Implementation of an Alternate Chemistry Library into OpenFOAM TM

Transcription

1 of an Alternate Chemistry Library into OpenFOAM TM Bernhard F.W. Gschaider 1 Markus Rehm 2 Peter Seifert 2 Bernd Meyer 2 1 ICE Strömungsforschung Leoben, Austria 2 Institute of Energy Process Enineering and Chemical Engineering, TU Bergakademie Freiberg, Germany Open Source CFD International Conference 4th/5th December bgschaid, mrehm alternatechemistrylibrary 1/47

2 Outline 1 Introduction/Motivation Current status Cantera Desired behaviour 2 Alternate Chemistry Library Transient Solver Steady Solver Other topics 3 Transient PFR Steady-state PFR 4 Results HM1 Discussion HM1 5 Summary Availibility Outlook/Future Work Acknowledgements bgschaid, mrehm alternatechemistrylibrary 2/47

3 The chemistrymodel-class Current status Cantera Desired behaviour Part of the OpenFOAM TM -distribution Solves a set of chemical reactions and returns a reaction-rate Used in the solvers dieselfoam and reactingfoam Chemical reactions can be specified in two formats: A generic OpenFOAM TM -format (allows the inclusion of non-standard reaction rates) The Chemkin TM -format which is very Fortran Both formats are converted to the same internal representation... and are solved using a number of different solvers bgschaid, mrehm alternatechemistrylibrary 3/47

4 Cantera Introduction/Motivation Current status Cantera Desired behaviour Cantera [1] is a open-source chemical kinetics package Very flexible and easy to handle Important functions for chemical species, (heterogeneous) kinetics, equilibrium, transport, diffusion, heat conduction are available and efficiently implemented Kernel written in C++; accessible via Python, Matlab, Fortran and of course C/C++ Good Chemkin TM -Lexer A set of ideal reactors is available Uses the CVODE-Solver of the Sundials Suite [2] for integration of the ODEs. bgschaid, mrehm alternatechemistrylibrary 4/47

5 Goals of this work Current status Cantera Desired behaviour As mentioned in the Milan presentation [3], Cantera would be benefitial for the development of solvers for reacting flows: Flexible exchange of chemistry engines without the necessity to change the solver Easy access to thermochemical functions Better stability (JANAF-error) Stationary chemistry solver bgschaid, mrehm alternatechemistrylibrary 5/47

6 Alternate Chemistry Library Transient Solver Steady Solver Other topics The Alternate Chemistry Library A Library that provides a framework for adding further chemistry solvers Should require only minimal changes to existing solvers Solvers should be implementation-agnostic (by using the RunTime-selection mechanism) This is achieved by replacing the chemistrymodel object (usually called chemistry) in the solvers with a autoptr<alternatechemistrymodel> Direct subclassing of chemistrymodel is not possible because necessary properties are private Usually every call chemistry.xxxx() is replaced by a chemistry().xxxx() bgschaid, mrehm alternatechemistrylibrary 6/47

7 Interface Introduction/Motivation Alternate Chemistry Library Transient Solver Steady Solver Other topics Only a small part of the original chemistrymodel-interface is needed by the solvers: solve Solve the reaction system RR Access the reaction rates tc Access the chemical time-scale Only these have to be implemented by a new chemical solver class In addition to these the methods calcdq for the calculation of dq tf Access the flow time-scale (described later) were added to the interface bgschaid, mrehm alternatechemistrylibrary 7/47

8 chemistrymodelproxy Alternate Chemistry Library Transient Solver Steady Solver Other topics Provides access to the functionality of the original chemistrymodel-class Has a private chemistrymodel-object Calls are passed through to the chemistrymodel-object solve() and tc() calcdq() implemented after example from original solvers With this chemistry model implementation the solver is equivalent to the original reactingfoam bgschaid, mrehm alternatechemistrylibrary 8/47

9 The Cantera-model Alternate Chemistry Library Transient Solver Steady Solver Other topics Implements interface to the Cantera-package Requires the thermotype to be a hmixturethermo<canteramixture> Instead of the usual hmixturethermo<reactingmixture> Currently required to have consistent calculations of the mixture and the chemistry Calculation of thermophysical properties is done in CanteraThermo a wrapper to the Cantera-class IdealGasMix Reactions are read from a cti-file (Cantera Input) Also the thermophysical data of the species. Thus the requirement on the thermotype Calculations in solve() are done via the Reactor/ReactorNet-classes of Cantera Solution of the ODEs is done via CVODE bgschaid, mrehm alternatechemistrylibrary 9/47

10 Class overwiew Alternate Chemistry Library Transient Solver Steady Solver Other topics bgschaid, mrehm alternatechemistrylibrary 1/47

11 alternatereactingfoam Alternate Chemistry Library Transient Solver Steady Solver Other topics Solver almost unchanged. Basis: reactingfoam. Remember the Chalmers PaSR-Model for turbulent combustion: κ = dt + tc dt + tc + tk Where tc is the chemical time scale and tk is the turbulent time scale. of tc()-method for the Cantera calculation necessary. For the Cantera solver no ODE solver from OF necessary (choice of the ODE solver in chemistryproperties doesn t have any effect). (1) bgschaid, mrehm alternatechemistrylibrary 11/47

12 alternatesteadyreactingfoam Alternate Chemistry Library Transient Solver Steady Solver Other topics Steady solvers are a must when simulating large-scale chemical reactors or burners. Necessary changes from the transient case: Usage of the SIMPLE-algorithm for p-u-coupling Coupling of chemistry to the flow-time Stabilization of the solution by reduction of κ if Yi old Yi new is too large (only at start-up!) bgschaid, mrehm alternatechemistrylibrary 12/47

13 Flow time scale Alternate Chemistry Library Transient Solver Steady Solver Other topics In unsteady calculations Co< 1 and integration time equals dt. For steady-state solvers Co>>1. So we limit the integration time by the residence time in the cell i: and hence we substitute dt in (1) by df : tf i = D i U i (2) κ = tf + tc tf + tc + tk We define the flow time scale tf. (3) bgschaid, mrehm alternatechemistrylibrary 13/47

14 CVODE Introduction/Motivation Alternate Chemistry Library Transient Solver Steady Solver Other topics An OF library that makes use of the CVODE-Solver of the Sundials Suite [2] was written. bgschaid, mrehm alternatechemistrylibrary 14/47

15 Transient PFR Steady-state PFR Plug flow reactor (PFR) test case Simple test case, which is easily comparable to ideal PFR models as available in CHEMKIN TM or Cantera: Ignitable premixed composition at inlet: Y H2 =.9, Y O2 =.26, Y Ar =.965, T = 16 K, p = 1 bar Tube idealized as 2D-domain x max = 1 m, y max =.1 m, 1x1 cells, inlet left patch, outlet right patch, upper and lower patch symmetry H 2 O 2 mechanism with 9 species and 27 reactions bgschaid, mrehm alternatechemistrylibrary 15/47

16 PFR alternatereactingfoam Transient PFR Steady-state PFR For the transient simulations we were interested in Behaviour of the Cantera chemistry library Choice of ODE solvers and their performance All plots were done along the x direction. The ideal Solution was obtained by calculating the time into a position with U and ρ. bgschaid, mrehm alternatechemistrylibrary 16/47

17 Transient PFR Steady-state PFR PFR alternatereactingfoam ODE-solvers (T) ODESIBS: ODE SIBS; EI: Euler Implicit; CV: CVODE T [K] CANTERA ideal PFR OF_UPFR_ODESIBS OF_UPFR_EI OF_UPFR_CV_Co=.5 CAN_UPFR x [m] bgschaid, mrehm alternatechemistrylibrary 17/47

18 Transient PFR Steady-state PFR PFR alternatereactingfoam ODE CPUTime 1 8 OF_UPFR_ODESIBS OF_UPFR_EI OF_UPFR_CV OF_UPFR_CVJ_Co=.5 CAN_UPFR CPU time [s] Simulation time [s] bgschaid, mrehm alternatechemistrylibrary 18/47

19 Transient PFR Steady-state PFR PFR alternatereactingfoam Summary Transient PFR results: With implementation of Cantera chemistry and tc() results for both chemistry engines give identical results for PFR. ODE SIBS results deviate sligtly from the others (EI, CV) (due to semi-impliciteness?) CVODE more stable than SIBS, EI but sometimes slower CVODE benefits from larger integration steps (internal time-stepping). So use higher Co, if results aren t affected. Compared to ideal reactor solvers are faster and there is a temperature difference of 5 K. Possibly induced by discretisation errors or idealisation. bgschaid, mrehm alternatechemistrylibrary 19/47

20 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam The following aspects for the runs of the steady-state simulations are of particular interest: A Performance and stability of the steady-state implementation in relation to transient case and the ideal case. B Influence of the implementation of κ C Influence of Cmix Plots are similar to the transient ones but scaling was changed. bgschaid, mrehm alternatechemistrylibrary 2/47

21 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (T) A Transient vs. steady T [K] CHEMKIN ideal PFR OF_UPFRT_CV 17 CAN_UPFRT CAN_UPFR_refine CAN_SPFRT OF_SPFRT_CV CAN_SPFRT_refine x [m] bgschaid, mrehm alternatechemistrylibrary 21/47

22 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (H2) A Transient vs. steady CHEMKIN ideal PFR OF_UPFRT_CV CAN_UPFRT CAN_UPFR_refine CAN_SPFRT OF_SPFRT_CV CAN_SPFRT_refine.85 yh2 [kg/kg] x [m] bgschaid, mrehm alternatechemistrylibrary 22/47

23 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (T) B Influence of κ T [K] CHEMKIN ideal PFR CAN_UPFR_refine CAN_SPFRT_kappa1_refine CAN_SPFRT_kappa2_refine x [m] bgschaid, mrehm alternatechemistrylibrary 23/47

24 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (H2) B Influence of κ.1.95 CHEMKIN ideal PFR CAN_UPFR_refine CAN_SPFRT_refine CAN_SPFR_kappaNeu_refine.9.85 yh2 [kg/kg] x [m] bgschaid, mrehm alternatechemistrylibrary 24/47

25 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (T) C Influence of Cmix T [K] CHEMKIN ideal PFR CAN_UPFR_refine 17 CAN_UPFR_refine_Cmix1 CAN_SPFR_kappa1_refine CAN_SPFR_kappa1_refine_Cmix=1 CAN_SPFR_kappa2_refine CAN_SPFR_kappa2_refine_Cmix= x [m] bgschaid, mrehm alternatechemistrylibrary 25/47

26 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (H2) C Influence of Cmix yh2 [kg/gk] CHEMKIN ideal PFR CAN_UPFR_refine CAN_UPFR_refine_Cmix1 CAN_SPFR_kappa1_refine.55 CAN_SPFR_kappa1_refine_Cmix=1 CAN_SPFR_kappa2_refine CAN_SPFR_kappa2_refine_Cmix= x [m] bgschaid, mrehm alternatechemistrylibrary 26/47

27 Transient PFR Steady-state PFR PFR alternatesteadyreactingfoam (OH) C Influence of Cmix CHEMKIN ideal PFR CAN_UPFR_refine CAN_UPFR_refine_Cmix1 CAN_SPFR_kappa1_refine CAN_SPFR_kappa1_refine_Cmix=1 CAN_SPFR_kappa2_refine CAN_SPFR_kappa2_refine_Cmix=1.5 yh2 [kg/gk] x [m] bgschaid, mrehm alternatechemistrylibrary 27/47

28 PFR Steady Summary Transient PFR Steady-state PFR A: Transient/Steady solutions do not agree totally but refinement reduces deviation B+C: Definition of new κ according to (3) brings no changes for small values of Cmix. But for Cmix=1 the solution seems closer to ideal solution. However, no clear conclusion could be drawn. bgschaid, mrehm alternatechemistrylibrary 28/47

29 Results HM1 Discussion HM1 Sydney Bluff-body Flame (HM1) Data from Sandia bluff-body flame HM1 conducted at Sydney university [4]. bgschaid, mrehm alternatechemistrylibrary 29/47

30 HM1 case set-up Results HM1 Discussion HM1 Wall (slip) Pressure outlet Bluff-body walls Air Inlet Fuel Inlet Symmetry axis bgschaid, mrehm alternatechemistrylibrary 3/47

31 HM1 set-up Introduction/Motivation Results HM1 Discussion HM1 Boundary Conditions: Diameters: D jet = 3.6mm; D channel = 15mm (3mm), D bluffbody = 5mm Turbulence:8.5 % (2.5%) turbulence intensity,.135 mm (5.625 mm) mixing length for jet (co-flow) Mesh: from blockmesh with 195 Inlet conditions: HM1: U jet = 118m/s U coflow = 4m/s Other models: k ɛ turbulence model, ATRMech (reduced GRI 3.) OF used KRR4 ODE solver for chemistry bgschaid, mrehm alternatechemistrylibrary 31/47

32 Results HM1 Discussion HM1 HM1 Plots & Points of Measurement Plots of radius against measurement values (Exp HM1) at different x-positions are shown together with simulation results. EDC referes to the EDCSimpleFoam solver [3]. CAN and OF mark Cantera- and OpenFOAM TM -solution K1 and K2 stands for κ 1 and κ 2 bgschaid, mrehm alternatechemistrylibrary 32/47

33 Results HM1 Discussion HM1 HM1 Plots & Points of Measurement bgschaid, mrehm alternatechemistrylibrary 33/47

34 HM1 results T Results HM1 Discussion HM1 T [K] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) T [K] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 34/47

35 HM1 results H2 Results HM1 Discussion HM1 Y_H2 [kg/kg] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) Y_H2 [kg/kg] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 35/47

36 HM1 results H2O Results HM1 Discussion HM1 Y_H2O [kg/kg] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) Y_H2O [kg/kg] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 36/47

37 HM1 results CO Results HM1 Discussion HM1 Y_CO [kg/kg] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) Y_CO [kg/kg] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 37/47

38 HM1 results CO2 Results HM1 Discussion HM1 Y_CO2 [kg/kg] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) Y_CO2 [kg/kg] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 38/47

39 HM1 results OH Results HM1 Discussion HM1 Y_OH [kg/kg] x/d_b=.26 (13mm) x/d_b=.6 (3mm) x/d_b=.9 (45mm) Y_OH [kg/kg] x/d_b=1.3 (65mm) x/d_b=2.4 (12mm) Legend 1) Exp HM1 2) EDC 3) CAN K1 4) CAN K2 5) OF K2 bgschaid, mrehm alternatechemistrylibrary 39/47

40 Discussion HM1 Results HM1 Discussion HM1 Why are all the resuls so similar? (No influence of κ-implementation or Cmix (not shown)) In case Cmix=.5 no difference in κ notable - too fast mixing as in PFR case. Furthermore, in the domain close to the inlets κ 1 in every case and influence of Cmix can only be seen further downstream where no measurements are available. Comparison with transient simulation of HM1 indicates that κ 2 seems more appropriate (see next slide) Still, further testing is needed bgschaid, mrehm alternatechemistrylibrary 4/47

41 Results HM1 Discussion HM1 HM1 comparison κ 1, κ 2 for Cmix=1 bgschaid, mrehm alternatechemistrylibrary 41/47

42 HM1 Results HM1 Discussion HM1 Flame was rendered with acceptable quality by the new solvers Steady solvers are very robust if the right ODE solvers are used (CVODE for Cantera, KRR4 for OF) Checking of dy i provides further stability and and prevents composition from being driven to unphysical states. Solutions for a finer grid (2 cells) are still not available - flame blows off very easily. bgschaid, mrehm alternatechemistrylibrary 42/47

43 Summary Introduction/Motivation Summary Availibility Outlook/Future Work Acknowledgements The implementation of the Cantera chemistry into OpenFOAM TM can be considered successful and provides: A flexible interface to access a wide range of thermophysical data,and chemistry functions Robust solvers for transient and steady-state calculations Valuable test-cases for validation and further investigation Collaborators are welcome to test the code and contribute own developments. bgschaid, mrehm alternatechemistrylibrary 43/47

44 How to get the library Summary Availibility Outlook/Future Work Acknowledgements The libraries and solvers presented here will be made available on the openfoam-extend-svn on sourceforge.net Libraries Two independent libraries: alternatechemistrymodel The general framework an the OpenFOAM TM -interface canterachemistrymodel The interface to Cantera Solvers The steady and the transient solver with the test-cases PFR and HM1 URL will be given on the Wiki Interfaces to other chemistry libraries, more solvers, critique, and improvements are encouraged bgschaid, mrehm alternatechemistrylibrary 44/47

45 Future Work Introduction/Motivation Summary Availibility Outlook/Future Work Acknowledgements There is always room for improvement: Both chemistry solvers read the same input files Converter between Cantera and OpenFOAM TM -format Make the canterachemistrymodel indepentent of canteramixture of reduction/tabulation techniques for more time-efficient solution. More combustion models. bgschaid, mrehm alternatechemistrylibrary 45/47

46 Acknowledgements Summary Availibility Outlook/Future Work Acknowledgements A part of the development and implementation work was done by the IEC in the context of the research project Grosstechnisch bertragbare Untersuchungen zur Weiterentwicklung der IGCC-Kohlekraftwerkstechnik mit CO2-Abtrennung - COORAMENT. This project is supported by the German Federal Ministry of Economics and Technology (Project ID B). Furthermore we thank N. Nordin for a valuable posting regarding the tc()- implementation, all developers who contributed to OpenFOAM TM, and people who inspired that work. bgschaid, mrehm alternatechemistrylibrary 46/47

47 Bibliography [1] DG Goodwin. Cantera: Object-oriented software for reacting flows. Technical report, California Institute of Technology, 22. [2] A.C. Hindmarsh, P.N. Brown, K.E. Grant, S.L. Lee, R. Serban, D.E. Shumaker, and C.S. Woodward. SUNDIALS: Suite of nonlinear and differential/algebraic equation solvers. ACM Transactions on Mathematical Software (TOMS), 31(3): , 25. [3] M. Rehm, P. Seifert, and B. Meyer. Towards a CFD Model for Industrial Scale Gasification Processes. In OpenFOAM Workshop 28, Milan, 28. [4] BB Dally, AR Masri, RS Barlow, and GJ Fiechtner. Instantaneous and Mean Compositional Structure of Bluff-Body Stabilized Nonpremixed Flames. Combustion and Flame, 114(1-2): , bgschaid, mrehm alternatechemistrylibrary 47/47