Regenerator design study combining numerical simulations and statistical tools

Transcription

1 14th Int. Seminar on Furnace Design - Operation & Process Simulation June Regenerator design study combining numerical simulations and statistical tools Z. Habibi, F. Bioul AGC Glass Europe Technovation Center Gosselies, Belgium 1

2 Introduction AGC and float glass We use heat from combustion for glass production: To melt raw materials and to refine the glass melt Raw materials Total energy Input = MW Molten glass to tin bath (float) (1100 C) Combustion air is preheated by means of regenerators 2

3 Design optimization What is the best design? In term of what? Is there only one? Could we reach it? How to reach it? 3

4 Regenerator optimization Key challenges Build a simple realistic model estimating regenerator efficiency Include in the model information related to clogging/ageing Objective Build a simplified tool in order to optimize regenerator design under constraints 4

5 Outline Methodology Numerical model, description and validation Statistical analysis and optimization Conclusions 5

6 Methodology: General overview in 5 steps Numerical validation Industrial measurements Statistical analysis Optimization Best possible designs DOE & Numerical simulations Model setup and validation Parameters identification Height Width Steam Insulation Design of experiment & validation 6

7 Model setup and validation 7

8 Study description Simulations Study based on a complete model Separate/Unique/Twin designs considered 7 burners Including combustion space 8

9 Assumptions and model inputs Steady simulation Glass No glass model (To reduce simulation cost) Fixed glass and batch T Incl. batch gases Combustion Global power is a tuning parameter Fire curve is fixed: gas flow rate distribution is fixed Lambda or air/gas ratio is fixed Air flow distribution Fixed for separated chambers Calculated for unique chambers Fumes distribution calculated Fixed boundary condition 9

10 Assumptions and model inputs Regenerators No detailed checkers geometry: approximated by porous wall model Reversal process modelling: Quasisteady simulation - Transient regenerator simulation is time consuming i.e. Checkers T is time averaged between firing and exhaust sides Porous wall T Fluid T 10

11 Model validation Generally good agreement between model and measurements for Eff T Air, out 1 TFumes, in T Air, out T Fumes, in 11

12 Identification of Parameters & Evaluation Criteria 12

13 Study description - Parameters Parameters included in the study Regenerator Configuration Unique / Separated / Twin Flow distribution Front / Back / Front & Back / Side Regenerator Insulation Fumes flow rate Selected parameters during the study to generate the optimisation tool Other parameters have not been included because either less impacting or with lower priority 13

14 Study description - Evaluation criteria Eff Eff 1 2 T T H H Selected evaluation criteria Air, out Fumes, in Air, out Fumes, in ( T ( T Regenerator efficiency (can reach 100%) Air, out Fumes, ) Cp ) in Cp Air Fumes q q Air Fumes T Air, out TFumes, in (usual, max ~ 70%) T Air, out H Air, out T Fumes, in H Fumes, in Combustion efficiency Clogging index cf. bellow. Eff comb Q Q Glass combustion Q combustion Q Glass 14

15 Study description - Clogging index Proposed sulfate clogging index Based on literature (Beerkens), a simplified index which can estimate deposition rate of sodium sulfate to the checkers has been defined An approximation of clogging function is implemented in GFM Thanks to the «field manipulation» tool of GFM Validation To confirm correlation of the index & reality, alkali concentration level difference between top and bottom level of the regenerator have been measured and compared with estimated index. Expectation: dc should be relatively smaller c:alkali concentration level dc Expectation: dc should be relatively larger due to more alkali loss by creating sodium sulfate deposition to the checkers 15

16 Design of Experiment Numerical simulation 16

17 Methodology: Design Of Experiment Step by step approach Step Goal Information Model Screening Modeling Selection key parameters First model (with most relevant key parameters) - General trend - Design well centered - KO for optimization - Full model - Interaction - OK for optimization = + = + + Sequential methodology to build model step by step by keeping previous trials Set of experiments Evaluation Criteria Case # Height Width N Case # N Efficiency Heat to glass Clog = + Y =

18 Methodology: DOE & Numerical simulation Set of experiments Evaluation Criteria Case # Height Width N Case # N Efficiency Heat to glass Clog = + Y = + + Numerical simulations Based on complete furnace model (incl. Combustion) 18

19 Study description Simulations Simulation cases Evaluation criteria: Regenerator efficiency (Temperature, Enthalpy) TRE =,, HRE = Combustion efficiency,, COMB = Sulfate clogging index Remark: Range of combustion efficiency 10% Significant impact 1) Selection of most impacting parameters 2) Generation of the simplified model 19

20 Statistical analysis 20

21 Statistical Analysis Set of experiments Evaluation Criteria Case # Height Width N Case # N Efficiency Y Heat to glass Clog = + = + + Numerical simulations Statistical tools Based on complete furnace model (incl. Combustion) Used to define the set of experiments and generate the simplified model 21

22 Statistical analysis Interactions Effect of one parameter depends on the level of other parameters Final approximation =

23 Physical understanding Conclusions are also derived from physical interpretation of the results Example: Fumes distribution between different chambers configuration 23

24 24 Best possible designs

25 Optimization Set of experiments Evaluation Criteria Case # Height Width N Case # N Efficiency Y Heat to glass Clog = + + Numerical simulations Statistical tools Optimization under constraints Based on complete furnace model (incl. Combustion) Used to define the set of experiments and generate the simplified model 25

26 Optimization Optimum is a balance between efficiency and ageing Efficiency Ageing Results show that there is not only one absolute optimum Optimization depends on constraints Dimensions (cold repair, building size ) Constraint 2 Civil work cost Refractories cost Optimization is constantly evolving in term of geo-economical factors, Investment strategies, know-how Low efficiency Constraint 1 Higher efficiency 26

27 Conclusions Thanks to numerical simulation (CFD) we were able to study the impact of 12 parameters simultaneously on regenerator efficiency and clogging index (Impossible to realize without simu). GFM simulation tool has a good ratio Accuracy x CPU-cost x Model setup Design of experiments allows us to optimize the number of simulation runs Statistical analysis shows the impact of the parameters and their interactions on regenerators efficiency. allows the building of a simplified function of efficiency in term of regenerator parameters Physical interpretation of the results improves our understanding of regenerators Optimization of the efficiency depends on constraints: civil works, refractories cost etc. no unique optimal design 27

28 Thank you for your attention Questions? 28