autoignition temperature. In general, the autoignition temperature for any flammable mixture is a function of fuel concentration, system volume and pressure, and flow conditions [80]. For normal alkanes, autoignition temperature decreases as hydrocarbon chain length increases, but reaches an asymptotic value of approximately 200 o C for alkanes larger than n-hexane. In contrast, boiling point increases as hydrocarbon chain length increases (Figure 1-5). Diesel fuel contains linear alkanes ranging from C 8 to C 20, most of which have higher boiling points than autoignition temperatures, contrary to those alkanes present in gasoline. This situation allows diesel engines to operate without an ignition source; the heat generated when compressing the fuel-air mixture inside the engine cylinder is enough to raise its temperature beyond the fuel autoignition temperature. Mixing of vaporized higher hydrocarbons and oxygen or air at partial oxidation stoichiometry can cause flames and explosions if mixing is not instantaneous. During mixing, the interface between the vaporized fuel and oxygen would have concentrations varying from pure fuel to pure air or oxygen and would include the combustion stoichiometry. Since local temperatures in regions near combustion stoichiometry may be above the autoignition temperature, the mixture could spontaneously ignite, producing flames and explosions. Partial oxidation experiments using n-decane, n-hexadecane, and diesel as the fuel are summarized in the next section. In these experiments, use of an automotive fuel injector to facilitate vaporization and mixing of the reactants before the catalyst section was found essential for reliable process operation Experiments with Fuel-injected CPO Reactor Use of a fuel injector with the SCTR has shown great promise for the efficient production of syngas [17]. CPO of n-decane and n-hexadecane with air over a Rhcoated monolith produces syngas with selectivities exceeding 80%, greater than 99% fuel conversion, and 100% oxygen conversion at catalyst contact times of 5 to 25 ms. 12

37 The reactor consists of a quartz tube with a 19-mm inside diameter, 22-mm outside diameter, and 55-cm length (Figure 1-6). Fuel is delivered at the top of the reactor using an automotive fuel injector. Air is also admitted at the top of the reactor through a separate side port. The conical dispersion of the µm size fuel droplets creates a thin film of liquid fuel on the heated inner walls of the reactor. The walls upstream of the catalyst are heated to between 250 and 400 C, depending on the fuel boiling temperature as opposed to the standard temperature profile for methane (Figure 1-2). This liquid film absorbs heat and vaporizes near the wall where there is little oxygen because of the fuel vapor boundary layer. Since vaporization and mixing of the fuel with air occur simultaneously, it is hypothesized that this avoids or reduces regions containing a combustible mixture at a temperature above the autoignition temperature. These experiments show that partial oxidation with higher alkanes, which possess lower autoignition temperatures than boiling points, is possible and can produce syngas in high yields. However, a quantitative explanation of why autoignition and flames do not occur has not been offered. In Chapter 8, detailed computational fluid dynamics modeling is presented to explain why flames are avoided by mapping out autoignition topography inside the reactor. In addition, these models can be useful in identifying the limits of successful operating conditions for future experiments. 1.3 Specific Aims The autothermal SCTR offers a promising alternative to conventional energyintensive processes that produce useful chemicals. However, there are still many questions about the reaction mechanism and process engineering that have remained unanswered. The motivation for this dissertation was to improve understanding of the kinetic phenomena and transient behavior of SCTRs with ceramic foam-based supports through the use of transient and steady-state characterization methods of reactor performance and by developing improved predictive models for reactor performance. Many of the unanswered questions described above have been addressed through the goals of this dissertation: (1) determine if modifying the standard SCTR temperature profile (Figure 1-2) can improve the yield for non-equilibrium C 2 products in the oxidative coupling of methane (Chapters 2 and 3), 13

38 (2) determine the validity of multi-step surface chemistry mechanisms and reactor models in predicting the dominating chemistry of methane CPO (Chapter 4), (3) use transient analysis to optimally design a SCTR reactor for the fast start-up of methane and study the effects of fuel and temperature on the governing surface chemistry during start-up with C 1 -C 16 alkanes (Chapters 5 and 6) and carbon formation (Chapter 7), (4) develop a numerical reactor model that can elucidate the operating conditions and upstream conditions necessary to prevent the possibility of upstream flames or explosions in the CPO of higher alkane fuels with a fuel-injected reactor (Chapter 8). 14

39 ceramic foam heat shields insulation fuel O 2 N 2 syngas olefins noble-metal coated ceramic foam catalyst Figure 1-1 Schematic of short-contact-time reactor. T liquid fuel vaporizer SCTR OCM quench SCTR liquid fuel or CH 4 and O 2 N 2 hot catalyst syngas or C 2 Figure 1-2 Desired axial temperature distributions for OCM (blue) and higher alkane CPO (red) compared with standard SCTR temperature distribution (black). 15

40 CH 4 C 2 H 6 CH 3 OO O 2 HO 2 OH. CH 3 C 2 H 5. CH 3. OH HO 2 CH 3 O O 2 C 2 H 4 H HO 2 OH O 2 CH 2 O C 2 H 3. OH CHO O 2 O 2 C 2 H 2 H HO 2 CO Figure 1-3 Hypothesized chemistry during OCM. Generally, reactions on the right half of the figure are favorable while the chemistry on the left half is unwanted. The solid arrow represents CH 3 addition. Although not shown, C 2 species can combine with methyl radicals to form higher hydrocarbons. Dotted arrows indicate hydrogen atom abstraction whereas dashed arrows are oxidation steps. Adapted from Mims et. al [81]. 16

41 1 0.8 Na/MgO La 2 O 3 SC SmO 3 S C2 = 1 - X CH4 0.2 Pt X CH4 Figure 1-4 Demonstration of the "rule of 100". The sum of methane conversion and total C 2 selectivity for any OCM catalyst is less than or equal to 100%. 17

42 temperature ( o C) T b T ai T m number of carbon atoms Figure 1-5 Melting (T m ), boiling (T b ), and autoignition (T ai ) temperatures for linear n- alkanes. Note that T ai and T bp switch hierarchy after n-decane so that alkane fuels larger than n- decane can spontaneously ignite before vaporization. 18

43 Fuel Injector Air g Heating tape Static mixer Pre-heat thermocouple Insulation Catalyst Heat shield Back-face thermocouple Products Figure 1-6 Schematic of fuel-injected short-contact-time reactor. 19

44 Chapter 2 Modeling of OCM 2.1 Introduction OCM has been the focus of a huge experimental and modeling effort over the last 20 years. Much of this effort has been expended in identifying novel catalytic surfaces that can produce C 2 compounds with high yields. Although many active catalytic surfaces have been found that operate at moderate to high temperatures, high C 2 selectivity is always accompanied by low methane conversion and vice versa. It has been hypothesized that unwanted post-catalyst C 2 oxidation converts the desirable intermediate products to unwanted equilibrium products. In an effort to avoid homogeneous chemistry, an exhaustive search for active lowtemperature catalysts has yielded little success. In contrast, very little attention has been given to finding methods to attenuate homogeneous chemistry downstream of the catalyst. The use of rapid and precise cooling to quench homogeneous chemistry after a high temperature catalyst operating at millisecond contact times may help to increase single-pass C 2 yields above 25%. This chapter discusses the viability of this technique via homogeneous and heterogeneous-homogeneous modeling of an SCTR with a Ptcoated monolith. Previous work has demonstrated that Pt is an active OCM catalyst at high GHSV with maximum C 2 yields of 9% [9]. If single-pass C 2 yields greater than 25% could be attained with the SCTR, its autothermal operation and high relative throughput would make it a highly favorable OCM process compared to other selective but less attractive processes. In section 2.2, purely kinetic transient simulations with isothermal temperature profiles are performed to determine maximum ethylene yields at a given temperature. Assumed axial temperature profiles are then used to show proof of concept of the quench system. OCM chemistry is simulated using a homogeneous gas-phase oxidation mechanism, which includes 120 species and 447 elementary reactions. While the simulations detailed in section 2.2 are a good starting point in determining the behavior of OCM gas-phase chemistry, homogeneous chemistry by itself does not explain experimental findings for the OCM process. The process is believed to be heterogeneously assisted with methyl radicals being created on the surface and coupling reactions occurring in the gas phase at moderate to high temperatures: 20

45 surface CH CH C H C H. Therefore, the next step in creating a realistic model is the introduction of surface chemistry. Plug-flow models coupled with both a homogeneous mechanism and Pt heterogeneous mechanism are presented in Section Kinetic Models with Detailed Homogeneous Chemistry SENKIN Solution Method and Governing Equations Initial homogeneous chemistry models were created using SENKIN. SENKIN computes the time evolution of a homogeneous reacting gas mixture in a closed system with the Chemkin differential equation solver [82]; therefore, it can simulate both batch and constant-density plug flow reactors. The model accounts for finite-rate elementary chemical reactions and performs kinetic sensitivity analysis with respect to the reaction rates. This program uses DASAC software [83] to solve both the nonlinear ordinary differential equations that describe the temperature and species mass fractions and the set of linear differential equations that describe the first-order sensitivity coefficients of temperature and species composition with respect to the individual reaction rates. The software is a modification and extension of Petzold's differential/algebraic equation solver termed DASSL [84]. DASAC handles the solution of the governing differential equations together with an efficient simultaneous computation of the first-order sensitivity coefficients. The numerical method is based on backwards differentiation formulas and is well suited for solving the stiff equations that are common in chemical kinetics applications. The program runs in conjunction with the Chemkin software package [85], which handles the chemical reaction mechanism. In this section, the equations for mass and energy conservation are described for the problem consisting of constant pressure and temperature as a known function of time [82]. The reacting mixture is treated as a closed system with no mass crossing the boundary, so the total mass of the mixture K m= m (2-1) k = 1 k 21

46 is constant, and dm/dt = 0. Here m k is the mass of the k th species and K is the total number of species in the mixture. The individual species are produced or destroyed according to dm k V ω W k k =, (2-2) dt where t is time, ω k is the molar production rate of the k th species by elementary reaction, W k is the molecular weight of the k th species, and V is the volume of the system, which may vary with time. Total system mass is constant, and the mass balance for each species can be written in terms of mass fractions: dyk = νω kwk, (2-3) dt where Y k = m k /m is the mass fraction of the k th species and ν = V/m is the specific volume. In this case temperature is input as a known function of time, so the energy equation is unnecessary, and the problem is completely defined by eqn The net chemical production rate of each species, ω k, results from a competition between all the chemical reactions involving that species. Each reaction proceeds according to the law of mass action, and the forward rate coefficients are in the modified Arrhenius form: E kf = kot β exp RT, (2-4) where the activation energy E, the temperature exponent β, and the pre-exponential constant k o are parameters in the model formulation Mechanism for OCM Homogeneous Chemistry The gas-phase chemistry is modeled using a detailed elementary step mechanism developed by Mims et al. [81] that was optimized for OCM. This mechanism contains 447 elementary reactions and 120 species and interfaces with the Chemkin kinetics software [85] for calculation of all thermodynamic properties Effect of Varying Temperature in the Isothermal Case Isothermal simulations, which have been previously performed in a limited parameter space [9], show the maximum ethylene yield possible at a given temperature. 22

47 The temperature range explored was C. Using the homogeneous chemistry model, these temperatures yield assumably tractable half-life times for CH 4 conversion (t 1/2 ). At 1200 C, t 1/2 is approximately 7.8 ms and at 1500 C, t 1/2 is approximately 0.3 ms. In this study, ethylene (C 2 H 4 ) is treated as the C 2 compound of interest for illustrative purposes and brevity. Results from a typical isothermal simulation are shown in Figure 2-1 for a CH 4 /O 2 feed ratio of 1.7, T = 1300 C, and P = 1atm (note: experimentally, feed ratios in this neighborhood produce maximum selectivity to ethylene [9]; however these results were attained using Pt-coated monoliths). The series reaction trend between ethane, ethylene, and acetylene is apparent. Ethylene selectivity reaches a maximum of 40% after approximately 1.5 ms, and the CH 4 conversion maximum lags behind the ethylene selectivity maximum. To maximize total C 2 H 4 production, it is necessary to maximize the product of the CH 4 conversion and C 2 H 4 selectivity (C 2 H 4 yield). The maximum C 2 H 4 yield (14.5%) is attained after approximately 2.3 ms. A summary of the isothermal simulations performed at a CH 4 /O 2 molar feed ratio of 1.7 and 1 atm is given in Table 2-1. At C, maximum C 2 H 4 yields are comparable; however, the times required to reach these maxima are quite different. As temperature approaches 1500 C, the maximum C 2 H 4 yield is 13.1% at 0.28 ms. Yield increases slightly with decreasing temperature, but the time required to reach that yield is even more dependent on temperature (as one would expect with an intermediate product) Effect of Varying Feed Composition in the Isothermal Case The effect of CH 4 /O 2 molar feed ratio on C 2 H 4 yield was also investigated. CH 4 /O 2 feed ratios between 1.0 and 2.5 were simulated at 1 atm and 1300 C (a temperature found reasonable in the previous isothermal simulations at 1.7 CH 4 /O 2 ). Table 2-2 summarizes the maximum C 2 H 4 yield and the time required to reach that yield for various molar feed ratios. As the feed ratio of methane to oxygen is increased from 1.0 to 2.5, the maximum ethylene yield predicted by the gas-phase modeling decreases significantly from 17.0% (at a feed ratio of 1.0) to 12.2% (at a feed ratio of 2.5). The time for maximal yield, on the other hand, increases as the molar feed ratio increases. 23

48 2.2.5 Effect of Pressure in the Isothermal Case Increasing system pressure impedes the maximum C 2 H 4 yield to a small extent. A simulation run at a CH 4 /O 2 ratio of 1.7, T = 1300 C, and P = 10 atm gave a maximum C 2 H 4 yield of 12.6% at 0.3 ms. Pressure decreases the time at which the maximal yield occurs for a given temperature (2.3 ms for P = 1 atm vs. 0.3 ms at P = 10 atm for T = 1300 C and CH 4 /O 2 ratio of 1.7) Quenching using Variable Temperature Profiles The results shown in the previous sections indicate that maximum C 2 H 4 yields occur in very short times. In contrast, CO and H 2 are the equilibrium products for a methane and oxygen feed at the ratios studied. In order to attain the maximum yield, reactant gases must be heated to C and then cooled back down below C within milliseconds. As a means of realistically attaining these yields, transient temperature profiles were used to show proof of concept of the thermal quench system. First order exponential heating and cooling temperature profiles were entered in SENKIN, and optimal heating and cooling times as well as optimal heating and cooling time constants were determined that effectively freeze the product gases at their maximum C 2 H 4 yield compositions. Figure 2-2 demonstrates the plausibility of using a thermal quench system to freeze the product gases at a favorable (high ethylene yield) composition. The temperature distribution parameters for Figure 2-2 are given in Table 2-3. Time constants for heating and cooling are denoted as τ hot and τ cool, respectively, while the heating and cooling time intervals are Δt heat and Δt cool, respectively. The initial temperature for Figure 2-2 is 273 K, while the model maximum temperature is 1573 K (1300 C), and the CH 4 /O 2 feed ratio is 1.7. Use of a transient temperature profile (Figure 2-2) demonstrates that after the temperature drops below 1000 C, conversion, selectivity, and yield remain unchanged over time scales on the order of 10 ms. However, in order to maximize the ethylene yield, a heating time must be used that ceases as yield is reaching its maximum value. In Figure 2-2, heating time is too long (6 ms) and yield passes its maximum before the product gases are effectively quenched. The ethylene yield after 10 ms is 10%. 24

49 The maximum ethylene yield in Figure 2-2 is attained at approximately 5.5 ms. If the heating time is adjusted to 5.5 ms and the cooling time to 4.5 ms while the time constants remain the same, the resulting ethylene yield is 14.5% at 10 ms (Figure 2-3), which is essentially the same as the maximum yield predicted by the isothermal modeling at 1300 C. Table 2-4 summarizes these optimized temperature profile parameters. 2.3 PFR Model with Homogeneous and Heterogeneous Chemistry PFR Solution Method and Governing Equations The Chemkin PLUG module simulates the behavior of plug-flow chemical reactors. More specifically, the program is designed to model the non-dispersive one-dimensional flow of a chemically reacting ideal gas mixture in a conduit of essentially arbitrary geometry. The program makes use of the Chemkin and Surface Chemkin [86] software packages to handle gas-phase and heterogeneous kinetics as well as the thermodynamic properties. In addition, the standard implicit numerical software DASSL is used to solve the set of differential/algebraic equations describing the reactor. The equations governing the behavior of the plug flow reactor (PFR) are simplified versions of the general relations for conservation of mass, energy, and momentum [87]. They can be derived by writing balances over a differential slice in the flow direction z, with the stipulations that (a) there are no variations in the transverse direction, and (b) axial diffusion of any quantity is negligible relative to the corresponding convective term. In this way, the overall mass balance (continuity equation) for the gas is found to be K g da du d ρ ρu + ρa + ua = a g W. (2-5) dz dz dz i k k gas Here ρ is the mass density and u the axial velocity of the gas, which consists of K g species; W k is the molecular weight of species k, and 25 g k is the molar production rate of this species by all surface reactions. The quantities A and a i are the cross-sectional (flow) area and the internal surface area per unit reactor length, respectively. Eqn. 2-5 simply states that the mass flowrate of the gas can change because of generation or consumption by surface reactions. A similar equation can be written for each species individually:

50 ρ K dy g k ua + Ykai gkwk = Wk gkai + ka dz gas ( ω ). (2-6) Here Y k is the mass fraction of species k, and ω k is its molar production rate by homogeneous chemistry. Such reactions cannot change the total mass of the gas, but they can alter its composition. The energy equation is not used for any of the models presented in this section since temperatures are either held constant or given as a function of axial distance in the reactor. The momentum equation for the gas expresses the balance between pressure forces, inertia, viscous drag, and momentum added to the flow by surface reactions. Thus K g dp du df A + ρua + + uai g kwk = 0, (2-7) dz dz dz gas where P is the absolute pressure and F is the drag force exerted on the gas by the tube wall, to be discussed below. Pressure is related to density via the ideal-gas equation of state: PW = ρrt. (2-8) Since the heterogeneous production rate g k will generally depend on gas and surface compositions, equations determining the site fractions Z k of the K s surface species are now needed. Assuming that these species are immobile, the steady-state conservation equations simplify to s k = 0 ; (2-9) that is, the net production rate of each surface species by heterogeneous reactions must be zero. There is, however, a complication. In PLUG, it is assumed that the total site density for each surface phase is a constant. As a result, the algebraic equations represented by eqn. 2-9 are not independent, and for each phase one of the equations must be replaced by the condition Z k = 1. (2-10) phase In order to minimize errors, eqn is used to replace eqn. 2-9 for the species having the largest site fraction. An analogous relation for the gas phase is not needed, since 26

51 eqn. 2-6 is a differential (as opposed to algebraic) equation. In fact, by summing eqn. 2-6 it can be shown that the Y k will automatically add up to unity at each point. The system of governing equations for the reactor is now mathematically closed. However, because the gas residence time (τ ) of the gas is often a quantity of interest, it is useful to include an equation that computes it automatically. This is simply dτ 1 =. (2-11) dz u The viscous drag force F is written in terms of a Fanning friction factor f as follows df 1 2 = a u f i ρ (2-12) dz 2 The friction factor can be expressed as a function of the local Reynolds number Duρ NRe =, (2-13) μ where D is the tube diameter and µ is the gas viscosity. For laminar flow (N Re < 2100) the analytical result for round tubes is 16 f =. (2-14) N Re Gas viscosity is computed by scaling the inlet value by (T/T in ) 0.5 and ignoring the composition dependence. It remains to specify the initial (inlet) conditions for the reactor. Clearly, values for ρ, u, T, P, and Y k at x = 0 are known from the ideal gas law and reactor geometry. Since there are no derivatives of the Z k in the governing equations, it might appear that no initial conditions are needed for them. However, DASSL requires that all algebraic equations be satisfied at the initial point, so eqns. 2-9 and 2-10 must be solved for the Z k using known values of the other variables. In PLUG this is accomplished in a separate preliminary calculation, in which DASSL is used to integrate the fictitious transient equations dzk s kσ k = (2-15) dt Γ in conjunction with eqn until steady state is reached. Here σ k is the site occupancy number for species k, and Γ is the total site density of the phase in question. 27

52 2.3.2 Mechanism for OCM Homogeneous Chemistry The homogeneous chemistry mechanism used for the plug-flow models is described in section Mechanism for OCM Heterogeneous Chemistry The surface chemistry is modeled using a detailed mechanism for ethane oxidation on a platinum surface developed by Zerkle et al. [88]. The ethane surface chemistry is based on the model of Wolf [89] for non-oxidative methane conversion on Pt and combined with oxidative steps. The mechanism consists of 82 elementary reactions and 19 surface species and interfaces with the Chemkin kinetics and Surface Chemkin software [85, 86] for calculation of all thermodynamic properties Geometry and Inlet Conditions for the Plug-flow Models The modeled geometry consists of a single pore of the variable-length catalyst monolith where homogeneous and heterogeneous surface chemistry apply (catalyst section) followed by a prescribed distance of blank monolith where only homogeneous chemistry applies (see Figure 2-4 for composite geometry). In typical experiments, the monolith contains 45 pores per linear inch giving a pore diameter of 0.5 mm based on a wall thickness of mm. Inlet velocities of 34 cm/s were used with a mixture of 1.7 CH 4 /O 2 (on a molar basis) for all temperature profiles. At standard temperature and pressure, a pore inlet velocity of 34 cm/s with a monolith void fraction of ~0.8 [90] is approximately 5 standard liters per minute (slpm) for the entire monolith Isothermal Catalyst Results First, to investigate the effect of temperature and residence time inside the catalyst on ethylene yield, isothermal simulations considering only the catalyst were performed. Essentially, there is a window of temperatures for the catalyst that gives attractive ethylene yields in reasonable times. At 900 C, a maximum ethylene yield of ~14.5% occurs after a residence time of 33 ms (1.2 cm downstream for this flowrate). At 1300 C, a maximum ethylene yield of ~9.7% occurs after a residence time of 0.26 ms (0.1 mm downstream for this flowrate). Typical results for temperatures in this range give maximum transient ethylene yields of 15-23%, which quickly decrease as residence time increases in the catalyst pore (Figure 2-5). 28

53 2.3.6 Isothermal Composite Model Results Within the allowable temperature operating window, small gains in ethylene yield can be obtained by increasing the catalyst operating temperature (23% at 1100 C versus 20% at 1000 C as shown in Figure 2-5). However, once the gases leave the catalyst, they are more reactive at higher temperatures, which would cause any optimized ethylene yields to quickly vanish. As an example, Figure 2-6 shows a composite model at 1050 C containing a catalyst pore followed by a noncatalytic section. The catalyst length has been optimized to approximately 1 mm so that the product gases leave the catalyst at a residence time (~3 ms) that gives the maximum ethylene yield of 22%. Unfortunately, once the gases leave the catalytic section, the optimized yield quickly drops to 9.6% after only an additional 17 ms or 0.65 cm of noncatalytic length (20 ms total residence time and 0.75 cm total length) Variable Temperature Composite Model The previous results show that product gases need to be immediately cooled after leaving the catalytic section, thereby realizing the optimized ethylene yield gained in the catalyst section. Figure 2-7 demonstrates how quenching the gases leaving the catalyst section performs this function. The catalyst section is 2.4 mm and is assigned a firstorder heating temperature profile. Gases enter the catalyst at 200 C and leave at 1100 C. Beyond the catalyst section, two sets of data are plotted for a non-catalytic section. The data denoted by dashed lines correspond to the post-catalyst gases maintained at 1100 C. The smooth lines correspond to the gases being cooled with a reasonable time constant of approximately 15 ms. The lines for ethylene yield show that quenching essentially freezes the yield at 21%. Without quenching, the yield drops to 8% after 20 ms (or 1.25 cm). 2.4 Conclusions Modeling of homogeneous chemistry with the use of millisecond heating and cooling of the reaction gases has been demonstrated to maximize ethylene yield to the same value predicted by isothermal modeling. For example, isothermal modeling at 1300 C and 1 atm for a CH 4 /O 2 feed ratio of 1.7 shows a maximum ethylene yield of 14.5% at 2.3 ms. Use of an optimized exponential heating and cooling cycle (see Table 29

54 2-4) results in the same maximum ethylene yield (14.5%) after 10 ms. However, homogeneous chemistry by itself does not explain experimental findings. The homogeneous-heterogeneous chemistry model shows that ethylene yields as high as 22% can be realized from the OCM process through the use of a short Pt catalyst section operating at higher temperature profiles ( C) than those traditionally used for this process ( C). A fast thermal quench of the gases leaving the catalytic section is necessary to maintain yields by eliminating gas-phase chemistry and trapping the valuable reactive intermediate (ethylene). 2.5 Nomenclature a i Internal surface area per unit length, m A Cross-sectional area, m 2 D E f Diameter, m Activation energy, J/kmol Fanning friction factor F Drag force, kg m/s 2 g k Molar production rate of gas species k by surface reactions, kg/(m 2 -s) k f Forward chemical rate coefficient, 1/s k o Pre-exponential factor, 1/s K K g K s m m k N Re P R s k S k Total number of species in gas phase Total number of gas species Total number of surface species Mass, kg Mass of species k, kg Reynolds number Pressure, atm Universal gas constant, x 10 3 J/(kmol-K) Net rate of production of surface species k, kg/(m 2 -s) Selectivity to species k 30

55 t t 1/2 T T i T in T max u Time, s Half-life for methane conversion, ms Temperature, K Initial temperature, K Inlet temperature, K Maximum temperature, K Mean axial velocity, m/s V Volume, m 3 W W k X k Y k Y k z Z k Average mixture molecular weight, kg/kmol Molecular weight of species k, kg/kmol Conversion of species k Yield of species k Mass fraction of species k Axial direction, m Surface site fraction of species k Greek β Temperature exponent for Arrhenius kinetic expression Γ Total site density, sites/m 2 ω k Total molar production rate of species k by elementary reaction, kmol/(m 3 -s) Δt hot Δt cool μ ν Heating time interval, ms Cooling time interval, ms Dynamic viscosity, kg/(m-s) Specific volume, m 3 /kg ρ Density, kg/m 3 σ k τ τ hot τ cool Site occupancy number for species k Residence time, ms Heating time constant, ms Cooling time constant, ms 31

56 Table 2-1 Summary for isothermal simulations at 1.7 CH 4 /O 2 molar feed ratio. T( C) maximum C 2 H 4 yield time for maximal yield (ms) % % % % 0.28 Table 2-2 Summary for variable feed ratio simulations at 1300 C. CH 4 /O 2 ratio maximum C 2 H 4 yield time for maximal yield (ms) % % % % % 3.3 Table 2-3 Temperature profile parameters for Figure 2-2. T i ( C) T max ( C) τ hot (ms) τ cool (ms) Δt heat (ms) Δt cool (ms) Table 2-4 Optimized temperature profile parameters for maximal C 2 H 4 yield. CH 4 /O 2 feed ratio = 1.7, P = 1 atm, and T max = 1300 C. T i ( C) T max ( C) τ hot (ms) τ cool (ms) Δt heat (ms) Δt cool (ms)

57 conversion / selectivity CH total C C 2 H 6 C H 4 C 2 H time (ms) Figure 2-1 Calculated methane conversion and C 2 selectivities versus time. CH 4 /O 2 feed ratio =1.7, T = 1300 C, and P =1atm. conversion / selectivity X CH S C2H Y C2H time (ms) T (K) Figure 2-2 Effect of transient temperature profile on product composition. Heating and cooling parameters are given in Table

58 conversion / selectivity S C2H4 Y C2H4 X CH time (ms) T (K) Figure 2-3 Use of millisecond heating and cooling to maximize C 2 H 4 yield. CH 4 O 2 products r z Figure 2-4 Diagram of modeled reactor geometry showing the catalytic section and heat shield (top) and a single pore of the catalyst and heat shield (bottom). 34

59 conversion / selectivity conversion / selectivity A X CH S C2H Y C2H tau (ms) B X CH S C2H4 Y C2H tau(ms) Figure 2-5 Demonstration of ethylene yield loss. (A) Ethylene yield for catalyst pore at 1000 C. A maximum yield of 20% is attained after 6.2 ms (0.22 cm downstream). (B) Ethylene yield for catalyst pore at 1100 C. A maximum yield of 23% is obtained after 1.39 ms (0.05 cm downstream). 35

60 conversion / selectivity end of catalyst (1 mm) X CH S C2H4 Y C2H tau (ms) Figure 2-6 Isothermal composite model consisting of 0.1 cm of catalytic section and 0.65 cm of noncatalytic section at 1050 C. Over 50% of the optimized ethylene yield is lost to homogeneous chemistry in the noncatalytic section. conversion / selectivity S C2H4 T X CH4 Y C2H end of catalyst (2.4 mm) tau (ms) T(K) Figure 2-7 Validation of thermal quenching of post-catalytic gases at high catalyst operating temperature to give maximum ethylene yields. Dotted lines correspond to constant temperature after the catalyst whereas solid lines correspond to a quench temperature profile. 36

61 Chapter 3 OCM over Pt-coated Foams: Experiments and Simulations 3.1 Introduction OCM has been attempted previously in the SCTR [9] with GHSV between 1x10 5 to 5x10 5 hr -1. The main C 2 products were acetylene (12-15% selectivity) and some ethylene (5-8% selectivity). Using the previous default experimental configuration, a baseline for OCM production was determined. In order to establish the effect of rapid cooling on post-catalyst gas composition, two additional configurations were employed. The quench control configuration, which serves as a control experimental apparatus, and the quench configuration, in which cold N 2 is added to the post-catalyst gases. Feed conditions for the experiments were 1.7 CH 4 /O 2 with 20% N 2 dilution and total feed flowrates of 3, 5, 7, and 9 slpm. In order to establish whether the preliminary OCM modeling (Chapter 2) gave quantitatively-correct results for the quench process, temperature information gathered from the OCM experiments was used with the Chapter 2 modeling techniques to simulate experimental results. Simulations in the previous chapter gave significantly larger C 2 yields than the experiments. One explanation for this discrepancy could be that the cooling time constant for experimental quenching is much larger than that assumed in the models. By using the Chemkin plug-flow modeling technique along with the same gas and surface chemistry mechanisms of Chapter 2, reproduction of the experimental results was attempted by approximating the axial temperature distribution observed during experiments. 3.2 Experimental Apparatus and Design Reactor Geometry and Configurations Reactants gases (CH 4, O 2, and N 2, > 99.9% pure) were supplied by Air Products (Allentown, PA). The gases were fed through Brooks 5890E series mass flow controllers to regulate the relative and total flowrates. All mass flow controllers were calibrated with a bubble column. Flowrates were accurate to within ± 1% of the maximum flow rating for each controller. The maximum flow ratings for the N 2, CH 4, and 37

62 O 2 controllers were 30 slpm, 5 slpm, and 5 slpm, respectively. All gases were fed to the reactor through 6.35 mm stainless steel tubes and were premixed in a Swagelok tee approximately 60 cm before the reactor entrance. The gases entered at the top of the reactor through a quartz endcap. The reactor consisted of a quartz tube loaded with a catalytic foam and front and back foams used as heat shields to prevent axial heat losses. The quartz tube was approximately 19 mm ID and 22 mm OD by 40 cm long. The three monoliths were wrapped in alumino-silicate paper (Unifrax Corp., Niagara Falls, NY) to make a good seal with the inside of the quartz tube and then placed in the quartz tube. The catalyst section of the reactor tube was wrapped with Fiberfrax ceramic blanket insulation (Unifrax Corp.) to attenuate radial heat losses. Connections to inlet and outlet tubes were made by Cajon fittings (Swagelok) and quartz endcaps. Leaks were prevented by sealing Viton o-rings between the endcaps and the reactor tube with clamps. The reactor was checked for leaks by flowing N 2, raising the internal pressure, and applying soap solution over all connections and fittings. A schematic for a typical SCTR is shown in Figure 3-1. Three reactor configurations were employed in the OCM experiments. These configurations were used to determine whether thermal quenching of post-catalyst gases was effective in increasing C 2 product yields. The default configuration (DC) is the standard configuration for the SCTR (Figure 3-2, left), and the design used in previous OCM experiments on Pt [9]. To determine the effects of gas phase quenching on product composition, the quench configuration (QC) and the quench control configuration (QCC) were designed (Figure 3-2, right and middle, respectively) and tested. In the DC, the catalytic monolith was placed between a front and back heat shield with no other reactor components. The QC incorporated a gas addition port that impinged room temperature N 2 directly onto the gases leaving the back heat shield. A wad of Nextel ceramic fibers was placed behind the back heat shield and formed a 25 mm long mixing section for the quench and product gases. A blank 80 pore per linear inch (ppi) monolith (1 cm long) was placed behind the ceramic fibers to ensure mixing of the quench and product gas. This monolith was denoted as the static mixer. Complete mixing of the gases was important not only to decrease the product gas temperature, but also to make sure the gas mixture was homogeneous so that product analysis was 38

63 accurate. The QCC was the same as the QC without the addition of the quench gas. The comparison of the QCC and QC determined whether the quench gas was really necessary, or if the downstream gas-phase chemistry could be abated through heat exchange with the non-insulated surfaces of the static mixer and downstream reactor tube. Product gases flowed out of a bottom endcap and through stainless steel lines to a lit Bunsen burner for incineration. The endcap, equipped with a sample port, was fitted with an self-sealing septum for product sampling via a gas-tight syringe. The product samples were analyzed with a Series II, Hewlett-Packard 5890 gas chromatograph (GC) employing a thermal conductivity detector. The experimental apparatus diagram for the DC and QCC is shown in Figure 3-3. The experimental apparatus diagram for the quench configuration is shown in Figure Reactor Components and Catalyst Preparation The catalyst for all experiments was Pt. The support in every case was an 80 ppi α-alumina foam monolith (92% alumina, 8% silica) obtained from Hi-Tech Ceramics (Alfred, NY). The diameter of the support was 17 mm, and the length was 3 mm (Figure 3-5). BET surface area of the support was approximately 1 m 2 /g [90, 91]. The front and back heat shields were also α-alumina and 80 ppi and 45 ppi, respectively. Their diameter was 17 mm, and their length was 1 cm. The porosity for all monoliths was approximately 0.8 [90]. To prepare the catalyst samples the unloaded support was weighed, and the amount of 8-wt% hydrogen hexachloro-platinate solution (Sigma-Aldrich, St. Louis, MO) in water to give 2-3 wt% Pt on the support was massed. The solution was applied evenly by a Pasteur pipette in drops onto the support until the support was saturated. The monolith was allowed to dry, turned over, and the application procedure was repeated until all the solution was spent. The loaded monolith was then calcined in air at 600 C for two hours to leave Pt in the metallic state. The sample was allowed to cool and then reweighed to determine the weight % of Pt on the monolith Temperature Acquisition and Thermocouple Placement Reactor temperatures at various locations were determined with chromel-alumel (Type K) thermocouples with an inconel sheath (Omega Engineering, Stamford, CT) that 39

64 were rated for use up to 1350 C. The thermocouple wires were connected to a digital voltmeter that read the voltage difference between the two wires. The measured voltage difference was then converted to temperature by an internal microprocessor that converted the voltage drop to temperature based on the thermocouple type ( >1% error). The number and placement of thermocouples inside the reactor depended on the reactor configuration. For the default configuration, a thermocouple was placed between the catalyst and the back heat shield to measure gas temperature leaving the catalyst (T bf ). No thermocouples were placed downstream of the catalyst section because there was significant radiation leaking from the back face of the back heat shield. Without a radiation shield behind the back heat shield, the temperature measured by a downstream thermocouple would not be indicative of the gas temperature at that point. For the QC and QCC, two additional temperature measurements were taken. The first measurement was taken behind the static mixer monolith (T sm ) and indicated how well the quench gas was cooling the product gases. The second measurement was obtained 28 cm downstream from the T sm thermocouple in the bottom endcap and indicated the exhaust temperature leaving the reactor (T ex ). Since almost no radiation was observed leaking through the static mixer in the QC and QCC, both T sm and T ex were indicative of the gas temperatures at these points. These temperature measurements also served as a comparison between the QC's and QCC's axial temperature profile Catalyst Contact Time The gas velocity in the reactor at STP is v STP o 4V =, (3-1) 2 π D where V is the volumetric flowrate of the feed, and D is the ID of the quartz tube. By the continuity equation, ρν is constant, and ρ P/T by the ideal gas law. Therefore, the inlet velocity to the front heat shield is v o To 1atm 4V = 2 298K P π D, (3-2) where T o is the temperature of the feed entering the front heat shield, and P is the reactor pressure. Velocity v through the catalyst is greater than v o because of the high catalyst temperatures during reaction and the large void fraction of the monolith. The 40

65 catalyst contact time based on inlet velocity conditions is the inverse of the gas space velocity or Lε τ =, (3-3) v o where L is the length of the catalyst, and ε is the porosity of the catalyst. For the experiments presented here, τ ranges from 12-4 ms based on feed flowrates from 3 to 9 slpm. 3.3 Experimental Procedure Startup, Steady-state Operation, and Shutdown All experiments were run autothermally. Experiments were started by flowing 3 slpm of feed gases with a CH 4 /O2 ratio of 1.7 with 20% N 2 dilution. The Fiberfrax blanket insulation was removed from the catalyst section, and the flame of a lit Bunsen burner was applied to the external surface of the quartz reactor next to the catalyst section. When T bf reached C, the catalyst section ignited causing the catalyst to glow orange-red. From that point on, exothermic reactions sustained the heat necessary to run the reactor. The Bunsen burner was removed, the insulation was placed back around the catalyst section with copper wire, and the feed flowrate was adjusted as necessary. Steady-state operation was determined by temperature readings and GC product analysis results. Typically, 45 minutes were needed for temperatures and product compositions to reach constant values for a fresh unreduced catalyst. Experiments were performed outside of the flammability limits and always on the fuel-rich side. For methane oxidation, the stoichiometric molar ratio for the combustion of a CH 4 /O 2 mixture is 0.5. In these experiments, the CH 4 /O 2 ratio was kept at 1.7 and never close to the combustion ratio. The flammability limits in a static system with air are CH 4 /O 2 = [80]. In addition to avoiding flames and possible explosions, fuel-rich operation is necessary for significant production of carbon-containing compounds other than CO and CO 2 [92]. The reactor was shut down by stopping the oxygen first and then the methane while maintaining N 2 flow. 41

66 3.3.2 Thermal Quenching of Post-catalyst Gases In the QC, product gases are cooled by a room temperature N 2 stream entering the reactor normal to the product flow direction. The necessary N 2 flowrate to properly quench product gases was determined based on a 200 C temperature drop of the product gases for a feed flowrate of 5 slpm. Experiments run in DC gave good estimates of product composition and temperature in the quench section of the reactor. By performing a sensible energy balance using product component heat capacities as a function of temperature, an N 2 quench flowrate of approximately 40% of the total feed to the reactor was estimated as sufficient. The quench gas was added to the reactor by a fourth Brooks 5890E mass flow controller. A Cajon connector was used to seal the connection between the N 2 line and the 6.35 mm OD quartz port Product Analysis A sample of the product gas was taken by a gas-tight syringe at a port in the bottom endcap that was sealed with an NMR septum. The sample was injected into a Hewlett-Packard 5890 GC equipped with a thermal conductivity detector (TCD) for product analysis. A Hayesep D fused-silica capillary column (9.2 m length and 0.32 mm ID) separated the products. Column retention times were determined by injected standardized samples of known species. Calibrated species in order of their increasing retention times were H 2, CO, O 2, CH 4, CO 2, C 2 H 2, C 2 H 4, and C 2 H 6. Total product detection took approximately 20 minutes as the last peak was water, which is an asymmetric peak that is unreliable for calibration. Analysis of the peaks (peak separation and integration) was performed using HP Chemstation 4.02 software. Nitrogen, an inert species in the reactor, was used as the calibration gas for atomic mass balances. Product selectivities were reported on an atomic basis. They were calculated as the molar ratio of a specific product to the sum of all products, scaled by the number of carbon atoms in the species. This definition accounts for the change in mole number during reaction and satisfies the requirement that all selectivities sum to one. The carbon atom selectivity to product j is S j = nf j j N, (3-4) nf i i i 42

67 where F j is the molar flowrate of species j, which contains n j carbon atoms, and N is the total number of species containing carbon. The conversion of reactant i is X F F io, i i =, (3-5) Fio, where the subscript o refers to the inlet value and i can refer to either methane or oxygen. The yield of product j is Y = S X. (3-6) j j i The F j are calculated from the GC peak areas by comparison with the N 2 peak area: FN 2 Fj = ( K j peak area of j). (3-7) peak area of N2 F N 2 is known from the feed conditions. The GC response factors K j were determined by injection of pre-analyzed gas samples, which were diluted in nitrogen (from Scotty II cylinders, Alltech, Deerfield, IL). The formula for the response factor for species j is peak area of N2 mol % of j in standard K j =. (3-8) peak area of j mol % of N2 in standard Carbon balances typically closed to within 1-5%. Water was used to close the oxygen balance. Since H 2 can be unreliable for linear calibration with a TCD, hydrogen was calibrated over a range of relative concentrations to give its response factor as a function of concentration in the standard. This extra step was necessary since hydrogen's response factor can change by a factor of two over a broad range of concentrations. Response factors for other species typically change by less than 10% over a broad range of concentrations so a linear calibration was used. 3.4 Experimental Results Effect of Quench Gas Flowrate Preliminary experiments were performed to study the effect of quench gas flowrate on the relative and total C 2 product selectivities. Feed conditions were 1.7 CH 4 /O 2 with 20% N 2 dilution and 5 slpm total flowrate. Although calculations showed that an N 2 quench flowrate of 40% of the feed flowrate was sufficient to cool the post-catalyst gases by 200 C, experiments with N 2 quench flows greater than this theoretical minimum 43

68 flowrate were also performed. Based on two duplicate experimental runs, C 2 selectivities were determined as a function of quench gas flowrate (Figure 3-6). The effect of switching from no quenching in DC to a 1.89 slpm quench flowrate in QC is large: total C 2 selectivity goes from 11% to 21%, C 2 H 2 selectivity goes from 8% to 16%, C 2 H 4 selectivity goes from 3% to 5%, and C 2 H 6 selectivities are under 1% for both experiments. Upon doubling the quench gas flowrate from 1.89 to 3.9 slpm, total C 2 selectivity remains approximately constant, C 2 H 2 decreases slightly, C 2 H 4 increases by approximately 30%, and C 2 H 6 approximately doubles. The ability to control the series reactions (CH 4 C 2 H 6 C 2 H 4 C 2 H 2 ) to give the C 2 product of choice by thermal quenching is discernable. However, limited conclusions can be drawn from the data based on the small sample size and the uncontrolled comparison between DC and QC. The two configurations have different components so comparisons between results from the two configurations are not strictly correct. At 5 slpm, adding quench gas in QCC produces more C 2 products than DC with no quench gas. However, the QC should be compared with QCC to determine the efficacy of the quenching process for increasing selectivity to C 2 products Effect of GHSV and Reactor Configuration In order to determine the effectiveness of the quench gas treatment, all three reactor configurations were run at feed gas flowrates of 3, 5, 7 and 9 slpm yielding GHSV of 3x10 5, 5x10 5, 7x10 5, and 9x10 5 hr -1, respectively. The quench gas flowrate used for QC was kept at 40% of the feed flowrate. Comparison of methane and oxygen conversion versus feed flowrate for the reactor configurations shows some expected and some unexpected trends (Figure 3-7A). For all three configurations, as GHSV is increased, both CH 4 and O 2 conversions decrease. This trend is much more pronounced for QC and QCC. As GHSV increases, the residence time decreases allowing less time for the reactants to react on the catalyst surface and less time for gas-phase chemistry before cooling occurs downstream. DC appears unaffected until 9 slpm. This result can be explained by the significant amount of radiation emitted in the downstream axial direction through the back heat shield for DC. The radiation could be keeping the downstream section of the tube hot and thus allowing gas-phase chemistry to continue. Very little to no radiation leaks through the 44

69 back of the static mixer for QC and QCC, so the downstream sections of these configurations should run at lower temperatures that minimize gas-phase chemistry. Static mixer and exhaust temperatures for QC and QCC progressively increase with GHSV (Figure 3-7B). As would be expected, T sm is higher for QCC than for QC at each feed flowrate. Therefore, the addition of quench gas significantly reduces the temperature of the post-catalyst gases. In fact, at 5 slpm the difference in T sm between QC and QCC is approximately 200 C as the original calculation for 40% quench gas predicts. Discrepancies for the differences in T sm between the two configurations point to the change in conversion as a function of GHSV and residence time in the static mixer section. The exhaust temperature is less for QCC than for QC at each GHSV. While at first this may seem counterintuitive, this result can be explained. The reactor section downstream from the catalyst section is not insulated. Through heat transfer with the ambient environment via the tube wall, post-catalyst gases should cool significantly while traveling the 28 cm from the T sm measurement location to the T ex measurement location. Results show that QCC exhaust temperature for each flowrate is lower than that for QC. It might be expected that the opposite result should occur. However, an additional amount of gas (40% of feed) is added to the post-catalyst gases in the QC, whereas no quench gas is added in the QCC. The quench gas makes a sensible heat contribution to decrease the overall gas temperature, which would increase its residence time for the rest of the tube. Nevertheless, by its mere addition it also decreases the residence time gases would otherwise experience. The downstream residence time for QC must be significantly less than that for QCC; therefore, the exhaust temperatures for QC are higher than those for QCC since there is less time for heat transfer. Of further note, back face temperatures are not shown in Figure 3-7B. For 3 slpm, a back face temperature of 1250 C was measured. For 5 slpm, the temperature went above 1400 C, which was higher than the maximum operating temperature for the thermocouple. Results for C 2 selectivities versus GHSV for each reactor configuration show promise for the quench concept. Overall C 2 selectivity is greater than 20% for QC at lower flowrates and is better than those for QCC at every flowrate investigated (Figure 3-8A). For flowrates above 3 slpm, DC shows a large boost in C 2 selectivity that is attributable to its increase in acetylene selectivity (Figure 3-8B). Interestingly, for 45

70 flowrates typically above 5 slpm for DC, yellow/white premixed flames were observed in the center of the reactor tube downstream of the catalyst section because of significant methane and oxygen breakthrough (Figure 3-9). These flames help to explain the increase in acetylene selectivity at higher GHSV for this configuration. Based on system geometry, it is appropriate to compare only QCC and QC C 2 selectivity performance since the DC geometry allows flames. Overall, at each flowrate the acetylene selectivity is significantly higher for QC than for QCC and decreases with increasing flowrate. Ethylene and ethane selectivities (Figure 3-8C and D) increase with increasing flowrate except for QCC at the two highest flowrates. The overall trends for selectivity show that decreasing the residence time the reactant gases spend in the catalyst section increases the selectivities to ethylene and ethane, and that cooling of the post-catalyst gases positively impacts overall C 2 selectivity significantly. However, based on this simple quench design, the cooling time constant for product gases does not appear optimized. The large amount of radiation that leaks through the back heat shield after the catalyst may keep post-catalyst gas temperatures higher than those necessary to effectively freeze C 2 species compositions. It is crucial that post-catalyst gases are cooled to temperatures where C 2 oxidation essentially ceases within milliseconds Carbon Formation For all experiments in which C 2 products were made, a dark film accumulated on the cool downstream walls of the reactor tube. In any feasible process, this carbon buildup would have to be eliminated. Currently, product gases leaving the quenching section of the reactor cool from 800 C down to 300 C in the section where the carbon accumulates. One way to decrease downstream carbon formation would be to immediately cool the post-catalyst gases to near room temperature before they reach the downstream section of the reactor. 3.5 Modeling of Quench Control Experiments Experimental Axial Temperature Distribution The experimental results provide an excellent opportunity to establish whether the Chemkin plug-flow model and accompanying chemical mechanisms can provide 46

71 accurate product compositions. Only by comparison with experiments can we establish confidence in the OCM plug-flow model and determine its utility as a predictive tool. In addition, experimental observations provided qualitative information about the axial temperature distribution in the reactor. For example, photographs taken during reactor operation in QCC showed the relative temperature distribution (Figure 3-10). Observations at 3 slpm feed flowrate showed that the front heat shield was relatively cold, and the front face of the catalyst did not glow appreciably (Figures 3-10A and B). The backside of the catalyst glowed intensely, and the heat released from the reaction was dissipated throughout the back heat shield and onto the ceramic Nextel fibers in the quench section. All of the radiation from the back face of the catalyst and the back heat shield was essentially blocked by the Nextel fibers (Figure 3-10C). The static mixer did not glow and appeared much cooler than the front of the fiber section (Figure 3-10D). Experimental observations of the axial temperature distribution at 5 slpm feed flowrate had similar qualitative results as those shown here for a 3 slpm feed with the QCC. These findings agree well with those of Hohn et al. [9], who found a very large temperature gradient across the length of the catalyst of as much as 800 C Proposed Axial Temperature Distributions Based on quantitative and qualitative experimental data, two axial temperature distributions were created to mimic the actual thermal environment inside the reactor for QCC operating at 5 slpm feed (Figure 3-11). These temperature distributions (TD1 and TD2) were then used in the Chemkin plug-flow model to determine how well the experimental product compositions could be reproduced. The distributions were similar, and their behavior differed only after the catalyst. For both distributions, temperature increased from 25 C at the front heat shield to 550 C by the front of the catalyst via first-order exponential heating. Through the 3 mm catalyst, the temperature increased linearly up to 1200 C. After the catalyst, temperature decreased slightly to 1150 C for TD1 and 1100 C for TD2 by the end of the back heat shield. Experimentally, temperature in this area was approximately constant since the system was insulated up to the end of the back heat shield to prevent radial heat losses. However, the quench area was not insulated so some heat losses most likely occurred from the end of the back heat shield, which probably affected the temperature distribution throughout the shield. 47

72 In the quench section, temperature behaved similarly for both distributions with similar time constants, cooling down to the experimentally-measured static mixer temperature for 5 slpm (820 C) Chemkin Plug-flow Modeling The temperature distributions described above along with the solution techniques and chemistry mechanisms described in section 2.3 were used to perform the simulations for 5 slpm flowrate. Homogeneous and heterogeneous chemistry was applied to the model catalyst section while only homogeneous chemistry was applied in all other sections. Since the half-life time for methane conversion in the homogeneous mechanism is one second or greater for temperatures at or below 900 C, no significant chemistry was assumed to take place after the static mixer. This time is much longer than the residence time for gases in the reactor. Therefore, simulations only included that part of the reactor geometry up to the end of the static mixer Comparison of Experimental and Simulation Results C 2 selectivities for the simulations and experiments were in good agreement (Table 3-1). However, experimental and simulated conversions for methane and oxygen differed by approximately 50% and 9%, respectively. TD2 gives a better relative C 2 selectivity distribution at 5 slpm. When conversions and selectivities are plotted versus axial distance, some interesting trends are apparent. For both TDs, no chemistry occurs before the catalytic section. This is expected because the total residence time predicted in the front heat shield is 12.5 ms, and the maximum temperature reached in this section is 800 C. Essentially, methane does not have enough time to react at these temperatures. Significant conversions of fuel and oxygen occur in the catalyst section as temperature increases from 800 C to 1200 C and the residence time in the catalyst is 2 ms (Figure 3-12). First, ethane is produced followed by its dehydrogenation to ethylene. Although not shown in Figure 3-12, very little CO 2 or CO is created in the catalytic section (S CO2 = 0.01 and S CO = 0.11 at the end of the catalyst for both simulations). In the back heat shield, ethylene selectivity is predicted to be quickly lost, and acetylene production increases. Significant CO and CO 2 production occurs in the back heat shield, and these products reach their final selectivity values by the time they leave the back heat shield. The 48

73 residence time predicted for the back heat shield for both models is approximately 4 ms. Very little reaction is predicted to occur in the quench sections of the simulations. 3.6 Quenching with Compact Heat Exchanger Apparatus and Methods In order to further test whether significant ethylene yields could be obtained using the quenching technique, the hot effluent from the catalyst was fed to a compact heat exchanger designed by Mesoscopic Devices LLC (Broomfield, CO). Figure 3-13A shows the physical dimensions of the heat exchanger, which was interfaced with the reactor via stainless steel headers and tubing (Figure 3-13B). The heat exchanger worked quite well at cooling the incoming OCM product gases from C to C. Inlet mass flowrates to the reactor were 2.9x x10-2 kg/min (3-13 slpm, 1.7 CH 4 /O 2 with 20% N 2 dilution). A recirculating pump supplied distilled water to the heat exchanger at a feed rate of 600 cc/min. Cooling water was supplied to the cooling coils in the pump reservoir to keep the inlet water temperature to the heat exchanger at or below 30 C. The outlet water temperature varied from 5-12 C higher than the inlet temperature depending on the reactor flowrate. Three reactor configurations were used to determine the effect of rapid cooling on the OCM product distribution: (1) default configuration, (2) "staggered quench" configuration where the catalyst is contained in the quartz tube, and the heat exchanger is 7.6 cm downstream (Figure 3-14), and (3) "direct quench" configuration where the catalyst is placed directly inside the front header. 3 or 5 mm Pt foam catalysts were used in the quench experiments. All experiments incorporated an 80 ppi, 80% porous, front heat shield but no back heat shield. Alumino-silicate paper (Unifrax Corp.) was used to seal the heat exchanger header tubes inside the 25-mm ID quartz tubes. For the "staggered quench" configuration, approximately 7 cm of the steel tube welded to the front header was removed in order to decrease the distance from the catalyst to the heat exchanger. For the "direct quench" configuration, the elliptical header entrance was partially sealed with alumino-silicate paper to make a circular entrance for reactants fed to the cylindrical catalyst therefore avoiding gas bypass. More paper was used to seal around the 5-mm thick front heat shield and catalyst inside the header. Catalyst lightoff in "direct quench" mode was accomplished by placing a blank 49

74 foam monolith in the upstream quartz tube and heating it with a Bunsen burner. Nitrogen was then passed through the reactor, which pre-heated the catalyst inside the header sufficiently for ignition Results Operation in "direct quench" mode produced no C 2 products over the range of catalyst contact times studied. The default and "staggered quench" configurations did produce significant C 2 products (mainly acetylene) at all flowrates. For the default and "staggered quench" configurations, methane and oxygen conversions were comparable at all flowrates (Figure 3-15). The conversions for the "direct quench" configuration were comparable to the other two configurations except at the highest flowrates (11-13 slpm), indicating that gas bypass was not an issue over the majority of flowrates studied. Differences in total C 2 and acetylene selectivity between the default and "staggered quench" configurations were statistically significant based on n 3 (Figure 3-16). This was not the case with ethylene or ethane. Since the "direct quench" configuration produced no C 2 products, some post-catalyst homogeneous chemistry must be necessary to produce C 2 products for Pt. The "direct quench" configuration product was mostly syngas with some CO 2. Selectivities with the default and "staggered quench" configurations were comparable to the results from previous experiments (section 3.4). The "staggered quench" configuration was comparable in C 2 production to the nitrogen quench system. Significant carbon accumulation was observed on the front and back side of the heat exchanger after operation in staggered quench mode (Figure 3-17), causing the exit water temperature to rapidly increase after ~60 min on-stream. Most of the flow channels appeared totally blocked and had to be irrigated with acetone and cleaned prior to the next run. The use of stainless steel (material of construction) for the heat exchanger was a major issue, and finding a material that can be used without major coking would be a significant advance. 3.7 Conclusions From the experiments performed on Pt at high GHSV, the effect of thermal quenching of post-catalyst gases has been established. Dramatically lowering the temperature of the product gases within 30 ms significantly improves C 2 selectivity. 50

75 However, the C 2 selectivity increase was not as high as that predicted by the simulations (section 3.5). Based on the N 2 QC, product gases are not cooled fast enough to realize an optimum yield of C 2 products. Simulations show that very little time is available to quench the product gases after the catalyst assuming the proposed temperature distributions are approximately correct. If ethylene selectivity is to be maximized, thermal quenching must occur as soon as possible after the catalyst. Of particular interest, if we sum the C 2 selectivities and the CH 4 conversion, the maximum value of approximately 111% occurs right at the end of the catalyst. However, there are some significant discrepancies between simulated and experimental results. While the selectivities predicted by simulations are in agreement with experiments, fuel and O 2 conversions are not. There are some possible reasons for these differences. First, there could be some catalytic activity by the ceramic fibers and monoliths, which is not accounted for. Second, the radial temperature profile is not constant and probably a large gradient exists in this direction since the walls of the quench section are not insulated. This gradient could cause lower fuel conversion than predicted by the simulations where the radial temperature profile is constant. The absence of C 2 production in direct quench mode with the compact heat exchanger was surprising. Based on previous experimental and modeling results, significant C 2 products were expected by quenching directly after the catalyst. While Pt is a poor catalyst, the thermal quench system in staggered quench mode does significantly increase C 2 yields over the DC. The main point to take away from these results is that some homogeneous chemistry is necessary after the catalyst to effect C 2 production; however, this homogeneous chemistry must be controlled to prevent further C 2 oxidation. Based on the work with the N 2 quench system and compact heat exchanger, it appears we have reached the upper limit for C 2 production over Pt-coated monoliths. This work shows that the quench system does offer improvements in C 2 production, and more successful OCM catalysts will be needed to test whether C 2 yield can be raised above the economically viable limit for this process (25%). 3.8 Nomenclature D Inside diameter of reactor tube, m DC Default reactor configuration 51

76 L QC QCC P Length of the foam catalyst, m Quench reactor configuration Quench control reactor configuration Pressure, atm slpm Standard liters per minute T Temperature, C T bf Back face temperature of catalyst, C T ex Temperature of exhaust gases leaving reactor, C T o Inlet temperature, C T sm Temperature of gases leaving the static mixer monolith, C v v o Velocity, m/s Inlet velocity, m/s STP v Inlet velocity at standard temperature and pressure, m/s o V Inlet volumetric flowrate, m 3 /s Greek ε Void fraction ρ Density, kg/m 3 τ Catalyst residence time, ms 52

77 Table 3-1 Comparison of experimental and simulation results. Mean values are shown for experimental data. Experiment TD1 Simulation TD2 Simulation X CH X O S CO S CO S C S C2H S C2H S C2H

78 CH 4, O 2, N 2 alumina heat shields catalyst ceramic insulation exhaust sampling port Figure 3-1 Schematic and photograph of SCTR reactor for OCM. 54

79 CH 4, O 2, N 2 CH 4, O 2, N 2 CH 4, O 2, N 2 default configuration quench control quench N 2 Nextel ceramic fibers 80 ppi monolith quench configuration Figure 3-2 Reactor configurations. 55

80 Reactor Pressure Incinerator Mixing Tee Reactor Mass Flow Controllers HP 5890 GC O 2 N 2 CH 4 Figure 3-3 Experimental apparatus diagram for default and quench control configurations. 56

81 Reactor Pressure Incinerator Mixing Tee Reactor Mass Flow Controllers HP 5890 GC O 2 N 2 CH 4 quench N 2 Figure 3-4 Experimental apparatus diagram for quench configuration. 57

82 A B Figure 3-5 Photographs of the catalyst and blank ceramic foam monoliths. (A) Side view from left-to-right: front heat shield (80 ppi), Pt catalyst (80 ppi), and back heat shield (45 ppi). (B) Top view in same order as (A). 58

83 selectivity C2 sel C2H2 sel C2H4 sel C2H6 sel quench N 2 SLPM Figure 3-6 Effect of quench N 2 flowrate on C 2 selectivities at 5 slpm feed. The selectivity values for 0 slpm quench flowrate were determined with the default configuration. The selectivity values for 1.89 and 3.9 slpm quench flowrate were determined with the quench configuration. 59

84 conversion T ( o C) O2 conv O 2 CH4 4 conv feed SLPM Tsm T sm T ex Tex feed SLPM A B Figure 3-7 CH 4 and O 2 conversions (A) and static mixer and exhaust temperatures (B) as a function of feed flowrate. Error bars represent the standard error of the mean for n 3. Data is sorted by reactor configuration according to the following legend: DC, QC, QCC 60

85 A 0.25 B C2 Csel C2H2 sel 2 C 2 H selectivity selectivity feed SLPM C feed SLPM D C2H4 C sel C2H6 sel 2 H 4 C 2 H selectivity selectivity feed SLPM feed SLPM Figure 3-8 C 2 selectivities versus feed slpm for reactor configuration. Error bars represent the standard error of the mean for n 3. Legend: DC, QC, QCC 61

86 A B Figure 3-9 Photograph of premixed methane/air flame downstream of catalyst. (A) The glowing section is the catalyst. (B) Zoom-in of flame. Flames were typical with flowrates greater than 5 slpm that caused significant methane and oxygen breakthrough. 62

87 A B 3 mm catalyst front heat shield back heat shield Nextel fibers static mixer C D Figure 3-10 Catalyst and quench sections of the OCM reactor during operation at 3 slpm feed in quench control configuration. Flow is from top to bottom. (A) Guide to reactor sections denoting reactor components inside Fiberfrax paper. The reactor is not operating in this photograph. (B) Side view. (C) View of reactor showing top of front heat shield. (D) View showing bottom of back heat shield. Notice that no visible radiation leaks through the heat shields. 63

88 T (C) distance (mm) flow in flow out Figure 3-11 Proposed axial temperature distributions for 5 slpm feed in quench control configuration. Distribution 1 is signified by the red line. Distribution 2 is given by the orange line. Solid red points represent actual experimental temperatures. 64

89 conversion selectivity A T distance (cm) B C 2 H 6 C 2 H 4 CH 4 C 2 H 2 O distance (cm) T (K) Figure 3-12 Calculated reactor product composition as a function of axial distance. (A) Conversions and temperature for temperature distribution 1. (B) C 2 selectivities for temperature distribution 1. 65

90 conversion C T CH 4 O distance (cm) T (K) selectivity D C 2 H 6 C 2 H 4 C 2 H distance (cm) Figure 3-12 Calculated reactor product composition as a function of axial distance. (C) Conversions and temperature for temperature distribution 2. (D) C 2 selectivities for temperature distribution 2. 66

91 A B Figure 3-13 Schematic of compact heat exchanger and interface. (A) Compact heat exchanger (all numbers in mm). (B) Schematic of heat exchanger with headers that interface with reactor. Courtesy of Mesoscopic Devices LLC, Broomfield, CO. 67

92 catalyst Figure 3-14 Photograph of operating reactor (front view) in "staggered quench" configuration with compact heat exchanger. Conversion CH4 conv. 0.1 O2 conv E E E E E+05 GHSV (1/hr) Figure 3-15 Methane and oxygen conversions versus space velocity for reactor configurations with heat exchanger. Direct quench configuration results not shown. Legend: default configuration, "staggered quench" configuration 68

93 C2 sel C2H2 sel selectivity selectivity E E E E E+05 GHSV (1/hr) E E E E E+05 GHSV (1/hr) C2H4 sel C2H6 sel selectivity selectivity E E E E E+05 GHSV (1/hr) Figure 3-16 Total and individual C 2 product selectivities versus GHSV. Direct quench configuration results were negligible. Legend: default configuration, "staggered quench" configuration E E E E E+05 GHSV (1/hr)

94 A B Figure 3-17 Front side of heat exchanger after operation in "staggered quench" configuration. There is significant carbon residue blocking most of the entrance slits. (A) Front side, and (B) back side. 70

95 Chapter 4 Modeling Steady-state Axial for Methane CPO on Rh-coated Foams 4.1 Introduction Methane conversion into hydrogen, liquid fuels, and chemicals is of growing economic importance since oil production may peak before year 2040 according to recent assessments [93]. Strong efforts in science and technology are underway to convert natural gas into liquids to open up an alternative petroleum resource that can compensate potential oil shortfalls [94]. Production of syngas is the first step in the conversion of methane to liquids. Methane catalytic partial oxidation (CPO) on Rhcoated foam monoliths is an efficient way to achieve this transformation [95]. The CPO reaction (eqn. 4-1) is slightly exothermic in contrast to highly endothermic steam reforming (eqn. 4-2): 1 o CH4 + O2 CO + 2H2 Δ H R = kj/mol CH4 (4-1) 2 CH4 + H2O CO + 3H2 Δ H = +206 kj/mol CH (4-2) o R 4 Steam reforming on Ni requires contact times in the range of 1 s to achieve sufficient methane conversion. Autothermal millisecond CPO reactors holds great promise for replacing steam reformers and steam crackers for production of hydrogen and olefins, respectively, because they operate at much shorter residence times, require much simpler equipment, and can be scaled up or down for different applications [11, 13, 94, 96]. For CPO on Rh, CH 4 conversion close to 100 % and > 90 % selectivities to H 2 and CO can be achieved in a few ms [4]. For process integration, methane CPO supplies a H 2 /CO ratio of 2 that is more favorable for downstream chemistry (methanol, Fischer- Tropsch synthesis) than the higher ratio of 3 obtained by steam reforming. In spite of several decades of research on millisecond reactors, many questions remain concerning the mechanism of the process. A major challenge is that the process cannot be decomposed into simple laboratory experiments from which kinetics and catalysts can be examined systematically. This is because realistic temperature and concentration profiles cannot be created without the presence of interphase mass and heat transfer that may change the process significantly. Temperatures above ~800 o C 71

96 are required to prevent total oxidation and carbon formation within the catalyst. Large gradients in temperature (>10 5 K/s) and concentration are necessary to prevent combustion before the catalyst and to quench primary products after the catalyst. There is an ongoing debate about the mechanism of methane CPO. Direct and indirect mechanisms have been postulated, discussed, and reviewed in the literature [46]. A brief summary of key experimental observations regarding CPO on Rh can be found in section Reactor Models and Multi-step Chemistry Detailed modeling with plug flow reactor (PFR) and two dimensional (2D) simulations has been used to simulate methane oxidation to syngas [3, ] and ethane oxidation to ethylene [88, 101, 102] over Rh and Pt catalysts using up to 100 steps for surface reactions and hundreds of steps for gas-phase reactions. However, while these simulations predict conversions and selectivities very accurately, for the most part only experimental exit compositions (integral data) were used for validating the postulated mechanisms because little information has been available regarding species compositions within the catalyst. Additionally, as with experimental studies, previous numerical studies differ in their conclusions about the reaction mechanism. Early work assumed a direct mechanism because a high temperature surface model (19 reactions) was developed that agreed with experimental integral conversion and selectivity data reasonably well [2, 3]. In this surface scheme, CH 4 dissociation on metallic Rh sites was lumped into a single step, and H 2 and CO were formed as primary products. Except for CO 2 adsorption, all adsorption-desorption steps in this original surface mechanism were reversible. Steam reforming was not excluded by definition, but CO 2 reforming was (low sticking coefficient of CO 2 on Rh). Using this mechanism with 2D simulations showed no steam reforming contribution [97]; it was too slow given the kinetic parameters used. A PFR study [98] using the 19-step surface model and large gas-phase mechanisms [103, 104] showed that gas-phase reactions were insignificant at atmospheric pressure for millisecond contact times but become important at elevated pressures (> 5 bar) [97, 105]. Experimental results corroborate these findings [53]. The mechanism was further improved by including steam reforming, water-gasshift (WGS), and CO 2 re-adsorption (38 reactions) [106]. These refinements are in 72

97 better agreement with experiments showing that steam reforming on Rh is possible at millisecond contact times [ ]. The 38-step surface mechanism has been validated against integral steady-state [99, 109] and transient experimental data [100, 110]. Coverage-dependent desorption energies for CO and O 2 were included [100] for better lightoff agreement. More recently, a 104 step C 1 mechanism on Rh was reported by Mhadeshwar and Vlachos and was partially derived from first-principles simulation [111]. All activation energies in this mechanism were either coverage or temperature dependent or both. Besides methane CPO, this mechanism also considered methane reforming by water and CO 2 and decomposition of oxygenates on Rh. According to the authors, this mechanism predicts distinct oxidation (CO 2 and H 2 O as products) and reforming (CO and H 2 as products) zones for methane CPO. No H 2 formation was found in the presence of gas-phase oxygen for experiments closely matching the present work based on the temperature profiles used in their simulations Previous Experimental/Numerical Spatial Profile Comparisons Numerous studies have been performed to determine axial temperatures (gas and surface) within working catalysts; however, high-resolution differential composition profiles were lacking until recently [15, 16]. Temperatures have been measured in extruded monoliths, fixed beds, or gauzes with IR thermography and thermocouples [50, ], but these experiments may introduce radiation and conduction losses compared to insulated catalysts. One-dimensional multi-phase (heterogeneous) simulations of a fixed bed coupling surface chemistry with interphase transport and external heat loss show that the experimentally measured gas and surface temperatures can be quantitatively reproduced [109]. Axial species profiles have been estimated by measuring exit compositions using different catalyst lengths with different thickness monoliths [107] or with sphere beds of different lengths [7], but these do not necessarily duplicate the profiles within a single foam catalyst. Axial fuel conversion profiles were measured for natural gas on Pt-coated honeycomb monoliths and simulated using detailed heterogeneous surface chemistry [116]. Species concentration profiles for methane oxidation on Rh at high pressure have been measured by gas chromatography sampling between multiple metal catalyst screen layers [56]. PFR simulations with C 1 chemistry on Rh [111] were compared to axial measured profiles [56] with the conclusion 73

98 that CO and H 2 O are the primary oxidation products in presence of gas phase O 2. H 2 was reported as secondary product formed by steam reforming after complete conversion of gas phase O 2. High-resolution spatial profile measurements have been used to distinguish between and validate heterogeneous and homogeneous mechanisms for methane oxidation in stagnation flow and catalytic channel reactors. A stagnation flow reactor (planar catalyst) with a quartz microprobe and mass spectrometer was used to study the catalytic combustion of methane at atmospheric pressure on a hexaaluminate catalyst [117, 118] and estimate global surface reaction rates. Raman and laser-induced fluorescence measurements over the catalyst boundary layer in a laminar channel flow reactor have been performed from 1-16 bar [119, 120]; these measurements were used to test the performance of state-of-the-art surface and gas phase chemistry mechanisms at high pressure. Additionally, axial variation of the oxidation state for a Pt/Rh catalyst [121] and axial/radial variation of the oxidation state for a Rh catalyst [122] were measured for methane CPO using X-ray absorption spectroscopy Motivation From these studies, it is clear that high-resolution spatially-resolved experimental data are key to discriminate between a number of multi-step surface mechanisms and reactor models that predict integral data equally well. An experimental method has been developed to measure axial species and temperature profiles within CPO foam monoliths at atmospheric pressure with 0.3 mm spatial resolution using a capillary sampling technique with a mass spectrometer [15, 16]. It was found that 2D simulations with a 38-step mechanism [99] fit experimental profiles well at C/O = 1.0, but some deficiencies were observed at C/O = 0.7 and 1.3 [15]. The goal of this work is the systematic performance analysis of two state-of-the-art surface mechanisms for methane CPO in comparison to high-resolution axial data measured in an adiabatically operated reactor with a Rh-coated ceramic foam. The quality of the experimental data has been improved significantly in comparison to the first spatial profiles reported [15, 16] by using a geometrically aligned reactor setup with fully automated capillary movement using a computer-controlled stepper motor. The two mechanisms used are a 38-step surface reaction mechanism without [99] and with [100] coverage dependencies and a 104-step mechanism with full coverage dependence [111]. The strengths and 74

99 weaknesses of both mechanisms are discussed over a range of feed stoichiometries (C/O = 0.7, 1.0, and 1.3). In addition, the role of the reactor models (1D and 2D) and gas and surface temperatures on model predictions are explored with the 38-step mechanism. 4.2 Numerical Methods PFR Model In one embodiment of the reacting system, a single cylindrical hypothetical pore of a 10 mm long 80-ppi foam monolith was simulated using a PFR model and a nominal pore diameter of 0.25 mm (area/volume = 160 cm -1 ), effectively neglecting the tortuous pore network present in ceramic foams. The differential algebraic equations resulting from the plug-flow treatment were solved using Chemkin PLUG [85, 123] or Detchem Plug code [124] D Pore Channel Model In a second embodiment of the system, the single hypothetical pore of the 10 mm long 80 ppi foam was simulated using computational fluid dynamics (CFD) in a 2D axisymmetric and elliptic PDE treatment solving the applicable transport equations for both solid and gas phases. This model is similar to that previously used to model ethane oxidative dehydrogenation (see Figure 1 of that work) [88]. The catalyst channel is 10 mm long as are the front and back heat shield channels. Channel diameter is 0.25 mm while, the wall thickness is mm. Solid phase was modeled using the thermal properties of alumina as a polynomial function of temperature (thermal conductivity varies from 36 to 5 W m -1 K -1 from 300 to 1500 K) [125, 126]. Solution procedures were the same as described in the next section D Porous Media Model The final embodiment of the reacting system consists of a 2D axisymmetric porous model to simulate reactor conditions and intra-construct species profiles using CFD; a treatment similar to a sub-grid scale model used previously to model catalytic converter chemistry [127]. This model incorporated the entire reactor geometry and dimensions. 75

100 Because of radial symmetry, only one half of the axial cross-section of the reactor tube was considered. A schematic of the computational grid is shown in Figure 4-1. The grid is comprised of the fluid zone (inner diameter = 19 mm) and the reactor wall (quartz tube, outer diameter = 22 mm). Gases (CH 4, O 2, and Ar) enter the grid under the inlet condition of plug flow and travel 5 cm (upstream region) to the front heat shield under developing laminar flow. Ceramic foam sections (front heat shield, catalyst, back heat shield, all 19 mm diameter by 10 mm long) were modeled using a homogeneous porous media approximation, which assumes local thermal equilibrium between gas and surface [128]. Porosity was taken as 0.81 [90], and the foam s solid component was simulated as polycrystalline α-alumina. Source terms were employed using the hydraulic diameter approach to account for the effect of the 80 ppi foam on the flow field by inputting isotropic inertial and viscous resistance terms for 80 ppi ceramic foams [129]. Thermal conductivity in the porous zones was treated as an average between gas and surface based on porosity. After leaving the back heat shield, gases travel 5 cm under developing laminar flow to the grid outlet. The boundary condition on external surfaces of the quartz tube was adiabatic. Surface chemistry in the porous catalyst section was included with a 38-step surface mechanism for CH 4 oxidation and reforming on Rh [99]. Temperaturedependent transport properties in the gas phase were calculated using kinetic theory, and effective thermal conductivities of the quartz (fused silica) and α-alumina were taken as polynomial functions of temperature [125, 126]. Fickian diffusion was included using a mixture-averaged dilute diffusion coefficient for each species. The effect of the porous medium on diffusion was included by multiplying the dilute diffusion coefficient by the ratio of porosity to tortuosity. A value of 1.5 was used for tortuosity based on foam measurements [130]. The effect of radiation in the calculation of the porous medium s effective thermal conductivity was neglected as it should be small on 80 ppi foams when compared to lower cell density foams (e.g., ppi foams) with a smaller extinction coefficient [131] or packed beds with a significantly larger thermal conductivity [132]. To demonstrate this point, Sweeting et al. [133] give an effective thermal conductivity for 45 ppi, 92% alumina foam (including the effects of conduction and radiation) of 1.28 W/m K at 1000 o C. Effective thermal conductivity in this work for a 45 or 80 ppi foam is 1.3 W m - 1 K -1 at 1000 o C. Neglecting the radiation effect becomes a better approximation as ppi 76

101 and extinction coefficient of the foam increase. A previous numerical study using a 1D heterogeneous code to model methane oxidation on Pt/MgO found no significant contribution from radiation [134]. A recent 2D analysis confirms this finding for methane CPO in a honeycomb Rh catalyst [135]. Converged solutions were acquired by solving the Navier-Stokes momentum, energy, and species continuity equations with a segregated, implicit solver [128] using a computational node on an IBM Power 4 system at the University of Minnesota. Species coverages were computed by coupling species continuity equations with wall surface reaction boundary conditions via a stiff, coupled solver [128]. A specific surface area of 160 cm -1 was used for the porous catalyst foam without additional tuning for comparison with PFR simulations. Geometric surface areas for 80 ppi foams up to 210 cm -1 have been estimated with no additional washcoat [90]. Using a standard underrelaxation method, approximately 100,000 iterations and hours were required for convergence. A criterion of 10-6 was used for each scaled residual component (continuity, r-velocity, z-velocity, energy, and species) to determine solution convergence. Solution results using convergence criteria ranging from 10-6 to 10-7 showed no significant difference in velocity, temperature, or concentrations (<1.0%), thereby validating the sufficiency of the criteria. Computational cell number in these simulations was varied to ensure the solution was grid independent. In addition, the second-order discretization scheme in Fluent was utilized to reduce the effects of numerical diffusion (discretization error) on the solution [128] Surface Mechanisms Step Mechanism Simulations (PFR or 2D) utilized a 38-step surface chemistry mechanism for methane oxidation on Rh [99, 100] that has been validated against integral steady-state [99, 109] and transient experimental data [100, 110]. The mechanism in its current form is an improved version of the original mechanism for high temperature oxidation [3]. The mechanism was revised to include steam reforming and WGS and CO 2 readsorption [99, 106], which gave excellent agreement with integral steady-state data [99]. Coveragedependent desorption energies for CO and O 2 were later included [100], because they are important for lightoff agreement. Gas-phase chemistry is negligible at atmospheric 77

102 pressure conditions for methane CPO [97, 98, 105] and was not included in the simulations. Intrinsic catalyst activity is included through the assumed site density for the Rh surface, which was 2.72x10-9 mol/cm 2 for the mechanism, and the geometric surface area-to-volume ratio of the catalyst (160 cm -1 ) [90]. Hereafter, the 38-step mechanism is designated mechanism Step Mechanism A much larger C 1 mechanism on Rh (104 reactions) was recently reported by Mhadeshwar and Vlachos [111]. Simulations were performed with this mechanism utilizing a PFR code based on CHEMKIN III architecture, which was provided by the mechanism s authors [136]. Kinetic parameter calculation with the UBI-QEP method is performed on-the-fly with the code causing solution time to increase significantly over the 38-step mechanism. In this work, the surface to volume ratio was kept at 7500 cm -1 as was used previously [111] to fit experimental spatial data [56], and Rh site density was kept at 2.49x10-9 mol/cm 2 as set by the authors. Hereafter, the 104-step mechanism is designated mechanism Results PFR Simulations Based on Experimental Temperature Profiles Figure 4-2 displays the species profiles generated by using the experimental temperature field measured under high-resolution sampling as model input for C/O = 1.0 and total flowrate = 5 slpm. The onset of O 2 conversion for both mechanisms, model 1 (0.3 mm) and model 2 (0.6 mm), is significantly delayed in comparison to the experiment where O 2 conversion starts right from the catalyst entrance. After onset, mechanism 1 and the experimental profile converge to complete O 2 conversion by 1.3 mm whereas mechanism 2 predicts total O 2 conversion slightly later at 1.5 mm (Figure 4-2A). CH 4 conversion behaves similarly for both mechanisms and both predict the same conversion at the catalyst exit, which is slightly lower than found by experiment (Figure 4-2A). Onset of syngas production is significantly delayed for mechanism 1 and 2 (0.5 vs. 1.4 mm, respectively) (Figure 4-2B). H 2 flow for model 2 reaches nearly the experimental value by 10 mm while mechanism 1 lags behind; however, initial H 2 production for mechanism 1 starts at 78

103 an earlier distance than mechanism 2. The same qualitative behavior is found for CO, but mechanism 1 better predicts the outlet CO value. In contrast to the experiment where H 2 and CO are formed at comparable rates in the oxidation zone, mechanism 1 predicts a faster CO formation. Both models capture peaking of H 2 O flowrate and the step like behavior of the CO 2 flowrate qualitatively, but quantitative agreement with the experimental profiles is poor (Figure 4-2C). The experimental data show that ~44% of the H 2 and ~67% of the CO are made in the presence of O 2. For mechanism 1, only about 6% of H 2 and 29% of CO are made in the presence of O 2. Even less H 2 and CO are made in the presence of O 2 according to mechanism 2 (6% and 9%, respectively). Figure 4-3 highlights species coverages (> 10-6 ) for both models. Over the catalyst length mechanism 1 shows a progression from initially O-covered to C- and Rh-covered with the major coverages (>10-2 ) being Rh, C, CO, and H within 10 mm (Figure 4-4A). Mechanism 2 shows qualitatively similar coverage behavior; however, the shift in coverages occurs much more rapidly with an impulse-type behavior found at ~ mm, the location of complete O 2 conversion (Figure 4-3B). The major outlet coverages for mechanism 2 are Rh, H, CH, and CH 2. Both mechanisms predict an entrance section of nearly full O coverage that corresponds to the delay in reactant conversion (Figure 4-2A) Adiabatic PFR Simulations Effect of Inlet Temperature on Oxidation Section Length Mechanism 1 was tuned based on surface/volume ratios of cm -1, which are typical values for the geometric surface area per unit volume of ppi foams. However, mechanism 2 was used with a surface/volume ratio of 7500 cm -1 [111] as a fitting parameter for previous experimental spatial data [56]. Figure 4-2 shows that both mechanisms give similar performance with the experimental temperature profile when surface/volume ratios of 160 cm -1 and 7500 cm -1 are used, respectively. Therefore, surface to volume ratios used previously for both mechanisms were not modified in this study. In order to compare the reactivity of the two mechanisms the effect of inlet temperature on the adiabatic PFR models has been investigated. Since an adiabatic 79

104 PFR model does not include the effect of upstream heat transport, the inlet temperature is a critical parameter for model performance. The oxidation length for mechanism 1 with coverage dependence (of the activation energy for CO and O 2 desorption) is shorter than for mechanism 2 for temperatures < 825 K (Figure 4-4). Above this temperature, this trend reverses. Furthermore, the oxidation length for mechanism 1 with coverage dependence is shorter than mechanism 1 without coverage dependence until ~800 K, where the curves converge (Figure 4-4). To enable a performance comparison with the adiabatic PFR model (Section ), the inlet temperatures for mechanism 1 with coverage dependence and mechanism 2 were chosen as ~770 K and 815 K, respectively, to guarantee an equivalent but arbitrary 0.5 mm oxidation length for both (see Figure 4-5) Temperature, Species, and Coverage Profiles: C/O = 1.0 Mechanisms 1 and 2 show similar qualitative performance under adiabatic conditions with pronounced exothermic oxidation and endothermic reforming zones; however, these zones develop much faster with mechanism 2 (Figure 4-5). The temperatures in the oxidation zone quickly rise to 1380 o C and 1580 o C for mechanisms 1 and 2, respectively (Figure 4-5A), in agreement with the disappearance of O 2 (Figure 4-5B). Temperatures in the reforming zone then decrease to ~800 o C for both (Figure 4-5A). Temperature in the reforming zone for mechanism 2 decays much faster than mechanism 1. CH 4 consumption occurs more quickly for mechanism 2. Both mechanisms overpredict the experimental CH 4 outlet conversion significantly (Figure 4-5B). Syngas production is drastically overpredicted by both models in comparison to experiment (Figure 4-5C) because both models overpredict the amount of steam reforming occurring (Figures 4-5B and 4-5D). This trend is even more pronounced for mechanism 2 than for mechanism 1. Mechanism 1 predicts CO 2 is formed mainly in the oxidation zone and is inert thereafter (Figure 4-5D). Mechanism 2 indicates some CO 2 reforming after the oxidation zone (Figure 4-5D inset), which is not experimentally observed. The relative amounts of syngas produced in the oxidation zone (direct production) for both mechanisms follow the same trend as for the PFR model with the experimental temperature profile. Only the absolute numbers are higher because of the higher 80

105 temperatures in the oxidation zone resulting from the adiabatic boundary conditions. As in Figure 4-2A, initial H 2 production for mechanism 1 starts at an earlier distance than mechanism 2. For mechanism 1, approximately 40% of H 2 and 60% of CO are made in the presence of O 2. For mechanism 2 only ~9% of H 2 and 30% of CO are made in the presence of O 2. Figure 4-6 highlights the coverages (> 10-5 ) for both mechanisms. Over the catalyst length mechanism 1 shows a progression from an initially O covered surface (0-0.2 mm) to a bare Rh surface (0.3-3 mm) to a partly C-covered Rh surface towards the end of the catalyst. Mechanism 2 again shows discontinuous changes in coverage at the location of complete O 2 conversion at 0.5 mm. Before the discontinuity the surface is essentially O covered, after the discontinuity the surface is to 50% free and the remaining sites are covered by H, CH, and CH 2 with H being the dominant surface species (coverage ~25%) Temperature, Species, and Coverage Profiles: C/O = 0.7 and 1.3 Examining mechanism performance at C/O = 0.7 (fuel-lean) and 1.3 (fuel-rich) elucidate the ability of multi-step chemistry to simulate catalyst behavior away from syngas stoichiometry at C/O = 1.0, where most mechanism validation has been performed. For C/O = 0.7, surface temperatures become much higher than C/O = 1.0 with peak temperatures of ~ o C for both mechanisms (Figure 4-7A). Even though mechanisms 1 and 2 predict similar CH 4, O 2, H 2 and CO exit flowrates under adiabatic conditions at C/O = 0.7, the development lengths of the oxidation and reforming zones are slightly different (Figures 4-7B-D). The oxidation and reforming zones for mechanism 2 are extremely sharp and all chemical reactions are finished within 0.7 mm of the catalyst. Temperature rapidly decays for mechanism 2 down to 1215 o C at mm along with complete consumption of CH 4 and O 2 (Figure 4-7B). H 2 and CO reach their exit flowrates within 0.7 mm (Figure 4-7C) accompanied by a rapid production and then consumption of H 2 O (Figure 4-7D). Partial and total oxidation as well as steam reforming occur practically simultaneously in mechanism 2. In addition, CO 2 flow reaches its final value within 0.75 mm (Figure 4-7D). In contrast, mechanism 1 shows continuous oxidation and reforming zones. Temperature in the oxidation zone reaches 1580 o C by 0.75 mm (Figure 4-7A) along with peak H 2 O flow (Figure 4-7D) and complete consumption of O 2 (Figure 4-7B). Steam 81

106 reforming continues until ~4 mm where methane conversion reaches completion. As indicated by the increasing CO 2 and H 2 flowrates and decreasing CO and H 2 O flowrates, WGS is predicted to occur to a much greater extent with mechanism 1 than found experimentally. For C/O = 1.3, peak surface temperatures are lower (~1225 o C for both mechanisms, Figure 4-7E) compared to C/O = 1.0. For mechanism 1 the temperature reaches its final value of 900 o C at 1.25 mm, whereas temperature continuously decreases over the remaining catalyst length for mechanism 2 in agreement with the experimental temperature after 3 mm. As the inlet temperatures were adjusted to give the same oxidation zone lengths at C/O = 1.0 (0.5 mm, Fig. 4), these inlet temperatures lead to similar oxidation zone lengths at C/O = 0.7 and 1.3 for the two mechanisms. Both mechanisms predict oxygen conversion too fast as well as too high CH 4 conversion (Figure 4-7F). As a result, H 2 and CO flowrates are overpredicted (Figure 4-7G). For C/O = 1.3, both mechanisms predict significantly more steam reforming than is experimentally observed (Figure 4-7F). CO 2 agreement with mechanism 1 is reasonably good with no CO 2 reforming or WGS apparent in agreement with the experimental profile (Figure 4-7H). Mechanism 2 predicts some CO 2 reforming (Figure 4-7H inset). Panels A-D of Figure 4-8 display the corresponding surface coverages (>10-6 ) for C/O=0.7 and 1.3, respectively. For C/O = 0.7 both mechanisms predict similar surface coverages, close to 100% O coverage at the catalyst entrance (0-0.5 mm) followed by an essentially bare Rh surface from mm. Mechanism 2 predicts about 15% H coverage in the latter region. The situation is more diverse at C/O = 1.3. Mechanism 1 predicts three distinct zones. An O-covered zone from mm, a bare Rh surface from mm and a C-covered surface from 1.5 mm until the end of the catalyst. Mechanism 2 predicts full O coverage until 0.4 mm followed by the typical discontinuous coverage change to a Rh surface covered partly with CH, H and CH D Simulations with 38-Step Mechanism Because neither of the PFR models (sections and 4.3.2) captured the experimental species and temperature profiles satisfactorily, 2D computations were used in the following to test whether the reaction mechanism or the reactor model was inadequate for the mathematical description of the system. Because of the inherent complexity of mechanism 2 and portability challenge in a CFD code, only mechanism 1 82

107 was tested to simulate species and temperature profiles on Rh for experimental GHSV of 2.6x10 5 hr -1 (total inlet flow = 5 slpm) and atom feed ratios of C/O = 0.7, 1.0, and 1.3. Figure 4-9 highlights the 2D temperature, velocity, and species profiles for mechanism 1 using the 2D porous model. Only half the domain given in Figure 4-1 is considered because of axial symmetry. Figure 4-10 presents a performance comparison of the 2D channel model vs. the 2D porous model. Significant gradients in the radial direction (Figure 4-9) are observed for the porous model profiles because of axial heat transport in the quartz reactor wall. Centerline profiles for C/O = 1.0 are displayed in Figure 4-10 allowing a direct comparison with the centerline experimental data. Figure 4-11 displays the predicted profiles for the superior 2D porous model at C/O = 0.7 and 1.3. Surface coverages for all C/O ratios are shown in Figure Temperature and Species Profiles: C/O = 1.0 Profiles for temperature, velocity magnitude, and species compositions for the 2D porous model reveal that radial gradients are noticeable because of axial heat transport through the quartz reactor wall (Figure 4-9). The quartz tube has some entrance effects on the flow as it conducts heat from the catalyst section upstream and warms up the entering gases close to the wall. Radial gradients smooth as the flow exits the catalyst. Comparison of the 2D channel and porous models (Figure 4-10) indicates the porous model matches the experimental data better than the channel model. This result is discussed in section However, in comparison to simple PFR models (Section and 4.3.2) both 2D models are in much better agreement with the experimental data. As shown for the porous model in Figure 4-10, the calculated surface temperature increases from ~600 o C at 2 mm within the front heat shield to 1000 o C at the catalyst entrance. All of the oxygen and 40% of methane are converted within 0.5 mm (oxidation zone) where H 2 O and CO 2 flowrates reach their maximum values. The oxidation zone is noticeably shorter than found experimentally (~1.3 mm). The amounts of H 2 and CO produced in the oxidation zone, 40% (0.035 mol H 2 /min) and 58% (0.026 mol CO/min) of total, respectively, are in good agreement with the experimental data, 45% (0.042 mol H 2 /min) and 67% (0.03 mol CO/min) of total. In the reforming zone, the CH 4 and H 2 O continue reacting to give significant production of more syngas through steam reforming accompanied by a decrease in the calculated temperature from 1000 o C 83

108 at the end of the oxidation zone to 735 o C at the catalyst exit. In agreement with the experiment, the simulations show no sign of CO 2 reforming or WGS under these conditions. The calculated surface temperature and experimental gas temperature equilibrate at ~2 mm, and calculated temperature falls below the experimental temperature for the remaining catalyst length Temperature and Species Profiles: C/O = 0.7 and 1.3 Comparison of the 2D porous model with the experimental data for the two other investigated stoichiometries is presented in Figure The simulations match the experimental data similarly well as with C/O = 1.0, and only few quantitative discrepancies are observed. For C/O = 0.7, the catalyst runs considerably hotter and the calculated surface temperature rises from ~700 o C 2 mm in front of the catalyst to ~1200 o C at the catalyst entrance (Figure 4-12). Steam reforming causes the temperature to decline to ~1050 C at the end of the catalyst. All of the oxygen and ~68% of methane are converted within 0.5 mm of the catalyst (oxidation zone). Oxidation zone length is still shorter than found experimentally; however, the simulation shows correctly that the length of the oxidation zone does not change by going from C/O = 1.0 to 0.7. The amounts of H 2 and CO predicted to be formed in the oxidation zone at these higher temperatures, 50% (0.042 mol H 2 /min) and 73% ( mol CO/min) of total, respectively, are again in close agreement with the experimental values, 56% ( mol H 2 /min) and 76% ( mol CO/min) of total, respectively. H 2 O flowrate peaks at the end of the oxidation zone and decreases in the steam reforming zone. The model overpredicts secondary H 2 O consumption, not by overpredicting steam reforming but rather by overpredicting WGS as reflected in the rising CO 2 flowrate. The prediction of WGS is the major disagreement between model and experiment for low C/O ratios. The calculated surface temperature and experimental gas temperature equilibrate after ~3 mm, but the exothermic WGS outweighs somewhat the endothermic steam reforming preventing the simulated temperature to fall below the experimental temperature in the steam reforming zone. For C/O = 1.3, the catalyst runs slightly cooler than at C/O = 1.0 and the temperature rises from ~550 C 2 mm in front of the catalyst to ~875 C at the end of the oxidation zone (0.5mm). As with C/O = 1.0 the calculated temperature falls below the experimental temperature in the steam reforming zone, and both values converge to 84

109 ~750 C at the end of the catalyst. The model predicts the same oxidation zone lengths (0.5 mm) as for C/O = 0.7 and 1.0. Independence of oxidation zone length from the C/O ratio is consistent with the experiments even though the difference between experimental and predicted oxidation zone lengths are for all C/O ratio ~0.8 mm. The amounts of H 2 and CO formed in the oxidation zone, mol H 2 /min (51% of total) and mol CO/min (63% of total), respectively, are in good agreement with those found experimentally, mol H 2 /min (45% of total) and mol CO/min (65% of total). The model predicts the outlet flowrates for CO, H 2 O and CO 2 well, but H 2 flowrate is underpredicted by 10% Species Coverages The calculated major species coverages show a clear trend with changing C/O ratio from 0.7 to 1.0 to 1.3 (Figure 4-12). For C/O = 0.7 characterized by calculated surface temperatures always above 1000 o C, the surface is essentially bare Rh with coverages varying from at the catalyst entrance to 0.95 at the outlet. Down the catalyst axis, oxygen coverage decreases in the first 2 mm from ~2% to a fraction of a percent, while CO, H, and C increase in coverage from below ~1% to 3% to 1%, respectively, at the reactor outlet. As C/O ratio is gradually increased (surface temperature decreases) it turns out that C and CO become the only two important surface species with their coverages increasing much steeper than for C/O = 0.7. In the first 0.5 mm where the surface is hot (oxidation zone) more than 98% of the Rh sites are predicted to be uncovered, similar to C/O = 0.7. However, C and CO coverages rise quickly with declining surface temperature reaching 30% and 50% respectively for C/O=1.0 and 88% and 7% respectively for C/O = 1.3. It can be summarized that the model predicts a clean surface at high temperatures and/or in the presence of O 2 and large amounts of H 2 O (C/O = 0.7). For low temperatures, absence of gas phase O 2 and low concentrations of water vapor, the surface is predicted to be carbon covered (C/O = 1.3). For syngas stoichiometry (C/O = 1.0), CO is predicted to be the major surface species in the second half of the catalyst with around 50% coverage. 4.4 Discussion Total agreement between experimental profiles and numerical simulations could only be achieved if the experimental data were accurate, reproducible and free of 85

110 artifacts, the reactor model mapped the experiment perfectly, and the multi-step microkinetic model was valid for all experimental conditions investigated. In reality, experimental data are always subject to error, and the reactor model can never reproduce the experiment in every detail. As the goal of the present numerical study is to assess the validity of the applied reactor models and chemistry mechanisms, experimental uncertainty and uncertainties in model parameters deserve review (sections and 4.4.2) before the predictive capabilities of the different models and the two mechanisms are discussed (sections to 4.4.6) Experimental Uncertainty A measure for experimental accuracy is the atom balance error at each axial location. At steady-state, species accumulation does not occur and the atom balance error should be zero. The experimental profiles presented in this work for the species H 2, CH 4, CO, O 2 and CO 2 are mass spectrometry measurements (see experimental details described elsewhere [16]). The H 2 O profile is calculated as a least squares optimization from the O and H atom balances. For the data presented in this work the carbon atom balance closed for all points between -1 to +3%. The hydrogen atom balance is slightly biased towards positive values and ranges in an interval between +2 and +5%. As the least squares calculation weights the H atom balance twice as high as the O atom balance, the O atom balance combines most of the experimental uncertainty. However, it closes for all axial locations between -13 and +5%. The thermocouple reading is accurate to ± 1%, but the measured temperature can be assigned neither to the catalyst temperature nor to the gas temperature. The thermocouple is enclosed in the quartz capillary and has direct thermal contact with the flowing gas but not with the catalyst surface. However, it sees radiation from the surface. The measured temperature is therefore largely biased towards the gas temperature, which has to be taken into account by comparing it to models, which have explicit surface and gas temperatures. The reproducibility of the experimental profiles has been verified by performing independent profile measurements on different catalysts and different reactor setups. Axial measures (e.g. length of the oxidation zone) can be reproduced to within 1 mm. Species concentrations at the reactor outlet were reproducible to ~5%. The fine plateaus, bumps and spikes observed in most of the profiles are only partly reproducible. 86

111 Some of the fine structure may be caused by the random structure and size distribution of the pores in the foams. Pockets or blocked pores along the capillary channel or even back mixing effects caused by the pore structure around the channel can lead to features in the profile that can never be captured by the model, which assumes a well defined or idealized flow profile. In the discussion of the performance of the investigated reactor models and mechanisms only deviations will be discussed that are significant with respect to the above mentioned experimental uncertainties Uncertainty in Model Parameters A number of parameters enter the model for which reasonable guesses have to be made. Whereas a comprehensive sensitivity analysis of their influence is beyond the scope of this work, the most important parameters and their chosen values will be discussed below Pore Diameter All reactor models require the specification of a pore diameter. In this work, a pore diameter of 0.25 mm was used, which is the mean pore diameter found from experimental visualization for an 80 ppi foam [90]. However, a foam possesses a large range of pore diameters, which are typically normally distributed for moderate ppi foams (10-65 ppi) [90]. Being restricted to a single channel diameter in these simulations is a drawback since the flow encounters many pore diameters as it traverses through the foam. Residence time distribution and heat transfer are strong functions of the pore diameter distribution, which most likely widens as ppi is increased. A larger pore diameter in a channel model for example increases the length that is required for the gas and surface to equilibrate. Figure 4-13 shows the results of simulations with the boundary-layer code Creslaf coupled with mechanism 1. As the channel diameter increases from 0.25 mm to 2 mm, the equilibration length increases from less than 0.5 mm to 7 mm. For the 2D models, inlet discrepancies are most likely caused by the assumption of thermal equilibrium in the catalyst, which is not valid in the oxidation zone. Since the gas-phase temperature is overpredicted, mass diffusivity is also overpredicted and some back-diffusion occurs in the model, which is observed for both the porous and channel models. 87

112 Foam Tortuosity Another critical parameter in the porous model becomes the tortuosity, which is the ratio between the actual length the gases travel through the foam to the length of the foam. The effect of tortuosity is to limit mass diffusion in the porous zones and to increase the pressure drop. The value of 1.5 used in this study was based on experimental measurements with 8-45 ppi foams [130], where there appears to be a strong power law dependence between ppi and tortuosity. However, the tortuosity for an 80 ppi foam may be much higher, and values between 1.5 and 5 have been measured for similar pore densities [137]. Increasing the tortuosity value in the porous model simulations from 1.5 to 4 improves the agreement between experiments and simulations in the first 3 mm of the catalyst (data not shown) Interphase Heat and Mass Transport It should be mentioned again that the porous model in this work has considered the catalyst as a homogeneous medium and the calculated species profiles shown here do not include the effect of interphase transport but only depict the performance of the multi-step chemistry. Mass and heat transfer correlations for lower ppi foams (15-45) have been measured, but may not be reliable for higher ppi foams because of the inconsistency in the pore structures as ppi increases. Nevertheless, inclusion of heat and mass transfer correlations in a two-zone (heterogeneous) porous model would most likely improve the agreement in the oxidation section Catalytic Surface Area and Site Density A final uncertainty is in the foam s catalytic surface area and the nature of the active sites. The BET surface area is typically found to be ~1-2 m 2 g -1 (specific surface area ~1.6x10 4 cm -1 ) for a blank foam [129] similar to those used in this work. The specific geometric surface area (4/diameter) for an average pore diameter of 250 µm is ~160 cm -1 (i.e. ~100 times smaller). In this work, mechanism 1 was used with 160 cm -1, whereas mechanism 2 was used with 7500 cm -1 (a factor of ~ 50 larger) as suggested by Mhadeshwar and Vlachos [111]. Mechanism 2 shows no reactivity at 160 cm -1 for typical inlet PFR temperatures shown in Figure 4-4. Mechanism 1 reacts much more rapidly at the higher 7500 cm -1 value. 88

113 The difference between BET surface and geometric surface is caused by the microstructure of the foam showing small cracks that increase the surface area. The addition of a metal with or without washcoat may lead to a film coverage with a catalytic surface close to the geometric surface area, as shown previously by surface characterization [138]. Assuming that every surface metal atom is a catalytic site and that the film is an extended Rh(111) surface leads to an estimate of the catalytic site density of ~2.7x10-9 mol/cm 2 as used in this work. However, metal dispersion on the surface can be a strong function of surface free energy of the support and the metal, the metal loading, and the operation temperature. If the metal sinters and forms crystals under operation then site density may change as a function of time on stream. It is also debatable if one single atom can be considered as an active site. A last fact that is not considered in the model at all is a loss of active sites due to multilayer carbon formation, in particular at the end of the catalyst. In fact, more characterization is needed to determine the catalyst surface area for foams under reaction conditions so that surface mechanisms can be tuned at the appropriate surface to volume ratio Comparison of PFR Models and 2D Models Both reactor model and surface mechanism are critical to simulate the species development in the catalyst correctly. A comparison between adiabatic 2D and PFR simulations show that considering axial heat transport in the wall and in the gas is crucial. The adiabatic PFR model (Figure 4-5) predicts a dramatically higher peak temperature than the 2D model (Figure 4-10) and therefore shows an unreasonably fast development of the species profiles to their final steady-state values. A PFR model might be applicable if a correct experimental surface temperature profile could be provided (e.g. pyrometer measurements). However, an experimental temperature profile composed of thermocouple readings is strongly biased towards the gas phase temperature, which is in the oxidation zone much lower and in the reforming zone higher than the surface temperature. In summary, neither the PFR model with the experimental thermocouple profile (Figure 4-2) nor the adiabatic PFR model (Figure 4-5) could resolve the measured species profiles using either mechanism. The PFR profiles develop either too slow or too fast. 89

114 4.4.4 Comparison of 2D Channel Model and 2D Porous Model Correct simulation of the surface temperature profile is mandatory to reasonably predict species development in an autothermal foam catalyst. Including effects of axial heat conduction in the alumina support and the quartz tube in the model is important. Even if the solution of a PFR model with a 1 cm long isothermal catalyst wall will approach the solution of a full 2D model by the reactor outlet [139], it never will for an autothermal foam catalyst that features large temperature gradients. Neglecting the effect of wall heat conduction seriously affects the surface temperature [134] and predicted reactor yields [140]. Axial heat conduction in the quartz tube and the alumina support is included in the 2D porous and channel model used in the present work; hence they show similar performance (Figure 4-10). The influence of the inlet temperature (25 C) is more severe on the channel model since the walls possess a higher thermal conductivity than in the porous model, and the channel model only comprises a hypothetical pore of the foam monolith stack (front heat shield, catalyst, back heat shield). Therefore, more heat is dissipated upstream and downstream resulting in a lower catalyst temperature and less steam reforming (CH 4 and H 2 O flowrates higher, H 2 and CO flowrates lower than experiment). Panel A in Figure 4-10 shows how the calculated temperatures of the 2D models compare to the experimental values measured with the thermocouple. As surface temperature dictates the chemistry only one energy equation is solved for both surface and gas (local thermal equilibrium). Together with the thermocouple measurements, the simulated profile and the experimental profile present together a set of surface and gas phase temperatures respectively. Exothermic reactions raise the surface temperature in the oxidation zone sharply. The surface temperature maximum occurs closely to the catalyst entrance, even before oxygen is fully consumed, because endothermic steam reforming sets in as soon as water is formed. The incoming gases need some time to adjust to the hot surface, therefore the temperature profile measured with the thermocouple falls below the calculated surface temperature. After the surface temperature peaks, endothermic steam reforming pulls the surface temperature down leading to a cross over between surface and gas temperature at ~1.8 mm. After cross over the gas temperature remains higher than the surface temperature because of heat transfer limitations from the hot gases to the surface that is constantly cooled by steam 90

115 reforming. These results are in line with previous simulations solving energy equations for both the gas and solid phases for fixed beds [109, 134]. The authors also report large temperature gradients between gas and surface in the catalyst entrance region (T sur >> T gas ) with a cross over and a slightly higher gas phase temperature in the reforming zone. For C/O = 1.0, the 2D porous model predicts the species and temperature profiles quantitatively within the experimental margins of error (Fig. 10, panels A-D). The channel model captures the trends qualitatively but the absolute values are outside the experimental margins of error (e.g. H 2, CO, CH 4 ). The 2D porous model is therefore superior Comparison of 2D Porous Model and Experimental Profiles The 2D porous model with the 38-step surface mechanism (mechanism 1) was applied to predict species and temperature profiles for C/O=0.7, 1.0, and 1.3 at a 5 slpm inlet flowrate (Figures 4-10 and 4-11). Quantitative species agreement between experiments and simulations for the 2D porous model is good for all three C/O ratios, and only few significant differences are observed. Probably the most pronounced weakness of mechanism 1 observed in this study is that it predicts more WGS to occur at C/O = 0.7 than is seen experimentally. This leads to an underprediction of the CO molar flowrate at the end of the catalyst (Figure 4-11B) and an overprediction of the CO 2 molar flowrate at the end of the catalyst (Figure 4-11C). At C/O = 1.3, the amount of steam reforming is slightly underpredicted leading to a higher H 2 O and lower H 2 molar flowrate at the catalyst exit than found experimentally. Simulations show clearly that surface mechanism 1 predicts a mixed mechanism for syngas production: CO and H 2 are produced in the oxidation zone and in the steam reforming zone. This is in agreement with the experiment. The model also predicts correctly that the higher the temperature, the higher the amount of H 2 and CO produced in the oxidation zone from C/O = 1.0 to C/O = 0.7 (see section 4.3.3). Direct H 2 formation is less pronounced than direct CO formation, again in agreement with the experimental trend. Species profiles at the catalyst exit compare well with the experimental selectivities found in this work and in previous integral data comparisons [99, 109]. Deviations between the experimental and predicted profiles in the first half of 91

116 the catalyst and the length of the oxidation zone cannot be considered significant as the experimental profiles show variability from catalyst to catalyst Comparison of Mechanism 1 and Mechanism 2 A significant part of this work was devoted to the comparison of two state-of-the-art surface reaction mechanisms for methane CPO on Rh [99, 111]. As mechanism 2 [111] could not be used within a 2D channel or porous model, the comparison was accomplished within a PFR reactor model using (i) experimental temperature profiles as input and (ii) treating the reactor as adiabatic. As the surface temperature is too low in case of using experimental thermocouple measurements as input (measures gas phase temperature) and too high in case of the adiabatic treatment, the species developments are either over- or underpredicted and only qualitative conclusions can be drawn. There are two important differences between mechanism 1 and 2. Whereas mechanism 1 [99] predicts continuously developing species and surface coverage profiles over the entire length of the catalyst, mechanism 2 [111] predicts extremely sharp changes in species flowrates and surface coverages at the point of total O 2 conversion. This nearly discontinuous behavior is obviously the result of the temperature and coverage dependencies of all surface reaction steps in mechanism 2. The second difference is that mechanism 2 predicts only small amounts of CO and almost no H 2 to be produced in the oxidation zone in contrast to mechanism 1, which predicts significant direct syngas formation in agreement with the experimental profiles (Figures 4-2B and 4-5C). The performance of mechanism 2 observed in this work is in agreement to what was found by Mhadeshwar and Vlachos themselves in an earlier comparison between simulation and experimental profiles [111]. In that work, the authors validated their mechanism against experimental spatial data [56] using the thermocouple temperature profiles as input. As in Figures 4-2 and 4-5, the mechanism predicted a distinct oxidation zone (CO 2, CO and H 2 O are products) and a reforming zone (CO and H 2 are products). The authors concluded that H 2 was produced strictly in the reforming zone whereas CO was produced in both oxidation and reforming zone. This assertion was in disagreement with the conclusions of the authors of the experimental spatial study [56]. In summary, mechanism 1 embedded in a realistic reactor model (e.g. 2D porous model) predicts the experimental profiles within their margins of error. The only significant difference found between model and experiment was the overprediction of 92

117 WGS at C/O = 0.7. Mechanism 2 was not used with a 2D model, however the PFR simulations indicate that mechanism 2 is inferior to mechanism 1 as it does not capture H 2 formation in the oxidation zone and predicts physically unlikely discontinuities at the point of total O 2 conversion. 4.5 Conclusions It has been shown in this work that axial species profiles for methane CPO on autothermal Rh coated foam catalysts can be modeled accurately by combining a multistep surface reaction mechanism with a 2D reactor model capturing axial heat and mass transport. Best results were obtained with a 38-step surface reaction mechanism and a 2D porous model. Three inlet stoichiometries (C/O = 0.7, 1.0, 1.3) at 5 slpm inlet flowrate were investigated and the predicted profiles agree within the margins of error with the experimental profiles. Only at C/O = 0.7 does the 38-step mechanism predict significantly more WGS than experimentally observed. A PFR model employing either an experimental gas phase temperature profile or an adiabatic boundary condition was shown to be unsatisfactory because it does not capture the right temperature profile of the surface. A 2D channel model was significant better than the PFR model but still inferior to the 2D porous model because it over emphasized heat transport in the channel wall. Another more extensive surface reaction mechanism consisting of 104 surface reaction steps was found to predict exit flowrates similar to the 38-step mechanism within the PFR embodiment but it did not predict significant hydrogen production in the oxidation zone in contrast to the 38 step mechanism and the experimental profiles. 93

118 quartz tube heat shields catalyst inlet axis outlet insulated outer wall Figure 4-1 Schematic of computational domain for 2D porous media model. Only the top half of the domain shown is simulated because of axisymmetry. The heat shields and catalyst (crosshatched regions) are modeled using a 1D sub-grid treatment that accounts for the effect of the porous media on flow and transport. Horizontal scale units are cm. 94

119 molar flowrate (mol/min) CH 4 O 2 (A) T exp (gas) T ( o C) distance (mm) (B) molar flowrate (mol/min) H 2 CO distance (mm) (C) 0.04 molar flowrate (mol/min) H 2 O 0 CO distance (mm) Figure 4-2 PFR species profiles based on experimental T profiles. Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. Inlet conditions: 5 slpm and C/O=

120 (A) 10 0 Rh(s) 10-1 O(s) CO(s) C(s) coverage H(s) coverage OH(s) CH (s) 3 H O(s) distance (mm) O(s) CH(s) Rh(s) H(s) Rh(s) CH (s) 2 C(s) CO(s) (B) 10-4 OH(s) H O(s) 2 O(s) distance (mm) Figure 4-3 PFR coverages based on experimental T profiles. (A) mechanism 1 and (B) mechanism 2. Inlet conditions: 5 slpm and C/O=

121 oxidation zone length (mm) mechanism 1 w/ cov. dep. mechanism 2 mechanism 1 w/o cov. dep T (K) inlet Figure 4-4 Effect of inlet temperature on oxidation zone length (99.9% oxygen conversion) for adiabatic PFR simulations with mechanisms 1 and 2. Reactor inlet conditions: 5 slpm total flowrate and C/O =

122 (A) 1200 T surf (sim) 1000 T( o C) T exp (gas) distance (mm) 0.08 (B) molar flowrate (mol/min) O 2 CH distance (mm) Figure 4-5 Adiabatic PFR species profiles (5 slpm and C/O=1.0. Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (A) Comparison of simulated surface temperature with experimental temperature. (B) CH 4 and O 2 flowrates. 98

123 0.12 (C) molar flowrate (mol/min) H 2 CO distance (mm) molar flowrate (mol/min) (D) H 2 O molar flowrate (mol/min) CO distance (mm) 0 CO distance (mm) Figure 4-5 (cont.) PFR species profiles based on adiabatic conditions (5 slpm and C/O = 1.0). Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (C) H 2 and CO flowrates. (D) H 2 O and CO 2 flowrates. 99

124 (A) 10 0 O(s) Rh(s) 10-1 Rh(s) CO(s) coverage CO(s) OH(s) C(s) H(s) H(s) CH (s) 3 OH(s) O(s) distance (mm) (B) O(s) Rh(s) CH(s) H(s) coverage Rh(s) CH (s) 2 C(s) CO(s) OH(s) H(s) CO(s) H O(s) OH(s) O(s) distance (mm) Figure 4-6 Adiabatic PFR coverages (5 slpm and C/O=1.0). (A) Mechanism 1. (B) Mechanism

125 T( o C) C/O = 0.7 T surf (sim) T exp (gas) (A) molar flowrate (mol/min) distance (mm) O 2 CH 4 molar flowrate (mol/min) O 2 CH 4 (B) distance (mm) distance (mm) Figure 4-7 Adiabatic PFR species profiles (5 slpm, C/O = 0.7). Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (A) Comparison of simulated surface temperature with experimental temperature. (B) CH 4 and O 2 flowrates. 101

126 0.12 C/O = 0.7 (C) molar flowrate (mol/min) CO H distance (mm) (D) molar flowrate (mol/min) H 2 O 0 CO distance (mm) Figure 4-7 (cont.) PFR species profiles based on adiabatic conditions (5 slpm, C/O = 0.7). Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (C) H 2 and CO flowrates. (D) H 2 O and CO 2 flowrates. 102

127 1600 C/O = 1.3 (E) T surf (sim) 1000 T( o C) T exp (gas) distance (mm) (F) 0.08 molar flowrate (mol/min) O 2 CH distance (mm) Figure 4-7 (cont.) PFR species profiles based on adiabatic conditions (5 slpm, C/O = 1.3). Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (E) Comparison of simulated surface temperature with experimental temperature. (F) CH 4 and O 2 flowrates. 103

128 0.12 C/O = 1.3 (G) 0.1 molar flowrate (mol/min) H 2 CO 0 molar flowrate (mol/min) distance (mm) H 2 O molar flowrate (mol/min) CO 2 (H) distance (mm) 0 CO distance (mm) Figure 4-7 (cont.) PFR species profiles based on adiabatic conditions (5 slpm, C/O = 0.7). Solid and dashed lines indicate simulations with mechanisms 1 and 2, respectively. (G) H 2 and CO flowrates. (H) H 2 O and CO 2 flowrates. 104

129 C/O = 0.7 C/O = 1.3 (A) (C) 10 0 O(s) Rh(s) 10 0 O(s) Rh(s) C(s) coverage Rh(s) CO(s) OH(s) H(s) C(s) O(s) OH(s) coverage Rh(s) CO(s) OH(s) H(s) C(s) CH 3 (s) O(s) OH(s) CO(s) H(s) distance (mm) (B) distance (mm) (D) 10 0 O(s) Rh(s) 10 0 O(s) CH(s) Rh(s) coverage H(s) Rh(s) OH(s) CH (s) 2 O(s) CO(s) H(s) H O(s) 2 CH(s) OH(s) H O(s) 2 C(s) distance (mm) coverage Rh(s) H(s) CO(s) OH(s) H(s) CH (s) 2 C(s) OH(s) CO(s) H 2 O(s) distance (mm) Figure 4-8 Adiabatic PFR species coverages (5 slpm, C/O = 0.7 and 1.3). Panels A-B correspond to C/O = 0.7; panels C-D correspond to C/O = 1.3. (A, C) Mechanism 1. (B, D) Mechanism

130 min max v T T CH 4 O 2 H 2 CO H 2 O CO 2 FHS CAT BHS Figure 4-9 Contour plots of temperature (T), velocity magnitude (v), and molar flows for adiabatic 2D porous media model (5 slpm and C/O = 1.0). Only top half of domain shown in Figure 4-1 is simulated because of axisymmetry. T and molar flows are shown as close up for the foam monolith stack. Min/max ranges: T: C, v: m/s, CH 4 : mol/min, O 2 : mol/min, H 2 : mol/min, CO: mol/min, CO 2 : mol/min, H 2 O: mol/min. 106

131 T num (surf) T exp (gas) (A) T( o C) distance (mm) (B) molar flowrate (mol/min) O 2 CH distance (mm) Figure 4-10 Centerline temperature and species profiles for 2D adiabatic models. Solid and dashed lines indicate simulations with the 2D porous media and pore channel models, respectively. Inlet conditions: 5 slpm and C/O = 1.0. (A) Comparison of simulated surface temperature with experimental temperature. (B) CH 4 and O 2 flowrates. 107

132 0.1 (C) molar flowrate (mol/min) H 2 CO distance (mm) (D) 0.04 molar flowrate (mol/min) H 2 O CO distance (mm) Figure 4-10 (cont.) Centerline temperature and species profiles for 2D adiabatic models (5 slpm and C/O = 1.0). Solid and dashed lines indicate simulations with the 2D porous media and pore channel models, respectively. (C) H 2 and CO flowrates. (D) H 2 O and CO 2 flowrates. 108

133 0.08 T surf (sim) C/O = 0.7 (A) molar flowrate (mol/min) O 2 CH 4 T exp (gas) T ( o C) distance (mm) (B) 0.1 molar flowrate (mol/min) H 2 CO distance (mm) (C) 0.04 molar flowrate (mol/min) H 2 O 0 CO distance (mm) Figure 4-11 Centerline temperature and species profiles for 2D adiabatic porous model (5 slpm and C/O = 0.7 and 1.3). Panels A-C correspond to C/O = 0.7. (A) Comparison of simulated surface temperature with experimental temperature and CH 4 and O 2 flowrates. (B) H 2 and CO flowrates. (C) H 2 O and CO 2 flowrates. 109

134 0.08 C/O = 1.3 T exp (gas) (D) 1000 molar flowrate (mol/min) O 2 CH 4 T surf (sim) T ( o C) distance (mm) 0.1 (E) molar flowrate (mol/min) H 2 CO distance (mm) (F) 0.04 molar flowrate (mol/min) H 2 O CO distance (mm) Figure 4-11 (cont.) Centerline temperature and species profiles for 2D adiabatic porous model (5 slpm and C/O = 0.7 and 1.3). Panels D-F correspond to C/O = 1.3. (D) Comparison of simulated surface temperature with experimental temperature and CH 4 and O 2 flowrates. (E) H 2 and CO flowrates. (F) H 2 O and CO 2 flowrates. 110

135 10 0 Rh(s) coverage C(s) H(s) O(s) CO(s) OH(s) H 2 O(s) C/O = distance (mm) Rh(s) CO(s) C(s) coverage H(s) OH(s) H 2 O(s) O(s) C/O = distance (mm) Rh(s) CO(s) C(s) coverage O(s) OH(s) H 2 O(s) H(s) C/O = distance (mm) Figure 4-12 Centerline species coverages for 2D adiabatic porous model (5 slpm and C/O = 0.7, 1.0, and 1.3). 111

136 T wall radial distance (cm) T( o C) T( o C) T axis 500 axial distance (cm) distance (cm) radial distance (cm) T( o C) T( o C) T wall T avg T axis 500 axial distance (cm) distance (cm) T wall radial distance (cm) T( o C) T( o C) T avg T axis 500 axial distance (cm) distance (cm) Figure 4-13 Effect of pore diameter on thermal entry length for methane CPO. Adiabatic Chemkin boundary layer calculation for pore diameters of 0.25 mm (top panels), 1 mm (middle panels), and 2 mm (bottom panels). Inlet conditions are 5 slpm total flowrate, C/O = 1.0, T in = 750 K, and v in = 80 cm/s. T avg and T axis denote the area average gas-phase temperature and centerline temperature, respectively. 112

137 Chapter 5 Rapid Lightoff of Syngas Production from Methane: A Transient Product Analysis 5.1 Introduction CPO of methane [2, 3] and larger alkanes [17, 63, 74] to produce syngas at millisecond contact times offers a pathway to hydrogen production for small mobile power applications such as solid oxide and proton exchange membrane fuel cells. The global reaction for methane CPO is 1 o CH4 + O2 CO + 2H2 Δ H R = kj/mol CH4 2 (5-1) and could be used instead of steam reforming: o CH4 + H2O CO + 3H2 Δ H = +206 kj/mol CH. (5-2) R 4 The partial oxidation reaction is exothermic so it can operate autothermally. In contrast, steam reforming is endothermic, thus requiring a furnace that is difficult to scale down and slow to start up. A key benefit of a CPO reactor is that it can operate at GHSV two to three orders of magnitude higher than a conventional steam reformer ( hr -1 vs.10 3 hr -1 ) while giving similar yields, so it should be two to three orders of magnitude smaller than a steam reformer for the same production rate. The relatively small footprint of CPO is attractive for applications that cannot accommodate a large fuel reformer, such as fuel cells to power automobiles or pollution abatement equipment for existing automotive engines. The positive impacts of hydrogen addition to the spark ignition engine on pollution abatement, especially during start-up, have motivated further research on both the onboard generation of hydrogen and the logistics of hydrogen delivery to the engine [77, 78]. Rapid start-up of these compact systems will be crucial. Given the frequent start-ups and process fluctuations automotive reforming systems will face, they will be required to produce hydrogen from room temperature mixtures in seconds and respond almost instantaneously to dramatic shifts in required power output. Therefore, design and optimization of a compact Chapter 5 is adapted from K.A. Williams, C.A. Leclerc, and L.D. Schmidt, Rapid Lightoff of Syngas Production from Methane: A Transient Product Analysis, AIChE Journal, 51, (2005) John Wiley & Sons, Inc. All rights reserved. 113

138 reforming system capable of producing hydrogen within several seconds is a critical step towards creating enabling technologies that can rapidly generate energy with lower harmful emissions than current practices. Studies on lightoff of methane CPO have yielded a range of start-up times. Use of an electrically-heated Pd-coated metallic monolith gave a lightoff time of ~90 s [141]. A Rh-coated honeycomb monolith preheated to 700 K with a furnace gave a start-up time of ~120 s after ignition [100]. In comparison, it was shown that CPO on Rh-coated alumina foams produced hydrogen from methane in ~10 s without feed preheat in a fast lightoff reactor that initially made use of the combustion reaction [142] CH + 2O CO + 2H O Δ H = -802 kj/mol CH (5-3) o R 4 by igniting a stoichiometric mixture of methane and air (CH 4 /O 2 = 0.5). The resulting premixed flame heated the catalyst to lightoff temperature (> 500 C), and then the fuel to air ratio was increased to partial oxidation stoichiometry (CH 4 /O 2 = 2.0) to extinguish the flame and allow autothermal syngas production. The time needed to reach steady-state syngas production was determined by the catalyst back face temperature and analysis of samples obtained from the reactor effluent by gas chromatography. The focus of the present study is the transient behavior of the catalyst foam during lightoff. In this work, the fast lightoff reactor is coupled with a fast response mass spectrometer and data acquisition system in order to quantify selectivities and methane conversion during the rapid start-up period. Additionally, numerical plug-flow models incorporating multi-step surface chemistry are used to simulate the transient species profiles and investigate how the monolith temperature develops with time. The combined experimental and numerical efforts supply useful information on the transient reactor behavior for various combustion times and identify a combustion time to avoid undershoot or overshoot in catalyst temperature and minimize start-up time. 5.2 Materials and Methods Experimental Apparatus and Procedure High purity reactant gases (CH 4, O 2, and Ar) were fed through calibrated mass flow controllers to regulate flowrates. Premixed gases were fed through 0.64 cm inside 114

139 diameter (ID) stainless steel tubing to the top of the reactor through a quartz endcap with a 0.64 cm ID side port. The reactor consisted of a quartz tube approximately 19 mm ID by 40 cm long (Figure 5-1A). Reactants passed through a blank ceramic monolith, which acted as a mixer and flame holder during the lightoff combustion phase, and entered the ignition compartment where electrodes of a spark generator were placed immediately upstream from the catalyst. The catalyst, mixer, and back heat shield (behind the catalyst to attenuate axial heat losses) were 80 ppi α-alumina foam monoliths. The catalytic monolith was 5 mm long by 18 mm diameter while the mixer and back heat shield were 10 mm long by 18 mm diameter. All three monoliths were wrapped in alumino-silicate paper to make a tight seal before being placed inside the quartz tube. Foam monoliths were coated with γ-alumina washcoat to increase effective surface area and then coated with Rh(NO 3 ) 3 solution to give final Rh loadings of ~5% by weight. The rest of the catalyst preparation was described previously [143]. The reactor temperature was measured at the catalyst back face with a 0.25 mm diameter type K thermocouple (< 0.3 s response time constant). The catalyst section was wrapped with ceramic blanket insulation to attenuate radial heat losses. Effluent gases flowed out of the reactor into heated stainless steel tubing. The outlet flow was split after the bottom reactor endcap so that the majority of the flow was directed to an incinerator while the remainder traveled to a differentially pumped quadrupole mass spectrometer (QMS) system for analysis. The gases for analysis entered a pre-chamber maintained at 0.07 kpa by a mechanical vacuum pump, passed into another chamber pumped down to 10-8 kpa by a turbomolecular pump, and then entered the QMS for mass separation and detection (Figure 5-1B). The QMS system scanned masses 1 through 50 at ~3 Hz. Argon instead of nitrogen was used as the calibration diluent because nitrogen overlapped with carbon monoxide at mass 28. QMS signals for water (mass 18) are not reported due to water condensation in the system. QMS signals were corrected using calibration curves that plotted the ratio of each component response over argon response versus the ratio of each species actual concentration over argon concentration. Accurate product concentrations for various reactant ratios were determined through daily analysis with a GC equipped with a thermal conductivity detector. Carbon and hydrogen atom balances typically closed to within 5% error. 115

140 Experimental control and data acquisition were automated using a computer workstation running Labview that allowed implementation of quick changes in reactant stoichiometry, control of the spark generator, combustion time input, and all experimental data to be written to a data file for analysis. Simulated air flowrates (79% Ar/21% O 2, or hereafter air ) were 5 to 9 slpm while methane flowrates were adjusted accordingly to create equivalence ratios (φ) of either 1.0 or 4.0 (CH 4 /O 2 = 0.5 or 2.0), which correspond to total or partial oxidation. Given the tortuous 80-ppi catalyst geometry with ~83% porosity [90], steady-state GHSV ranged from 4.0x10 5 to 7.3x10 5 hr -1. Experimental procedure has been described previously [142]. Initially, an ambient mixture of methane and air was fed to the reactor at combustion stoichiometry. A spark ignited the mixture forming a premixed flame upstream of the catalyst that heated the foam monolith to catalytic lightoff (T > 500 C) in seconds. After the combustion phase (1 to 20 s in this study), the methane/oxygen ratio was increased to partial oxidation stoichiometry, which extinguished the flame and caused immediate autothermal syngas production. Experimental variability was determined by repeating experiments 5 times at each combustion time for each flowrate. A total of ~50 monoliths were used in this work with reproducible results Determination of QMS System Response Time Accurate measurement of transient species compositions leaving the reactor required that the response time of the analytical system was rapid, and axial mixing of the products was minimized. Quantification of the product response and lag times from the reactor outlet to the QMS system was determined through a set of diagnostic experiments performed under both room temperature and steady-state reactor temperature conditions. First, a mixture at CPO stoichiometry (5 to 15 slpm) was fed to the cold reactor, step changes in each component (CH 4, O 2, Ar) were made sequentially, and QMS responses were recorded. Figure 5-2 shows the mass flow controller and QMS system responses to a step change in methane flowrate. Note that there are three time constants involved in the QMS system (mass flow controller response, pre-chamber mixing, and QMS chamber mixing). However, the pre-chamber mixing response (at 0.07 kpa) dominates the QMS chamber mixing response (at 10-8 kpa). 116

141 To ascertain an individual response time constant for the QMS system, the mass flow controller response was removed from the overall measured response. Mass flow controller response times were calculated using a first-order fit of the output voltage versus time. Differences in mass flow controller response time constants between species or flowrates were insignificant (0.53 s average ± 0.11 s standard deviation). A second-order plus time delay model was used to fit the overall measured response, and error was minimized in a least-squares sense. Lag time was determined visually. The fit was calculated by holding one time constant at the predetermined flow controller value while calculating the other. Model fit was excellent for all cases giving a QMS time constant that was not statistically significant between species or flowrates. There was no significant difference between the QMS system response (0.55 s average ± 0.10 s standard deviation) and the flow controller response reported above. A plot of inverse flowrate versus lag time was linear as expected. The above procedure was repeated under steady-state CPO conditions (~750 C and 1.1 atm) for the same flowrates (5-15 slpm) and a step change in argon flowrate. Although lag times markedly decreased (the product flowrate under CPO conditions increased approximately fourfold over the reactant flowrate), the change in the QMS response time constant was statistically insignificant compared to previous experiments under room temperature conditions. This observation indicated that the effect of axial mixing on system response time was insignificant compared to the pre-chamber mixing response Numerical Simulations Numerical computations were used to test the capability of a multi-step elementary kinetic model to accurately simulate species profiles during the transient lightoff of methane on Rh for an experimental GHSV of 4.0x10 5 hr -1 (total inlet flow = 7.1 slpm) and combustion times of 1 and 5 s. A single straight hypothetical pore of a 5 mm long 80-ppi foam monolith was modeled using a nominal diameter of 0.25 mm, effectively neglecting the tortuous pore network found in ceramic foams. To generate numerical species profiles, model feed conditions were matched to the recorded mass flow controller outputs from experiment at time t. Simulations in this work utilized a 1D plug-flow model with a detailed 38-step surface chemistry mechanism for methane oxidation on Rh [100]. Gas-phase chemistry is negligible at low pressure conditions for methane CPO [97, 98, 117

142 105] and was not included in the simulations. See Chapters 2 and 3 for more information and governing equations of the plug-flow model. The catalyst s activity is included through the assumed site density for the Rh surface, which was 2.72x10-9 mol cm -2 for the mechanism. The plug-flow assumption was verified using two criteria. If only radial mass transfer is considered, then plug-flow can be assumed if Dθ Fo = > 1, (5-4) 2 r where Fo is the Fourier number, D is the characteristic diffusion coefficient, θ is the residence time, and r is the reactor radius [144]. A more complete assessment treating both the axial and radial transport assumptions of plug-flow yields the following acceptable bounds: d L << Re d Sc <<, (5-5) L d where d is the reactor diameter, L is the reactor length, Re d is the Reynolds number based on reactor diameter, and Sc is the Schmidt number [139]. Using eqn. 5-5, a 5 mm catalyst length and 0.25 mm pore diameter give a practical range of for the mass transfer Peclet number. The Chemkin module TRANSPORT [145] was used to calculate binary transport coefficients based on temperature and species, while a mixture average treatment was employed to calculate characteristic transport coefficients for eqns. 5-4 and 5-5. Based on the average GHSV and stoichiometry used in this study (4x10 5 hr -1 and CH 4 /O 2 = ) and assigned temperature history of the catalyst, which is presented in the next section, the radial mass transfer Fourier number ranged from ~2-25 while the mass transfer Peclet number range was 2-9 over the transient lightoff period validating the plug-flow assumption. Under the experimental conditions used, reactor residence times of 1 to 10 ms are typical and can be considered a representative time scale for reactive flow. A more relevant time scale for surface chemistry would be Rh surface desorption times for pertinent species that range from approximately 10-3 to s (e.g., 10-3 s for O 2, 10-9 s for H 2, and s for H 2 O). However, the reaction time scales are many orders of magnitude smaller than the monolith thermal response time, O(1 s). Therefore, the time scales for reactive flow and catalyst thermal response are assumed decoupled in these 118

143 simulations. Taking advantage of this simplification, transient temperature profiles were employed in the simulations so the energy equation was not solved Temperature Profiles for Numerical Simulations Since a premixed flame initially heats the catalyst, only by solving the Navier- Stokes equations with homogeneous chemistry and radiation treatment can a rigorous numerical solution for the transient catalyst temperature profile be attained after the combustion phase. To simplify the model, all uncertainty about the axial temperature profile in the first few seconds of lightoff was placed into parameterized catalyst temperature profiles, which were employed in simulations. In order to isolate parameters for the temperature profiles, previous experimental and numerical investigations focusing on the monolith temperature profile under steadystate and transient CPO conditions were examined. At steady-state CPO conditions similar to conditions in this work, the maximum catalyst temperature (1200 C) is found within the first millimeter of the monolith [107] and is indicative of a short, exothermic oxidation section where oxygen is completely consumed. The endothermic reforming section after the first millimeter is characterized by a roughly exponential temperature decrease to the catalyst back face (~400 C temperature drop); an approximately 300 C linear temperature drop is observed from 1 mm to 5 mm [107]. Data from previous lightoff experiments and simulations suggest that for very short initial combustion times (~1 s), no temperature overshoot (compared to the steady-state profile) occurs [142]. For slightly longer combustion times (e.g., 5 s), temperature overshoot presumably occurs in the oxidation section, and for even longer combustion times it can occur in the reforming section. To determine the kinetic model s ability to reproduce experimental species profiles with and without the overshoot phenomena, both 1 s and 5 s combustion time temperature profiles were developed. Based on the above considerations, three parameters were used to generate the temperature profiles for the 1 s combustion time model: (1) steady-state catalyst front face temperature; (2) oxidation section length ( ox ); (3) heating time constant for the catalyst front face (τ ff,h ). Since radial mixing is pronounced in the foam, temperature was lumped in the radial direction and distributed axially. A first-order exponential heating profile was used to estimate how the front face temperature developed. The oxidation 119

144 section length was made isothermal for simplicity and assumed to develop instantaneously at the feed switch. The remainder of the catalyst temperature profile, which simulates the reforming section, followed an exponential decay to the experimentally measured back face temperature at each time point: z ox T( z > ox, t) = T( z,0) + [ T(0, t) T( z,0) ] Exp, (5-6) τ s () t where T(,0) z is the initial monolith temperature from experiment, and T(0, t) is the front face temperature at time t. The spatial decay constant τ s (t) was determined by inserting T(,0) z and T(0, t) into eqn. 5-6 and solving for τ s (t) to make T(5 mm, t) equal to the measured back face temperature at time t. The spatial decay constant at each time point was defined completely by the catalyst front face temperature, oxidation section length, and measured back face temperature and was not an independent parameter. For the 5 s combustion time model, two additional parameters were used since temperature overshoot occurred in the oxidation section based on experimental results. These parameters were the theoretical overshoot temperature in the oxidation section at 5 s (T max ), and the catalyst front face cooling time constant (τ ff,c ), which was applied at the front face after 5 s to lower temperature to its steady-state value. Optimized values for the oxidation section length and steady-state front face temperature were determined by minimizing the least-squares error between simulation output and averaged steady-state species data. Optimized values for the remaining parameter(s) for the 1 and 5 s combustion time models were obtained by fitting the transient simulation output to experimental data in a least-squares sense for the 1 s combustion time case and visually for the 5 s combustion time case. 5.3 Results First, experimental findings for product species profiles and catalyst temperature are presented for 1 to 20 s combustion times and various feed flowrates. Next, the general behavior of the surface chemistry mechanism used in simulations is outlined over a range of experimentally-relevant temperatures. Last, results from plug-flow simulations incorporating the transient temperature profiles are compared to ~1 and 5 s combustion time experiments. 120

145 5.3.1 Temperature and Species Profiles for Short Combustion Times For a 5 slpm air flowrate, species profile results from experiments employing 1 and 5 s combustion times are shown in Figure 5-3. Similar trends exist for both 1 and 5 s combustion time experiments. Standard deviations of approximately 5-10% of the local 1 s average are found for each species. Upon sparking the combustion mixture (t = 0 s), oxygen and methane flowrates rapidly decrease toward zero while carbon dioxide begins to increase. Hydrogen and carbon monoxide production is near zero. The catalyst back face temperature rises rapidly after an approximate 2 s delay. Upon switching to CPO stoichiometry, methane flow is increased fourfold while air flow is held constant. After this feed switch, hydrogen and carbon monoxide flowrates start to rapidly increase. Note that the methane mass flow controller response time for 99% gain is ~2 s and contributes to the observed delay. The carbon dioxide flowrate peaks immediately after the switch and then descends to its steady-state CPO value. Oxygen flow remains at zero while methane flow reaches a minimum near zero then increases to its new steady state value. Steady-state back face temperatures are C. However, significant differences in species response exist between the two combustion times. For a ~1 s combustion time (Figures 5-3A and B), overshoot occurs in the methane flow upon switching to φ = 4.0 indicating breakthrough, and that catalyst preheat was below optimum (undershoot in catalyst temperature compared to steadystate CPO conditions). Temperature, hydrogen, carbon monoxide, and methane reach their steady-state values within s. A 5 s combustion time (Figures 5-3C and D) proves much more effective in producing a quick steady-state for syngas. Hydrogen and carbon monoxide reach steady-state within 10 s (within 5 s from the switch) while temperature monotonically increases to steady-state. Slight overshoot in carbon monoxide at the feed switch indicates that temperature overshoot in the catalyst may occur during the combustion phase. The methane behavior indicates that the front portion of the catalyst possesses a higher temperature for the 5 s combustion time than the 1 s combustion time upon feed switch. Methane does not exhibit breakthrough and rises smoothly to steady-state (Figures 5-3C and D). A comparison of the transient fuel conversion and syngas selectivities marks the difference between 1 s and 5 s combustion times more clearly (Figure 5-4). For the 1 s combustion time, methane conversion peaks near 100% at 2 s during the combustion phase (Figure 5-4A). Once stoichiometry is switched, fuel conversion drops to below 121

146 60% then slowly rises back to a steady-state conversion of ~70%. Approximately 85% carbon monoxide and hydrogen selectivities are reached within 15 s. Using a 5 s combustion time initially gives near 100% fuel conversion from 3-5 s (Figure 5-4B). Upon switching the feed composition, methane conversion quickly descends to 70% exhibiting no breakthrough compared to steady-state performance. Hydrogen and carbon monoxide selectivities rise to ~85% within the same time span. Experiments performed with 7 and 9 slpm air flowrates give the same general results although peak back face temperatures increase over the 5 slpm air flow experiment as expected. For this reason, combustion times greater than 10 s were not performed with these higher flows as the peak back face temperature for 10 s combustion time experiments exceeded 1200 C risking Rh sintering and pronounced metal migration. A 5 s combustion time for 7 and 9 slpm air flowrates yields steady-state syngas selectivities of 85% within 10 s (data not shown). Steady-state fuel conversions (~70%) are also similar to the lower flowrate experiments Temperature and Species Profiles for Longer Combustion Times For combustion times 10 s and above, back face temperature overshoots steadystate and is accompanied by pronounced overshoot in hydrogen and carbon monoxide after the feed composition is switched (Figure 5-5). As combustion time is increased from 10 to 20 s, back face temperature and CPO product overshoot become more dramatic. These phenomena indicate that the catalyst becomes increasingly superheated with increasing combustion times compared to the temperature profile for steady-state partial oxidation. Longer combustion times decrease the initial hydrogen and carbon monoxide response times and increase the response time for unconverted methane to rise to its steady-state value Mechanism Behavior at Steady-state Figure 5-6 displays isothermal catalyst performance for the detailed surface chemistry mechanism [100] over the range of temperatures typically found in a lightoff sequence and GHSV applicable to this study (2.0 to 7.3x10 5 hr -1 ) and equilibrium data. Equilibrium values were calculated using the Chemkin module Equil, which minimizes the Gibbs free energy of the chemical system [146]. Three distinct kinetic regimes are evident in this mechanism. In the first regime, the catalyst lights off between 420 and 122

147 500 C yielding combustion products (CO 2 and H 2 O). The second regime (between 500 and 700 C) gives mostly combustion products. Methane conversion is approximately constant (~25%) in this regime, since methane is fed in excess of combustion; oxygen conversion is complete after 500 C. In the third regime (above 700 C), syngas quickly becomes the major product as temperature and methane conversion increase. At the highest temperatures, the lowest GHSV (2.0x10 5 hr -1 ) gives product compositions that approach equilibrium composition Temperature and Species Profiles for Short Combustion Time Simulations Table 5-1 summarizes the optimized parameters for the 1 s combustion time temperature profile. Results from the 1 s combustion time simulation (Figures 5-7 and 5-8) show that a transient, distributed temperature profile can be combined with a steadystate chemistry mechanism to reproduce experimental profiles. The temperature of the catalyst oxidation section quickly rises in the first second (Figure 5-7A) above the lightoff temperature for methane on Rh (~500 C). Upon switching to CPO conditions after ~1.5 s, the catalyst front face heats further, albeit at a slower rate, to steady-state CPO temperature. The temperature difference between the front and back of the catalyst peaks after a few seconds and then slowly decays to its 200 C steady-state value, which effectively mimics transient heat transport to the catalyst back face. The combination of the oxidation section temperature, experimental catalyst back face temperature, and exponential decay for the remainder of the catalyst simulates the temperature development of the reforming section in time and space after the stoichiometry switch (Figure 5-7B). Physically, the temperature drop in the reforming section is initially due to latency in heat transport, while at steady-state it is caused by endothermic reactions. The first 1 s of the experiment cannot be rigorously compared to the model since fuel consumption is most likely homogeneous in experiments yet heterogeneous in simulations. However, good agreement is still found in both fuel conversion (Figure 5-8A) and carbon dioxide production (Figure 5-8B). Upon switching to CPO conditions at ~1.5 s, the simulation qualitatively predicts the methane breakthrough experimentally observed. Experimental and simulated peak methane breakthrough values are not the same; however, the comparison is not entirely valid until the transition from φ = 1.0 to 4.0 is complete, and the upstream flame is extinguished (at ~2-3 s). In addition, the QMS system may not be able to resolve the rapid methane breakthrough peak that occurs in 123

148 less than 2 s. After 4 s, simulated and experimental methane conversions match well. Transient and steady-state agreements are excellent between model and experiment in syngas production, which occurs after ~3 s (Figure 5-8C). To test the temperature profile assumptions for longer combustion times, simulations were also performed for the 5 s combustion time case. In the first 5 s of lightoff, the oxidation section reaches a maximum temperature that overshoots steadystate CPO conditions (Figure 5-9A). The oxidation section heating profile is equivalent to the heating profile for the 1 s combustion model over the first 1.5 s to maintain consistency between both models. After 5 s, the oxidation section experiences firstorder cooling to the steady-state CPO temperature from the 1 s model. As in the 1 s combustion simulation, an exponentially-decaying profile for the catalyst from 1 to 5 mm is coupled with the assumed oxidation section temperature profile and experimental back face temperatures to simulate the reforming section thermal history (Figure 5-9B). Temperature profile parameters for the 5 s model are shown in Table 5-2. For this case, simulated species profiles are shown only after 5 s since simulated and experimental chemistries would be fundamentally different between 0 and 5 s. Upon switching to CPO feed at 5 s, methane flow does not exhibit overshoot in experiments (as in the 1 s combustion time case). Instead, it slowly approaches its steady-state value (Figure 5-10A). This phenomenon is consistent with a superheated catalyst front section as assumed in the temperature profile and reflected in the simulation. However, the simulation predicts a quicker response than experiments show for both methane and carbon dioxide (Figure 5-10A and B). This response disparity is also noticeable for hydrogen and carbon monoxide production (Figure 5-10C). The simulation predicts overshoot in both hydrogen and carbon monoxide, whereas overshoot in only carbon monoxide is found for 5 s combustion time experiments. 5.4 Discussion In this work, rapid start-up of an SCTR to produce syngas from methane has been demonstrated within a time scale relevant to mobile power applications (5-10 s) and in the absence of any external heating source. The reactor employs a Rh-impregnated ceramic foam monolith and an initial homogeneous combustion phase (duration of 1-5 s) to preheat the catalyst front face above the autocatalytic ignition temperature. Steadystate hydrogen yield is ~60% within 10 s for a CH 4 /O 2 ratio = 2.0; yield could be 124

149 increased significantly by running at a leaner feed composition (CH 4 /O 2 1.8) as shown in previous steady-state experiments [107]. Partial oxidation stoichiometry was employed only to show proof of concept of the reactor design. That transient species profiles can be reproduced using an established surface mechanism typically used to model steady-state chemistry can be demonstrated by 1D plug-flow simulations of short combustion time experiments (1 and 5 s). Combining empirically-based assumptions for the thermal catalyst history with the plug-flow model resulted in the development of a transient, axially-distributed temperature profile that offers insight into how the catalyst temperature develops during lightoff. This modeling flexibility is realized because reaction time scales for this process are at least several orders of magnitude smaller than the catalyst temperature response time. Therefore, chemistry and temperature during reactor start-up can be decoupled Minimizing Start-up Time by Modifying Combustion Time Lightoff experiments at a constant 5 slpm air feed elucidated how initial combustion time affects the overall time in which steady-state syngas is obtained. A 1 s combustion time yields steady-state syngas production within approximately 15 s after the switch (Figure 5-3A). Increasing combustion time to 5 s dramatically decreases syngas response time to 4-5 s (Figure 5-3C); however, adding the initial 5 s combustion time to the syngas response gives a total duration of 10 s from the start of the experiment. At 5 s combustion times and greater, the catalyst yields syngas as fast as the mass flow controller can switch the stoichiometry. Since the controller response is ~2 s, any decrease in syngas response time after switching by using combustion times longer than 5 s cannot be distinguished by the analytical system. If an optimal start-up time is desired, a trade-off exists between the length of combustion time and the catalyst response time after switching to CPO conditions. By using an optimal combustion time, methane breakthrough (at short combustion times) and syngas overshoot (at longer combustion times) are avoided. A combustion time of ~3 s gives steady-state syngas production in less than 7 s from the start of the experiment (data not shown), effectively heating the catalyst to a temperature profile similar to the steady-state CPO profile. Initial axial heat transport resistance is hypothesized to facilitate start-up so that by the feed switch, the temperature profile closely matches the steady-state CPO profile, which is characterized by a short, high temperature oxidation section and a longer, lower 125

150 temperature reforming section. At higher feed flowrates (7 and 9 slpm air), optimal combustion time decreases slightly for the same catalyst dimensions because of larger heat release Hydrogen Response Time in the QMS Concern over longer hydrogen response compared to other species and its effect on the accuracy of the transient profiles is warranted. Gases used in QMS diagnostic analyses had molecular weights of 16 or higher. Hydrogen poses a greater challenge for rapid QMS response. For a turbomolecular pump, pumping speeds generally decrease as compound molecular weight decreases. Specifically, the logarithm of the compression ratio is proportional to the square root of the gas molecular weight. The importance of this issue can be ascertained by analyzing transient hydrogen selectivity from experiments. For a 1 s combustion time (Figure 5-3A), experimental hydrogen and carbon monoxide response times after feed switch are comparable (approximately 10 s) and much longer than the QMS system response times measured in diagnostic work. Therefore, accurate results are assured. In longer combustion time experiments, overshoot in syngas is observed rapidly (Figure 5-5). Peak hydrogen selectivity is 80-90% after feed switch (based on methane input) demonstrating that the QMS system may not accurately resolve the large spikes in hydrogen production upon adjusting the feed composition. However, these inconsistencies in the hydrogen balance are only found at the peak response for longer combustion time experiments. The QMS response may also explain the initial quantitative discrepancies between experimental and simulated methane breakthrough peaks for the 1 s combustion time case (Figure 5-8A) and experimental and simulated carbon monoxide and hydrogen peaks for the 5 s combustion time case (Figure 5-10C) Catalyst Deactivation Fast lightoff experiments have been conducted using ~50 monoliths with reproducible results. For individual monoliths, no significant difference is found in catalyst performance over 20 lightoff sequences at the flowrates presented in this work as long as combustion times less than 10 s are used. At longer combustion times, some deterioration in syngas yield and catalyst mechanical integrity is observed after as few as five lightoff experiments. At longer combustion times, Rh in the catalyst front section 126

151 reaches temperatures well above 1200 C causing sintering and metal migration and agglomeration that decrease the active catalyst area and effective catalyst length. However, catalyst deterioration at long combustion times is not an issue since combustion times longer than 5 s are not needed to rapidly effect syngas production Kinetics and Transport Phenomena in Methane CPO The kinetic mechanism used in this study does not predict equilibrium products over the majority of the temperature range of interest in this study, as shown in Figure 5-6. There has been considerable debate on the role of transport phenomena in methane CPO on Pt and Rh. Previous experimental and numerical efforts with Rh and Pt foams suggest that the system is flux-limited by methane adsorption rather than by external mass transfer through a boundary layer since foams have very high interphase mass transfer rates [3, 105]. This phenomenon is manifested by the very small sticking coefficients used for methane adsorption in the mechanisms mentioned above and the similar mechanism used in this study. However, a performance comparison of several multi-step mechanisms for methane oxidation with in-situ experimental observations indicated a large uncertainty for the methane adsorption sticking coefficient on Pt [147]. For fixed bed supports, experimental and numerical studies with Rh-coated sphere beds have suggested that the process is oxygen mass transfer limited through a boundary layer [113, 148]. However, surface adsorption steps were not considered in the simulations. A more detailed numerical study of Rh-coated fixed beds utilizing the same mechanism as this work concluded that the importance of external mass transfer compared to methane adsorption increased with GHSV and particle size [109]. The studies presented above for methane CPO on Pt and Rh agree on the qualitative details of the catalyst s axial temperature and species profiles at steady state, regardless of support structure. The catalyst entrance or oxidation section is characterized by a maximum temperature where oxygen is completely consumed by both total and partial fuel oxidation. The remaining section of the catalyst allows endothermic reforming reactions (CO 2 and H 2 O reacting with the remaining fuel) and possesses a monotonic decrease in temperature. 127

152 5.4.5 Transient Simulations The goal of the transient models was to investigate how the axial temperature of the monolith may develop during lightoff by simulating a single hypothetical pore of the foam catalyst. In reality, foam monoliths are a highly porous and tortuous network of interconnected cells with a distribution of pore sizes unlike traditional honeycomb monoliths. Honeycomb structures consist of straight parallel channels of well-defined and uniform cross-section that are segregated convectively yet linked thermally. Modeling the lightoff of honeycomb structures has shown great promise [100] and benefits from the clearly defined geometry. These structures offer fertile ground for CFD model development of catalytic structures. Experimentally, however, many chemical reaction processes perform better with foam structures than conventional sphere beds and honeycomb structures. These foams offer benefits over conventional supports such as pronounced radial mixing, lower pressure drop, and higher effectiveness factors [90], although the heat transfer properties of sphere beds allow them to operate at higher GHSV without blowout for methane CPO [132]. The foam monolith geometry is quite complex, random, and therefore ill-defined for rigorous computational mesh development. However, assuming the development of a computational mesh possessing this intricate pore structure, accurate solution of the conservation equations coupled with even a simplified chemistry mechanism would be intractable with current computational resources considering the mesh refinement necessary. In contrast, solution of the energy equation with transport treatments more complex than plug-flow is of little quantitative benefit to intra-catalyst study if a pore is approximated as a single, straight channel since critical aspects of the foam geometry are ignored. If the averaged effluent species compositions are of primary concern, the high degree of radial mixing in tortuous foams likely limits the development of boundary layers typically found in honeycomb monoliths and aids the use of simplified solution treatments such as plug-flow. In view of these factors, a plug-flow model was coupled with empirical temperature profiles to simulate effluent species profiles during the lightoff of a foam monolith. Because the complex foam geometry is ignored in the straight channel simulations, differences between gas phase conditions and those at the gas-catalyst interface were neglected. Previous and more rigorous simulations for methane CPO on a Pt monolith showed that the gas and solid phases are thermally equilibrated within the first millimeter 128

153 [105]. Empirical temperature profiles were used in lieu of a 1D transient energy balance because of the complex nature of the combustion phase. Numerical simulations in this work attempted to capture the surface chemistry occurring after the switch from total to partial oxidation stoichiometry and draw useful conclusions about catalyst temperature development during lightoff. A 1D transient energy balance applied after combustion would only be as accurate as the initial conditions for the monolith s axial temperature profile once the stoichiometry was switched. Since this profile has not been well-defined experimentally or numerically through more sophisticated physical treatment, the presumed effect of the combustion phase on the monolith was lumped in the empirical temperature profiles Temperature Profile Parameters The temperature parameters iteratively determined in the simulations give good agreement between experimental and numerical species profiles. However, the temperature profiles and their accompanying best-fit parameters should be additionally bounded by theory. The length chosen for the CPO oxidation section (1 mm) is an upper bound and agrees well with more rigorous simulations solving the 1D energy equation [105]. For short-time combustion experiments exhibiting no temperature overshoot (e.g., ~1 s combustion time), heating of the catalyst back face to steady-state temperature should be dominated by two time constants: the time constant for the front face temperature to reach steady-state, and the effective diffusion time constant for the remainder of the monolith. The first time constant can be approximated by the actual combustion time itself (e.g., 1 to 1.5 s). From diffusion theory, an upper bound for the second time constant can be approximated as α/δ 2, where α is the thermal diffusivity of the alumina monolith and δ is the catalyst s axial length. A typical value for the alumina monolith thermal diffusivity (including monolith porosity) is 6.7x10-6 m 2 /s at 25 C [149]. Note that this value is an upper bound since transport inside the porous monolith is advective and not just diffusive. Combining this value with the catalyst length (5 mm) gives a time constant of 3.7 s. These theoretical time constants can be tested by fitting experimental back face temperature data to a second-order plus time delay fit. This approach should be reasonably robust since the thermocouple used to measure back face temperature possesses a rapid response time (< 0.3 s). For 1 s combustion time experiments, 129

154 minimizing the least-square error between the second-order fit and back face temperature data gives average time constants of 1.3 s and 2.2 s with a time delay of 2.2 s. The first time constant, 1.3 s, compares well with the average 1.5 s combustion time used in experiments and the 1.1 s time constant used as a simulation parameter (Table 5-1). The 2.2 s effective diffusion time constant is significantly less than 3.7 s from conduction theory, but this is expected since monolith heat transport has a significant convective component. The above approach is of limited value in justifying model parameters for longer combustion times where the initial flame heats the front face above the steady-state CPO temperature causing overshoot. In this case, three time constants (heating and cooling of front face and diffusion) will control the catalyst temperature profile and retrieving them accurately from a third order fit becomes difficult. Also, the high front face temperatures at feed switch cause the product species to respond so quickly (<< 1 s), that the mass spectrometer likely misses the peak response. For this reason, errors between species profiles for the 5 s combustion time simulation and experiments were not minimized using least squares criteria but were determined visually. 5.5 Summary The rapid start-up (< 10 s) of an SCTR to produce steady-state syngas from methane has been demonstrated without external preheat by initially using homogeneous combustion. Analysis of the reactor products with a rapid response mass spectrometer allowed the determination of the optimal combustion time that minimized time delay in syngas production. That transient product profiles can be accurately reproduced using an established surface mechanism originally used to model steadystate chemistry can be demonstrated by 1D plug-flow simulations of short combustion time experiments (1 and 5 s combustion times). Combining parameters bounded by theory for the thermal catalyst history with the plug-flow model resulted in the development of a transient, axially-distributed temperature profile that offers insight into how the catalyst temperature develops during lightoff. This work helps further the stateof-the-art in fast start-up reforming technologies for demanding fuel cell-based mobile applications. 130

155 Table 5-1 Temperature profile parameters for 1 s combustion time simulation. Input parameters Oxidation section length 1 mm Catalyst ΔT at steady-state 200 C τ ff,h a 1.1 s T(5 mm, t) experimental value Profile summary T(0-5 mm, 0 s) 40 C T(0-1 mm, ) 915 C T(5 mm, ) 715 C a Oxidation section heating time constant Table 5-2 Temperature profile parameters for 5 s combustion time simulation. Input parameters Oxidation section length 1 mm Catalyst ΔT at steady-state 200 C τ ff, h a τ ff, c b 2.0 s 2.0 s T max (0-1 mm, 5 s) c 1265 C T(5 mm, t) experimental value Profile summary T(0-5 mm, 0 s) 40 C T(0-1 mm, 5 s) 1166 C T(0-1 mm, ) 915 C T(5 mm, ) 715 C a Oxidation section heating time constant b Oxidation section cooling time constant c Theoretical overshoot temperature in oxidation section at 5 s if τ ff, h 0 s 131

156 CH 4, O 2, and Ar A Static Mixer Spark Generator Catalyst Back Heat Shield Thermocouple Products to QMS system Products to Incinerator B P = 10-8 kpa QMS P = 0.07 kpa Pre-chamber prechamber Products to QMS system Turbo Pump Mechanical Pump Figure 5-1 Fast lightoff reactor and QMS system schematics. (A) A spark ignites the combustion feed forming a flame above the catalyst. The resulting premixed flame heats the catalyst for a predetermined time (1 to 20 s), and then the fuel flowrate is increased to partial oxidation stoichiometry. During partial oxidation, carbon monoxide and hydrogen are made exclusively by surface chemistry. (B) Products are measured with a differentially pumped QMS in which the products pass from the reactor outlet at 102 kpa to a pre-chamber at 0.07 kpa to the QMS at 10-8 kpa. 132

157 1 0.8 normalized signal MFC response MFC step input overall response time (s) Figure 5-2 Normalized response of mass flow controller (MFC) and QMS system to step change in methane flowrate. At t = 0 s, a step change from 0 to 5 slpm methane is made (denoted by the solid line) to a steady 5 slpm argon flow giving 10 slpm total flow at steady state. An initial lag (0.3 s) in the MFC response is followed by an ~1.5 s total response time. The total analytical system response (MFC plus QMS) is ~2.5 s after an initial 2.6 s lag time. 133

158 0.14 ignition A 1200 molar flowrate (mol/min) O 2 φ = 1 φ = 4 t comb = 1.5 s H 2 T CO T ( o C) 0.02 CH 4 CO time (s) panel B ignition B φ = 1 t comb = 1.5 s φ = 4 H molar flowrate (mol/min) O 2 CH 4 CO T T ( o C) 0.01 CO time (s) Figure 5-3 Effect of short combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst. Markers represent output from the QMS system with lag time removed while lines are a locally weighted least-squares fit to the data. Ignition of the combustion mixture occurs at t = 0 s. (A) ~1.5 s combustion time. (B) Magnified view for ~1.5 s combustion time over -1 to 5 s. 134

159 0.14 ignition C 1200 molar flowrate (mol/min) φ = 1 t comb = 5 s O 2 φ = 4 H 2 T CO T ( o C) 0.02 CH 4 CO molar flowrate (mol/min) time (s) panel D ignition O 2 CH 4 t comb = 5 s φ = 1 φ = time (s) Figure 5-3 (cont.) Effect of short combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst. Markers represent output from the QMS system with lag time removed while lines are a locally weighted least-squares fit to the data. Ignition of the combustion mixture occurs at t = 0 s. (C) 5 s combustion time. (D) Magnified view for 5 s combustion time over -1 to 10 s. 135 H 2 T D CO CO T ( o C)

160 1 A 1200 X CH4 S CO S H conversion or selectivity T T( o C) time (s) B 1200 S CO S H2 conversion or selectivity X CH4 T T( o C) time (s) Figure 5-4 Effect of combustion time on CH 4 conversion (X CH4 ) and syngas selectivities (S CO and S H2 ) versus time for 5 slpm air flowrate on a 5 mm Rh catalyst. Markers represent output from the QMS system with lag time removed while lines are a locally weighted least-squares fit to the data. Ignition of the combustion mixture occurs at t = 0 s. (A) 1 s combustion time. (B) 5 s combustion time. 136

161 0.14 ignition φ = 1 φ = 4 A 1200 molar flowrate (mol/min) O 2 t comb = 10 s H 2 T CO T ( o C) 0.02 CH 4 CO time (s) 0.14 ignition φ = 1 φ = 4 B t comb = 15 s H molar flowrate (mol/min) O 2 T CO T ( o C) 0.02 CH 4 CO time (s) Figure 5-5 Effect of longer combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst. Markers represent output from the QMS system with lag time removed while lines are a locally weighted least-squares fit to the data. Ignition of the combustion mixture occurs at t = 0 s. (A) 10 s combustion time. (B) 15 s combustion time. 137

162 0.14 ignition φ = 1 φ = 4 C 1200 H t comb = 20 s 1000 molar flowrate (mol/min) O 2 T CO T ( o C) 0.02 CH 4 CO time (s) Figure 5-5 (cont.) Effect of longer combustion times on product flowrates versus time for 5 slpm air flowrate on a 5 mm Rh catalyst. Markers represent output from the QMS system with lag time removed while lines are a locally weighted least-squares fit to the data. Ignition of the combustion mixture occurs at t = 0 s. (C) 20 s combustion time. 138

163 1 A CH 4 equil Conversion x10 5 hr x10 5 hr x10 5 hr T ( o C ) 1 B O 2 equil 0.7 Conversion T ( o C ) Figure 5-6 Steady-state conversions and selectivities versus isothermal reactor temperature for an atmospheric plug-flow simulation with a methane/air feed at φ = 4.0. Results are shown for GHSV relevant to this study and equilibrium. In each row, GHSV values denoted in the left panel also correspond in the same trend to the data in the right panel. (A) Methane conversion. (B) Oxygen conversion (note that equilibrium oxygen conversion is 100% from 400 to 1200 C for φ = 4.0). 139

164 1 C Selectivity H 2 O equil 7.3x10 5 hr x10 5 hr x10 5 hr T ( o C ) 1 D Selectivity CO 2 equil T ( o C ) Figure 5-6 (cont.) Steady-state conversions and selectivities versus isothermal reactor temperature for an atmospheric plug-flow simulation with a methane/air feed at φ = 4.0. Results are shown for GHSV relevant to this study and equilibrium. In each row, GHSV values denoted in the left panel also correspond in the same trend to the data in the right panel. (C) Water selectivity. (D) Carbon dioxide selectivity. 140

165 Selectivity H 2 equil E 7.3x10 5 hr x10 5 hr x10 5 hr T ( o C ) 1 F CO equil 0.7 Selectivity T ( o C ) Figure 5-6 (cont.) Steady-state conversions and selectivities versus isothermal reactor temperature for an atmospheric plug-flow simulation with a methane/air feed at φ = 4.0. Results are shown for GHSV relevant to this study and equilibrium. In each row, GHSV values denoted in the left panel also correspond in the same trend to the data in the right panel. (E) Hydrogen selectivity. (F) Carbon monoxide selectivity. 141

166 1000 A T ff T( o C) T bf Δ T time (s) 1000 B 900 t = 30 s 800 t = 2.2 s t = 10.8 s 700 t = 1.4 s T( o C) t = 0.9 s t = 7.6 s t = 4.2 s t = 0 s distance (cm) Figure 5-7 Temperature profiles for 1 s combustion time simulation. Profile parameters are given in Table 5-1. (A) Assumed catalyst front face temperature (T ff ) and experimental back face temperature (T bf ) versus time. The first 1 mm of the catalyst (oxidation section) experiences first order heating and reaches steady state temperature in approximately 5 s. (B) Combining the front and back face temperature profiles with an exponentially decaying profile for 1-5 mm (reforming region) gives an axially-distributed, transient temperature profile for the 5 mm catalyst. 142

167 0.06 A 0.05 O 2 molar flowrate (mol/min) CH time (s) 0.06 B molar flowrate (mol/min) H 2 O CO time (s) Figure 5-8 Transient species profiles for 1 s combustion time simulation. Experimental species profiles (markers) and numerical profiles (solid lines) for (A) methane and oxygen; (B) carbon dioxide and water (only simulation shown). 143

168 0.12 C molar flowrate (mol/min) H 2 CO time (s) Figure 5-8 (cont.) Transient species profiles for 1 s combustion time simulation. Experimental species profiles (markers) and numerical profiles (solid lines) for (C) hydrogen and carbon monoxide. The model captures the transient profiles after switching the feed stoichiometry at ~1.5 s. 144

169 T( o C) T( o C) A T ff T bf Δ T time (s) 1200 B t = 15.1 s t = 10.0 s t = 8.2 s 400 t = 7.0 s 300 t = 6.2 s 200 t = 5.0 s 100 t = 0 s distance (cm) Figure 5-9 Temperature profiles for 5 s combustion time simulation. Profile parameters are given in Table 5-2. (A) Assumed catalyst front face temperature (T ff ) and experimental back face temperature (T bf ) versus time. The first 1 mm of the catalyst (oxidation section) experiences first-order heating (0 to 5 s) then cooling (5 to 13 s) to its steady-state temperature. Note that the oxidation section temperature for this profile is consistent with the 1 s combustion model through the first 1.5 s. (B) Combining the front and back face temperature profiles with an exponentially decaying profile for 1-5 mm (reforming region) gives an axially-distributed, transient temperature profile for the 5 mm catalyst. 145

170 0.05 A molar flowrate (mol/min) CH time (s) 0.05 B molar flowrate (mol/min) H 2 O CO time (s) Figure 5-10 Transient species profiles for 5 s combustion time simulation. Experimental species profiles (markers) and numerical profiles (solid lines) for (A) methane, and (B) carbon dioxide and water (only simulation shown). 146

171 0.12 C molar flowrate (mol/min) H 2 CO time (s) Figure 5-10 (cont.) Transient species profiles for 5 s combustion time simulation. Experimental species profiles (markers) and numerical profiles (solid lines) for (C) hydrogen and carbon monoxide. 147

172 Chapter 6 Catalytic Autoignition of Higher Alkane Partial Oxidation on Rh-coated Foams 6.1 Introduction Catalytic reforming of heavy hydrocarbon fuels (e.g., gasoline, diesel, or jet fuel) to produce a hydrogen-rich reformate has generated great interest for NO x abatement in diesel engines and electricity production in fuel cells. Transportation fuels are attractive because of their high energy density and widespread distribution infrastructure. While traditional steam reforming (SR) may be suitable for fixed power applications, its slow start-up and endothermic operation are not attractive for efficient mobile applications. CPO at short contact times, another route to syngas production, can be carried out at millisecond contact times on Rh-based catalysts with greater than 90% fuel conversion and over 80% hydrogen selectivity for large alkanes (n-octane, i-octane, n-decane, and n-hexadecane) and diesel fuel [17, 63]. Autothermal reforming (ATR), which combines CPO and SR, has been studied for many of the aliphatic (i-octane, n-decane, n- dodecane, and n-hexadecane) and aromatic (hexene and toluene) components of transportation fuels as well as combinations thereof [68, 69, 71, 150]. Besides improved start-up and transient performance, thermodynamic analyses have shown CPO and ATR can be more efficient than SR alone for reforming applications [64], leading most mobile application developers to concentrate on ATR. Multiple criteria have been deemed crucial for the efficient start-up of mobile fuel processors including start-up time as well as energy usage and carbon formation during start-up [151]. These issues are related to the lightoff temperature of the catalyst and the fuel used. Multiple strategies have been devised to reduce start-up time including heating the catalyst electrically [141] and using homogeneous combustion to heat the catalyst from room temperature to lightoff in seconds, which has been demonstrated for alkanes from methane to i-octane [73, 74, 142]. The latter technique has been shown to effect steady-state production of syngas within 10 s from start-up [110]. While this Chapter 6 is adapted from K.A. Williams and L.D. Schmidt, Catalytic Autoignition of Higher Alkane Partial Oxidation on Rh-Coated Foams, Applied Catalysis A: General, 299, (2006) Elsevier, Inc. All rights reserved. 148

173 technique produces a rich hydrogen stream that is suitable for a solid oxide fuel cell, the reformate s high CO content is not suitable for a polymer electrode membrane (PEM) fuel cell. Research is ongoing to produce a PEM fuel-cell grade hydrogen stream in a mobile reformer within 30 s from start-up [152, 153]. Fuel effects on start-up energy and reactor volume have been investigated for a number of liquid fuels and methane [154]. Use of an upstream cool-flame mode to vaporize and mix liquid fuels with air and water to lower external pre-heat requirements has also been proposed [65, 66, 72]. Microscale reactor experiments with an ATR feedstock (fuel/air/water) and a Pt/Rh catalyst indicate that aliphatic hydrocarbons possess lower lightoff temperatures than aromatic hydrocarbons, and short-chain aliphatics have lower lightoff temperatures than longer-chain aliphatics [151]. Lightoff tests for both alkane and aromatic fuels with and without water co-feed reveal the production of some surface carbon (0.5 3% of carbon fed in the first 30 s of start-up) [151]. To reduce start-up time, it may be favorable to start the fuel processor in CPO mode (exothermic) and then add in water to transition to ATR mode (thermally neutral) [153]. However, in order to partially oxidize transportation fuels (which contain a large percentage of higher alkanes), the fuels must be vaporized and mixed with air before the catalyst while avoiding homogeneous autoignition. For normal alkanes, the minimum homogeneous autoignition temperature (T ai,min ) decreases asymptotically from ~600 o C for methane to ~200 o C for alkanes larger than undecane (C 11 H 24 ), which is lower than the normal boiling point (T b ) for fuels larger than undecane (Figure 6-1). Saturation temperatures for combustion (C/O = variable) and fuel reforming stoichiometries (C/O = 1, 2) indicate that, given enough time (equilibrium time), fuel can be vaporized and mixed with air below T ai,min. However, during start-up the mixing time is constrained to be extremely short. In the absence of exothermic homogeneous chemistry, subsequent surface chemistry requires the catalyst to be initially heated near the heterogeneous lightoff temperature (T L/O,surf ). On Rh, avoiding homogeneous chemistry presents a challenge for alkanes larger than hexane since their T L/O,surf are higher than T ai,min ; this phenomenon implies that the ignition bifurcation diagrams for higher alkane gas-phase and surface chemistry overlap, a situation not found for lower alkanes such as methane and ethane [ ]. To avoid homogeneous ignition before the catalyst, the fuel must be vaporized and mixed with air on a time scale shorter than the gas-phase ignition delay time. 149

174 To better understand the surface processes at work during higher alkane CPO, transient studies of catalyst performance (such as during lightoff) must complement steady-state characterization. Although global mechanisms have been proposed for the steady-state high temperature surface kinetics of higher alkanes and aromatics [70, 71, 150, 158], there exists a significant knowledge gap for the surface mechanism of higher alkane versus small alkane CPO on noble metals, especially during ignition. For example, while Pt and Rh mean-field surface chemistry and the accompanying interactions between surface and gas-phase chemistry are known relatively well for methane [3, 97, 100, 116, ] and ethane [88, 89, 102], no analogous treatment has been proposed for higher alkanes. In this work, the start-up and ignition behaviors of higher alkane fuels (i-octane, n- octane, n-decane, and n-hexadecane) on Rh-coated foams were investigated under CPO conditions using online mass spectrometry to better understand the chemistry governing lightoff. In particular, the minimum surface autoignition temperature as well as the ignition delay time as a function of initial surface preheat temperature were determined for each fuel in a near adiabatic reactor. From this analysis, the kinetic parameters governing lightoff for each liquid fuel were determined for the first time and compared with those of methane. Additionally, the effect of carbon surface coverage on lightoff and the roles of gas-phase and surface chemistry during start-up were studied. 6.2 Materials and Methods Experimental Apparatus High-purity reactant gases (O 2 and Ar) were fed through calibrated mass flow controllers, premixed through 0.64-cm inside diameter (ID) stainless steel tubing, and delivered to the top of the reactor through a quartz endcap with a 0.64-cm ID side port at ~108 kpa. The reactor consisted of a quartz tube approximately 19 mm ID x 21 mm OD x 40 cm long (Figure 6-2). The first 20 cm of the reactor outer wall were wrapped with resistive heating tape (1.27 cm x 122 cm) supplying 210 W at 120 V and 1.75 A. Liquid fuels (> 99% pure) were admitted at the top of the reactor using a calibrated automotive fuel injector that sprayed fuel in a hollow conical pattern on the reactor inner wall, which vaporized on the heated inner wall and mixed partially with air in the first 18 cm of the reactor. Computational fluid dynamics simulations show that temperature and 150

175 concentration gradients develop in the upstream section of the reactor that allow fuel vaporization and partial mixing with air but avoid fuel autoignition (see Chapter 8). Vaporized fuel and air were mixed by flowing through a blank ceramic monolith (static mixer) placed approximately cm downstream. The static mixer also largely equilibrates the radial temperature gradient in the gas-phase at 19 cm (see Chapter 8). Reactants entered the front heat shield at 20 cm downstream and then the catalyst at 21 cm. Products left the catalyst and entered a back heat shield at 21.5 cm. The catalyst, mixer, and heat shields were 80 ppi α-alumina foam monoliths. The catalytic monolith was 5 mm long x 18 mm diameter, whereas the mixer and heat shields were 10 mm long x 18 mm diameter. All four monoliths were wrapped in alumino-silicate paper to make a tight seal with the reactor wall. Catalysts were prepared by coating foams with γ-alumina washcoat (3-5% by weight) to increase effective surface area and then by repeated coating with Rh(NO 3 ) 3 solution to achieve final Rh loadings of about 5% by weight. A detailed account of catalyst preparation has been previously reported [143]. Reactor temperature was measured at the catalyst front (T ff ) and back face (T bf ) with a 0.25-mm-diameter type K thermocouple (< 0.3-s response time constant). Thermocouples were also placed between the heating tape and quartz tube approximately 10 cm downstream to monitor the reactor wall temperature and between the static mixer and front heat shield to measure upstream preheat temperature. The entire external side of the reactor was wrapped with ceramic blanket insulation to attenuate radial heat losses. Effluent gases flowed out of the reactor into heated stainless steel tubing. The outlet flow was split after the bottom reactor endcap so that the majority of the flow was directed to an incinerator, whereas the remainder traveled to a differentially pumped QMS system for rapid analysis (~ 0.5-s response time constant). A detailed description of the QMS system and its calibration as well as diagnostic analysis of the measurement lag and response times can be found in Chapter 5 [110]. Experimental control and data acquisition were computer automated using LabVIEW that allowed quick changes in individual flowrates and QMS scan rate as well as all experimental data to be electronically recorded in a data file for analysis. Simulated air flow rates (79% Ar / 21% O 2, or hereafter air ) were 5 to 10 slpm, and fuel flow rates were adjusted accordingly to create C/O ratios between 0.6 and 1.4. Assuming a fully developed velocity profile in the reactor tube, delay time before injected fuel reached the catalyst was estimated as s for air flowrates between

176 slpm. Given the tortuous 80-ppi catalyst geometry with ~83% porosity [90], steady-state GHSV ranged from 2.8x10 5 to 8.1x10 5 hr Experimental Procedure Before the start of ignition experiments, air was preheated resistively by the reactor wall upstream of the catalyst, which raised the catalyst temperature to C. After pre-heating to the desired temperature, masses 1-50 were scanned at 5 Hz for approximately 10 s to establish baseline signals for O 2 (mass 32) and Ar (mass 40). After 10 s, fuel was injected (t = 0 s), and the start-up behavior of the catalyst was recorded for approximately 60 s by which time chemical and thermal steady-states were attained. Other masses recorded included H 2 (mass 2), CH 4 (15), C 2 H 4 (26), CO (28), and CO 2 (44). Steady-state product compositions were also measured by gas chromatography [17]. In all experiments, catalyst temperatures greater than 1200 o C were avoided to minimize metal sintering and vaporization of Rh metal oxide. Typically, steady-state back face temperatures between o C were attained, and a measurable amount of carbon was formed on the catalyst (< 1 to 5% of catalyst weight depending on alkane size) based on carbon burnoff measurements with the mass spectrometer. To shut down the reactor, O 2 flowrate was first cut in half to lower the back face temperature to 800 C, and then fuel flow was stopped causing a o C exotherm over the s carbon burnoff. Oxygen flowrate was then reset back to its original value, the catalyst was cooled in air to the desired pre-heat temperature, and the next run was performed. Although stopping the fuel flow in the presence of oxygen may render the reactant mixture explosive, no flames or explosions were observed since the time fuel and air spend completely mixed upstream of the catalyst is shorter than the ignition delay time as briefly mentioned in section 2.1 (results to be published). The effect of surface carbon on catalyst start-up performance was also studied by preserving catalyst carbon coverage after a 60-s run. After 60 s of runtime, O 2 and fuel input were stopped, and the catalyst was allowed to cool in Ar only, thus avoiding the carbon burnoff exotherm. Once temperature was reduced to the desired pre-heat value, O 2 flow was re-initiated, and the next run was performed. For each fuel and flowrate studied, 2-3 experiments were run where each consisted of start-ups with one catalyst over a range of pre-heat temperatures (typically between o C). 152

177 6.2.3 Analysis of Ignition Parameters Three measured parameters were used to characterize the two-stage surface ignition of the various fuels on Rh: (1) minimum catalytic autoignition temperature (T cai,min ), (2) lightoff temperature (T L/O ), and (3) ignition delay time (t idt ). Determination of these parameters is demonstrated for methane (Figure 6-3); methane was fed to the reactor with a fast-response mass flow controller, and preheat was provided by a controllable tube furnace. T cai,min was defined as the lowest catalyst preheat temperature that supported exothermic surface chemistry leading to lightoff within 60 s (Figure 6-3A). For an initial 305 o C preheat temperature, exothermic ignition chemistry occurred at the back face once methane flow was activated (t = 0 s in Figure 6-3A); an induction period was observed where back face temperature continued to rise slowly with time until T L/O (~350 o C) was reached at ~14 s. The front face temperature decreased during the induction period indicating ignition occurred at the back of the catalyst. After lightoff, the back face temperature rose rapidly towards steady state marking the transition from a kinetically limited to mass-transfer or flux limited regime. In contrast, preheating to 295 o C did not support ignition: once methane flow was activated (t = 0 s) both the front and back face continuously cooled over the next 20 s (Figure 6-3A inset). Determination of T cai,min is strongly dependent on the experimental set-up and heuristic definition of the allowable transition time. In these experiments, the maximum time allowed for the system to ignite was 60 s; if lightoff was not observed within 60 s, the run was stopped, and a higher temperature was selected for the next run. Ignition delay time, t idt, was defined as the time required for 10% oxygen conversion. To measure t idt for a given preheat temperature, QMS oxygen signal was transformed into oxygen conversion and plotted versus time (Figure 6-3B). For each experiment, the near linear portion of the oxygen conversion data during lightoff was identified, regression was performed, and t idt was calculated. Signal noise (standard deviation in O 2 conversion from -5 s to 0 s) for oxygen conversion was typically 4-6%; uncertainty (error bar) in the t idt calculation was estimated as the signal noise divided by the slope of the least-squares fit line. As an example, for 305 o C initial preheat methane t idt was ~13.7 s ± 0.6 s (Figure 6-3B) in good agreement with the lightoff induction period determined from temperature data (~14 s from Figure 6-3A). 153

178 6.2.4 Determination of Ignition Kinetics Assuming plug flow, first-order reaction with respect to oxygen, quasi-steady state in surface coverages before lightoff, decoupling of flow and temperature from steadystate chemistry [31], and constant surface temperature, the rate constant governing lightoff can be approximated as k ( XO ) ( ) ln 1 ln t t t 2 = =, (6-1) idt idt idt where t idt represents the residence time needed for 10% oxygen conversion. The kinetic parameters governing ignition were calculated from an Arrhenius plot of t idt versus an exponential function of the catalyst preheat temperature using nonlinear regression: t idt E = exp. (6-2) k ko RTi Since t idt was determined from the 10% oxygen conversion criterion, accuracy of the constant surface temperature assumption was up to ~8% in error for a given run depending on the proximity of T i to T L/O (e.g., Figure 6-3) Statistical Analysis Confidence intervals (α = 0.05) for E and k o were determined using nonlinear regression. Systematic deviations between the curve fit and data were tested by residual analysis through the use of a runs test (p < 0.05) [162]. Analysis of covariance [163] was used to test if the E and k o determined from regression were significantly different (p < 0.05) between fuels and flowrates used. At T i greater than 400 o C for the higher alkanes and 525 o C for methane, t idt values systematically deviated from the linear trend for the lower temperature data in a given run indicating that the measured ignition response time may have been significantly limited by the analytical response time of the QMS. As a result, these data were not used in determining kinetic parameters. 154

179 6.3 Results Effect of Preheat Temperature on Start-up i-octane and n-octane Figure 6-4 displays the reactor start-up behavior for an initial 300 o C preheat and 10 slpm air flowrate with i-octane as fuel. Air is fed solely to the reactor until t = 0 s, when the fuel injector is actuated, and fuel is fed to the reactor. From 0 to 3 s, back face temperature rises from 300 to 310 o C and then rises rapidly from 3 to 20 s reaching ~1080 o C at steady state (Figure 6-4A). Once fuel is admitted, O 2 conversion reaches completion within 5 s and is accompanied by CO 2, CO, and H 2 production (Figure 6-4B). CO 2 production reaches a maximum at s then decreases slightly towards its steady-state value (Figure 6-4C). H 2 and CO reach steady-state at ~8 s (Figure 6-4B). i-octane conversion is near 100% at steady-state (data not shown). Initial preheat temperature was also varied to determine minimum temperature needed for surface ignition (Figure 6-5). For 230 o C preheat, no catalyst reactivity is observed, and T bf drops slightly after 20 s (Figure 6-5A). For 240 o C preheat, the back face lights off after approximately 6 s, and as preheat is further increased, lightoff occurs after 1-2 s. T ff lagged T bf during ignition for all preheat temperatures capable of lightoff indicating ignition started at the catalyst back face; however, for preheat above 300 o C this difference was negligible (data not shown). H 2 and CO production tracked T bf quite well and reached steady-state within 15 s for even the lowest preheat temperature capable of ignition (240 o C). For higher preheat, steady-state syngas production could be attained in as little as 6 s after the fuel in turned on (Figure 6-5B and C). For all preheat temperatures studied, the maximum in CO 2 production always occurred before syngas reached steady-state then CO 2 decayed to steady-state. From the i-octane preheat experiments T cai,min was found to be ~242 o C, and T L/O was 250 o C (Table 6-1). Experiments with n-octane showed similar behavior to i-octane as preheat temperature was varied (data not shown). Oxygen consumption after ignition was followed by a peak and then decay in CO 2 production followed by CO and H 2 reaching steady-state in less than 10 s. However, differences in the minimum ignition temperatures between i- and n-octane were observed (Table 6-1). T cai,min and T L/O for n- octane were ~258 and 265 o C, respectively. 155

180 n-decane Figure 6-6 displays the reactor start-up behavior for an initial 295 o C preheat and 10 slpm air flowrate with n-decane. From 0 to 2 s T bf rises from 295 to 315 o C and then rises rapidly from 2 to 10 s reaching ~970 o C at steady state (Figure 6-6A). Species profiles during this time span are similar to the results shown for i-octane; however, ethylene is also a major product of decane at C/O = 1.0. Once the fuel injector is actuated, O 2 flow immediately starts to decrease and is accompanied by production of CO 2, H 2 O, CO, H 2, and C 2 H 4 (Figure 6-6B). CO 2 production peaks at ~2.5 s then decreases to steady-state (Figure 6-6C). Syngas production reaches steady-state within 10 s. C 2 H 4 response is significantly slower than H 2 and CO and levels off after 20 s. n-decane conversion is approximately 85-90% at steady-state. Varying initial preheat temperature gave results qualitatively similar to i-octane (Figure 6-7). For 230 o C preheat, no catalyst reactivity is observed, and T bf drops slightly after 20 s (Figure 6-7A). For 250 o C preheat, the back face lights off after approximately 5 s, and as preheat is further increased, lightoff occurs after 1-2 s. For preheat above 250 o C, H 2 and CO production reaches steady-state within 10 s (Figure 6-7B and C). C 2 H 4 flowrate typically lags syngas production by 2-5 s and reaches steady-state no quicker than 20 s for preheat between o C (Figure 6-7D). Ignition behavior for n- decane was very similar to n-octane and i-octane for all preheat temperatures studied as the maximum in CO 2 production occurred first followed by syngas during lightoff. Furthermore, T ff lagged T bf during lightoff indicating ignition started at the catalyst back face; however, for preheat above 300 o C the difference was negligible (data not shown). From these preheat experiments T cai,min was 241 o C, and T L/O was 252 o C (Table 6-1) n-hexadecane Figure 6-8 displays the reactor start-up behavior for an initial 315 o C preheat and 5 slpm air flowrate with n-hexadecane as fuel. From 0 to 2 s T bf rises from 315 to 340 o C and then rises rapidly from 2 to 10 s reaching over 1000 o C at steady state (Figure 6-8A). Species profiles develop similarly to the other alkanes with major products being CO 2, H 2, and CO at steady-state. Once the fuel injector is actuated, O 2 flow immediately starts to decrease and is accompanied by production of CO 2, H 2 O, CO, and H 2 (Figure 6-8B). CO 2 production peaks at ~3.5 s then decreases to steady-state (Figure 6-8C). 156

181 Syngas production reaches steady-state within 10 s. n-hexadecane conversion is ~90% at steady-state. In contrast to the other alkanes investigated, n-hexadecane was reactive at temperatures below 240 o C preheat for this experimental set-up. Below 240 o C all other alkanes showed a drop in T bf after 20 s from fuel admission, and no lightoff occurred within 60 s; n-hexadecane showed an increase in T bf on the route to lightoff for preheat temperatures as low as 220 o C (Figure 6-9A). For a 227 o C preheat, after fuel admission T bf continually rose during the first 30 s from 227 to 245 o C (Figure 6-9A). O 2 conversion rose from 0 to 5-10% during this same period, and some CO and CO 2 production was observed (data not shown). An energy balance confirmed that 5-10% O 2 conversion to CO and CO 2 agreed well with the observed temperature rise. After 38 s, lightoff occurred (not shown). For 235 o C preheat, T bf rose from 235 to 255 o C from 0 to 20 s, and an accompanying 10% O 2 conversion was attained after 5 s, which stayed constant until lightoff at 22 s (Figure 6-9A). H 2 and CO production tracked T bf quite well and reached steady-state values within 15 s for preheat temperatures above 280 o C (Figure 6-9B and C). Control experiments substituting a blank monolith for the Rh-coated monolith reproduced the same temperature and oxygen conversion behavior over a large range of preheat temperatures ( o C): typically T bf rose o C over the 60 s after fuel admission, and an accompanying 10% O 2 conversion was observed after 3-6 s, which stayed constant over the next 60 s. Therefore, the observed exotherm was kinetically controlled and did not make the transition to a mass-transfer limited condition (indicative of a cool-flame type regime). Control experiments indicated that the initial 10% O 2 conversion observed was not due to Rh surface chemistry but gas-phase chemistry instead (slow oxidation) assisted either directly (surface-assisted initiation) or indirectly (enhanced mixing of fuel and air) by the tortuous blank alumina foam. From these preheat experiments T ign,min was found to be ~220 o C regardless of whether the catalyst was coated with Rh; T L/O was ~ o C for the Rh-coated foam (Table 6-2) Effect of Temperature vs. Fuel Overall, there is little difference in the range of temperatures presented in Figures 6-4 through 6-9 (< 5%). The only appreciable difference is the air flowrate (5 slpm) for hexadecane experiments compared to the other alkane experiments (10 slpm). This 157

182 flowrate modification was made because the steady-state temperature for hexadecane with 10 slpm air goes above 1200 o C. If a quantitative comparison of surface ignition behavior is desired between fuels, Figures 6-10 and 6-11 highlight the differences between fuels and flowrates used (see section 3.2). In contrast, Figures 6-4 through 6-9 show a quantitative comparison between temporal species and temperature profiles as initial preheat temperature is varied for a given fuel Ignition Kinetics Methane, n-octane, i-octane, and n-decane on Rh From the variable preheat experiments, the kinetic parameters governing ignition were determined by regression (Figure 6-10). Results from this analysis are shown in Table 6-3. For a 5 slpm air flowrate (Figure 6-10A), no significant difference was found for the apparent activation energy (slope) between methane, n-octane, and n-decane; however, the preexponential factors (ordinate intercept) were significantly different. For a 10 slpm air flowrate (Figure 6-10B), no significant differences were found for either kinetic parameter between n-octane, i-octane, and n-decane. Since activation energy was statistically invariant to the fuel or flowrate studied, the pooled value was 78 kj/mol n-hexadecane: Gas-phase vs. Surface Ignition Initial gas-phase chemistry, which consumed approximately 10% O 2 with either a Rh or blank foam over a large range of preheat temperatures (section ), had a significant effect on the analysis of the ignition kinetics for n-hexadecane. Further analysis with a 25% O 2 conversion criterion was performed to decouple the possible gas-phase chemistry from the kinetics of surface ignition. For C/O = 1.0, no difference in the ignition kinetics was measured with either the 10% or 25% O 2 conversion criteria. To further suppress gas-phase chemistry, a C/O ratio of ~0.6 was used; typically gas-phase ignition delay times increase as C/O ratios are decreased because of the feed dilution effect. For a 5 slpm air flowrate, 10% O 2 conversion, and C/O = 0.6, there was no significant difference in the activation energy between a Rh foam and blank foam confirming that initial O 2 consumption was not due to Rh surface chemistry (Figure 6-11A and Table 6-4). When using the 25% O 2 conversion criterion, two linear regions 158

183 (high T and low T) were found with significantly different activation energies and preexponentials. Activation energy for the high temperature region was statistically similar to the previous 10% O 2 conversion curves. However, in the low temperature region the activation energy (Table 6-4) was identical (p > 0.05) to E calculated for C 1 - C 10 alkanes (Table 6-3) indicative of Rh chemistry. Subtracting t idt,10% from t idt,25% decoupled the initial gas-phase chemistry from surface chemistry and gave E and k o values (Table 6-4, data set # 7) similar to those reported in Table 6-3 for the other alkanes. Similar behavior was observed for 10 slpm air flow (Figure 6-11B and Table 6-4). The intersection of the two linear regimes for 25% O 2 conversion occurs at the approximate T L/O for n-hexadecane; above T L/O the characteristic time for surface ignition (τ surf ) is much less than the time for 10% O 2 consumption in the gas-phase (τ gas ). Below T L/O, the magnitudes of τ surf and τ gas become similar to causing the two distinct linear regimes Effect of Previous Burnoff on Subsequent Start-up with n-decane For all alkanes investigated the activation energy for surface lightoff on a Rh catalyst that had undergone carbon burnoff subsequent to ignition has been shown statistically invariant. Experiments were also performed to see if carbon coverage had a significant effect on ignition kinetics for C/O = 1.0 and 1.4. Not burning off carbon accumulated on the catalyst surface after a 60 s run had a significant effect on the lightoff kinetics (Figure 6-12 and Table 6-5). However, kinetic parameters were not statistically different between C/O feed ratios of 1.0 and 1.4 for ignition experiments performed on the same type of surface. 6.4 Discussion The present study has shown there is little difference in the qualitative lightoff behavior of short- (methane) and long-chain alkanes (n- or i-octane, n-decane, and n- hexadecane) on Rh-coated monoliths. During the induction period CO 2 and H 2 O (H 2 O not shown) production occurs, which is more apparent at very low preheat temperatures near T cai,min. After CO 2 reaches a maximum, H 2 and CO rapidly rise towards steady-state while CO 2 decreases to steady-state. Ignition starts at the back face of the catalyst for all fuels investigated in agreement with the numerical study of methane lightoff using 159

184 multi-step surface chemistry [100]. Initially heating the catalyst o C above T cai,min allows the production of steady-state production of syngas within 5 s after beginning fuel admission. While these results were obtained on ceramic foam supports, use of a metallic support that possesses a faster thermal time constant may further decrease start-up time. Previous lightoff/ignition experiments [100, 151, 155, 164] and associated numerical models [155, 165, 166] do not address the induction period observed during autothermal ignition studies. These studies give indirect estimates of the kinetics governing surface chemistry during ignition since the data sets contain no transient information on the induction time leading to lightoff as a function of initial preheat temperature. No quantitative analysis of the ignition kinetics for higher alkanes is available in the literature, nor has an experimental transient assessment of the ignition kinetics been performed directly with small alkanes to the best of our knowledge. In this work, the ignition delay time, which is the time span between when fuel and oxygen surface competition begins and lightoff (defined here as 10% oxygen conversion), has been measured as a function of initial surface temperature. The fit of eqn. 2 to lightoff data possessed no systematic deviation from the data (p > 0.05) suggesting that surface coverages do not change appreciably over the ignition delay time (Figures 6-10 through 6-12). This also implies that the ignition delay times were accurately measured by the QMS over a large range of initial surface temperatures. Furthermore, kinetic parameters were extracted from the data with high confidence. From the regression analysis, the apparent E ign on Rh was 77 ± 15 kj/mol, which was statistically invariant between fuels (methane, n- and i-octane, n-decane, and n-hexadecane) and flowrates. A significant difference was found between the ignition k o for methane, O(10 4 1/s), and the other large alkanes, O(10 6 1/s) Controlling Step for Surface Lightoff A mathematical model [166] has been previously proposed to fit lightoff temperature versus feed stoichiometry data from stagnation flow or catalytic wire experiments, and it is helpful to examine the present results in light of this surface reaction model. Assuming the site coverages are in quasi steady-state, the apparent reaction rate prior to lightoff can be expressed as (see [166] for derivation of equations 3-7) 160

185 1 rsurf = rf, A rf, D = ( ro, A ro, D ). (6-3) σ The rates of adsorption (without activation) and desorption (with activation) for species i can be defined respectively as S px N = = (6-4) 2πWRT nia, io, i Av nia, ia, kia, θv θv r E r = k =ΓA RT nid, id, nid, id, id, θi id, exp θi i. (6-5) For an initially oxygen covered surface (θ O 1, θ F << 1, θ V 0), the reaction rate for O 2 consumption prior to lightoff is nf, A surf 2σ kf, AθV r =. (6-6) If n O,A and n F,A are equal, and n O,A is 2, then the apparent reaction rate is r ΓA k E 2σ exp. (6-7) OD, F, A OD, surf = koa, σ kfa, RT Note the units of the above rate are molecules/m 2 /s. To convert this rate to an apparent homogeneous rate (mol/m 3 /s) corresponding to the measured rate constant (1/s) in eqn. 1 the following conversion is made r app S = N v Av r surf O2 O2 Fo, S ΓA RT γ W E 2σ exp W γ W v O, D F O, D = N S Av Oo, σ S Fo, RT k E exp RT OD, = app, o C S Comparing eqn. 6-1 with eqn. 6-8 leads to the analogy that k o and E in eqn. 6-2 are k app,o and E O,D in eqn In order for lightoff to occur, a sufficient number of empty sites must be available [167]. With methane, it has been proposed that before ignition O 2 readsorption is fast (no significant activation barrier for adsorption), and methane adsorption is the controlling ignition step [164]. While methane adsorption may be slightly activated (up to 22 kj/mol [161]), typically O 2 and CH 4 adsorption is treated as non-activated [166]. This 161 F C O2. (6-8)

186 would indicate that the controlling activation energy for methane ignition is the desorption of O 2 to free surface sites. Since E was invariant between fuels in this study, O 2 desorption appears to be the controlling energetic event for alkane lightoff. Previously, the apparent kinetics for O 2 desorption have been measured for a Pt/Rh catalyst over a range of surface coverages with dramatic results [168]. Decreasing the oxygen surface coverage from 1.0 to 0.95 caused the apparent activation energy and preexponential for desorption to increase from 69 to 219 kj/mol and from 10 7 to /s, respectively [168]. The rapid change in surface bonding energy between adsorbed oxygen and metal surface with changing surface coverage has been confirmed by density functional theory calculations for Rh(111) [169]. E measured in this work (77 kj/mol) is in good agreement with the apparent O 2 desorption energy measured when the Pt/Rh surface is saturated with O 2 (69 kj/mol from [168]) indicating that the surface in the present work may be close to O 2 saturation prior to ignition. In addition, E in this work matches extremely well with the value of activation energy for O 2 desorption at saturation coverage (75.2 kj/mol) as part of multi-step methane surface chemistry on Rh [100]; this value was used to accurately simulate methane T L/O experimental data. A similar value of E (79 kj/mol) for O 2 desorption at saturation coverage was used in multi-step H 2 and CH 4 surface chemistry on Pt [161, 170], based on the net repulsive interaction (134 kj/mol) found between adsorbed oxygen on Pt at high θ O [171]. The two order of magnitude difference between k o for methane and the large alkanes may be explained by differences between the sticking probabilities and/or activation energies in the fuel adsorption steps. T L/O is 360 o C for methane and ~260 o C for all other alkanes investigated. In order to give the same rate constant for the higher alkanes as with methane at a 100 o C lower temperature for the same feed stoichiometry (C/O = 1.0, σ = 2, and γ = depending on fuel) with eqn. 8, S F,o for the higher alkanes must be times higher than S CH4,o assuming fuel adsorption is not activated. If fuel adsorption is assumed unequally activated between methane and the higher alkanes (methane activation energy higher), energetic compensation occurs and the calculated difference between the sticking coefficients decreases. However, since no difference was detected in the apparent ignition activation energy between fuels, large differences in the sticking coefficients seem to better explain the differences between k o for methane and the large alkanes. 162

187 6.4.2 Comparison with Previous Experimental Ignition Studies The terms ignition and lightoff are often used interchangeably in the literature. In this work, a surface temperature capable of catalytic autoignition possesses enough energy to either directly lightoff or indirectly produce a sufficient amount of exothermic chemistry to further raise the surface temperature high enough to undergo lightoff. Starting the ignition experiment at a catalyst temperature below T cai,min results in catalyst cooling because of the increased heat load from the added fuel, and no reaction occurs. T L/O is defined as the temperature where a stepwise transition occurs from kinetically limited to mass-transfer limited operation [164, 172]. Previous studies have measured T L/O with C 1 to C 4 alkanes on various noble metals (most commonly with Pt, Rh, or a mixture of both). Measurements with a Pt foil (in stagnation flow geometry) [157] and with a Rh-coated honeycomb monolith [100] found T L/O was 430 o C for methane at C/O 1.0; similar measurements in a micro-scale reactor with Rh gave 480 o C, while in a nearadiabatic reactor with a Rh-coated metallic monolith T L/O was o C [164]. Although some mention is made about the limited exothermic chemistry occurring before lightoff [100, 164], no estimates for t idt were made in any of these studies. Lightoff temperature measurements were also conducted with various large aliphatics and aromatics with water co-feed in an isothermal micro-scale reformer (with heating ramp, not autothermal) and adiabatic (autothermal) reactor with Pt/Rh catalyst on a Yttria Stabilized Zirconia foam [151]. However, it is unclear what the quantitative criterion was for lightoff in this study. Experiments with the micro-scale reformer showed that lightoff temperature varied from 155 to 180 o C for straight chain n-alkanes from C 6 to C 10. i-octane possessed a higher lightoff temperature (210 o C) than any of the n- alkanes. Differences between the apparatus/procedure in some of the above experiments and the setup in the present work may explain the discrepancies between the respective results. In this work the temperatures of the incoming gaseous feed and catalyst surface are in equilibrium before fuel is added similar to the adiabatic reactor in [164]; for methane, excellent agreement is observed for methane T L/O ( o C) between [164] and the present study. Other experimental geometries [100, 157] incorporated a cold (25 o C) feed impinging on a resistively heated surface and report higher temperatures for methane T L/O on Rh (430 o C) possibly because of the thermally non-equilibrated condition. Not surprisingly, whether the catalyst is heated in a premixed fuel/air feed 163

188 [164] or only an air feed until just a few seconds before the ignition experiment (this work) makes no difference in T L/O for a rich methane/air feed. This observation further suggests that the surface is covered with O 2 prior to ignition for small alkane fuels as has been previously discussed [100, 105, 155, 157, 164, 173]. Little quantitative comparison can be made between the higher alkane lightoff results in the present work and [151] because of differences in feed composition and reactor configuration. While both studies agree that higher alkane/air mixtures will catalytically ignite for an initial preheat temperature of 325 o C, the minimum autoignition temperatures on the Pt/Rh catalyst were not characterized in [151] Comparison with Previous Numerical Ignition Studies Numerical analyses have been used to extract ignition kinetics indirectly from the T L/O versus feed stochiometry data in [155]. Using a Langmuir-Hinshelwood mechanism coupled with energy balance and mathematical description of ignition (turning point), a simplified model was used to deduce the apparent preexponential (k o ) and activation energy (E) of ignition for C 1 to C 4 alkanes on Pt [155]. Apparent activation energy (apparent E for the overall reaction plus hydrocarbon heat of adsorption minus oxygen heat of adsorption) varied from 110 kj/mol (methane) to 80 kj/mol (n-propane and i- butane). Inclusion of E for the overall surface reaction in E ign contradicts other numerical studies that show the lightoff temperature to be independent of surface reaction rate parameters (directly shown in [166] and indirectly mentioned in [167, 173]). For rich methane ignition on Pt, studies indicate that the important kinetic parameters governing ignition temperature are those for adsorption of methane and adsorption/desorption of O 2 [165, 166, 173]. A more sophisticated analysis [166] of the data set for methane ignition on Pt [155] used non-activated second-order adsorption for fuel and oxygen and second-order desorption of oxygen; model results indicated that E and k o for O 2 desorption are ~190 kj/mol and 7.5x /s, and the ratio of the zero coverage sticking coefficients for O 2 / CH 4 is approximately 5.9. Although the model fits the data well, use of lightoff temperature data (steady-state) as a function of stoichiometry to determine ignition kinetics still presents a challenge in determining the initial kinetics governing lightoff since typically O 2 desorption is a strong function of surface coverage [171], and coverages are unknown experimentally at the ignition temperature. The difference 164

189 between the ignition activation energy found in the present work (77 kj/mol) and in [166] (190 kj/mol) in light of the results of [168] suggests there may be up to a 5% decrease in O 2 coverage between ignition and lightoff (i.e., between the non-reactive O 2 covered surface and the surface at the time when the turning point criterion occurs) Effect of C/O Ratio and Carbon Burnoff with n-decane For an initially oxygen covered surface prior to lightoff, it is expected that increasing the fuel to oxygen ratio in the feed should generally decrease T L/O up to a certain limit, as shown for alkanes from C 1 to C 4 on Pt [155]. However, experiments with decane showed that changing the C/O ratio from 1 to 1.4 produced no difference in T cai,min, T L/O, or the ignition kinetics in this study (Table 6-5). Since an injector was used to deliver higher alkane fuels, the fuel is vaporized, mixed with air, and then contacts the catalyst. During the initial seconds of the experiment, the C/O ratio the catalyst experiences is a broad range; the catalyst sees ratios from 0 up to the steady-state value. Based on this assessment, the C/O ratio was not a constant quantity during reactor start-up, and therefore similar ignition kinetics were obtained from different steady-state C/O ratios (Table 6-5). Experiments testing the effect of carbon burnoff on the subsequent lightoff run were significant (Figure 6-12 and Table 6-5). A partially carbon covered catalyst lights off faster at lower preheat temperatures and slower at higher preheat temperatures than a burned-off catalyst. This phenomenon may appear at first to be the homogeneous oxidation of the carbon occurring before surface lightoff occurs. Once the carbon is burned off, free sites may become available for O 2 adsorption and then surface ignition can occur. However, the measured activation energy (~35-40 kj/mol) is much lower than those measured for the homogeneous ignition of coke compounds and decoking of reformers ( kj/mol) [ ]. A more plausible explanation is that by not performing burnoff, the next ignition run is forced to be covered by fuel species instead of the typical oxygen coverage found with alkanes. Oxygen is then forced to compete for sites with the adsorbed unsaturated (possible polyolefin) carbon deposits (the opposite of the situation discussed in section 4.1). For example, the activation energies in Table 6-5 for ignition experiments without previous carbon burnoff are in good agreement with those measured for ethylene and propylene on Pt (35-50 kj/mol) [155]; in addition, the ignition preexponential without burnoff drops ~4 orders of magnitude as 165

190 compared to the preexponential for a burned off surface (Table 6-5). This difference is similar to the measured differences in the ignition preexponential between using alkanes (initially oxygen covered surface) and olefins (initially fuel covered surface) as fuel on Pt [155]. 6.5 Conclusions This work has demonstrated that steady-state production of syngas (CO and H 2 ) can be attained within 5 s after admitting large alkanes (i-octane, n-octane, n-decane, or n-hexadecane) and air to a short-contact-time reactor by using an automotive fuel injector and initially preheating the Rh-coated catalyst above each fuel s respective catalytic autoignition temperature. T cai,min with Rh was C for n-octane, i-octane, and n-decane, and ~300 o C for methane. In contrast, ignition of n-hexadecane occurred at lower temperatures (220 o C and greater) because of an indirect two-stage process where exothermic homogeneous reactions preheated the catalyst by o C to temperatures (~280 o C) sufficient for surface lightoff. The catalytic autoignition kinetics for large alkanes were determined experimentally and compared with those of methane using rapid-response mass spectrometry. The controlling step for surface ignition possessed an apparent activation energy of ~77 kj/mol, which was not significantly different between fuels (p > 0.05), and a preexponential on the order of /s for higher alkanes and /s for methane. Catalytic autoignition from T cai,min is a visible two-stage process: starting the reactor between T cai,min and T L/O effects sufficient exothermic surface chemistry to raise the catalyst temperature high enough to undergo lightoff (transition from kinetically limited to mass-transfer limited operation). The differences between T cai,min and T L/O varied from ~60 o C for methane to 10 o C for higher alkanes with the ceramic foam monoliths used in this study. Through this work, some of the similarities and differences in ignition behavior between alkanes of various size have been demonstrated. Transient characterization has helped in understanding the surface processes at work during the ignition of higher alkane CPO. However, transportation fuels are not single components but rather a diverse mixture of aliphatics and aromatics. Further transient experiments are needed to better understand the differences in the lightoff kinetics between aliphatics and aromatics as well as aromatic-aliphatic mixtures. Future work will employ temperature 166

191 programmed oxidation and reduction experiments coupled with transient characterization to identify the reactivity of the catalyst carbon coverage versus temperature and its composition (C or C and H) as a function of start-up time. This information is crucial towards building mean-field descriptions of the surface chemistry governing these compounds that can be used in predictive models of reactor performance. 6.6 Nomenclature A i,d E E ign preexponential factor for desorption of species i, molecules/site/s activation energy, J/mol activation energy for ignition, J/mol k apparent first-order rate constant, 1/s k o preexponential, 1/s k app,o apparent preexponential, 1/s k i,a rate of adsorption of species i at zero coverage, molecules/m 2 /s k i,d rate coefficient for desorption of species i, molecules/m 2 /s n i,a n i,d N Av p P order of adsorption for species i order of desorption for species i Avogadro constant, x10 23 molecules/mol statistical probability used in significance testing pressure, Pa r app apparent rate, mol/m 3 /s r i,a rate of adsorption for species i, molecules/m 2 /s r i,d rate of desorption for species i, molecules/m 2 /s R ideal gas constant, J/mol/K S i,o zero coverage sticking coefficient for species i S v specific surface area, m 2 /m 3 t t idt t idt,10% t idt,25% T T ai,min time, s ignition delay time, s ignition delay time based on 10% O 2 conversion, s ignition delay time based on 25% O 2 conversion, s temperature, K minimum gas-phase autoignition temperature, K 167

192 T b T bf T cai,min T ff T i T ign,min T L/O T sat W i normal boiling point, K back face catalyst temperature, K minimum catalytic autoignition temperature, K front face catalyst temperature, K initial preheat temperature, K minimum ignition temperature, K lightoff temperature, K saturation temperature, K molar mass of species i, kg/mol Greek α statistical significance level γ ratio of oxygen to fuel in gas phase Γ active site density of Rh catalyst, 1.64x10 19 sites/m 2 θ i σ τ gas τ surf site coverage of species i stoichiometric ratio of oxygen to fuel for surface reaction during ignition characteristic time for gas-phase ignition, s characteristic time for surface ignition, s subscripts F O surf V fuel oxygen surface vacant 168

193 Table 6-1 Minimum catalytic autoignition and lightoff temperatures for C 1 -C 10. fuel T cai,min ( o C) a,b T L/O ( o C) a,b methane 305 ± ± 9 i-octane 242 ± ± 5 n-octane 258 ± ± 5 n-decane 241 ± ± 10 Data reported are for a 5% by weight Rh-coated foam with ~3-5% wash coat and C/O =1.0. a Values are mean ± standard deviation. b No significant differences between air flowrates of 5-10 slpm (p > 0.05). Table 6-2 Minimum ignition and lightoff temperature for C 16 H 34. C/O surface T ign,min ( o C) a,b,c T L/O ( o C) a,b 0.63 Rh 223 ± ± 5 1 Rh 225 ± ± blank (control) 219 ± 4 N/A 1 blank (control) 222 ± 5 N/A Data reported are for a 5% by weight Rh-coated foam with ~3-5% wash coat and C/O =1.0. N/A: not applicable. a Values are mean ± standard deviation. b No significant differences between air flowrates of 5-10 slpm (p > 0.05). c T ign,min is used for n-hexadecane instead of T cai,min since a blank foam gave the same minimum ignition temperatures (for 10% O 2 conversion) as a Rh-coated foam. 169

194 Table 6-3 Kinetic parameters for surface ignition of C 1 C 10 alkanes. fuel air flow (slpm) k o,mean (1/s) k o 95% CI a E mean (kj/mol) ± 95% CI a methane 5 3.7x10 4 ** (1.3x10 4, 1.0x10 5 ) 74 ± 6.0 n-octane 5 2.8x10 6 ** (9.2x10 4, 8.6x10 7 ) 85 ± 16 n-octane x10 6 (3.8x10 5, 5.7x10 7 ) 83 ± 12 i-octane x10 5 (2.5x10 4, 3.2x10 7 ) 75 ± 17 n-decane 5 2.3x10 6 ** (1.1x10 5, 4.9x10 7 ) 80. ± 14 n-decane x10 6 (2.5x10 4, 4.2x10 7 ) 75 ± 17 pooled value N/A 78 Feed stoichiometry was C/O = 1.0, and ignition criterion was 10% oxygen conversion. ** Indicates value is significantly different from other 5 slpm air flow values in column (p < 0.05). a 95% CI (confidence interval). 170

195 Table 6-4 Kinetic parameters for ignition of C 16 H 34. E data set # C/O air flow (slpm) X O2 crit. / comment k o,mean (1/s) k o 95% CI a mean (kj/mol) ± 95% CI a % 4.4x10 0 (1.4x10 0, 1.4x10 1 ) 21 ± % 3.9x10 0 (1.6x10 0, 9.3x10 0 ) 22 ± % 1.8x10 0 (2.5x10 0, 1.3x10 1 ) 16 ± % 1.1x10 1 (3.4x10 0, 3.7x10 1 ) 26 ± % / high T b 4.6x10 1 (7.9x10 0, 2.7x10 2 ) 35 ± % / low T c 4.1x10 5 (3.7x10 4, 4.6x10 6 ) 77 ± 11 ** % - 10% / all T 3.9x10 6 (4.7x10 5, 3.2x10 7 ) 85 ± 9.6 ** % / high T b 2.6x10 0 (5.2x10-1, 1.3x10 1 ) 20 ± % / low T c 2.3x10 6 (1.3x10 5, 4.1x10 7 ) 83 ± 13 ** % / blank d 2.5x10 0 (6.6x10-2, 9.2x10 1 ) 20. ± % / blank d 3.4x10 0 (4.9x10-1, 2.4x10 1 ) 21 ± 8.8 a 95% CI (confidence interval). b High T indicates T i > 290 o C t idt values were used in the regression analysis. c Low T indicates T i < 290 o C t idt values were used in the regression analysis. d Blank indicates the data corresponds to a blank foam instead of a Rh-coated foam. ** E for data set # 6, 7 and 9 are significantly different (p < 0.05) from all other experiments (1-5, 8, 10, and 11) but are not significantly different from each other. 171

196 Table 6-5 Effect of surface burnoff on kinetic parameters for n-decane ignition. C/O with burnoff? k o,mean (1/s) a k o 95% CI a E mean (kj/mol) ± 95% CI a 1 Yes 3.6x10 6 (1.3x10 5, 9.9x10 7 ) 84 ± Yes 1.8x10 6 (1.3x10 5, 2.5x10 7 ) 81 ± 12 pooled value 2.4x No 4.2x10 2 (1.5x10 1, 1.2x10 4 ) 42 ± No 2.0x10 2 (2.9x10 1, 1.4x10 3 ) 39 ± 9.0 pooled value 2.7x Feed flowrate was 5 slpm, and ignition criterion was 10% oxygen conversion. Kinetic ignition parameters for burned off surfaces were significantly different (p < 0.05) than those for surfaces that did not receive burnoff. No significant difference was found between C/O = 1.0 and 1.4 for same surface condition prior to start-up. a 95% CI (confidence interval). 172

197 T ai,min (gas-phase) temperature ( o C) T (C/O=1) L/O,Rh T b T (C/O=1) L/O,Pt 0 T (C/O=2) sat -100 T (C/O=1) sat T (C/O=comb.) sat number of carbon atoms Figure 6-1 Temperatures pertinent to successful vaporization and mixing of C 1 to C 16 alkanes with air and alkane CPO ignition. Normal boiling points [177] for alkanes monotonically rise from -162 o C (methane) to 281 o C (hexadecane). However, for alkanes that are liquids at 25 o C the saturation temperatures (T sat ) needed to maintain C/O ratios between combustion and C/O = 2.0 in air are much less than the corresponding boiling points. Minimum autoignition temperatures for gas-phase chemistry asymptotically decrease from near 600 o C (methane) to ~200 o C (hexadecane) [178, 179]. Surface ignition temperatures for alkanes on Pt (C/O = 1.0) range from ~430 o C (methane) to 175 o C (butane) [155, 157]; on Rh these temperatures asymptotically decrease from 430 o C (methane) to 250 o C (hexane) [63, 73, 100, 157]. 173

198 Fuel Injector Air Heating tape g Static mixer Catalyst Thermocouple Heat shields Insulation Products Figure 6-2 Schematic of fuel-injected short-contact-time reactor. Permanent gases and liquid fuels were delivered to the top of the reactor and entered the catalyst, while products left the catalyst and entered a back heat shield at 21.5 cm. Effluent gases flowed out of the reactor into heated stainless steel tubing for either analysis with the QMS system or incineration. Temperature was measured with thermocouples at the catalyst front (T ff ) and back face (T bf ), and the external side of the reactor was wrapped with ceramic blanket insulation to attenuate radial heat losses. 174

199 temperature ( o C) ~T cai,min temperature ( o C) T bf lower preheat T L/O T ff time (s) time (s) (dt/dt) bf T bf T ff A (dt/dt) ff dt/dt ( o C/s) Figure 6-3 Demonstration of measured ignition parameters with methane (C/O = 1.0, 310 o C preheat, and 5 slpm air flowrate). (A) Before methane addition (t < 0 s), the back face temperature (T bf ) is ~308 o C and the front face temperature (T ff ) is 300 o C. Upon introducing methane (t > 0 s), T bf rises from 310 to 350 o C during the first 14 s whereas T ff decreases from 300 to 285 o C in the same time frame. Upon reaching T L/O (~350 o C), T bf rapidly rises to steady-state and makes the transition from kinetically limited to mass-transfer limited as noted by (dt/dt) bf (the derivative of T bf with respect to time calculated using a central difference approximation). (A inset) No ignition occurs below 300 o C. 175

200 O 2 conversion CH 4 = fuel X X O2 = = t O2 R 2 R= = 0.95 C/O = 1.0 t idt time (s) Figure 6-3(cont.) Demonstration of measured ignition parameters with methane (C/O = 1.0, 310 o C preheat, and 5 slpm air flowrate). (B) O 2 conversion reaches 10% conversion 13.7 s (t idt ) after CH 4 is introduced to the catalyst in good agreement with the time for lightoff (~14 s) determined from the temperature data in panel A. B 1200 A backface temperature ( o C) FI on i-c 8 H 18 = fuel time (s) Figure 6-4 Effect of 300 o C preheat on catalyst temperature and effluent species profiles during start-up for i-octane/air feed (C/O=1) at 10 slpm air flowrate. (A) T bf. 176

201 0.2 B H 2 molar flowrate (mol/min) O 2 CO CO time (s) panel C 0.15 C H 2 molar flowrate (mol/min) O 2 CO CO time (s) Figure 6-4 (cont.) Effect of 300 o C preheat on catalyst temperature and effluent species profiles during start-up for i-octane/air feed (C/O=1) at 10 slpm air flowrate. (B) O 2, CO 2, CO and H 2 molar flowrates. (C) Magnified view of panel B from 0 to 5 s. 177

202 1200 A T bf backface temperature ( o C) FI on preheat T 230 o C 240 o C 260 o C 300 o C 330 o C i-c 8 H 18 = fuel time (s) Figure 6-5 Effect of varying preheat temperature on start-up time for i-octane fuel (same feed stoichiometry and flowrate as Figure 6-4). (A) Preheating the catalyst to 230 o C is not sufficient to produce ignition chemistry within 20 s after fuel is fed to the reactor, and T bf decreases slightly in the same time span. Above 240 o C the catalyst lights off on a time scale exponentially dependent on preheat temperature. 178

203 0.2 B H 2 molar flowrate (mol/min) preheat T 240 o C 260 o C 300 o C 330 o C time (s) 0.2 CO C molar flowrate (mol/min) preheat T 240 o C 260 o C 300 o C 330 o C time (s) Figure 6-5 (cont.) Effect of varying preheat temperature on start-up time for i-octane fuel (same feed stoichiometry and flowrate as Figure 6-4). (B and C) H 2 and CO flowrates track T bf and reach steady-state within 10 s for preheat temperatures above 260 o C. 179

204 1200 A backface temperature ( o C) FI on n-c 10 H 22 = fuel time (s) Figure 6-6 Effect of 295 o C preheat on catalyst temperature and effluent species profiles during start-up for n-decane/air feed (C/O=1) at 10 slpm air flowrate. (A) T bf. 180

205 0.1 B O 2 molar flowrate (mol/min) CO C 2 H 4 H 2 CO time (s) panel C 0.06 C 0.05 O 2 molar flowrate (mol/min) CO H CO 2 C 2 H time (s) Figure 6-6 (cont.) Effect of 295 o C preheat on catalyst temperature and effluent species profiles during start-up for n-decane/air feed (C/O=1) at 10 slpm air flowrate. (B) O 2, C 2 H 4, CO 2, CO and H 2 molar flowrates; (C) Magnified view of panel B from 0 to 5 s. 181

206 1200 T bf n-c 10 H 22 = fuel A backface temperature ( o C) FI on preheat T 250 o C 280 o C 295 o C 320 o C 230 o C time (s) 0.1 H 2 preheat T B molar flowrate (mol/min) o C 280 o C 295 o C 320 o C time (s) Figure 6-7 Effect of varying preheat temperature on start-up time for n-decane fuel (same feed stoichiometry and flowrate as Figure 6-6). (A) Preheating the catalyst to 230 o C is not sufficient to produce ignition chemistry within 20 s after fuel is fed to the reactor, and T bf decreases slightly in the same time span. Above 240 o C the catalyst lights off on a time scale exponentially dependent on preheat temperature. (B ) H 2 flowrate tracks T bf and reach steady-state within 10 s for preheat temperatures above 260 o C. (D) C 2 H 4 flowrate typically lags syngas production by 2-5 s and reaches steady-state no quicker than 20 s for preheat between o C. 182

207 0.1 CO C molar flowrate (mol/min) preheat T 250 o C 280 o C 295 o C 320 o C time (s) 0.05 C 2 H 4 D molar flowrate (mol/min) preheat T 250 o C 280 o C 295 o C 320 o C time (s) Figure 6-7 (cont.) Effect of varying preheat temperature on start-up time for n- decane fuel (same feed stoichiometry and flowrate as Figure 6-6). (C) CO flowrate tracks T bf and reach steady-state within 10 s for preheat temperatures above 260 o C. (D) C 2 H 4 flowrate typically lags syngas production by 2-5 s and reaches steady-state no quicker than 20 s for preheat between o C. 183

208 1200 A backface temperature ( o C) FI on n-c 16 H 34 = fuel time (s) Figure 6-8 Effect of 315 o C preheat on catalyst temperature and effluent species profiles during start-up for n-hexadecane/air feed (C/O=1) at 5 slpm air flowrate. (A) T bf. 184

209 O 2 H 2 B molar flowrate (mol/min) CO 2 CO time (s) panel C 0.05 O 2 C molar flowrate (mol/min) H 2 CO CO time (s) Figure 6-8 (cont.) Effect of 315 o C preheat on catalyst temperature and effluent species profiles during start-up for n-hexadecane/air feed (C/O=1) at 5 slpm air flowrate. (B) O 2, CO 2, CO, and H 2 molar flowrates; (C) Magnified view of panel B from 0 to 6 s. 185

210 1200 A T bf n-c 16 H 34 = fuel backface temperature ( o C) FI on preheat T 280 o C 315 o C 360 o C 390 o C 235 o C 227 o C time (s) Figure 6-9 Effect of varying preheat temperature on start-up time for n-hexadecane fuel (same feed stoichiometry and flowrate as Figure 6-8). (A) Preheating the catalyst to 227 o C is sufficient to produce ignition chemistry within 20 s after fuel is fed to the reactor, and T bf increases in the same time span by 20 o C. Lightoff eventually occurs at ~38 s (not shown). For 235 o C preheat, T bf rises from 235 to 255 o C from 0 to 20 s, and an accompanying 10% O 2 conversion is attained after 5 s (not shown), which stays constant until lightoff at 22 s. 186

211 0.06 B molar flowrate (mol/min) H 2 preheat T 280 o C 315 o C 360 o C 390 o C time (s) 0.06 C molar flowrate (mol/min) CO preheat T 280 o C 315 o C 360 o C 390 o C time (s) Figure 6-9 (cont.) Effect of varying preheat temperature on start-up time for n- hexadecane fuel (same feed stoichiometry and flowrate as Figure 6-8). (B and C) H 2 and CO flowrates track T bf and reach steady-state within 15 s for preheat temperatures above 280 o C. 187

212 ~586 o C ~479 o C ~395 o C ~328 o C ~274 o C ~228 o C 100 A n-c 8 H CH 4 R 2 = 0.98 R 2 = 0.91 t idt (s) n-c 10 H 22 R 2 = /RT i (mol/kj) Figure 6-10 Surface ignition kinetic parameters for C 1 -C 10 alkanes from Arrhehius plots. Feed stoichiometry was C/O = 1.0, and ignition criterion was 10% O 2 conversion. Ignition delay times are plotted as an exponential function of 1/T with high confidence using nonlinear regression; curve deviations from the data are insignificant (p > 0.05). See Table 6-3 for results of statistical analysis. (A) Least-squares analysis for 5 slpm air flowrate on CH 4 (E = 74 kj/mol, k o = 3.7x10 4 1/s), n-c 8 H 18 (E = 85 kj/mol, k o = 2.8x10 6 1/s), and n-c 10 H 22 (E = 80. kj/mol, k o = 2.3x10 6 1/s). 188

213 ~395 o C ~328 o C ~274 o C ~228 o C 100 B n-c 10 H 22 x 5 R 2 = i-c 8 H 18 x 2.5 t idt (s) R 2 = n-c H 8 18 R 2 = /RT (mol/kj) i Figure 6-10 (cont.) Surface ignition kinetic parameters for C 1 -C 10 alkanes from Arrhehius plots. Feed stoichiometry was C/O = 1.0, and ignition criterion was 10% O 2 conversion. Ignition delay times are plotted as an exponential function of 1/T with high confidence using nonlinear regression; curve deviations from the data are insignificant (p > 0.05). See Table 6-3 for results of statistical analysis. (B) Least-squares analysis for 10 slpm air flowrate on n-c 8 H 18 (E = 83 kj/mol, k o = 4.6x10 6 1/s), i-c 8 H 18 (E = 75 kj/mol, k o = 8.9x10 5 1/s), and n-c 10 H 22 (E = 75 kj/mol, k o = 1.0x10 6 1/s). Kinetic parameters were statistically invariant (p > 0.05) between fuels at 10 slpm air flowrate so t idt values for i-octane and n-decane (and best-fit regression line) are scaled by a constant for ease of viewing. 189

214 ~395 o C ~286 o C ~208 o C 100 A t idt,25% (high T) t idt,25% (low T) R 2 = 0.97 R 2 = t idt (s) 1 t idt,10% R 2 = 0.85 t (blank) t t idt,10% idt,25% idt,10% R 2 = 0.74 R 2 = /RT (mol/kj) i Figure 6-11 Gas-phase vs. surface chemistry during ignition for n-hexadecane. C/O = 0.63, and ignition criterion was 10% or 25% O 2 conversion. See Table 6-4 for results of statistical analysis. (A) t idt values for 10% conversion with a blank foam are denoted by white squares. For a 5 slpm air flowrate and 10% O 2 conversion, there is no significant difference in the activation energy between a Rh foam and blank foam indicating initial O 2 consumption is due to exothermic gas-phase chemistry. When using the 25% O 2 conversion criterion, two linear regions (high T and low T) are found with significantly different activation energies (p < 0.05). E for the high T region is statistically similar to the previous 10% O 2 conversion curves. However, in the low T region the activation energy is identical (77 kj/mol, p < 0.05) to E calculated for C 1 -C 10 alkanes indicating surface chemistry. Subtracting t idt,10% from t idt,25% decouples the initial gasphase chemistry from surface chemistry and gives E and k o values (85 kj/mol and 3.9x10 6 1/s, respectively) similar to those reported in Table 6-3 for C 1 -C

215 ~395 o C T L/O ~ 286 o C ~208 o C 100 B Rh-coated foam 25% X O2 Rh-coated foam 25% X O2 R 2 = R 2 = 0.78 t idt (s) 1 blank foam 10% X O2 R 2 = 0.70 higher T i lower T i τ >> τ gas surf τ = or > τ surf gas /RT (mol/kj) i Figure 6-11 (cont.) Gas-phase vs. surface chemistry during ignition for n-hexadecane. C/O = 0.63, and ignition criterion was 10% or 25% O 2 conversion. See Table 6-4 for results of statistical analysis. (B) t idt values for 10% conversion with a blank foam are denoted by white circles. The same phenomenon is observed at 10 slpm air flow. The intersection of the two linear regimes for 25% O 2 conversion occurs at the approximate T L/O for n-hexadecane; above T L/O the characteristic time for surface ignition (τ surf ) is much less than the time for 10% O 2 consumption in the gas-phase. Below T L/O the opposite case is observed. 191

216 ~395 o C ~328 o C ~274 o C ~228 o C 10 n-c 10 H 22 (no burnoff) R 2 = 0.88 t idt (s) 1 n-c H (with burnoff) R 2 = /RT (mol/kj) i Figure 6-12 Effect of previous surface burnoff on subsequent ignition kinetics with n- decane. C/O = 1.0, air flowrate was 5 slpm, and ignition criterion was 10% O 2 conversion. Leastsquares analysis on previously burned off surface gives E = 84 kj/mol and k o = 3.6x10 6 1/s. Least-squares analysis for no burnoff condition gives E = 42 kj/mol and k o = 4.2x10 2 1/s. These results indicate that carbon burnoff has a significant effect (p < 0.05) on the kinetic parameters for n-decane. See Table 6-5 for results of statistical analysis. 192

217 Chapter 7 Transient Analysis of Integral Carbon Formation for CPO on Rh Foams: Methane and Decane 7.1 Introduction Reforming of natural gas and logistic fuels has been explored as a means to produce hydrogen for on-board fuel cell applications. Use of steam reforming, CPO, and autothermal reforming has received much attention as discussed in the previous chapter. These reactions are accompanied by the formation of undesired carbon compounds (coke) that can deactivate the catalyst [180, 181]. The structure, morphology, and reactivity of the coke depends heavily on the reforming reaction and the reaction conditions [ ]. For example, autothermal reforming could produce a much different form of coke than CPO. The three major types of carbon that may form/occur in hydrocarbon reforming are polymeric, filamentous, and graphitic [185]. Whereas polymeric coke is generally produced from homogeneous hydrocarbon decomposition, filamentous and graphitic cokes are typically formed on the catalyst and require the presence of metal sites. Temperature programmed oxidation (TPO) has been used to quantify the amount of carbon present on a catalyst [ ]. Improvements have been made by deriving kinetic schemes that simulate TPO data by making assumptions about coke morphology, particle size, and reaction order [189]. Additionally, multi-step surface mechanisms have been used to model the carbon formation and catalyst deactivation using the mean field approximation [190, 191]. Use of a pulse TPO setup to differentiate between coke oxidized by the catalyst and coke oxidized by the support has been demonstrated [183]. Reaction schemes and kinetic parameters for carbon oxidized from cracking catalysts have also been validated from CO 2 and CO evolution rates in TPO experiments [192, 193]. Most TPO studies for fuel reforming have considered Ni and Pt. Methane CPO on Ni-MgO [194], propane steam reforming on Co-Ni [195], and methane dry reforming on Ni [190, 196] have provided new insight into the carbon burnoff kinetics for Ni based systems. Likewise, carbon on Pt with liquid fuels [197] and with the addition of sulfur and aromatics [198] has been explored. Studies have also been devoted to studying carbon formation directly in solid oxide fuel cells with methane [199] and liquids [200]. 193

218 The effect of coke on start-up with liquid fuels in a microreactor with a Pt/Rh catalyst showed that ~0.5 3% of the fuel carbon formed on the catalyst within the first 30 s [151]. The effect of Rh on catalyst formation has not been studied as extensively. Previous work has shown Rh forms little coke compared to Ni [201]. In this chapter, the amount of carbon formed on a Rh CPO catalyst as a function of time on stream was determined by rapid response online mass spectrometry. Current state-of-the-art methods to model surface kinetics on Rh foams employs a mean-field approximation that assumes monolayer coverages. In this chapter, one goal is to determine whether monolayer carbon or multilayer carbon is present to determine the validity of the meanfield assumption. 7.2 Materials and Methods Experimental Apparatus The apparatus used in this chapter is similar to that used in Chapter 6 except for the addition of a capillary sampling system, which improved the system response time (Figure 7-1). A thin quartz capillary (~0.6 mm) ran along the downstream section of the reactor. The open end was placed behind the Rh catalyst via a back heat shield with a small hole (~0.6 mm) drilled through its axial length. More information on the capillary sampling system and its interface with the mass spectrometer can be found elsewhere [15, 16]. The orientation of the sampling capillary relative to the catalyst is shown in Figure 7-2A. In this analysis, it is assumed that a sample taken during the carbon titration from the center of the catalyst is representative of the coverage at all radial locations (i.e., there are no radial gradients in carbon coverage). The response time constant of the system is improved to ~50-80 ms compared to ~0.5 s response time for the system in Chapters 5 and 6 (see Figure 7-2B) Experimental Procedure Before the start of each experimental run, air was preheated resistively by the reactor wall upstream of the catalyst, which raised the catalyst temperature to C. After pre-heating to the desired temperature, masses 1-50 were scanned at 5 Hz for approximately 10 s to establish baseline signals for O 2 (mass 32) and Ar (mass 194

219 40). After 10 s, fuel was injected (t = 0 s), and the start-up behavior of the catalyst was monitored for approximately s by which time chemical and thermal steady-states were attained. Other masses recorded included H 2 (mass 2), CH 4 (15), CO (28), and CO 2 (44). Steady-state product compositions were also measured by gas chromatography. In all experiments, catalyst temperatures greater than 1200 o C were avoided to minimize metal sintering and vaporization of Rh metal oxide. Typically, steady-state back face temperatures between o C were attained, and a measurable amount of carbon was formed on the catalyst. To prepare for a titration, O 2 and fuel flow were stopped, and the reactor was purged with Ar for a few seconds. O 2 flow was then reinitiated, and the carbon oxidized in the form of CO 2 and CO was measured. The catalyst was cooled in air to the desired pre-heat temperature, and the next run was performed. Run times studied ranged from 10 s to 12 min for methane and 1-3 minutes for n- decane. For each fuel and flowrate studied, 2-3 experiments were run where each consisted of 2-4 titrations with one catalyst. Catalysts were prepared by coating foams with γ-alumina washcoat (3-5% by weight) to increase effective surface area and then by repeated coating with Rh(NO 3 ) 3 solution to achieve final Rh loadings of about 5% by weight. A detailed account of catalyst preparation has been previously reported [143] Titration Analysis To eliminate signal fluctuations caused by changes in the total molar flowrate or by instrument instabilities, the peak areas of the species of interest (H 2 at 2 amu, CH 4 at 15 amu, CO at 28 amu, O 2 at 32 amu, Ar at 40 amu, CO 2 at 44 amu) were calculated and divided by the area of the Ar peak. Molar flowrates were obtained by multiplication of the Ar normalized peak areas with the ratio of the Ar flows used for measurement and calibration, respectively, and division by the sensitivity obtained from calibration as in Chapter 2. The only cross sensitivity considered was the contribution of CO 2 to 28 amu. H 2 O could not be measured directly as adsorption in the stainless steel lines and the MS chamber could not be fully suppressed even though all parts were heated resistively. H 2 O was calculated by closing the oxygen atom balance with it. Using this method of quantification the carbon balances closed to ± 5% while the hydrogen balance closed to ± 10%. 195

220 Carbon on the catalyst was quantified by recording the CO and CO 2 signals from the mass spectrometer during a titration. A major challenge with this technique was accounting for the CO and CO 2 non-zero baselines that depended heavily on the O 2 partial pressure in the vacuum chamber. Because of the presence of higher alkanes and other fuels typically used with this apparatus, it was difficult to keep the chamber entirely clean. Admitting O 2 to the mass spectrometer caused the creation of CO and CO 2 and slight consumption of O 2 inside the vacuum chamber. These readings were most likely the result of contamination on the ionization rods and walls of the vacuum chamber being oxidized by the incoming O 2. For example, step input and output commands are given to the mass flow controller for O 2 in the presence of Ar (Figure 7-3A). The mass flow controller and mass spectrometer output follow with approximate first order responses. Note that the output from the mass spectrometer is slightly lower than the output from the mass flow controller in the first few seconds. The corresponding flows from the mass spec for H 2 O, CO 2 and CO are shown in Figure 7-3B. While the H 2 O signal is not completely reliable because of H 2 O adsorption in the vacuum chamber, Figure 7-3B indicates that the contamination in the mass spec could contain H atoms as well as C atoms. Also, the CO 2 and CO flows mirror well the O 2 partial pressure in the vacuum chamber giving a non-constant baseline. Plotting least-square fit lines through the CO 2 and CO flows shows that the species typically respond in a first order fashion to the admittance of O 2 (Figure 7-4A and B). In order to get accurate estimates of the catalyst carbon amounts, these baselines had to be adjusted to zero. To do this, first order response curves were created to account for the response of CO and CO 2 to the changing partial pressure of O 2. These response curves could then be subtracted from the measured molar flows of CO and CO 2 to give standard zero baselines for analysis. An example response curve for CO 2 is given in Figure 7-5. Carbon amounts on the catalyst were quantified by first recording the molar flowrates of CO and CO 2 evolved during a titration (Figure 7-6A). Then the flows were adjusted using the baseline curves so that the CO 2 and CO flows started at 0 before O 2 was readmitted for the burnoff (Figure 7-6B). Once the flowrates were adjusted, the flows were plotted vs. time, and the trapezoid rule was used to estimate the total moles 196

221 of CO and CO 2 evolved during the titration (Figure 7-6C). The moles of CO and CO 2 were then converted to mg of carbon using their respective molecular weights. To complete the titration analysis and enable decisions about the statistical significance of the results to be made, limits of detection (LOD) and quantification (LOD) were determined using the 3σ and 9σ criteria, respectively [202]. Blank baselines for CO and CO 2 were modified with the adjustment procedure outlined in the previous paragraph and then integrated over a typical time span for a titration experiment (60 s). This was done for 10 samples, and the standard deviation of the molar flowrates for CO and CO 2 were calculated and multiplied by 3 and 9 to get the LOD and LOQ (Table 7-1). From this analysis, the minimum amount of carbon that could be reliably measured with 99% confidence quantitatively was 0.57 and 0.13 mg of CO and CO 2, respectively. In addition, one-way analysis of variance was used (α = 0.05 level) to test the hypothesis that the carbon amount from the shortest run time was significantly different from the carbon amount at the longest run time Adsorption Model The transient amounts of C from CO 2 and CO were fit to a simple first-order kinetic expression for adsorption [203] of the form () = 1 exp ( / ) mt meq tτ, (7-1) where m(t) is the mass of C at time t, m eq is the equilibrium amount of C, and τ is the time constant for m eq. Fit optimization was performed by nonlinear least squares regression using the Microsoft Excel solver. The objective function to minimize was the RMS error between the experimental data points and the model equation. 7.3 Results The next sections detail the carbon titration results for a 10 mm, 5 wt% Rh catalyst and a 5 mm, 20 wt% Rh catalyst Methane on 10 mm 5 wt % Rh Catalyst Figure 7-6 shows the 10 mm long, 5 wt% Rh catalyst before use (fresh, Figure 7-7A) and after use (~ 5 hours, Figure 7-7B). Initially, the fresh oxidized catalyst appears 197

222 black. After a few hours of use, the catalyst appears a much duller gray after being chemically reduced under reaction conditions where metal dispersion also decreases C/O = 1.0 After a 12 minute run time, a noticeable amount of carbon can be oxidized from the catalyst mainly in the form of CO 2. H 2 O also is evolved, and a small peak is noticeable before the H 2 O signal decays back to baseline (Figure 7-8). The effect of run time (20 s to 12 minutes) on the carbon titrations is shown in Figure 7-9, where the results of one set of titrations are displayed. Run time increases the extent of the exotherm observed at the catalyst back face until ~ 3 minutes where subsequent time on stream appears to have a diminishing effect (Figure 7-9A). CO 2 comes off in a unimodal distribution (Figure 7-9B). Once O 2 is reintroduced to the hot catalyst, CO 2 immediately peaks and then decays to zero. CO displays a similar behavior, which is hard to discern from the experimental noise because of its low evolution rate (Figure 7-9C). A summary of the run time experiments (n = 4 for each run time) discerns the apparent trends from the carbon titrations (Figure 7-10). Carbon evolved from CO 2 and CO increases initially and then asymptotes as run time increases. Notice that carbon from CO 2 is always above the limit of quantification (Figure 7-10A), whereas carbon from CO is below the limit of quantification until 1.5 minutes of run time (Figure 7-10B). Firstorder adsorption models fit the data well with randomly distributed residuals, giving equilibrium amounts of C from CO 2 and CO of 2.5 and 0.9 mg carbon, respectively (Tables 7-2 and 7-3). Combining the two carbon curves gives the total carbon curve where a total carbon amount after 12 minutes is 3.3 mg based on the adsorption model. Carbon from CO 2 at 20 s run time was significantly lower than carbon from CO 2 at 12 minutes (p = ). This analysis was not done with CO since the mean carbon amount from CO at 20 s (0.37 mg) was below the limit of quantification C/O = 2.0 The effect of run time (20 s to 12 minutes) on the carbon titrations is shown in Figure 7-11 where the results of one set of titrations are displayed. Run time increases the extent of the exotherm observed at the catalyst back face over all the run times (Figure 7-11A). CO 2 comes off in a unimodal distribution (Figure 7-11B). Once O 2 is reintroduced to the hot catalyst, CO 2 immediately peaks and then decays to zero. CO 198

223 displays a similar behavior, which is hard to discern from the experimental noise because of its low evolution (data not shown). A summary of the run time experiments (n = 4 for each run time) discerns the apparent trends from the carbon titrations (Figure 7-12). Carbon evolved from CO 2 and CO increases initially and then asymptotes as run time increases. Notice that carbon from CO 2 is always above the limit of quantification (Figure 7-12A), whereas carbon from CO is below the limit of quantification until 1.5 minutes of run time (Figure 7-12B). Firstorder adsorption models fit the data well with randomly distributed residuals, giving equilibrium amounts of carbon from CO 2 and CO of 5.6 and 1.3 mg carbon, respectively (Tables 7-2 and 7-3). Combining the two carbon curves gives the total carbon curve where a total carbon amount after 6 minutes is 6.9 mg based on the adsorption model. This amount is approximately twice the amount of carbon on the 5 wt% Rh catalyst after 12 minutes of run time at C/O = 1.0. Carbon from CO 2 at 20 s run time was significantly lower than the carbon from CO 2 at 12 minutes (p = ). This analysis was not done with CO since the mean carbon amount from CO at 20 s (0.37 mg) was below the limit of quantification Methane on 5 mm 20 wt % Rh Catalyst Figure 7-13 shows the 5 mm long, 20 wt% Rh catalyst before use (fresh, Figure 7-13A) and after use (~ 5 hours, Figure 7-13B). Initially, the fresh oxidized catalyst appears black. The higher loading of this catalyst compared to the 5 wt% Rh catalyst can be discerned from the high agglomeration of metal observed for the 20 wt% loading. After a few hours of use, the catalyst appears a much duller gray after being reduced under reaction conditions where metal dispersion also decreases C/O = 1.0 After a 6 minute run time, a noticeable amount of carbon can be oxidized from the catalyst mainly in the form of CO 2. Also, H 2 O is evolved, and a small peak is noticeable before the H 2 O signal decays back to baseline (data not shown). The effect of run time (15 s to 4 min.) on the carbon titrations is shown in Figure 7-14, where the results of one set of titrations are displayed. Run time increases the extent of the exotherm observed at the catalyst back face until ~ 3 minutes where subsequent time on stream appears to have a diminishing effect (Figure 7-14A). In contrast with the 5 wt% Rh catalyst, the 199

224 shape of the CO 2 curves change as run time is increased. Within the first minute of run time, the CO 2 curve possesses a unimodal shape similar to the 5 wt% Rh catalyst experiments. After 1 minute of run time, CO 2 comes off in a bimodal distribution (Figure 7-14B). By 4 minutes of run time a clear pattern emerges where CO 2 immediately peaks, decreases slightly, and then peaks again before decaying to zero. CO displays a unimodal behavior similar to the 5 wt% Rh loading for all run times (data not shown). A summary of the run time experiments (n = 2 for each run time) discerns the apparent trends from the carbon titrations (Figure 7-15). Carbon evolved from CO 2 and CO increases initially and then asymptotes as run time increases. The CO 2 data fits a first-order adsorption model well with randomly distributed residuals, but significant error arises for the CO curve. Carbon from CO 2 is always above the limit of quantification (Figure 7-15A), whereas carbon from CO is below the limit of quantification until 2 minutes of run time (Figure 7-15B). Equilibrium amounts of carbon from CO 2 and CO predicted by the model are 7.1 and 1.0 mg carbon, respectively (Tables 7-2 and 7-3). Combining the two carbon curves gives the total carbon curve, where a total carbon amount after 6 minutes is 8.0 mg from the adsorption model. Carbon from CO 2 at 15 s run time was significantly lower than carbon from CO 2 at 4 minutes (p = 0.017). This analysis was not done with CO since the mean carbon amount from CO at 20 s (0.31 mg) was below the limit of quantification C/O = 2.0 After 6 minutes of run time, a noticeable amount of carbon can be oxidized from the catalyst mainly in the form of CO 2, but H 2 O is also evolved, and a small peak is noticeable before the H 2 O signal decays back to baseline (data not shown). The effect of run time (15 s to 6 minutes) on the carbon titrations is shown in Figure 7-16 where the results of one set of titrations are displayed. Run time increases the extent of the exotherm observed at the catalyst back face as run time is increased (Figure 7-16A). Similar to the results in the previous subsection for C/O = 1.0, the shape of the CO 2 curves changes as run time is increased, and this phenomenon can be seen even more clearly (Figure 7-16B). Within the first minute of run time, the CO 2 curve possesses a unimodal shape similar to the 5 wt% Rh catalyst experiments. After 1 minute of run time, CO 2 comes off in a bimodal distribution (Figure 7-16B). At 6 minutes of run time a clear pattern emerges where CO 2 immediately peaks, decreases slightly, and then 200

225 peaks again before decaying to zero. CO displays a unimodal behavior similar to the 5 wt% Rh loading for all run times (Figure 7-16C). A summary of the run time experiments (n = 2 for each run time) discerns the apparent trends from the carbon titrations (Figure 7-17). Carbon evolved from CO 2 and CO increases initially and then asymptotes as run time increases. Both the CO 2 and CO data fit a first-order adsorption model well with randomly distributed residuals as carbon from CO 2 and CO are always above the limit of quantification (Figures 7-17 A and B). Equilibrium amounts of carbon from CO 2 and CO predicted by the models are 40.0 and 2.7 mg carbon, respectively (Tables 7-2 and 7-3). Combining the two carbon curves gives the total carbon curve where a total carbon amount after 4 minutes is 42 mg from the adsorption model, which is ~ 5.3 times the total carbon amount for the 20 wt% Rh catalyst at C/O = 1.0 after 4 minutes of run time and ~6 times the total carbon amount for the 5 wt% Rh catalyst at C/O = 2.0 after 12 minutes of run time. Carbon from CO 2 at 15 s run time was significantly lower than carbon from CO 2 at 4 minutes (p = ). In contrast, carbon from CO at 15 s run time was not significantly lower than carbon from CO at 4 minutes (p = 0.054) n-decane on 5 mm 20 wt % Rh Catalyst, C/O = 2.0 n-decane titrations were performed with the 20 wt% Rh catalyst at C/O = 2.0 to compare the relative amounts of carbon made by methane and n-decane. The result of a titration for n-decane after 3 minutes of run time is shown in Figure In the methane titration experiments, oxygen is never completely consumed after being reintroduced for the burnoff. The rate of CO 2 and CO evolution is not enough to remove all of the incoming O 2. This is not the case for the n-decane titration. In the first few seconds of the titration when CO 2 reaches a maximum, O 2 is completely consumed. The CO 2 curve appears similar to the methane results for the 20 wt% Rh catalyst at C/O = 2.0. It is bimodal, and the two local maxima are almost equal in magnitude. The CO curve also exhibited a bimodal distribution with a quick peak followed by another peak that took much longer to reach its maximum. Integration of the CO 2 curve yielded 136 mg carbon from CO

226 7.3.4 Catalyst Overheating and Metal Migration A major concern when removing carbon from the catalyst autothermally is overheating that can lead to pronounced sintering, lowered dispersion, and migration of the catalyst metal. In the titration experiments outlined so far, the reactor was purged with Ar after the desired run time was reached to remove any remaining hydrocarbon from the reactor before turning the O 2 back on and burning carbon off. This technique prevents the alkane from oxidizing along with the catalyst carbon. However, another benefit is that catalyst temperature is also lowered before the titration preventing the catalyst from reaching temperatures over 1200 o C. Figure 7-19 displays the adiabatic isotherm for a titration with n-decane at C/O = 0.6 after 3 minutes where fuel is turned off in the presence of O 2 rather than purging with Ar first. This titration was with a fresh catalyst (~5 min total time on-stream). The back face temperature reaches ~1330 o C (approximately the maximum for the K-type thermocouple used). Afterwards, the catalyst stack was examined to see the effect of the high temperature excursion (Figure 7-20A). Upon inspection, the front face of the catalyst had a large spot that was almost white (Figure 7-20B), and the back heat shield, which should be white and free from carbon after a titration, had a large black spot on its front face (Figure 7-20B). This is a clear demonstration of the metal migration where Rh that has been overheated moves down the catalyst and into the back heat shield. Closer inspection of the catalyst shows that the back face of the catalyst was also discolored (Figure 7-20C). It appears that metal lost from the back of the catalyst gets replaced by metal lost from the front of the catalyst. The migration of the carbon lost by the back of the catalyst is seen upon inspection of the back heat shield (Figure 7-20D). The front face of the back heat shield contains a large area of the metal but examination of the back face of the back heat shows that some of the Rh reached the back face. In just one titration, much of the metal from the catalyst migrated to the back heat shield. 7.4 Discussion Examination of the total carbon amounts as a function of run time for two different loadings (5 and 20%) and two C/O ratios (1 and 2) shows some marked trends (Figure 7-21). For 5% loading, doubling the C/O ratio approximately doubles the amount of carbon on the catalyst for 6 minutes of run time. For 20% loading, doubling the C/O ratio increases the amount of carbon on the catalyst by a factor of 5 for 4 minutes of run time. 202

227 The effect of loading on carbon formation is clearly nonlinear considering that the 20% Rh catalyst is only half the mass of the 5% Rh catalyst yet forms 8 times more carbon for C/O = 2.0 after 4 minutes of run time. Table 7-5 gives the weight percent carbon for each of the experimental conditions investigated Monolayer or Multilayer Carbon? A key goal of these experiments was to determine if carbon deposited on the foam catalysts forms a mono- or multilayer. Multi-step surface chemistry only considers monolayer coverages, and it is important to determine if the monolayer assumption holds for methane. BET surface area has been measured as a function of washcoat loading for 30 ppi foams [129]. For foams, BET surface area is typically dominated by the microstructure that possesses cracks instead of the geometric surface area. For example, the calculated geometric surface area is ~200 times smaller than the measured BET surface area. A blank 30 ppi (no washcoat) had a BET surface area of ~2 m 2 /g, while adding 5% washcoat increased the area to ~ 3 m 2 /g [129]. The active catalytic surface area is much lower with the addition of ~ 5% Rh metal, which forms a film [138] with the washcoat once heated to ~1000 o C (typical reaction conditions in this work). Therefore, the geometric surface area more likely represents the true catalytic area instead of the BET surface area. For an 80 ppi foam, the geometric surface area is ~ cm -1 ( m 2 /g) [90]. To estimate the number of carbon monolayers on the catalysts in this work, it is helpful to bound the catalytic surface area by the geometric area (low estimate) and BET area (high estimate). Assuming a Rh site density of 2.72x10-5 mol/m 2 and 1 atom of carbon bound to 1 atom of Rh, the mg carbon per monolayer for a 10 mm long catalyst is based on the geometric surface area and 1.57 based on the BET surface area (2 m 2 /g). For a 5 mm long catalyst, the mg carbon per monolayer is based on the geometric surface area and based on the BET surface area (2 m 2 /g). For the 10 mm, 5 wt% Rh catalyst, the maximum carbon measured was ~ 6 mg, which translates to monolayers of carbon based on the upper bounds for the catalytic surface area. For the 5 mm, 20 wt% Rh catalyst, the maximum carbon measured was ~ 40 mg, which translates to monolayers of carbon based on the upper bounds for the catalytic surface area. Therefore, methane multilayer carbon formation would be 203

228 expected if the catalytic surface area approaches the geometric area of the foam. For decane, the carbon could approach tens of thousands of layers Chemical Nature of the Carbon TPO is a common technique for probing the chemical nature and amount of a carbon deposit. As catalyst temperature is increased in the presence of O 2, one or more peaks in CO 2 or CO evolution may be observed, indicating that the carbon is comprised of one or more types that are reactive at different temperatures. Carbon titrations are different in this work since the catalyst is allowed to adiabatically oxidize in the presence of O 2 starting at temperatures near CPO ( o C). While this technique is useful to quantify the reactive carbon amount (non-graphitic), discerning different types of carbon from an adiabatic burnoff may be more difficult since all carbon may burn off at one time at high temperatures. In this case, the carbon may be burnt off at temperatures higher than those needed to cause the carbon to become reactive, and different forms of carbon, which usually react at very different temperatures, may react at the same time for the high starting temperatures used in this work. However, even in these adiabatic oxidation experiments, structure is found in the CO 2 evolution curves. Plotting back face temperature vs. CO 2 evolution rate gives an adiabatic oxidation curve and shows the relative reactivity of the carbon versus temperature. This type of analysis assumes that carbon oxidized throughout the catalyst is at the back face temperature during the titration. Two types of curves are found from the experiments in this work (Figure 7-22). For a 5% Rh catalyst (curve 1), carbon is oxidized all at once, and one peak is observed. For a 20% Rh catalyst (curves 2 and 3), two peaks are observed suggesting that at least two types of carbon are found at this higher loading. Previous literature suggests that the first peak (more reactive) could be carbon on the metal while the second peak (less reactive) could be carbon on the support (alumina) [183, 197]. For curve 2, oxygen is never completely consumed during the titration ruling out the possibility that two peaks are caused by initial oxygen starvation during the evolution of the first peak. 204

229 Table 7-1 Limits of detection and quantification for C from CO and CO 2. species sd (mg) LOD (mg) LOQ (mg) CO CO Table 7-2 Summary of adsorption model for C from CO 2. loading (wt % Rh) C/O max run time (min) m eq C from CO 2 (mg) τ (s) RMS error (%) Table 7-3 Summary of adsorption model for C from CO. loading (wt % Rh) C/O max run time (min) m eq C from CO (mg) τ (s) RMS error (%) Table 7-4 Summary of adsorption model for total C. loading (wt % Rh) C/O max run time (min) m eq total C (mg) τ (s) RMS error (%) Table 7-5 Total carbon as a percentage of catalyst weight. loading (wt % Rh) C/O max run time (min) wt % total carbon % % % % 205

230 Fuel Injector Air Heating tape g Static mixer Catalyst Thermocouple Heat shields Insulation Sampling Capillary Products to QMS Figure 7-1 Schematic of experimental apparatus with capillary sampler. 206

231 A Flow direction integrated signal (m/z = 4) time (s) B Figure 7-2 Orientation and response of sampling capillary. (A) Photograph of capillary sampler positioning relative to foam monoliths. (B) Response time example for capillary with He (60 ms in this figure). 207

232 mol flowrate (mol/min) A O2 QMS output O2 MFC output O2 MFC setpt time (s) mol flowrate (mol/min) B CO CO2 H time (s) Figure 7-3 Effect of O 2 on baseline response. (A) Response of mass flow controller and mass spectrometer to step input of O 2. (B) Response of H 2 O, CO 2 and CO baselines to step input of O

233 0.003 CO (mol/min) molar flowrate (mol/min) A time (s) CO2 (mol/min) molar flowrate (mol/min) B time (s) Figure 7-4 Variation of CO and CO 2 internal flowrates with O 2 partial pressure. Least-squares fit lines to the response baselines of (A) CO 2 and (B) CO from Figure 7-3B. Both CO 2 and CO baselines respond in the same first-order fashion as the O 2 mass flow controller. 209

234 CO2 (mol/min) CO2 baseline model (time dependent) O2 MFC setpt O2 MFC out O2 response model time (s) MFC signal (slpm) Figure 7-5 Example of CO 2 response adjustment model used to correct fluctuations in CO 2 baseline from changing partial pressure of O 2 in mass spectrometer. 210

235 molar flow rate (mol/min) molar flow rate (mol/min) CO CO time (s) CO B CO time (min) A C moles x10-3 mol COCO2 2 CO 2.73x10-4 mol CO time (s) Figure 7-6 Demonstration of baseline correction and titration analysis. (A) Uncorrected data from the mass spectrometer. (B) Corrected data giving a zero baseline before the titration. (C) Integration of the data from panel B yielding the moles of CO 2 and CO produced during the burnoff. 211

236 A B Figure 7-7 Photographs of 10 mm long, 5 wt% Rh catalyst with 5 wt% washcoat. (A) fresh. (B) Run for ~ 5 hours. 212

237 mol flowrate (mol/min) H20 CO2 CO Tbf temperature ( o C) time (s) Figure 7-8 Temperature, H 2 O, CO 2, and CO flows during a burnoff at 12 min for C/O = 1.0: Catalyst was 10 mm long, 5 wt% Rh foam. 213

238 run time A T ( o C) s run time 45 s run time 1.5 min run time 3 min run time 6 min run time 12 min run time CO2 (mol/min) time (s) B 20 s run time 45 s run time 1.5 min run time 3 min run time 6 min run time 12 min run time time (s) C CO (mol/min) s run time 45 s run time 1.5 min run time 3 min run time 6 min run time 12 min run time time (s) Figure 7-9 Effect of run time on titration at C/O = 1.0 for a 5 wt % Rh foam catalyst. (A) Back face temperature. (B) CO 2. (C) CO. 214

239 mg C from CO2 mg C from CO mg C total A time (s) B time (s) C time (s) Figure 7-10 Effect of time on stream on carbon amounts for 5 wt % Rh foam, C/O = 1.0. (A) Carbon from CO 2. (B) Carbon from CO. (C) Total carbon. 215

240 run time A T ( o C) s run time burn off 45 s run time burn off 1.5 min run time 3 min run time 6 min run time 12 min run time time (s) B CO2 (mol/min) time (s) Figure 7-11 Effect of run time on titration at C/O = 2.0 for a 5 wt % Rh foam catalyst. (A) Back face temperature. (B) CO 2. (C) CO. 216

241 mg C from CO2 mg C from CO A time (s) B time (s) C mg C total time (s) Figure 7-12 Effect of time on stream of carbon amounts for 5 wt % Rh foam, C/O = 2.0. (A) Carbon from CO 2. (B) Carbon from CO. (C) Total carbon. 217

242 A B Figure 7-13 Photographs of 5 mm long, 20 wt% Rh catalyst with 5 wt% washcoat. (A) fresh. (B) Run for ~ 5 hours. 218

243 T (oc) run time 15 s run time 30 s run time 60 s run time 120 s run time 240 s run time time (s) A CO2 (mol/min) run time B time (s) Figure 7-14 Effect of run time on titration at C/O = 1.0 for a 20 wt % Rh foam catalyst. (A) Back face temperature. (B) CO

244 mg C from CO2 mg C from CO mg C total A time (s) B time (s) C time (s) Figure 7-15 Effect of time on stream on carbon amounts for 20 wt % Rh foam, C/O = 1.0. (A) Carbon from CO 2. (B) Carbon from CO. (C) Total carbon. 220

245 CO2 (mol/min) T ( o C) run time 15 s run time 30 s run time 60 s run time 120 s run time 240 s run time time (s) run time time (s) A B Figure 7-16 Effect of run time on titration at C/O = 2.0 for a 20 wt % Rh foam catalyst. (A) Back face temperature. (B) CO

246 mg C from CO2 mg C from CO mg C total A time (s) B time (s) C time (s) Figure 7-17 Effect of time on stream on carbon amounts for 20 wt % Rh foam, C/O = 2.0. (A) Carbon from CO 2. (B) Carbon from CO. (C) Total carbon. 222

247 molar flow rate (mol/min) O2 CO CO2 H time (s) Figure 7-18 H 2 O, CO 2, and CO flows during a burnoff at 3 min for C/O = 1.0 with n- decane as fuel. Catalyst was 5 mm long, 20 wt% Rh foam. T ( o C) back face T front face T time (s) Figure 7-19 Overheating the catalyst during a carbon titration at C/O =

248 A B Flow direction Flow direction C D Catalyst front face Back heat shield front face Catalyst back face Back heat shield back face Figure 7-20 Photographs of front heat shield, catalyst, and back heat shield after burnoff with temperature history shown in Figure

249 mg C total % Rh, C/O = 2 20 % Rh, C/O = 1 5 % Rh, C/O = 2 5 % Rh, C/O = time (s) Figure 7-21 Summary of total carbon amount versus run time parameterized by C/O ratio and catalyst loading for methane as fuel CO2 (mol/min) n-decane C/O = % Rh 1 3 methane C/O = % Rh methane C/O = 2.0 5% Rh back face T ( o C) Figure 7-22 Adiabatic oxidation diagram demonstrating the reactivity and amount of carbon as a function of back face temperature time

250 Chapter 8 Numerical Analysis of Upstream Mixing and Autoignition Criteria in a Fuel-injected CPO Reactor 8.1 Introduction CPO of higher alkanes (n-decane, n-hexadecane, and diesel fuel) with air has been successfully performed in a SCTR over a wide range of fuel concentrations with the aid of an automotive fuel injector. While other methods of fuel delivery and fuel/air mixing are quite common, the SCTR typically employed with an automotive fuel injector enjoys advantages: it is simple, starts up quickly [110, 204], and takes advantage of equipment already available in an automotive or mobile application. Syngas (H 2 and CO) selectivities greater than 80% and fuel and oxygen conversions of ~100% have been achieved at catalyst contact times of 5-15 ms [17]. Although the process has been shown operable under carefully chosen conditions, very little information is available on the range of allowable operating conditions. Indeed, CPO of higher alkanes offers a distinct process challenge: fuel vaporization without autoignition that could cause flames or an explosion depending on the system pressure. In order to partially oxidize transportation fuels (which contain a large percentage of higher alkanes), fuels must be vaporized and mixed with air before the catalyst while avoiding homogeneous autoignition. For normal alkanes, the minimum homogeneous autoignition temperature (T ai,min ) decreases asymptotically from ~600 o C for methane to ~200 o C for alkanes larger than undecane (C 11 H 24 ), which is lower than the normal boiling point (T b ) for fuels larger than undecane (Figure 8-1). Typically, autoignition temperature for any flammable mixture is a function of fuel concentration, system volume and pressure, and flow conditions [80]. Saturation temperatures for combustion (C/O = variable) and fuel reforming stoichiometries (C/O = 1, 2) indicate that, given enough time (equilibrium time), fuel can be vaporized and mixed with air below T ai,min. However, in the SCTR the mixing time is constrained to be extremely short. To avoid homogeneous ignition before the catalyst, flammable fuel/air mixtures must reach the catalyst on a time scale shorter than the gas-phase ignition delay time. While using air as the oxidant source is sufficient for the production of syngas, the use of pure oxygen may be necessary for the efficient production of valuable olefins. However, use of pure oxygen would considerably increase the risk of flames and 226

251 explosions. Therefore, a model that can explain why the current process works and can predict operating limits for various operating conditions would be extremely useful to reduce the number of experiments necessary to fully characterize the system. In this work, preliminary studies towards the development of a predictive model of the upstream mixing and autoignition phenomena in the reactor are presented. To characterize the transport and chemistry involved in the steady-state mixing of fuel and air and identify regions capable of autoignition, 2D axisymmetric and 3D computational fluid dynamics (CFD) models are used. 8.2 Computational Methods Reactor Geometry In the experimental apparatus, reactant gases (O 2 and Ar or N 2 ) are premixed and delivered to the top of the reactor through a quartz endcap with a 0.4 cm ID side port at ~108 kpa. The reactor consists of a quartz tube approximately 19 mm ID x 21 mm outside diameter (OD) x 40 cm long (Figure 8-2). The first 20 cm of the reactor outer wall are wrapped with resistive heating tape (1.27 cm x 122 cm) supplying 210 average W at 120 VRMS and 1.75 A ACRMS current. Liquid fuels (> 99% pure) are admitted at the top of the reactor using a calibrated automotive fuel injector that sprays fuel in a hollow conical pattern on the reactor inner wall. Fuel vaporizes on the heated inner wall and mixes partially with air in the first 20 cm of the reactor. Vaporized fuel and air are further mixed by flowing through a blank ceramic monolith (static mixer) placed approximately 20 cm downstream. A detailed account of the reactor setup can be found in previous work [204]. In the following sections, a description of the models used to simulate the mixing of fuel and air in the preheat section upstream of the catalyst is given D and 3D Reactor Models Typical flowrates in the reactor range from 2-10 standard liters per minute (slpm). Given the tortuous 80-ppi catalyst geometry with ~81-83% porosity [90], steady-state GHSV ranged from 1.6x10 5 to 8.1x10 5 hr -1. Based on the 19 mm inside diameter (ID) of the quartz tube, laminar flow conditions should prevail at all flowrates typically used (Table 8-1). However, based on the 4 mm ID side port, the transition to turbulent flow 227

252 should theoretically occur as flowrate is increased from 5 to 10 slpm (Table 8-2). Based on this information, both 2D and 3D models of the reactor were used to validate and elucidate the fluid dynamics and transport in the short contact time reactor. For the 2D model, air enters the solution domain at 25 C and undergoes developing laminar flow under the inlet condition of plug flow (Figure 8-3). The grid is comprised of the fluid zone (ID = 19 mm) and the reactor wall (quartz tube, outer diameter = 22 mm). The acceleration of gravity acts in the flow direction. Because of the cylindrical reactor geometry, the conditions inside the reactor should be approximately symmetric in the azimuthal direction. Therefore, the domain is modeled as 2D axisymmetric. Up to 21 cm, the thermal boundary condition at the outer wall is assigned to be a constant heat flux. Beyond 21 cm, an adiabatic boundary condition is applied. The ceramic foam section (static mixer, 19 mm diameter by 10 mm long) was modeled using a homogeneous porous media approximation, which assumes local thermal equilibrium between gas and surface [128]. Porosity was taken as 0.81 [90], and the solid component of the foam was simulated as polycrystalline alumina. Momentum sink terms were employed using the hydraulic diameter approach to account for the effect of the 80 ppi foam on the flow field by inputting isotropic inertial and viscous resistance terms for 80 ppi ceramic foams [129]. Thermal conductivity in the porous zones was treated as an average between gas and surface based on porosity. The effective thermal conductivities of the quartz (fused silica) and alumina were supplied as polynomial functions of temperature [125]. In order to capture the effects of turbulence as the flow transitions from the side port to the reactor at high flowrates, a 3D model incorporating the geometry of the entire experimental apparatus was used. This grid incorporated the 4 mm side port and 19 mm reactor tube and the transition between them with the correct eccentricity Vaporization Model The fuel is delivered at the top of the reactor using an automotive fuel injector. The conical dispersion of µm size fuel droplets creates a thin film of liquid fuel on the heated inner wall of the reactor, which is heated to between 100 and 400 C, depending on the fuel boiling temperature (Figure 8-2). A steady-state fuel film length is reached when the fuel vaporization rate is equal to the fuel deposition rate on the reactor wall. Depending 228

253 on the wall temperature, this fuel can either immediately flash or more slowly evaporate by convective effects. In this analysis, all of the fuel is assumed to boil at the wall and is introduced into the solution domain via a volumetric mass source term with the assumption that the external resistance to heat transfer dominates. The coupling of the fluid/solid energy balance is accomplished using volumetric energy source/sink terms for the gas and solid phases, respectively (Figure 8-4). n-decane was used as the model fuel. The vaporization length was calculated by vap where Pneeded P = P needed input, (8-1) is estimated as T boil P = n Δ H (298 K) + c ( gas) dt. (8-2) needed fuel vap p, fuel 298 The first and second terms on the right hand side of eqn. 8-2 represent the latent and sensible heats, respectively. For example, the energy necessary to bring 1 mol of liquid n-decane from 25 o C to o C and then vaporize it is 93.2 kj at 1 atm. For a 5 slpm total air flowrate and feed ratio of C/O = 1.0, the power necessary to vaporize the fuel is 13.3 kw. From this computation, vaporization length is implicitly dependent on wall heat flux at the outer wall. As in the experiment, this value becomes the only adjustable parameter for a given feed flowrate and stoichiometry. Once the vaporization length is calculated from the power input, computational cells over this length immediately adjacent to the inner reactor wall on both the gas and solid sides are assigned volumetric energy source/sink terms to simulate the heat transport to the gas phase and cooling in the wall (Figure 8-4). The value of P needed is divided uniformly over the vaporization length as an energy sink. Since the reference state for the decane in the CFD model is 298 K in the gaseous form, only the sensible energy component of eqn. 8-2 is applied to the gas-phase energy source on the gasside. Computational cells adjacent to the wall are assigned a volumetric mass source that emits n-decane at syngas stoichiometry (C/O = 1.0) and at a constant rate over the vaporization length. According to the following CPO reaction ( ) n C H + 5 O N 10CO+ 11H N (8-3)

254 if all of the decane emitted at the wall were to mix with the air entering the solution domain, fuel concentration would be 4.03 mol % n-decane (syngas stoichiometry). For the fuel and air flowrates studied in this work, the effect of the boiling fuel's radial momentum on the solution is negligible and was not included in the model Solution Method and Constitutive Equations In order to calculate steady-state velocity, concentration, and temperature distributions upstream of the catalyst, the commercial CFD code Fluent (version 6.1, Lebanon, NH) was used to solve the governing integral equations for momentum, mass, and heat transfer using a finite-volume based technique. This software has been previously used to model SCTR performance under various operating conditions [88, 97, 205]. The finite-volume method [128] consists of: (1) division of the solution domain into discrete control volumes using a computational grid, which is accomplished with Gambit software (Fluent, Lebanon, NH), (2) integration of the governing equations on the individual control volumes to construct algebraic equations for the discrete dependent variables ("unknowns") such as velocities, pressure, temperature, and conserved scalars, (3) linearization of the discretized equations and solution of the resultant linear equation system to yield updated values of the dependent variables. A segregated, implicit numerical scheme was employed to linearize the non-linear governing equations and solve for the dependent variables. Specifically, a point-implicit (Gauss-Seidel) linear equation solver in conjunction with an algebraic multigrid (AMG) method is used to solve the resultant scalar system of equations for the dependent variables in each cell. For laminar flow in an inertial (non-accelerating) reference frame, the general form of the mass continuity equation [128] is ρ + i ( ρv) = Sm (8-4) t and is valid for incompressible and compressible flow. The source S m is mass that is added to the solution domain through any user defined source. For 2D axisymmetric geometries, eqn. 8-4 can be written as ρ ρvr + ( ρvx ) + ( ρvr) + = Sm. (8-5) t x r r 230

255 Conservation of momentum in an inertial reference frame is described by [206] t ρv + ρvv = p+ τ + ρg+ F, (8-6) ( ) i( ) i( ) where p is the static pressure, τ is the stress tensor, and ρ g and F are the gravitational body force and external body forces, respectively. The stress tensor is 2 τ μ T = + ( ) v v i vi, (8-7) 3 where μ is the molecular viscosity and I is the unit tensor. For 2D axisymmetric geometries, the axial and radial momentum conservation equations are 1 1 ( ρvx) + ( rρvxvr) + ( rρvrvx) t r x r r p 1 vx 2 1 vx vr = + rμ 2 ( iv) + μ r F x r x x r r r x x (8-8) and 1 1 ( ρvr) + ( rρvxvr) + ( rρvrvr) t r x r r p 1 vr vx 1 vr 2 = + rμ + + rμ 2 ( iv) 3, (8-9) r r x x r r r r vr 2 μ 2μ + 2 ( iv) + Fr r 3 r where the divergence of the velocity vector is vx vr vr iv = + + x r r, (8-10) and swirl velocity is neglected. The energy conservation equations for a compressible gas may be written as [128] ( ρ ) + ( ρ + ) = E i v E p i k T hj i i + ( τiv) + Sh, (8-11) t i where k is the thermal conductivity, and J i the diffusion flux of species i. Pressure and kinetic energy work terms on the left hand side are typically negligible for an incompressible gas. The first three terms on the right-hand side of eqn are heat transfer due to conduction, species diffusion, and viscous dissipation, respectively. S h 231

256 includes the heat of chemical reaction, radiation, and any user-defined volumetric energy sources. Energy (E) is defined as 2 p v E = h +, (8-12) ρ 2 where sensible enthalpy h for an ideal gas is defined as and h= Yh i i (8-13) i T hi = cp, idt (8-14) Tref with T ref = K. The energy equation in the solid phase is ( ρh) = i ( k T) + Sh. (8-15) t To solve for species concentrations in the solution domain, the general species continuity equation is solved for each species i [128]: ( ρyi) + i( ρvyi) = iji + Ri + Si, (8-16) t where R i is the net production rate of species i by chemical reaction, and S i is the rate of creation of species i through user-defined sources. Eqn is solved for N-1 species where N is the total number of chemical species present. The mass fractions must sum to unity so the mass fraction of the N th species is one minus the sum of the N-1 solved mass fractions. To minimize error, the N th species should be selected as the species with the largest overall mass fraction. In the models presented here, N 2 is the N th species since it is present in the highest amount. Handling of the species diffusion flux, J i, which arises from concentration gradients in the flow field, is accomplished with the dilute approximation of Fick's Law: J = ρd Y, (8-17) i i, m i where D i,m is the diffusion coefficient of species i in the multicomponent system (decane, oxygen, and nitrogen) that is computed by 232

257 D im, = 1 X i j, j i X D j ij. (8-18) For the 3D models including the effects of turbulence, an RNG k-ε model was used. More information can be found in [128] Physical Properties of Components When modeling a multicomponent system with large temperature and concentration gradients, good estimates for the physical properties of each species must be provided. All pure component properties were specified as a function of temperature, and the mixture properties were allowed to vary with the local composition in the reactor. The incompressible gas density was calculated using the ideal gas law: ρ = RT p i op Yi M wi,, (8-19) where p op is the operating pressure. The heat capacity of each compound is specified as a polynomial function of temperature, and the mixture heat capacity is then calculated as a simple mass average. The viscosity, thermal conductivity, and binary mass diffusion coefficients are calculated using kinetic theory [128, 207]. In particular, the binary diffusion coefficients are calculated using a modified form of the Chapman-Enskog formula [208]. In order to use kinetic theory, values of the Lennard-Jones (LJ) parameters (σ and ε) for each species must be specified. Values for permanent gases such as N 2 and O 2 are tabulated [209] and based on experimental viscosity measurements [210]. However, for large molecules such as n-decane and n-hexadecane, tabulated values are not readily available. To approximate the LJ parameters for these compounds, the method of Chung et al. was used, which relates the parameters to the compound's critical constants [211, 212]: σ = 0.809V (8-20) 1/3 c ε T = c, (8-21) k b 233

258 where V c and T c were taken from the literature [179]. It should be noted that any pair of values for the LJ parameters is not unique and other pairs of values for any compound can reproduce experimental property data quite well if the temperature range is not too broad (reduced temperatures in the range of 0.3 to 1.2) [209]. Of further note, while the transport properties are not particularly sensitive to the shape of a gas molecule, they are sensitive to the molecule s length to diameter ratio [207]. For example, the temperature dependence for molecules such as n-decane is slightly greater than that predicted by kinetic theory for spherical molecules Empirical Flammability Limits In order to identify regions in the upstream mixing zone capable of autoignition and flame formation, accurate values for the flammability limits of the fuel of interest must be used. In general, Jones [213] found that for n-alkane vapors the lower flammability limit (LFL) and upper flammability limit (UFL) at 25 C are approximately scalar functions of the stoichiometric mol% of fuel (C st ): LFL(25 C) = 0.55C st (8-22) UFL(25 C) = 3.50C st. (8-23) For hydrocarbons, the combustion reaction is y CH x y + zo2 xco2 + HO 2. (8-24) 2 It follows from stoichiometry that z = x+y/4, where z is moles O 2 /moles fuel. Substituting z into the definition of C st and then inserting this relation into eqns and 8-23, respectively, gives 0.55(100) LFL(25 C) = 4.76x y + 1 (8-25) 3.5(100) LFL(25 C) = 4.76x y + 1. (8-26) For n-decane, x = 10 and y = 22 so the LFL and UFL based on Jones general formula are 0.735% and 4.68%, respectively. More specific experimental values for the LFL and UFL at 1 atm and 25 C are and 4.73 mol% n-decane, respectively [214]. These latter values were used in the post-processing of the model results. 234

259 Furthermore, the variation in the flammability limits as a function of temperature must also be accounted for. In general, the flammability range increases with temperature [215]. The following empirically-derived equations are available for hydrocarbon vapors [80]: where ( T 25) LFL( T) = LFL(25 C) o ΔHc ( T 25) UFL( T ) = UFL(25 C) o ΔH c (8-27), (8-28) o Δ H c is the net heat of combustion (kcal/mol) at 25 C and T is in C Homogeneous Chemistry Mechanisms To account for ignition delay and provide better estimates of the flammability limits and autoignition temperatures as a function of concentration, two chemistry mechanisms for n-decane oxidation were used. A one-step model with 4 species predicts experimental flammability limits and flame speeds but over predicts steady-state temperatures for rich mixtures since only combustion products are considered [216]. A multi-step mechanism consisting of ~600 reactions among 67 species predicts autoignition and cool-flame behavior quite well and has been validated against a range of autoignition and ignition delay time experimental data [217]. Both mechanisms were used with the Chemkin Plug code [123] to estimate ignition delays times as a function of preheat temperature Computational Mesh Generation Gambit meshing software (Fluent Corp., Lebanon, NH) was used to create the computational grid. A simple hexahedral grid of approximately 10,000 cells was initially used before adaption of the grid for the 2D axisymmetric model. Accurate CFD solutions for momentum, heat, and mass transport require that numerical solutions be independent of the numerical grid size and structure. This is achieved with a refined grid containing a large number of nodes [128]. However, reducing computational grid size increases computational time. Since concentration gradients in the fuel-injected reactor vary most near the reactor walls, these regions were assigned a finer mesh. In the center of the reactor, grid size was increased without loss of flow-field accuracy. 235

260 Truncation error was controlled by limiting the change in grid density from the highdensity to low-density regions of the mesh [128]. Gradients of contiguous cell areas in the models were constrained to change by less than 150%. In addition, the optional second-order discretization scheme in Fluent was utilized to reduce the effects of numerical diffusion (discretization error) on the solution [128]. The 3D computational grid consisted of approximately 100,000 cells dispersed in a hybrid fashion (Figure 8-5). Hexahedral cells were used in the majority of the volume of the side port and reactor tube. Tetrahedral cells were used to mesh the junction between the two where non-ideal eccentricity exists CFD Computations Velocity, temperature, and concentration profiles upstream of the catalyst section were calculated using a computational node on the IBM Regatta supercomputer at the University of Minnesota. The effects of viscous heating and thermal diffusion on the solutions were negligible. A criterion of 10-6 was used for each scaled residual component (continuity, x-velocity, y-velocity, energy, O 2, and n-decane) to determine solution convergence. Solution results using convergence criteria ranging from 10-5 to 10-6 showed on average no significant difference in velocity or concentrations (<1.0%), thereby validating the sufficiency of the criteria. For flame calculations, a standard underrelaxation method was used to achieve a converged solution (for more information see [218] for a good description of the underrelaxation method). 8.3 Results Heat Transfer Model Temperatures were measured at 3, 7, 11, 15, and 19 cm on the outside reactor wall and at 22 cm inside the reactor after the static mixer for a range of flowrates and preheat settings for an air feed in order to determine the validity of the heat transfer model and constant heat flux boundary conditions in the preheat section of the reactor. These temperatures are compared with the model results in the next sections. 236

261 Effect of Inlet Velocity Condition for 2D Axisymmetric Model The effect of the inlet boundary condition on wall temperature is noticeable in the preheat section. For a 5 slpm air flowrate and 1825 W/m 2 heat flux, wall temperature for the fully developed (FD) velocity inlet condition is higher than the wall temperature for the plug flow (PF) velocity inlet condition until they equilibrate ~16 cm downstream (Figure 8-6A). The PF velocity inlet condition allows incoming gas to draw more heat from the wall into the gas than the FD inlet condition because of the developing thermal boundary layer for PF inlet condition. Comparison of the experimental wall temperatures for a Variac setting of 65 VRMS with the 1825 W/m 2 model heat flux gives RMS errors of 3.0% and 2.9% for the PF and FD velocity inlet conditions, respectively. The mass average temperatures at 22 cm downstream for the two inlet conditions underpredict the well-mixed experimental temperature after the static mixer (Figure 8-6B). Comparison of the experimental gas temperature at 22 cm for a Variac setting of 65 VRMS with the 1825 W/m 2 model heat flux gives RMS errors of 14% and 15% for the PF and FD velocity inlet conditions, respectively. From these results, the PF velocity inlet condition was used in further 2D axisymmetric simulations Effect of Porous Media Zone in 2D Axisymmetric Model A computational zone from cm was added to the numerical simulation with the PF velocity inlet condition to reproduce the effect of the 80 ppi ceramic foam used in experiments to mix the reactants and better equilibrate temperature. Addition of this zone to the 2D axisymmetric model gives a better qualitative and quantitative fit with the experimental temperatures (Figure 8-6). The main quantitative differences between the temperatures from this simulation and the temperatures from the previous section are the more rapid decrease in the wall temperature from 17 to 20 cm and increase in gas temperature at 22 cm in better agreement with the experimental results The porous zone decreases the RMS error between the simulation and experimental wall temperature to 1.8% for a 5 slpm air flowrate and 65 VRMS Variac setting. In addition, experimental/simulation error for the gas temperature at 22 cm is decreased to 0.8% (Figure 8-6B). The mechanism by which error improvement in the heat transfer model with the porous media zone is attained can be observed by examining the velocity and temperature fields with and without the porous zone (Figure 8-7). 237

262 Experimentally, addition of the static mixer causes radial mixing of the gas. The numerical model simulates this through blunting of the velocity profile observed from cm in the preheat section (Figure 8-7A). The porous zone also distributes the heat at the wall into the gas-phase through the effective thermal conductivity used in this zone. The numerical simulation shows that the porous zone significantly distributes the heat from the wall into the gas phase and significantly decreases the wall temperature from cm (Figure 8-7B) Heat Transfer Validation Summary A comparison of the final heat transfer model (2D axisymmetric, laminar flow, PF velocity inlet condition, and porous media zone) with experimental temperatures measured at 5 and 10 slpm air flowrate and a range of Variac voltages gives excellent agreement (Figures 8-8 and 8-9). In these simulations, only the heat flux is varied for a given flowrate just as in the experiment. No other simulation parameters are adjusted. For 5 slpm air flowrate, wall temperature agreement between experiment and simulation is excellent for Variac settings of 35, 50, and 65 VRMS (Figure 8-8A). Two different simulation lines are shown in comparison with each experimental data set to highlight the sensitivity of the results to the heat flux parameter input. RMS error between the simulation lines and experimental data remains < 2% for all comparisons. Error between the best-fit simulation and experiment for the well-mixed gas temperature at 22 cm is similarly < 1% for all comparisons (Figure 8-8B). Based on the error analysis, simulations heat flux values of ~750, 1100, and 1900 W/m 2 fit the experimental data best for Variac settings of 35, 50, and 65 VRMS, respectively. For 10 slpm air flowrate, wall temperature agreement between experiment and simulation remains good for Variac settings of 35, 50, 65, and 75 VRMS (Figure 8-9A). Two different simulation lines are shown in comparison with each experimental data set to highlight the sensitivity of the results to the heat flux parameter input. RMS error between the simulation lines and experimental data remain < 2% for all comparisons. However, error between the best-fit simulation and experiment for the well-mixed gas temperature at 22 cm increased over the comparisons made at 5 slpm air flowrate (Figure 8-9B). Error increased from 3.6% to 7.0% as Variac voltage was increased from 35 to 75 VRMS. Based on the error analysis, simulations heat flux values of ~800, 1300, 238

263 2200, and 2900 W/m 2 fit the experimental data best for Variac settings of 35, 50, 65, and 75 VRMS, respectively. Calculating average power in average watts from the product of 60 Hz ACRMS current and RMS voltage and dividing by the preheat surface area gives an estimate for the theoretical heat flux to the reactor. Dividing the best-fit heat flux from the simulations by the theoretical heat flux for each Variac setting gives the heat transfer efficiency of the heating tape. For both 5 and 10 slpm flowrates, efficiency significantly increases as the Variac voltage increases (Figure 8-10). A slight difference also appears in efficiency for a given Variac voltage as flowrate is increased from 5 to 10 slpm Vaporization Model Figure 8-11 displays the estimated vaporization length as a function of heat flux for different total air flowrates based on eqn The feed stoichiometry is set at C/O = 1.0. As expected, for a given heat flux the vaporization length increases as the total feed flowrate increases. If a set vaporization length is desired then the heat flux must increase as total flowrate is increased. To test the vaporization model, walls temperatures were measured experimentally and compared to simulations for two Variac settings and 10 slpm air flowrate (Figure 8-12). For a Variac setting of 65 VRMS, simulations exhibit the same qualitative features as the experiment: wall temperature increases, plateaus, and then decreases (Figure 8-12 A). However, simulations largely underestimate the wall temperature until ~15 cm downstream. An inflection point in the experimental data seems to occur at ~ 7 cm, while the simulations predict a much more pronounced peak temperature than the experimental data shows. By 20 cm, the 3000 W/m 2 simulation matches the experimental temperature. RMS error between the 3000 and 3500 W/m 2 simulations and wall temperature data for 65V was 16 and 13%, respectively. Both qualitative and quantitative agreement between the vaporization model and experimental data improve for 75 VRMS (Figure 8-12B). As the voltage is increased, the downstream length in which the simulations appear to underpredict the experimental temperature data decreases. The inflection point in the experimental data seems to occur again near ~7 cm for the 75 VRMS Variac setting, and differences in the inflection point between the 65 and 75 VRMS Variac settings are impossible to detect with the limited data resolution. RMS error between the 4500, 5000, and 5500 W/m 2 simulations 239

264 and wall temperature data for 75 VRMS was 11, 10, and 12%, respectively. Heat transfer efficiencies from the experiment/best-fit simulation comparisons give 43 and 53% for 65 and 75 VRMS, respectively. These are significantly higher than reported in section where only single-phase flow occurs Creating an Autoignition Map The 2D axisymmetric simulation, which was validated in sections and , yields temperature and concentration data as a function of radial and axial distance for various flowrates, Variac settings, and feed stoichiometries. These data allow the exploration of the extent of upstream mixing and possibility of autoignition before the catalyst. For example, the 5000 W/m 2 heat flux simulation for 10 slpm air flowrate and C/O = 1.0, which corresponded best to a 75 VRMS Variac setting in the previous section, is displayed in Figure 8-13 without the porous media section. In the first 40 cm, the temperature field develops faster than the decane concentration field. A maximum decane mol fraction of ~23% occurs at ~ 8 cm downstream of the reactor entrance at the end of the simulated film region. In this region, the momentum boundary layer is developing and convection is quite limited because of the no-slip condition at the wall. This maximum dissipates a few centimeters downstream as the fuel diffuses radially. In order to assess the extent of downstream mixing and the possibility of autoignition, it is helpful to examine radial profiles of temperature and concentration at various downstream locations in comparison with empirical flammability limits and minimum autoignition temperature. A qualitative example for a representative downstream location is displayed in Figure Zero mm on the abscissa represents the centerline of the reactor tube where the composition is fuel-lean and oxygen-rich, and gas-phase temperature is a minimum. Conversely, at the inner walls (± 9.5 mm) the composition is fuel-rich and oxygen-poor and gas-phase temperature is maximal. Orange-shaded sections indicate radial regions within the temperature-dependent flammability limits, whereas the red boxes show regions above the empirical autoignition temperature. As long as the temperature of any flammable region is below the autoignition temperature, there is no risk of flames or explosions. In this qualitative example, this is the case. For a quantitative demonstration, Figure 8-15 shows the radial temperature and decane and O 2 profiles for 5, 10, 15, and 20 cm downstream generated from the data in 240

265 Figure For 5 cm downstream, there is a region within the flammability limits but nowhere is the temperature above the autoignition temperature. For 10 and 15 cm downstream, there are regions above the autoignition temperature and inside the flammability limits; however, these regions do not overlap. However, at 20 cm downstream, these regions finally coincide. For these operating conditions, 20 cm represents a worst case estimate of the flame initiation point where autoignition could occur. Therefore, within the first 20 cm, fuel autoignition is avoided. Placing the catalyst before 20 cm downstream but after 7.7 cm (evaporation length) would be sufficient to avoid autoignition based on this empirical criteria and effect complete vaporization. Performing this type of analysis for a range of flowrates and heat flux values provides data for an autoignition map that gives flame initiation points as a function of heat flux for C/O = 1.0 (Figure 8-16). For instance, a better resolved analysis of the flame initiation point for 10 slpm air, C/O = 1.0, and 5000 W/m 2 heat flux shows that the possibility of autoignition occurs at > 18.5 cm. As another example, if a flame initiation point greater than 20 cm is required then heat flux must be kept < 800 W/m 2 for a 2.5 slpm air flowrate and < 4600 W/m 2 for a 10 slpm air flowrate Ignition Delay Time from Chemistry Mechanisms Results in the previous section indicate a worst-case estimate of the autoignition initiation point. This is because in reality the flame initiation point would occur further downstream because of a number of factors the previous method did not include. First, autoignition temperature exhibits a strong concentration dependence, whereas in the previous section the minimum autoignition temperature is employed. Second, the method does not account for the ignition delay time which is the time required to build a sufficient radical pool to initiate self-sustaining homogeneous chemistry. To include this missing model component, homogeneous chemistry mechanisms, which have been validated against experimental ignition delay and flammability data, can be integrated with the reactor model in the previous section to give a more accurate picture of the autoignition map. Inclusion of a validated chemistry mechanism with the reactor model would move the curves in Figure 8-16 up. Significant differences are observed between the ignition delay times predicted by the one-step and multi-step chemistry mechanisms as a function of feed composition and initial temperature (Figure 8-17). The multi-step mechanism shows a marked 241

266 difference in the ignition delay times predicted at combustion (C/O = 0.322) and CPO (C/O = 1.0) feed stoichiometries predicted as a function of initial temperature (Figure 8-17A). It also reproduces the cool-flame regime at low temperatures extremely well. The one-step mechanism shows no such dependence or cool-flames behavior (Figure 8-17B). Between o C, the one-step mechanism shows good agreement with the multi-step mechanism but deviates drastically at higher and lower temperatures. If the only desired objective function is ignition delay time, then the one-step mechanism could be used at much less computational expense to predict autoignition behavior if the temperature range of operation was within o C. For a premixed flame, the effect of temperature on ignition delay time can be seen clearly from Figure Here, the steady-state location of a pre-mixed n-decane/air flame is compared for wall temperatures from 450 to 700 o C and an inlet temperature of 200 o C. For a 500 o C wall temperature, the beginning of the flame occurs at ~ 20 cm downstream from the inlet. When the wall temperature is increased to 550 o C, the beginning of the flame migrates upstream to ~4 cm. The difference in the ignition delay times between the two simulations is the residence time from one flame location to another (~ 0.5 s). 8.4 Discussion The 2D axisymmetric CFD model presented has lent insight into how fuel autoignition is avoided in the CPO reactor coupled with a fuel injector. In this model, all of the introduced liquid fuel is assumed to vaporize from a fictitious film on the inner reactor wall in the preheat section. In reality, fuel injectors emit small droplets of fuel over a large solid angle. Over the experimental reactor dimensions, some of the fuel droplets, within a small range of solid angles, vaporize in air and travel downstream in the gas-phase to the catalyst (without physical contact with the reactor walls). Other droplets, over a broader solid angle range, impact the reactor walls before they can vaporize and accumulate on the heated walls. The low flowrate fuel injector used in this study gives a hollow cone pattern with a large cone angle so that most of the fuel is directed at the heated reactor wall. Experimentally, initial fuel composition gradients are probably less severe since possible vapor-phase fuel sources come from droplet evaporation in the gas core as well as thin film evaporation from the heated walls. Therefore, the length predicted over 242

267 which fuel autoignition is not possible may be overestimated using this model. Another issue with the current reactor configuration is that flow is not always a laminar flow over the range of flowrates used experimentally. In fact, the side port causes significant local turbulence in the transition region between the side port and main reactor tube. For a 5 slpm flowrate, the flow remains laminar but local recirculation and stagnation flow can be observed in the transition region (Figure 8-19A). For 10 slpm total flowrate, flow becomes turbulent (Figure 8-19B). This flow pattern creates swirl and most likely an uneven wetting of the inner walls of the heat reactor tube and channeling. However, the 2D axisymmetric model presented here captures at least part of the mechanism by which fuel autoignition is abated in the reactor pre-heat section. The modeling shows that a favorable disparity exists between the rate at which the temperature and concentration fields develop. For the example given in Figure 8-15, over the first 20 cm the temperature near the inner reactor wall quickly goes above the minimum autoignition temperature; however, the fuel composition is much richer than the UFL near the wall and autoignition is not possible (Figure 8-15 A-C). In contrast, after 20 cm the flammable region overlaps the region above the minimum autoignition temperature (Figure 8-15 D). While the worst-case autoignition map is useful didactically, combining the 2D model with multi-step chemistry would serve as a more accurate method of determining the flame initiation location than using the empirical flammability limits and minimum autoignition temperature. Combining transport and vaporization models with the multistep chemistry would allow solution of the transient flame propagation problem. The solution would determine the original flame location and track the migration of the premixed flame, which would depend on the total flowrate. Future work should combine the 3D fluid dynamics with a vaporization and chemistry model to achieve a more realistic picture of the actual physics involved. These improvements should include interaction between gas phase and droplets, two-phase flow, and spray wall interaction through empirical heat transfer coefficients that would overcome the assumption that the controlling heat transfer resistance occurs on the outer side of the reactor. These improvements have been performed in considering a urea-water DeNO x system [219]. Interestingly, the increase in ease with which autoignition is avoided as the hydrocarbon chain length increases does not seem readily apparent because of competing effects. As chain length increases, hydrocarbon gas-phase diffusivity 243

268 decreases so the development length of the concentration field would increase. However, the disparity between the autoignition temperature and boiling point of the fuel also increases as chain length increases. Therefore, the relative disparity in the development of temperature and concentration fields as chain length increases is a complex function of temperature-dependent physical properties, wall vaporization, and wall heating rate. This phenomenon is further complicated when fuel mixtures are used. These complex issues dictate that tools such as CFD be used to map out autoignition regimes. Further improvements to the models will be helpful in the development of a predictive mixing model for liquid fuels. Although the results shown in this chapter help to explain experiments already performed, the goal is to predict reactor autoignition topography so that possibly dangerous and costly experiments are avoided. For example, syngas production from large alkanes allows the use of air as the oxidizer. However, if olefins are desired from these fuels, oxygen may have to be used instead of air. Use of oxygen would make experiments much more prone to explosion. Use of a predictive CFD model for scoping studies could limit the number of necessary experiments and drive reactor design for similar processes. 8.5 Conclusions CFD modeling has allowed the quantitative investigation of how flames and explosions are avoided in an experimental reactor for liquid-fuel partial oxidation. Longchain liquid fuels used in experiments possess higher boiling points than autoignition temperatures, a factor that makes processing them in the vapor phase a delicate and possibly dangerous process. Modeling results show a large disparity between the rates at which the gas-phase temperature and concentration fields develop when using an automotive fuel injector. This disparity keeps flammable regions from overlapping regions above the stoichiometric autoignition temperature in the entrance region of the reactor. 8.6 Nomenclature AIT Autoignition temperature, K C p,i Heat capacity at constant pressure of species i, J/(kg K) 244

269 C st Mol % fuel at combustion stoichiometry d Driving force for mass diffusion, 1/m i D i,j Binary mass diffusion coefficient of component i in component j, m 2 /s D i,m Mass diffusivity of species i for dilute mixture, m 2 /s D T,i Thermal diffusion coefficient for species i, m 2 /s E Energy per unit mass, J/kg F External body force vector F x Magnitude of external body force in x-direction, kg m/s 2 F r Magnitude of external body force in r-direction, kg m/s 2 g Gravitational force vector g Magnitude of gravitational force, m/s 2 h i Specific enthalpy, J/kg I Identity tensor J Diffusion flux vector for species i i k Thermal conductivity, W/(m K) k b Boltzmann constant, 1.38 x J/(molecule K) Vaporization length, m vap L e LFL M w M wi N N Re p p op Hydrodynamic entrance length based on laminar flow, m Lower flammability limit, mol % fuel in air Average molecular weight, kg/kmol Molecular weight of species i, kg/kmol Number of species in mixture Reynolds number Pressure, Pa Operating pressure, Pa P Power needed to vaporize fuel feed, W needed P Power input to outer wall of reactor per unit length of heating tape, W/ m input r Radial coordinate, m R Universal gas constant, x 10 3 J/(kmol K) R i Net rate of production of species i by chemical reaction, kg/(m 3 s) 245

270 S i User-defined mass source for species i, kg/(m 3 s) S h User-defined energy source, J/(m 3 s) S m User-defined mass source, kg/(m 3 s) t Time, s T Temperature, K T ai,min T b T c T ref T sat UFL v v v x v r V i V c x X i Y i Greek Minimum autoignition temperature, K Boiling temperature, K Critical temperature, K Reference temperature, K Saturation temperature, K Upper flammability limit, mol % fuel in air Velocity vector Velocity magnitude, m/s Velocity magnitude in x-direction, m/s Velocity magnitude in r-direction, m/s Diffusion velocity vector of species i Critical volume, cm 3 /mol x-direction coordinate, m Mole fraction of species i Mass fraction of species i o Δ H Standard heat of combustion, J/kmol c Δ H vap Heat of vaporization, J/kmol ε Characteristic energy, J/molecule μ Dynamic viscosity, kg/(m s) ρ Density, kg/m 3 σ Collision diameter, o A τ Stress tensor Vector Operators Vector differential operator or del Dot product 246

271 Table 8-1 Reynolds numbers for 19 mm ID tube. VFR (slpm) MFR (kg/s) v avg (m/s) N Re L e (mm) x x x x x x x x x x Table 8-2 Reynolds numbers for 4 mm ID tube. VFR (slpm) MFR (kg/s) v avg (m/s) N Re L e (mm) x x x x x x x x x x

272 T ai,min (gas-phase) 400 temperature ( o C) number of carbon atoms T b T (C/O=2) sat T (C/O=1) sat T (C/O=comb.) sat Figure 8-1 Temperatures pertinent to successful vaporization and mixing of C 1 to C 16 alkanes with air for CPO. Normal boiling points [177] for alkanes monotonically rise from -162 C (methane) to 281 C (hexadecane). However, for alkanes that are liquids at 25 C the saturation temperatures (T sat ) needed to maintain C/O ratios between combustion and C/O = 2.0 in air are much less than the corresponding boiling points. Minimum autoignition temperatures for gas-phase chemistry asymptotically decrease from near 600 C (methane) to ~200 C (hexadecane) [178, 179]. 248

273 Fuel Injector Air Heating tape g Static mixer Catalyst Thermocouple Heat shields Insulation Products Figure 8-2 Schematic and photograph of fuel-injected short-contact-time reactor. Temperatures were measured along the outside wall of the upstream heating section and at locations inside the heat shield/catalyst stack. 249

274 z 21cm : constant heat flux g z > 21cm : adiabatic Air: T = 300 K P = 1 atm cm: porous zone 40 cm long 22 mm OD x 19 mm ID fused silica Figure 8-3 Schematic of 2D axisymmetric model geometry. wall energy sink term: latent + sensible heat sink fluid-phase energy source term: sensible heat source (power required to heat decane vapor from 25 o C to 174 o C) fluid-phase mass source term: constant fuel vaporization rate along vaporization length corresponding to fuel flowrate Figure 8-4 Schematic of volumetric source/sink cells for vaporization model. 250

275 g Figure 8-5 Computational grid for 3D model. 251

276 W m -2 fully developed inlet 400 temperature ( o C) W m -2 plug flow inlet 1900 W m -2 plug flow inlet w/ porous media distance (cm) A temperature ( o C) sim: 700 W m -2 exp: 65 V sim: 750 W m -2 sim: 1825 W m -2 inlet: plug flow sim: 1825 W m -2 inlet: fully developed comparison at 22 cm exp: 65 V sim: 1900 W m -2 inlet: plug flow porous media mixer sim: 1900 W m -2 B Figure 8-6 Comparison of the effect of inlet boundary conditions and static mixer on temperature profile. Experimental data for 5 slpm air flowrate, Variac = 65 V. (A) Wall temperature comparison. (B) Gas temperature comparison at 22 cm downstream. 252

277 v (m/s) A empty tube 22 cm w/ porous mixer T ( o C) B empty tube 22 cm w/ porous mixer Figure 8-7 Effect of porous media model on velocity and temperature. Conditions are 10 slpm total flow, 1100 W/m 2, aspect ratio not to scale. (A) Velocity magnitude. (B) Temperature. 253

278 temperature ( o C) A 1900 W m W m W m W m W m W m distance (cm) B 250 temperature ( o C) exp: 35 V sim: 700 W m -2 sim: 750 W m -2 exp: 50 V sim: 1000 W m -2 sim: 1100 W m -2 exp: 65 V sim: 1825 W m -2 sim: 1900 W m -2 0 A B C comparison at 22 cm Figure 8-8 Heat transfer validation for 5 slpm total air flowrate. (A) Wall temperature comparison. Experimental data sets correspond to 35, 50, and 65 V Variac settings. (B) Gas temperature comparison at 22 cm downstream. 254

279 500 A temperature ( o C) distance (cm) B temperature ( o C) exp: 35 V sim: 750 W m -2 sim: 800 W m -2 exp: 50 V sim: 1200 W m -2 sim: 1300 W m -2 exp: 65 V sim: 2100 W m -2 sim: 2200 W m -2 exp: 75 V sim: 2800 W m -2 A B C D comparison at 22 cm sim: 2900 W m -2 Figure 8-9 Heat transfer validation for 10 slpm air flowrate. (A) Wall temperature comparison. Experimental data sets correspond to 35, 50, 65, and 75 V Variac settings. Simulation lines correspond to 750, 800, 1200, 1300, 2100, 2200, 2800, and 2900 W/m 2. (B) Gas temperature comparison at 22 cm downstream. 255

280 efficiency slpm air 10 slpm air variac setting (V) Figure 8-10 Heating efficiency for all heat transfer validation experiments. evap length (cm) slpm air 5 slpm air 7.5 slpm air 10 slpm air wall flux (W/m 2 ) Figure 8-11 Evaporation length vs. heat flux. 256

281 A temperature ( o C) W m cm evap. length W m cm evap. length distance (cm) B 5500 W m cm evap. length temperature ( o C) W m cm evap. length 4500 W m cm evap. length distance (cm) Figure 8-12 Validation of vaporization model at 10 slpm total air flowrate, C/O = 1.0. (A) 65 V Variac setting. (B) 75 V Variac setting. 257

282 T ( o C) 40 cm n-decane mol fraction Figure 8-13 Concentration and temperature profiles from 2D autoignition model for 10 slpm air, C/O=1, and wall flux = 5000 W/m mol fraction 5 0 decane O2 LFL UFL T ( C) AIT T ( o C) radius (mm) Figure 8-14 Qualitative example of temperature, n-decane, and O 2 radial profiles at a fictitious downstream location in the preheat section. The autoignition temperature at combustion stoichiometry is denoted by the dotted black line. Orange shading indicates radial regions within the flammability limits whereas red boxes indicate regions above the autoignition temperature. 258

283 mol fraction cm downstream A decane O2 LFL UFL T ( C) AIT T ( o C) radius (mm) cm downstream B 600 mol fraction decane O2 LFL UFL T ( C) AIT T ( o C) radius (mm) Figure 8-15 Quantitative example of temperature, n-decane, and O 2 radial profiles. (A) 5 and (B) 10 cm downstream in the preheat section. Conditions are 10 slpm air, C/O=1, and wall flux = 5000 W/m

284 mol fraction cm downstream C decane O2 LFL UFL T ( C) AIT T ( o C) radius (mm) cm downstream D 600 mol fraction decane O2 LFL UFL T ( C) AIT T ( o C) radius (mm) Figure 8-15 (cont.) Quantitative example of temperature, n-decane, and O 2 radial profiles. (C) 15 and (D) 20 cm downstream in the preheat section. Conditions are 10 slpm air, C/O=1, and wall flux = 5000 W/m

285 flame initiation point (cm) slpm 5 slpm 7.5 slpm 10 slpm heat flux (W/m 2 ) Figure 8-16 Autoignition map for C/O =

286 idt (s) Bikas reduced-step A C/O=1 C/O= T i ( o C) idt (s) One-step B C/O=1 C/O= idt (s) T i ( o C) One-step vs multi-step at C/O=1 One-step Multi-step C T i ( o C) Figure 8-17 Dependence of ignition delay time on feed stoichiometry and initial temperature in a PFR model. 262

287 1850 T wall 450 o C o C 550 o C o C 700 o C cm 200 T( o C) Figure 8-18 Pre-mixed n-decane flame with one-step chemistry. Inlet: 10 slpm air, C/O=1,and T inlet = 200 o C, constant wall T = parameter. 263

288 Figure 8-19 Velocity pathlines (m/s) for (A) 5 slpm total flow at 25 o C. 264

289 Figure 8-19 (cont.) Velocity pathlines (m/s) for (B) 10 slpm total flow at 25 o C. The turbulent RNG k-ε model is used. 265

290 Chapter 9 Conclusions and Future Work 9.1 Conclusions OCM and Quenching in Short-contact-time Reactors The work contained in Chapters 2 and 3 investigated the role of the temperature profile in the production of non-equilibrium products (e.g., ethylene) with short-contacttime reactors. Modeling of homogeneous chemistry with the use of millisecond heating and cooling of the reaction gases has been demonstrated to maximize ethylene yield to the same value predicted by isothermal modeling. For example, isothermal modeling at 1300 C and 1 atm for a CH 4 /O 2 feed ratio of 1.7 shows a maximum ethylene yield of 14.5% at 2.3 ms. Use of an optimized exponential heating and cooling cycle (see Table 2-4) results in the same maximum ethylene yield (14.5%) after 10 ms. However, homogeneous chemistry by itself does not explain experimental findings. The homogeneous-heterogeneous chemistry model shows that ethylene yields as high as 22% can be realized from the OCM process through the use of a short Pt catalyst section operating at higher temperature profiles ( C) than those traditionally used for this process ( C). A fast thermal quench of gases leaving the catalytic section is necessary to maintain yields by eliminating gas-phase chemistry and trapping the valuable reactive intermediate (ethylene). From experiments performed on Pt at high GHSV, the effect of thermal quenching of post-catalyst gases was established (Chapter 3). Dramatically lowering the temperature of product gases within 30 ms improved C 2 selectivity significantly. However, the C 2 selectivity increase was not as high as that predicted by simulations in Chapter 2. With the current N 2 quench configuration, product gases are not cooled fast enough to realize an optimum yield of C 2 products. Accompanying simulations show that very little time is available to quench product gases after the catalyst assuming the proposed temperature distributions are approximately correct. If ethylene selectivity is to be maximized, thermal quenching must occur as soon as possible after the catalyst. Of particular interest, if we sum C 2 selectivities and the CH 4 conversion, the maximum value of approximately 111% occurs right at the end of the catalyst. 266

291 However, there were some significant discrepancies between simulated and experimental results. While selectivities predicted by simulations are in agreement with the experiments, the fuel and O 2 conversions are not. There are some possible reasons for these differences. First, there could be some catalytic activity by the ceramic fibers and monoliths, which is not accounted for. Second, the radial temperature profile is not constant and a large gradient likely exists in this direction since the walls of the quench section are not insulated. This gradient could cause lower fuel conversion than predicted by simulations where the radial temperature profile is isothermal. The absence of C 2 production in direct quench mode with the compact heat exchanger was caused primarily by the olefins adsorbing on the stainless steel fins. Based on previous experimental and modeling results, significant C 2 products by quenching directly after the catalyst were expected. While Pt is a poor catalyst, the thermal quench system in staggered quench mode does significantly increase C 2 yields over the default configuration. The main point to take away from these results is that some homogeneous chemistry is necessary after the catalyst to effect C 2 production; however, this homogeneous chemistry must be controlled to prevent further C 2 oxidation. Based on previous work with the N 2 quench system and this heat exchanger, it appears the upper limit for C 2 production over Pt-coated monoliths has been reached. This work shows that the quench system does offer improvements in C 2 production, and more promising OCM catalysts will be needed to test whether C 2 yield can be raised above the economically viable limit for this process (25%) D Reactor Model for Methane CPO Numerical simulations for methane catalytic partial oxidation on Rh-coated foam monoliths were performed and tested against high-resolution species and temperature profiles taken along the catalyst axis (Chapter 4). A systematic comparison considering 2D models with heat and mass transport and a 1D reactor model with two multistep surface chemistry mechanisms were presented for feed conditions of 5 standard liters per minute, atmospheric pressure, and C/O ratios of 0.7, 1.0, and 1.3. Agreement between the plug flow model and experimental data is for all conditions inferior to 2D models. 2D simulations and measured profiles agree qualitatively for all experimental conditions. Quantitative agreement was best for syngas stoichiometry (C/O=1.0), while some differences were observed for C/O = 0.7 and 1.3. The importance of spatial 267

292 profiles for mechanism and reactor model validation was highlighted. The occurrence of direct versus indirect syngas production for both mechanisms depends strongly on catalyst temperature. When gas and surface temperatures are correctly differentiated during surface chemistry calculations, both mechanisms predict hydrogen and carbon monoxide are produced partially in the presence of oxygen in agreement with experimental spatial profiles. The current work reinforces the need to compare numerical simulations to sufficiently resolved experimental profiles in order to verify the validity of the reaction mechanism and the reactor model and motivates further work on mechanism refinement Start-up, Transients, and Ignition Kinetics The rapid start-up (< 10 s) of a short-contact-time reactor to produce steady-state syngas from methane was demonstrated without external preheat by initially using homogeneous combustion (Chapter 5). Analysis of the reactor products with a rapid response mass spectrometer allowed the determination of the optimal combustion time that minimized time delay in syngas production. One-dimensional plug-flow simulations of short combustion time experiments (1 and 5 s combustion times) demonstrated that transient product profiles can be accurately reproduced using the established surface mechanism for C 1 chemistry on Rh. Combining parameters bounded by theory for the thermal catalyst history with the plug-flow model resulted in the development of a transient, axially-distributed temperature profile that offers insight into how catalyst temperature develops during lightoff. This work has helped further the state-of-the-art in fast start-up reforming technologies for demanding fuel cell-based mobile applications. Experiments in Chapter 6 extended the transient study in Chapter 5 to higher alkane fuels and allowed the governing ignition kinetics for C 1 -C 16 alkanes to be determined. This work demonstrates that steady-state production of syngas (CO and H 2 ) can be attained within 5 s after admitting large alkanes (i-octane, n-octane, n- decane, or n-hexadecane) and air to a short-contact-time reactor by using an automotive fuel injector and initially preheating the Rh-coated catalyst above each fuel s respective catalytic autoignition temperature. T cai,min with Rh was C for n-octane, i-octane, and n-decane, and ~300 o C for methane. In contrast, ignition of n-hexadecane occurred at lower temperatures (220 o C and greater) because of an indirect two-stage process 268

293 where exothermic homogeneous reactions preheated the catalyst by o C to temperatures (~280 o C) sufficient for surface lightoff. The catalytic ignition kinetics for large alkanes were determined experimentally and compared with those of methane using rapid-response mass spectrometry. The controlling step for surface ignition possessed an apparent activation energy of ~77 kj/mol, which was not significantly different between fuels (p > 0.05), and a preexponential on the order of 10 6 s -1 for higher alkanes and 10 4 s -1 for methane. Catalytic autoignition from T cai,min is a visible two-stage process: starting the reactor between T cai,min and T L/O effects sufficient exothermic surface chemistry to raise the catalyst temperature high enough to undergo lightoff (transition from kinetically limited to mass-transfer limited operation). The differences between T cai,min and T L/O varied from ~60 o C for methane to 10 o C for higher alkanes with the ceramic foam monoliths used in this study. Through this work, some of the similarities and differences in ignition behavior between linear alkanes of various size have been demonstrated. Transient characterization has helped in understanding the surface processes at work during the ignition of higher alkane CPO Carbon Formation versus Time On-stream Significant carbon formation (1-40 mg) was measured on Rh foam catalysts with methane CPO (Chapter 7). Carbon evolves mainly as CO 2 with little CO. Time constants for steady-state carbon formation were ~60 s, and carbon amount varied linearly with C/O ratio with a 5 wt% Rh loading. However, carbon formation seemed to non-linear fashion as C/O was increased with a 20 wt% Rh loading. The actual catalytic surface area of the catalyst foams is unknown and a complex function of washcoat loading, Rh loading, and time on-stream. However, it appears that more than monolayer carbon is formed on these catalysts at higher C/O ratios and Rh loadings. Two types of carbon (support and metal) are suggested by the adiabatic oxidation diagrams Modeling Upstream Autoignition in CPO Reactor for Higher Alkanes CFD modeling has allowed the quantitative investigation of the route in which flames and explosions are avoided in an experimental reactor for liquid-fuel partial oxidation (Chapter 8). The liquid fuels used in experiments possess higher boiling points than autoignition temperatures, a factor that makes processing them in the vapor phase 269

294 a delicate and possibly dangerous process. Modeling results show a large disparity between the rates at which gas-phase temperature and concentration fields develop when using an automotive fuel injector. This disparity keeps flammable regions from overlapping regions above the stoichiometric autoignition temperature in the entrance region of the reactor Towards an Integrated Reactor Model All of the work in this thesis has been geared towards understanding the axial temperature profiles developed in short-contact-time reactors and exploiting transient analytical techniques to gain insight into the kinetics of linear alkane fuels on noble metal catalysts. The results of this work have helped towards the goal of a complete reactor model for linear alkanes (C 1 -C 16 ) that encompasses the catalyst and reactor conditions upstream of the catalyst. While the results of Chapter 4 show that it is now possible to accurately model short-contact-time reactors for methane CPO, further work will be needed to develop an analogous model for higher alkanes. Chapters 6 and 7 have offered experimental support for developing a mechanism for higher alkanes. Chapter 8 details the development of a numerical model that accurately simulates the upstream mixing and vaporization in a fuel-injected reactor for higher alkanes. The next section summarizes techniques that may be useful in building a surface chemistry mechanism for higher alkanes and build upon the understanding gained from this dissertation. 9.2 Future Work The following sections highlight opportunities to improve or utilize the methods in this dissertation to enhance understanding of the chemical and physical phenomena occurring in short-contact-time reactors. Because reaction conditions are so severe and previous experimental efforts have been limited to integral measurements, it has been difficult to isolate one particular process from another and truly characterize the kinetics at work. However, coupling of the transient methods presented in this thesis with spatial capabilities opens up a new realm of diagnostic and theoretical possibilities. 270

295 9.2.1 Perturbation Methods to Probe Surface Kinetics for CPO: Deconvolution of Mass Transfer from Apparent Kinetics Use of transient analysis has been shown effective in dissertation to ascertain start-up time (Chapter 5), probe the kinetics of alkane ignition (Chapter 6), and quantify catalyst carbon formation with time on-stream (Chapter 7). In addition, the comparison of 2D detailed kinetic simulations with high-resolution spatial data has validated C 1 surface chemistry for methane CPO (Chapter 4). However, combining spatial and transient analysis results in a powerful quantitative tool that can be used to characterize surface chemistry at different axial locations in the catalyst. It has been shown that by periodically switching the C/O ratio, the chemical history of the catalyst at any axial location can be recorded as the catalyst relaxes in temperature [16]. For example, Figure 9-1 shows temperature and species histories in the catalyst caused by switching the C/O ratio from 0.6 to 1.4 every 20 s. By comparing two different locations in the catalyst, the capillary technique enables the measurement of the H 2 and CO formation rate in a differential slice of the catalyst. As discussed in Chapter 5, the chemistry adjusts almost instantaneously to a switch in the inlet stoichiometry (ms timescale) but catalyst cooling or heating takes seconds; therefore the dependence of the H 2 and CO production rate on temperature can be followed like in a series of isothermal experiments [220]. The rest of this section outlines a theoretical technique that may allow deconvolution of the chemical kinetics from the external mass transfer at each axial location in the catalyst. The combination of spatio-temporal data with theory can be a powerful method to determine reaction kinetics: one perturbation experiment can replace dozens of isothermal experiments, which are not possible with CPO at high conversions. More information on the experimental technique can be found elsewhere [220]. Figure 9-2 shows the formation rates calculated from one set of experiments switching from C/O = 0.6 to 1.4 and back with a 20 s dwell time. The slice of the catalyst that has been picked for the analysis is from x = 4.13 mm to 4.76 mm. In both cases, gas phase O 2 is still present in this slice (Figure 9-1). The reason for picking a slice in the middle of the catalyst is to have a reasonable value for the surface temperature. According to numerical simulations (Chapter 4), the gas phase needs 3-4 mm in the catalyst to equilibrate with the surface. These temperature measurements are most likely biased towards the gas phase as the thermocouple is not in direct contact with the 271

296 surface. By picking a slice in the middle of the catalyst the assumption of thermal equilibrium can be made. The upper panels in Figure 9-2 (set A and B) show the molar flowrates (F) of H 2 and CO at 4.13 mm and 4.76 mm in the catalyst. The difference of these molar flowrates (ΔF) is the rate of formation of H 2 and CO in the catalyst slice (lower panels of Figure 9-2, set A and B). The temperature used is the mean of the temperatures measured at both positions and is a good value for the mean surface temperature in this catalyst slice. The temperature difference between 4.76 mm and 4.13 mm is on average 180 C. The rate of formation for CO can be converted to an atomic turn over frequency (TOF) via the following equation: 1 1 1min 1 3 TOF ( s ) =Δ F ( mol min ) S( m ) V ( m ), (9-1) CO CO 60 s bed where S is the surface area per unit volume of bed, and V bed is the volume of the catalyst. S was taken as 1.6x10 4 m -1 as per Chapter 4 and V bed was 4.3x10-9 m 3 for the 16.5 mm diameter by 1 cm long catalyst bed. The atomic TOF for H 2 is calculated the same way except the right hand side of equation 9-1 is multiplied by an additional factor of 2 to account for 2 atoms of atomic hydrogen per 1 molecule of H 2. By plotting the natural logarithm of TOF versus 1/T an Arrhenius analysis supplies apparent activation energies that allow a conclusion whether the H 2 and CO formation rates are controlled by the same rate limiting step. Figure 9-3 shows the Arrhenius plot for the data of Figure 9-2 (set A, ) whereas 9-4 gives the plot for the data in Figure 9-2 (set B, ). TOF data from the foam catalysts may not only be influenced by kinetic phenomena but mass transfer as well [221]. To account for this, TOF data were simulated by either eqn. 9-2 or 9-3: k k o o 1 = k S k 1 = k S k v c ko, v c (9-2), (9-3) where k ko, = 1 1 k + k. (9-4)

297 In these equations, k o is the observed rate constant (s -1 ), S v is the area per unit volume of solid in the catalyst bed (9.3x10 4 m 2 m -3 bed), k c is the film mass transfer coefficient (m s -1 ), and k, k 1, and k 2 represent kinetic rate constants (s -1 ). A zero reaction order was assumed because the rate appears independent of reactant and product concentration (linear decrease of CH 4 and O 2 and the linear increase of CO and H 2 ) with distance in the oxidation zone (spatial profiles of Chapter 4). Eqn. 9-2 considers only one controlling kinetic step whereas eqn. 9-2 considers two distinct kinetic steps. The kinetic rate constants were all considered to have the Arrhenius form E RT k = A e, (9-5) where A is the preexponential factor and E is the activation energy. The film mass transfer coefficient was modeled as kc 1.5 = BT, (9-6) where B, the mass transfer coefficient prefactor, has units of m s -1 K This expression has been shown acceptable in previous analysis of mass transfer in ceramic foams [221]. Fitting either eqn. 9-2 or 9-3 to the TOF data as a function of temperature consisted of optimizing the values of A, E, and B (eqn. 9-2) or A 1, A 2, E 1, E 2, and B (eqn. 9-3) using the solver tool in Microsoft Excel. The objective function to minimize was the sum of the squared error between the TOF data and the k o values at each temperature. Since values of A, E, and B can differ by several orders of magnitude, each one was scaled by a constant to cause the manipulated values to be near unity and allow more robust optimization. For example, a typical value for E is on the order of 10 4 J/mol, so the manipulated variable for E considered by the non-linear optimization software is scaled by 10 4 (i.e., if E = 10 4 J/mol, E a,manipulated = 1 J/mol). In Figure 9-3, fitting the data for the switch from C/O = (Figure 9-2 set A) to eqn. 9-2 works well for both H 2 (6% RMS error) and CO (10% RMS error). Residual values appear normally distributed around the mean. Optimized values for A, E, and B are reported in Table 9-1. For this experiment the activation energy is different between H 2 and CO, and both mass transfer and kinetics appear to play a role in the observed catalyst performance. 273

298 For the TOF data in Figure 9-4, eqn. 9-2 does not fit the data for the switch from C/O = (Figure 9-2 set B). Instead of a straight line, the data (Figure 9-4) display two distinct regimes as temperature changes. H 2 TOF changes slope as temperature is increased and CO TOF changes from a positive slope to near 0 slope. Instead of eqn. 9-2, eqn 9-3 fits the TOF data well for both H 2 (5% RMS error) and CO (7% RMS error) with normally distributed residuals. Optimized values for A 1, A 2, E 1, E 2, and B are reported in Table 9-2. It appears that two very different kinetic steps dominate the chemistry as temperature changes. From the negative activation energy calculated for k 2,CO, it appears that one or both of the idealized rate constants may be effective rate constants that contain a number of reaction steps. Interestingly the ratio of the B H2 /B CO for both sets of data is ~3 (Tables 9-1 and 9-2). The mass transfer coefficient typically represents a diffusion coefficient divided by an unknown film length, and the diffusion coefficient is inversely proportional to the square root of the compound s molecular weight from kinetic theory. The value of 3 observed is in fair agreement with the ratio of the square roots of the molecular weights for H 2 and CO, (28/2) 0.5 = The effective diffusion coefficients for CO and H 2 are mixture averaged and hence the ratio of their dilute diffusion coefficients will not be exactly the ratio of square root of their molecular weights. The values for B determined here for an 80 ppi foam (10-6 to 10-5 m s -1 K -1.5 ) are in reasonable agreement with B estimated in previous work on 30 ppi foams for CO oxidation(2x10-5 m s -1 K -1.5 ) [221]. Assignment of the obtained apparent activation energies to values reported for elementary steps in the literature cannot be performed until the effect of (1) mass transport limitations, (2) surface coverages, (3) surface temperature, and (4) changing gas phase concentrations are quantified. The methods in this section have been devoted to addressing the mass transfer limitation. All of these issues are discussed next. (1) Mass transport limitation: when a mass transport limitation is present, the apparent activation energy obtained from an Arrhenius analysis is lowered. The change in slope for the H 2 TOF data during heating at C/O = 0.6 (Figure 9-4 A) appears to be a classic example of a change from mostly kinetic control at low temperature to a combination of mass transfer/ kinetic control at higher temperature. The influence of mass transfer 274

299 limitation on the CO TOF data during heating at C/O = 0.6 (Figure 9-4 B) is observed as mass transfer seems to dominate the overall rate. (2) Surface coverages: Activation energies for surface or desorption reactions can be coverage dependent. As surface coverages in the catalyst slice analyzed during the transient experiment are unknown, obtained values cannot be compared to literature values. However, these coverages could have profound effects on the measured energetics since when switching the C/O ratio a composition comes into contact with a surface that is never at that temperature with that composition at steady-state. (3) Surface temperature: The temperatures measured with the thermocouple inside the sampling capillary are biased towards gas phase temperature. According to the numerical simulations of the foam catalyst (Chapter 4), surface and gas phase are largely equilibrated in the middle of the catalyst. However, in the case that surface temperature is higher than the measured temperature, the apparent activation energy would come out too low. Vice versa, if surface temperature was lower than measured by the thermocouple (e.g. by endothermic surface chemistry), the obtained apparent activation energy may be too high. (4) Changing gas phase concentrations: During the transients (Figure 9-2 and analyzed in Figures 9-3 and 9-4), the reactant and product concentrations in the examined catalyst slice are continuously changing. If an activation energy has to be determined, changes in species concentrations are usually kept small so as not to alter the reaction rate. This is of course impossible for a transient in an autothermal foam as conversions are high. The Arrhenius analysis presented returns only straight lines because the reaction rate in the oxidation zone appears to be independent of any species concentration in the gas phase. This is indicated by the linear decrease of CH 4 and O 2 and the linear increase of CO and H 2 with distance in the oxidation zone (see steady state profiles in Chapter 4). The reaction rate is obviously nearly constant in this zone. More experiments need to be performed to test this assumption. For instance, measuring CH 4 and O 2 conversion versus time could also test whether all of the rates of reactants and products both behave zero-order. 275

300 The analysis outlined in this chapter helps one to better understand the surface chemistry at work during alkane partial oxidation and holds great promise. While the surface mechanism for methane could be further validated with this technique, it would also allow the determination of Arrhenius parameters for larger alkane fuels at various locations along the catalyst, information that is only beginning to be tackled by computational methods such as density functional theory Microkinetic Surface Mechanism for Higher Alkane Partial Oxidation While Chapter 4 highlighted the spatial validation of CH 4 chemistry on Rh, surface mechanisms for alkanes larger than C 2 on noble metals are non-existent in the scientific literature. As alkane chain length increases, the number of possible reactions increases dramatically. Mechanism development for C 1 chemistry has benefited from the vast surface science literature for the reactivity of H 2, CO, and CH 4 on single crystal. Indeed, most of the energetic parameters for mechanism 1 in Chapter 4 are based on data from these experiments (e.g, see [3, 222]). For C 2 alkanes and higher, the body of literature dealing with their reactivity on surfaces vastly diminishes. In addition, at atmospheric pressure methane gas-phase chemistry is negligible. For C 2, this is not the case and surface mechanisms have been developed in coordination with gas-phase chemistry mechanisms [88, 102]. Use of the Unity Bond Index-Quadratic Exponential Potential (UBI-QEP) method has been extremely helpful in providing highly-accurate estimations of activation energies for the surface reactions of C 1 and C 2 molecules on noble metal surfaces [223, 224]. With increasing computer speed, the ability to do density functional theory calculations is becoming increasingly quicker with small molecules [ ]; however, for large molecules computations such as these with molecules from C 8 to C 16 are still quite a long way off. For large alkanes, the interplay between surface and gas-phase chemistry becomes increasingly important for short-contact-time reactors as the primary products are not just H 2,CO, CO 2, and H 2 O, but olefins as well. The evolution of C 2 surface mechanisms for Pt and experimental understanding has led to the theory that most of the olefins are produced in the gas-phase, and syngas and total oxidation products are produced on the surface [7, 88, 101, 102, 230]. Extending this idea to higher alkanes such as C 8, simulations with large gas-phase mechanisms for rich oxidation have demonstrated that olefin production at short-contact- 276

301 times can be explained almost completely by homogeneous chemistry [231]. However, in these models heterogeneous and homogeneous chemistry zones are artificially constrained into a two-zone model where in the first zone (surface chemistry) fuel and O 2 are consumed and the presumed products (H, CO, CO 2, and H 2 O) are created. This fictitious zone is followed by a zone where more fuel is consumed without oxygen and olefins are formed through gas-phase chemistry. H 2, CO, CO 2, and H 2 O are mostly inert without O 2 at the millisecond contact times typical in short-contact-time reactors according to the gas-phase mechanism. While these simulations indicate that olefins may be formed solely through gasphase chemistry, the study offered no insight into the surface chemistry since the extent of reaction in the first zone was based solely on experimental data. The amounts of H 2, CO, CO 2, and H 2 O leaving the first zone of the model were based entirely on the experimental values at the end of the catalyst. A more rigorous approach considered next relaxes the two-zone approximation and considers surface and gas-phase chemistry occurring simultaneously. In the rest of this section, a preliminary surface mechanism for the reaction of an n- octane molecule on the surface is highlighted, which extends the methane on Rh mechanism. The simplest way to simulate the reaction of larger molecules on the surface is to assume they behave like methane. More specifically, once the hydrocarbon is adsorbed, it falls apart on the surface and becomes a source of C and H atoms: CH () s 8 Cs () + 18 H() s (9-7) 8 18 Cascading alkyl abstraction is not considered, and the C and H react with surface O to form products. Combining this reaction step with the mechanism for CH 4 on Rh would be the simplest pseudo-methane model for octane reaction on a surface. However, with the methane surface mechanism used in Chapter 4, both methyl (CH 3 ) and methylene (CH 2 ) are considered as intermediates in the decomposition of CH 4 on the surface. These steps are the result of a non-oxidative methane coupling study on Pt [89]. Since this pathway is available and simulates methane reactivity quite well, the next way of modeling octane dissociation is to assume it breaks down into CH 3 and CH 2 groups: CH () s 6 CH () s + 2 CH() s (9-8)

302 This type of scheme was tried by combining the methane on Rh mechanism with additional steps for octane adsorption and reaction on the surface. Additional steps were added based on analogous steps found important for C 2 surface chemistry on Pt [102]. Reaction steps in the mechanism were largely developed by trial and error, and kinetic parameters were estimated by comparison with C 2 chemistry. The C 1 surface mechanism is shown in Table 9-3, and the added steps for C 8 decomposition and oxidation are shown in Table 9-4. The C 1 mechanism in Table 9-3 consists of the same reaction steps as mechanism 1 [100] from Chapter 4; however, some of the kinetic parameters are different because of attempts to tune the mechanism to be thermodynamically consistent for low temperature water-gas-shift [106, 232, 233]. Modifications consisted of modifying preexponential factors and varying the surface coverage dependent activation energies for R10, R20, and R21. Some preexponential factors were further modified in the present work to better fit experimental data for n- octane CPO experiments. The added steps for octane (Table 9-4) consist of fuel adsorption (R39), desorption (R40), and oxidative (R42 and R43) and non-oxidative (R44 and R45) fuel decomposition reactions. Additional steps for C, CO, and CO 2 oxidation were added based on [234], and the activation energies for these steps were left as calculated by UBI-QEP for Rh(111). The desorption of OH radical (R41) was included as it was found to significantly affect H 2 O and CO 2 behavior. The sticking coefficient for octane adsorption (R39) was set to 0.4 based on the work in Chapter 6, which indicates the sticking coefficient for octane should be ~two orders of magnitude larger than that for methane. The desorption activation energy for n-octane was set at 75 kj/mol based on physisorption experiments with transition metals [235]. The desorption energy for R41 was taken from [222]. The remaining parameters (preexponentials for R40-R49 and activation energies for R42-R45) were treated as fit parameters. Therefore, the current mechanism has not been constructed in a thermodynamically consistent fashion. Other reaction paths were considered but appeared unimportant based on the fitting analysis performed. The nonoxidative abstraction of H atoms from the adsorbed n-octane molecule had little effect on the mechanism performance. R42 and R43 had a strong effect on the light-off temperature and ignition. Tuning of the parameters was performed while using an adiabatic plug-flow simulation with multi-step gas phase chemistry [236] (2411 reactions 278

303 among 473 species) coupled with the surface chemistry in Chemkin Plug [123]. Any deficiencies in the gas-phase chemistry obviously make their way into the surface mechanism. The C/O ratio was varied between 0.7 and 2.0 (fuel in air), inlet temperature was kept at 750 K, and catalyst length was 1.5 cm (tortuosity = 1.5). The integral (catalyst exit) experimental data used to compare with the simulation results was taken from [231, 237]. Agreement between simulations and experimental data is good for O 2 conversion, but simulations overpredict fuel conversion and back face temperature. Agreement deteriorates as C/O is increased (Figure 9-5 A). Qualitative fit of the H 2 and CO data is excellent with the major trends being well reproduced as a function of C/O ratio (Figure 9-5 B), while the fit of H 2 O and CO data worsens as C/O increases. Qualitative agreement also exists for the major olefins (Figure 9-5 C). Figure 9-6 displays the spatial development of the major species and temperature for C/O = 1.0. Exothermic and endothermic zones are predicted (Figure 9-6 A), however the reforming zone is much less endothermic than with CH 4 as fuel. Gas-phase consumes some CO in the high-temperature oxidation zone, while CO 2 and H 2 O increase then decrease indicating reverse water-gas-shift and some CO 2 reforming are predicted. Interestingly, these spatial profiles are qualitatively similar to spatial profiles measured for n-octane on Rh [237]. It should be noted that the C 8 mechanism was tuned over a far greater C/O range than for the C 1 mechanism as syngas is a desired product at C/O = 1.0 and olefins are desired at C/O = 2.0. In contrast, methane CPO is always operated near 1.0 to maximize syngas production. After tuning, the surface mechanism was run with methane as fuel to determine if the mechanism still simulated methane CPO well. Ideally, as the complexity of a surface mechanism increases to accommodate larger fuels, its performance with small fuels should not deteriorate. Figure 9-7 compares the spatial profiles for CH 4 CPO generated with the C 8 mechanism against profiles generated with mechanism 1 from Chapter 4 [100]. Good qualitative agreement exists between the two mechanisms, but performance of the C 8 surface mechanism appears inferior to mechanism 1 as CO 2 reforming is predicted with the new C 8 mechanism. The C 8 mechanism presented in this section offers many opportunities for improvement. The method presented in section could give insight into kinetic parameters for higher alkane surface chemistry at various axial locations. The 279

304 assumption of thermal equilibrium between gas and surface plays a crucial role when simulating higher alkanes since temperatures are comparatively higher than with methane, and homogeneous chemistry becomes very important even at atmospheric pressure. Therefore, heat transfer should also be considered in the plug-flow models to reduce predicted surface temperature and better model gas-phase temperatures. Further modifications to the mechanism should include attention to thermodynamic consistency [238]. While the C 1 backbone of the mechanism models low temperature water-gas-shift well, it is still not thermodynamically consistent for water-gas-shift above 1000 o C. This deficiency must be improved if the distribution of H 2, CO, CO 2 and H 2 O products is to be modeled correctly with higher alkane fuels that typically run hotter than 1000 C on Rh. Adding surface pathways for olefin formation would also serve to determine the relative amounts of olefin production from the gas and surface. Tuning of the surface kinetic parameters with an automated optimization routine, similar to the technique in [238], would probably improve the fit with experiments, but care should be taken to draw meaning from the optimized values. 280

305 Table 9-1 Optimized parameters for Figure 9-3. A (s -1 ) E (kj mol -1 ) B (m s -1 K -1.5 ) H 2 8.8x x10-6 CO 1.9x x10-6 Table 9-2 Optimized parameters for Figure 9-4. A 1 (s -1 ) E 1 (kj mol -1 ) A 2 (s -1 ) E 2 (kj mol -1 ) B (m s -1 K -1.5 ) H 2 6.0x x x10-5 CO 7.9x x x

306 Table 9-3 Methane on Rh mechanism used for C 8 surface simulations. Reaction A (cm, mol, s) s o E a (kj/mol) Adsorption R1 H Rh 2 H(s) 0.01 R2 O Rh 2 O(s) 0.01 R3 CO 2 + Rh CO 2 (s) 5.0E-04 a R4 CO + Rh CO(s) 0.50 R5 H 2 O + Rh H 2 O(s) 0.1 R6 CH 4 + Rh CH 4 (s) 8.0E-03 Desorption R7 2 H(s) H Rh 3.00E R8 2 O(s) O Rh 1.30E θ 0 R9 CO 2 (s) CO 2 + Rh 1.00E R10 CO(s) CO + Rh 1.00E+13 a θ C0 R11 H 2 O(s) H 2 O + Rh 3.00E R12 CH 4 (s) CH 4 + Rh 1.00E Hydrogen Oxidation R13 H(s) + O(s) OH(s) + Rh 3.00E+23 a 83.7 R14 OH(s) + Rh H(s) + O(s) 3.00E R15 H(s) + OH(s) H 2 O(s) + Rh 3.00E R16 H 2 O(s) + Rh H(s) + OH(s) 7.00E+22 a R17 2 OH(s) H 2 O(s) + O(s) 3.00E R18 H 2 O(s) + O(s) 2 OH(s) 3.00E Carbon Oxidation R19 C(s) + O(s) CO(s) + Rh 1.00E+23 a 97.9 R20 CO(s) + Rh C(s) + O(s) 2.50E θ C0 R21 CO(s) + O(s) CO 2 (s) + Rh 1.00E+23 a θ C0 R22 CO 2 (s) + Rh CO(s) + O(s) 1.00E+23 a Pyrolytic Decomposition R23 CH 4 (s) + Rh CH 3 (s) + H(s) 3.70E R24 CH 3 (s) + H(s) -> CH 4 (s) + Rh 3.70E R25 CH 3 (s) + Rh CH 2 (s) + H(s) 3.70E R26 CH 2 (s) + H(s) CH 3 (s) + Rh 3.70E

307 Table 9-3 (cont.) Methane on Rh mechanism used for C 8 surface simulations Reaction A (cm, mol, s) s o E a (kj/mol) R27 CH 2 (s) + Rh CH(s) + H(s) 3.70E R28 CH(s) + H(s) CH 2 (s) + Rh 3.70E R29 CH(s) + Rh C(s) + H(s) 3.70E R30 C(s) + H(s) CH(s) + Rh 3.70E O Assisted Decomposition R31 CH 4 (s) + O(s) CH 3 (s) + OH(s) 1.70E R32 CH 3 (s) + OH(s) CH 4 (s) + O(s) 3.70E R33 CH 3 (s) + O(s) CH 2 (s) + OH(s) 3.70E R34 CH 2 (s) + OH(s) CH 3 (s) + O(s) 3.70E R35 CH 2 (s) + O(s) CH(s) + OH(s) 3.70E R36 CH(s) + OH(s) CH 2 (s) + O(s) 3.70E R37 CH(s) + O(s) C(s) + OH(s) 3.70E R38 C(s) + OH(s) CH(s) + O(s) 3.70E Rh site density = 2.72x10-9 mole cm -2. Pre-exponential units are s -1 for first-order reactions and cm 2 mol -1 s -1 for second order reactions. a Pre-exponential factor changed in order to fit experimental data. 283

308 Table 9-4 Added steps for C 8 CPO surface mechanism on Rh. Reaction A (cm, mol, s) s o E a (kj/mol) Adsorption R39 C 8 H Rh(s) C 8 H 18 (s) 0.4 Desorption R40 C 8 H 18 (s) C 8 H Rh(s) 1.00E R41 OH(s) OH + Rh(s) 2.00E Reaction R42 C 8 H 18 (s) + O(s) C 8 H 17 (s) + OH(s) 5.00E R43 C 8 H 17 (s) + OH(s) C 8 H 18 (s) + O(s) 5.00E R44 C 8 H 17 (s) + 6 Rh(s) 7 CH 2 (s) + CH 3 (s) a 2.00E R45 7CH 2 (s) + CH 3 (s) C 8 H 17 (s) + 6 Rh(s) b 5.00E R46 C(s) + OH(s) CO(s) + H(s) 4.00E R47 CO(s) + H(s) C(s) + OH(s) 3.00E R48 CO(s) + OH(s) CO 2 (s) + H(s) 3.00E R49 CO 2 (s) + H(s) CO(s) + OH(s) 3.00E Rh site density = 2.72x10-9 mole cm -2. Pre-exponential units are s -1 for first-order reactions and cm 2 mol -1 s -1 for second order reactions. a 1 st order in Rh(s). b 1 st order in CH 2 (s). 284

309 Figure 9-1 Spatio-temporal experiment for methane on Rh. Temperature (T) and species profiles (H 2, CH 4, CO, H 2 O, CO 2 ) at 15 points within the catalyst/heat shield stack were measured for a stepwise switch of C/O = with 20 s dwell time at each C/O. The smooth response surfaces were generated by interpolating between the data points using Matlab. Adapted from [16]. 285