|Title||Demonstrated Implementation of Detailed Chemistry into CFD Simulations for Tailored Commercial Evaluation of Industrial Combustion Equipment|
|Contributor||Wang, D.H.; Adams, B.R.|
|Spatial Coverage||Houston, Texas|
|Subject||2014 AFRC Industrial Combustion Symposium|
|Description||Paper from the AFRC 2014 conference titled Demonstrated Implementation of Detailed Chemistry into CFD Simulations for Tailored Commercial Evaluation of Industrial Combustion Equipment by M. Cremer.|
|Abstract||CFD simulation of industrial combustion equipment for trouble shooting, optimization, and design related to emissions or performance has become an ever increasingly utilized tool along with testing and empirically based methods. Depending on the fuels and the conditions under which they are burned, the chemical kinetics involved in the combustion process along with the turbulent mixing and heat transfer must be sufficiently accurate in order to predict performance and emissions at a level useful for decision making, but also balance model complexity and sophistication with existing computational resources and time. This paper will discuss how CFD models have been developed, tailored, and utilized to address combustion related problems associated with: 1) NOx and CO emissions in low NOx process heaters, 2) quality and extent of reaction in hydrocarbon reforming applications, and 3) heat transfer and emissions associated with an air to oxy conversion in special purpose steam generators. All three applications involve the use of REI's internally developed Reynolds Averaged Navier Stokes (RANS) based ADAPT CFD code. ADAPT accommodates local mesh refinement based on an unstructured Cartesian based grid. The associated chemical mechanisms that have been incorporated into these simulations are based on a robust mechanism reduction approach using quasi steady state assumptions and use of the Computer Assisted Reduction Method (CARM). The paper will describe how this approach yields robust and accurate representation of the problem specific detailed chemistry within the CFD simulations. In all three examples, parametric simulations are carried out on a time scale short enough to support design efforts within actual commercial projects using REI's in-house computer resources.|
|Rights||No copyright issues exist.|
AFRC 2014 Industrial Combustion Symposium - 1 - REACTION ENGINEERING INTERNATIONAL Demonstrated Implementation of Detailed Chemistry into CFD Simulations for Tailored Commercial Evaluation of Industrial Combustion Equipment Cremer, M.A., Wang, D.H., Denison, M.K., Adams, B.R. Reaction Engineering International http://www.reaction-eng.com Abstract CFD simulation of industrial combustion equipment for trouble shooting, optimization, and design related to emissions or performance has become an ever increasingly utilized tool along with testing and empirically based methods. Depending on the fuels and the conditions under which they are burned, the chemical kinetics involved in the combustion process along with the turbulent mixing and heat transfer must be sufficiently accurate in order to predict performance and emissions at a level useful for decision making, but also balance model complexity and sophistication with existing computational resources and time. This paper will discuss how CFD models have been developed, tailored, and utilized to address combustion related problems associated with: 1) NOx and CO emissions in low NOx process heaters, and 2) impacts of combustion modifications for NOx control in a gas fired power boiler. The two applications involve the use of REI's internally developed Reynolds Averaged Navier Stokes (RANS) based ADAPT CFD code. ADAPT accommodates local mesh refinement based on an unstructured Cartesian based grid. The associated chemical mechanisms that have been incorporated into these simulations are based on a robust mechanism reduction approach using quasi steady state assumptions and use of the Computer Assisted Reduction Method (CARM). The paper will describe how this approach yields robust and accurate representation of the problem specific detailed chemistry within the CFD simulations. In both examples, parametric simulations are carried out on a time scale short enough to support design efforts within actual commercial projects using REI's in-house computer resources. 1. Introduction Since the time of the clean air act amendments in the 1970s, air emissions from industrial equipment have become increasingly regulated and air emission standards have become increasingly stringent. Strict emissions requirements for NOx and CO require either post combustion or combustion based approaches for achieving the specified standards. In many cases, post combustion remedies such as selective catalytic reduction (SCR) are very expensive to retrofit to existing equipment, so relatively less proven, but less expensive combustion modifications are often considered. Low NOx burner strategies can provide a relatively inexpensive solution to NOx reduction requirements, but reductions must be achieved without compromising the heat release profiles or flame ‘quality' of the burners. It is not uncommon for ultra-low NOx burners to have wider and longer flames than conventional burners or even, in the worst case, for flames to impinge directly on the heat absorbing surfaces. Similarly, the delayed combustion that is associated with air staging, can lead to elevated CO emissions. AFRC 2014 Industrial Combustion Symposium - 2 - REACTION ENGINEERING INTERNATIONAL CFD modeling provides a potentially cost-effective method for evaluating emissions and furnace performance for new burner technologies, thus minimizing furnace start-ups times and unforeseen performance impacts. However, accurate predictions of emissions requires integration of several key capabilities within the CFD tool including: · sufficient computational points to resolve the detailed burner geometry, multiple fuel mixing zones, flame interactions between burners, and full furnace geometry; · combustion sub-models to predict fuel-lean, premixed, turbulent combustion; · radiative heat transfer sub-models to describe gas-wall-tube heat exchange; · flow and chemistry sub-models to account for turbulence-chemistry interactions; and · finite-rate chemical kinetics sub-models to account for NOx reactions. Developing software that provides all of these capabilities in a single modeling tool is challenging due to limitations in computer capacity and descriptions of detailed turbulence and chemistry. This paper describes the application of ADAPT, REI's CFD software that was specifically designed to model gas fired premixed and non premixed combustion in ultra-low NOx burners in process furnaces and boilers. This code uses Adaptive Mesh Refinement (AMR) to capture the fine detail of the burner and near-burner mixing. Two turbulent combustion models - an Eddy- Dissipation-Concept (EDC) model and a joint Probability Density Function (PDF) model are implemented, which allow for using finite-rate chemistry in CFD calculations. An in-situ Adaptive Tabulation (ISAT) algorithm is used to accelerate the computation of reaction source terms. The radiative heat transfer is calculated using a discrete ordinates method. Commercial application of this software to the evaluation of emissions and performance issues in utility gas boilers and process heaters is discussed. 2. Description of CFD Software - ADAPT This section provides a relatively brief overview of the fundamental basis of the ADAPT software and the typical sub models used in ADAPT for commercial application to simulations of industrial problems. Additional details regarding the basis of ADAPT are provided elsewhere.1,2 2.1 Unstructured Grid Flow Solver with Adaptive Mesh Refinement The governing equations are the Favre-averaged Navier-Stokes equations. The k-e turbulence model is employed to close the solution for the Reynolds stress terms. An AMR flow solver developed by REI is used for the 3-D finite volume solver. The solver is an adaptive grid code that uses an orthogonal grid with local mesh refinement. A structured base grid is input and grid refinements are performed. Cells that are tagged for refinement are divided into eight smaller cells. Cells are tagged for refinement either manually via user input, automatically to resolve 1 Q. Tang, M. Denison, B. Adams, D. Brown, "Towards comprehensive computational fluid dynamics modeling of pyrolysis furnaces with next generation low-NOx burners using finite-rate chemistry," Proceedings of the Combustion Institute, 32 (2009) 2649-2657. 2 Q. Tang, M. Bockelie, M. Cremer, M. Denison, C. Montgomery, A. Sarofim, "Towards comprehensive CFD modeling of lean premixed ultra-low NOx burners in process heaters," AIChE 2005 Spring National Meeting. AFRC 2014 Industrial Combustion Symposium - 3 - REACTION ENGINEERING INTERNATIONAL geometry, or based on gradients in field variables in the solution. A typical AMR grid is shown in Figure 1. The flow model of the AMR code uses a SIMPLEM algorithm3. The momentum equations are solved in a collocated manner so that the velocity field is stored at the cell centers and velocities from the discretized equations, without the pressure gradient source term, are interpolated to the cell faces via a momentum interpolation. The pressure field, also stored at cell centers, is solved from the continuity equation discretized over the cells. The resulting pressure gradient sources are then added to the interpolated face velocities resulting in a mass satisfying flow field at the faces. The discretized partial differential equations are solved using an algebraic multi-grid (AMG) strategy. 2.2 Turbulent combustion models The use of a Favre-averaged Navier-Stokes approach for simulation of turbulent reacting flows requires the calculation of time-averaged chemical source terms. Simplified approaches involve the use of either "fast mixing" or "fast reaction" assumptions. A fast mixing assumption allows the time average of the reaction rate to be represented by the Arrhenius function of the average concentrations and temperature. A fast reaction assumption allows the mean temperature and species concentrations to be determined from simple scalars such as mixture fractions. In situations where the mixing and chemistry time scales are of the same order of magnitude, neither of these simplifications are valid. In ADAPT, two turbulent combustion models have been implemented: 1) the EDC model, and 2) the joint scalar PDF model. Both models are capable of incorporating multi-specie reaction mechanisms and without changing formulations, the models are applicable to both non-premixed and premixed combustion. Both combustion models have been tested and compared with data obtained from a single burner test facility firing an ultra low-NOx burner1. Although the results showed some improved accuracy in NOx and CO emissions predictions with the joint scalar PDF model, they also showed a factor of four increase in run-time compared with the EDC model. Subsequent commercial application of the EDC model in ADAPT has shown that the small potential inaccuracy associated with use of the EDC model is an acceptable trade-off with the improved turn-around time. 3 S. Acharya, and F.H. Moukalled, Numerical Heat Transfer, Part B, 15:131-152, 1989. Figure 1. ADAPT solution for an injector study. AFRC 2014 Industrial Combustion Symposium - 4 - REACTION ENGINEERING INTERNATIONAL The EDC model implemented in ADAPT follows the idea outlined by Gran and Magnussen4 and has been widely used for combustion modeling. The model gives an empirical expression for the mean reaction rate based on a steady perfectly stirred reactor (PSR) concept, which assumes that chemical reactions take place in turbulent fine structures whose characteristic dimensions are of the order of the Kolmogorov length scale. The fluid state is determined by the fine structure state, the surrounding state and the fraction of fine structures. In ADAPT, a novel pseudo time-splitting scheme is developed to solve the scalar transport equations. 2.3 Combustion chemistry Detailed chemical mechanisms for hydrocarbon combustion are becoming increasingly available and accurate. However, the direct use of these mechanisms, which typically involve of the order of 30 species for methane, and up to 1000 species or more for higher hydrocarbons, is computationally prohibitive in turbulent flow computations in which the 3D composition fields have to be represented either directly or statistically. In ADAPT, a combination of dimension reduction and storage/retrieval is adopted to produce a tractable and efficient computation of combustion chemistry. The dimension reduction strategy is the reduced mechanism constructed using the quasi-steady-state assumption where given a detailed mechanism with ns chemical species, a small number of nc species are retained to represent the dynamics of the detailed system, while the other nss = ns - nc species that are associated with fast processes are assumed to be in steady state with their net chemical production rates being set to zero. Then, in the turbulent combustion calculation, the relevant equations are solved for the nc unsteady-state species instead of for the ns species. Because the computational cost increases at least linearly with the number of species represented, substantial gains can be achieved for nc << ns. The technique has been used extensively for hydrogen and methane combustion. The computer assisted reduction method (CARM) is a computer code that automates this process5. It uses a set of input test problems, typically PSR and PFR solutions, to evaluate the error in the steady state assumption for each species in the detailed mechanism under conditions of interest. REI has developed a fully integrated computational problem solving environment (PSE) based around CARM for creating, testing and optimizing reduced chemical kinetic mechanisms for combustion simulation and implementing these mechanisms into commercial and research CFD codes6. The CARM-produced reduced mechanisms represent the original detailed mechanisms faithfully over a broad thermo-chemical parameter space. REI has been quite successful in applying reduced mechanisms for simulations of combustion and pollutant formation for a range of fuels including coal, oil, and high order hydrocarbons7,8,9. 4 I.R. Gran, B.F. Magnussen, Combust. Sci Technol. 119 (1996) 191-215. 5 J.-Y. Chen, Workshop on Numerical Aspects of Reduction in Chemical Kinetics, CERMICS-ENPC, Cite Descartes -Champus sur Marne, France, Sept. 2, 1997. Available at http://www.reaction-eng. com/downloads/carmpaper.pdf 6 Montgomery, C. J., Swensen, D. A., Harding, T. V., Cremer, M. A., and Bockelie, M. J., A Advances in Engineering Software, 33:59-70, 2002. 7 Cremer, M.A., Montgomery, C.J., Wang, D.H., Heap, M. P., Chen, J.-Y., Proc. Combust. Inst. 28:2427-2434, 2000. AFRC 2014 Industrial Combustion Symposium - 5 - REACTION ENGINEERING INTERNATIONAL The storage/retrieval strategy adopted in ADAPT is the insitu adaptive tabulation (ISAT) algorithm. ISAT is a storage/retrieval algorithm for the efficient implementation of combustion chemistry. In a typical turbulent combustion calculation, the combustion chemistry has to be computed at least 109 times. The idea of storage/retrieval is to avoid performing these expensive chemistry calculations 109 times by storing the information from many fewer calculations, and then to retrieve the information 109 times. A detailed description of the ISAT algorithm can be found elsewhere10,11. Some of the important features of ISAT as implemented into ADAPT are as follows: · The table is built up in situ (as the turbulent combustion calculation is performed) rather than in a pre-processing step. In this way, only the region of the composition space that is accessed in the calculation is tabulated. · Interpolation in the table is performed using a linear approximation (retrieval). · The interpolation error ε is controlled relative to a specified tolerance εtol. If a table entry is found such that ε is less than εtol, the linear approximation is used (retrieval). Otherwise, a growth of the ellipsoid of accuracy is performed, or a new table entry is generated (addition). 2.4 Radiative heat transfer Radiation is typically the most significant mode of heat transfer in practical combustion equipment. Accurately simulating radiation heat transfer to specific regions in a system requires a model that can account for both absorbing-emitting radiation processes and complex system geometries, including arbitrary structures such as convective tube passes. REI's model utilizes the discrete-ordinates method (DOM) that has proven to be a viable choice for modeling radiation in combustion systems, both in terms of computational efficiency and accuracy12. The model has been implemented into the ADAPT flow solver. When the EDC model is in use, the mean enthalpy transport equation includes a radiative energy source term calculated by the DOM model. However, the effects of radiative heat transfer on the fine scale structure is neglected. 3. Application to Simulation of Industrial Combustion Equipment Pilot scale verification of the ADAPT software has previously been carried out through comparisons with temperature and species measurements in the Sandia D flame2 and in a single burner test furnace1 (SBTF) located at John Zink Company's test center in Tulsa, OK. Since that time, REI has applied the ADAPT software to simulation of numerous full scale furnaces and boilers in order to predict performance and emissions. These applications have been carried out 8 Cremer, M.A., Wang, D.H., Montgomery, C.J., Adams, B. R., 2001 Joint AFRC/JFRC/IEA International Combustion Symposium, Kauai, Hawaii, September 9-12, 2001. 9 Montgomery, C. J., Cremer, M. A., Chen, J.-Y., Westbrook, C. K., and Maurice, L. Q., Journal of Propulsion and Power, 18:192-198, 2002. 10 Pope, S.B., Combust. Theo. Modelling, 1:41-63, 1997. 11 Fox, R.O., Computational Models for Turbulent Reacting Flows, Cambridge University Press, Cambridge, 2003. 12 Adams, B.R., and Smith, P.J., Combust. Sci. and Tech., 109:121-140, 1995. AFRC 2014 Industrial Combustion Symposium - 6 - REACTION ENGINEERING INTERNATIONAL in order to solve industrial problems on a commercial time scale requiring case turn-around times on the order of several days to a week rather than weeks to months. REI typically uses in house computer resources for these projects, although cloud computing resources have been utilized in cases of high demand. Typical ADAPT simulations of furnaces and boilers are carried out using up to six processors and requiring up to 20 GB of RAM. Case sizes typically range from 1 to 3 million grid cells. Two examples of these applications are discussed in this section. 3.1 Simulation of Tangentially Fired Gas Boiler The first example involves application of ADAPT to the simulation of a natural gas fueled tangentially fired boiler with a heat input of approximately 800 MBtu/hr. The focus of the evaluation was the investigation of proprietary burner modifications in combination with the use of targeted water injection for NOx reduction. Initial CFD based evaluations carried out by others proved unsatisfactory with regards to the predicted NOx emissions in that the impacts of the burner modifications and targeted water injection on NOx emissions were overestimated. REI was contracted to work with the NOx control equipment supplier to use CFD based modeling of this boiler to better understand the boiler behavior limiting the NOx reduction. Figure 2 shows the overall geometry of the radiant section of the boiler. The burner columns are slightly offset from the corners and the two elevations of gas burners are located on the two side walls. Figure 3 shows the arrangement of the gas burners in the baseline (pre-retrofit) geometry CFD Simulation Domain Furnace Depth: 6706 mm Furnace Width: 7315 mm Figure 2. Overall geometry of simulated natural gas fueled T-fired boiler. AFRC 2014 Industrial Combustion Symposium - 7 - REACTION ENGINEERING INTERNATIONAL Gas nozzle 45o swirler Figure 3. Baseline gas burner arrangement showing the actual and the simulated geometry along with the model representation. Simulated burner modifications associated with the proprietary retrofit geometry included changes to both the burner air supply as well as the gas poker arrangement. The ADAPT simulations of the baseline and retrofit conditions utilized computational grids with a total of 1.5 million cells. As shown in Figure 2, the inlets for secondary air were located upstream of the burners within the air supply registers to predict the impact of the duct turns on the air velocity distribution and subsequent mixing with the fuel. The fuel injectors were represented in great detail with grid resolution sufficient to accurately represent fuel tip inlet areas to within 1%. The adaptive mesh refinement within AMR allows for multiple levels of grid refinement within the burner region and subsequently less refinement away from the burners where gradients are smaller. The vertical model exit plane was located downstream of the rear wall screen tubes, where temperatures are low enough that NOx formation and destruction rates are negligible. A uniform thermal resistance and fixed backside temperature boundary condition was used to represent the waterwalls. The reduced mechanism that was used for these simulations is a 20 specie mechanism (H2, H, O2, OH, H2O, HO2, H2O2, CH3, CH4, CO, CO2, CH2O, CH3OH, C2H2, C2H4, C2H6, NH2, NO, HCN, N2) developed with CARM from the detailed mechanism of GRI Mech 3.0. This mechanism has been well validated for fuel lean combustion conditions involving natural gas and blends of natural gas and hydrogen, including NOx emissions from these flames. AFRC 2014 Industrial Combustion Symposium - 8 - REACTION ENGINEERING INTERNATIONAL Figure 4 shows the predicted average results of gas temperature, CO, O2, and NOx near the radiant furnace exit for the baseline simulation. Predictions were compared with measurements obtained through HVT measurements along a single traverse near the radiant exit as shown in Figure 5. This comparison represents the results of a blind test in which CFD predictions were reported prior to knowledge of the HVT test result, so no tuning of the CFD predictions was carried out. Qualitative and quantitative agreement is very good. Table 1 shows the comparison of Plane 1 - HVT Elevation Plane 2 - Model Exit Baseline Plane 1 Gas Temperature, oC 1,370 CO Concentration, ppm, dry 285 O2 Concentration, vol. % dry 1.68 NOx Concentration, ppm dry 108.7 Model Exit Plane Gas Temperature, oC 1,300 CO Concentration, ppm dry 101 O2 Concentration, vol. % dry 1.58 NOx Concentration, ppm dry 109.3 NOx Concentration@3% O2, ppm dry 101.3 Figure 4. Predicted furnace exit results for baseline conditions. 1200 1250 1300 1350 1400 1450 1500 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Gas Temperature (oC) Depth into boiler HVT Measurement Prediction North South 0 20 40 60 80 100 120 140 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% NOx Concentration (ppm, dry) Depth into boiler HVT Measurement Prediction North South 0 200 400 600 800 1000 1200 1400 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% CO Concentration (ppm, dry) Depth into boiler HVT Measurement Prediction North South 0 0.5 1 1.5 2 2.5 3 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% O2 Concentration (%, dry) Depth into boiler HVT Measurement Prediction North South Figure 5. Comparisons of model predictions and HVT measurements along a single traverse (as shown by red line in the image at top-left) for baseline conditions. AFRC 2014 Industrial Combustion Symposium - 9 - REACTION ENGINEERING INTERNATIONAL predicted and measured average values at the HVT measurement plane. Very good quantitative agreement was seen, particularly for the NOx emissions. Table 1. Comparison of average measured and predicted furnace exit conditions for the tangentially fired boiler for baseline conditions Measured Predicted Temperature (oF) 2526 2498 CO (ppm,dry) 266 285 NOx (ppmv, 3%O2, dry) 94 101 Further verification of the results of the CFD model were obtained through comparison of model predicted NOx emissions for the first burner retrofit condition. Both the boiler NOx measurements and the CFD model predictions yielded 82 ppmv, 3% O2, dry, in remarkable agreement. In comparison with the baseline NOx emissions, the CFD model showed a predicted NOx reduction associated with the retrofit configuration of 18% while the measurements showed 12%. Again, this agreement is quite good in that the CFD model showed relatively poor overall NOx reduction associated with the combustion modifications, in agreement with the measurements. Subsequent simulations were carried out for this boiler to evaluate alternate burner modifications and the addition of targeted water injection for NOx control. The concept of the targeted water injection is to inject and vaporize water in the regions of peak heat release and high temperature where thermal NOx formation rates are highest. Figure 6 shows the results of the parametric simulations for two different combustion retrofit conditions with and without targeted water injection. As can be seen with the retrofit simulations, NOx reduction was predicted to improve with the subsequent burner modifications as well as with the addition of water, but at the expense of increased CO and gas temperature at the furnace exit. It can be seen that the predicted incremental NOx reduction associated with the water injection was 5.2% and 7.0%, respectively for cases 2 and 3. The increased benefit of the water injection in case 3 was found to be partly Plane 1 - HVT Elevation Plane 2 - Model Exit Baseline Retrofit 2 Retrofit 2_H2O Retrofit 3 Retrofit 3_H2O Plane 1 Gas Temperature, oC 1,370 1,434 1,435 1,427 1,435 CO Concentration, ppm, dry 285 5,041 3,350 5,217 5,424 NOx Concentration, ppm dry 108.7 83.7 79.8 84.9 78.7 Model Exit Plane Gas Temperature, oC 1,300 1,372 1,367 1,374 1,368 CO Concentration, ppm dry 101 2,438 1,932 2,324 2,189 O2 Concentration, vol. % dry 1.58 1.21 1.18 1.21 1.19 NOx Concentration, ppm dry 109.3 85.9 80.0 87.0 79.1 NOx Concentration@3% O2, ppm dry 101.3 77.3 72.0 78.3 71.2 NOx Reduction@3% O2, ppm dry N/A 23.7% 28.9% 22.7% 29.7% Figure 6. Predicted furnace exit results for parametric simulations. AFRC 2014 Industrial Combustion Symposium - 10 - REACTION ENGINEERING INTERNATIONAL due to shortening of the flames in case 3 due to the combustion modifications, leading to more effective targeting of the injected water into the high temperature flame zone. Figure 7 shows the predicted spatial relationship between the injected water droplet trajectories and the injected fuel. As shown, the majority of the injected water vaporizes within the envelope of the injected fuel, thereby limiting the quantity of water vaporization that is occurring in the high temperature shear layer where fuel is mixing with combustion air. The modeling also showed that the tangentially fired geometry can increase the challenge associated with targeting water into the high heat release zones due to the degree of burner-to-burner interactions that occur, where combustion air and flue gases from the "upstream" burners mix with fresh fuel and air in the downstream burners. 100% 0% H2O Remaining in Droplet iso-CH4 mole fraction = 0.3 0.4 0 Iso-CH4 colored by CH4 mole fraction iso-CH4 mole fraction = 0.2 iso-CH4 mole fraction = 0.1 Figure 7. Spatial relationship between targeted water droplets and injected fuel for one of the retrofit simulations The ADAPT simulations were carried out in parallel fashion running on 4 processors on a 40 processor linux cluster with 64 bit AMD opteron processors equipped with 224 GB of RAM. The simulations that did not incorporate water injection required approximately 6 days of CPU time to obtain converged solutions including NOx. Simulations including water injection required approximately 8 days of CPU time. These times can be reduced by approximately 40% if NOx predictions are not required. 3.2 Simulation of Ethylene Furnaces An application of significant focus for ADAPT has involved simulation of radiant heaters for ethylene furnaces. REI has worked extensively with Technip Stone and Webster and ethylene AFRC 2014 Industrial Combustion Symposium - 11 - REACTION ENGINEERING INTERNATIONAL producers to use CFD modeling to evaluate burner and furnace design issues within new furnaces as well as to trouble shoot existing problems with performance or emissions. The increasing regulation of NOx emissions from these furnaces has led to the growing necessity for accurate predictions of NOx emissions from these, particularly during the design phase. This necessity has largely motivated the REI development of the ADAPT software and the physical and chemical models that it includes, particularly with regards to turbulence-chemistry interactions and the use of accurate chemical mechanisms. Since REI's initial development of ADAPT and verification with results and measurements from pilot scale modeling, REI has applied ADAPT to over 25 different ethylene furnaces. These furnaces have been equipped with floor and wall burners designed by multiple burner suppliers typically incorporating low NOx combustion strategies such as air and fuel staging, in-furnace flue gas recirculation, and partial fuel premixing. Overall, the application of the model to the large number of units fired with differing burner configurations and fuels ranging from refinery gas, natural gas, and high hydrogen containing gases has proven the reliability of the predictions regarding heat transfer distributions to the process fluid and radiant efficiency, flame rollover and overall flame shape, furnace effects such as shield walls and end wall effects, as well as CO and NOx emissions. In simulation of an ethylene heater, REI's typical approach involves set up and simulation of a single burner column model involving one narrow section of the radiant furnace centered on one pair of floor burners (on either side of the process coils) as well as any wall burners located above, as shown in Figure 8. In this model, significant focus is placed on construction of the Single Burner Column Model Furnace Model Furnace Symmetry Both Symmetries Figure 8. Model representation of the single burner column model and furnace model of an ethylene furnace AFRC 2014 Industrial Combustion Symposium - 12 - REACTION ENGINEERING INTERNATIONAL computational mesh associated with the burners (see Figure 9) as well as representation of the coil geometry in relationship to the burners. This model is useful in verifying acceptable representation of the burners and local furnace geometry prior to devoting significant CPU time to simulation of the full furnace. In addition, the single burner column model is useful for prediction of combustion air distribution between floor and wall burners depending on their respective damper settings, as shown in Figure 10. Since this is a reacting flow model, this approach ensures that the local pressure distribution that develops as a result of combustion within the furnace is taken into account in the calculation of induced air flow rates. Figure 9. CFD geometry of a lean premixed low NOx burner and local mesh refinement Gauge Pressure (mm of w.c) -7.5 mm of w.c specified Figure 10. Predicted pressure distribution as it is related to prediction of air distribution between floor and wall burners in induced draft burners AFRC 2014 Industrial Combustion Symposium - 13 - REACTION ENGINEERING INTERNATIONAL Once the set up associated with the single burner column is verified, we proceed with simulation of the full furnace. Rather, it is typical that a one-half furnace model is established with a symmetry plane through the center. In most ethylene furnace geometries, a one half furnace model is a complete representation of the full furnace given that geometrical symmetry is typical. Although operational issues may lead to operational asymmetry, this behavior would not be represented in the model unless the boundary or inlet conditions leading to this asymmetry were imposed. Items of particular importance for furnace simulations are prediction of any burner to burner interactions or furnace geometry issues that could create unforeseen deleterious impacts on heat flux distribution or emissions. Thus, the analysis of full furnace model results includes contour plots and iso-surfaces of gas temperature and CO distributions to evaluate tendency for flame rollover or flame impingement on the process tubes. Figure 11 shows predicted CO isosurfaces for a 40 floor burner and 40 wall burner fired ethylene heater, where flame rollover is not a concern. Gas Temp (oC) Figure 11. Predicted CO isosurfaces colored by gas temperature in a floor and shelf fired ethylene heater In some circumstances, even though flame rollover may not be occurring, an imbalance between floor and wall firing and/or between floor and wall burner stoichiometries may lead to excessive local heat flux and tube metal temperatures. Figure 12 shows the predicted incident radiative heat flux distribution to the process tubes under such a circumstance. AFRC 2014 Industrial Combustion Symposium - 14 - REACTION ENGINEERING INTERNATIONAL Incident Heat Flux (W/m2) > Symmetry Plane Figure 12. Predicted incident heat flux to the process tubes where the floor vs. wall firing distribution is out of balance With regards to NOx predictions, the commercial simulations have shown agreement within 10% of measured emissions. This level of agreement is considered to be acceptable, given differences in designed geometry and operation vs. actual operation. This level of quantitative accuracy has also been found acceptable and useful for design and trouble shooting associated with burner and furnace operation. For simulation of the half furnace associated with 40 floor and 40 wall burner ethylene heater shown in Figure 11, a computational grid of approximately 3.3 million cells was used. Simulations were carried out on four processors of a 64 bit Intel quad-core Xeon workstation with 100 GB of RAM. Each simulation converged in approximately 8 days, including NOx calculations. 4. Conclusions and Future Work In this paper an advanced CFD modeling tool for simulating ultra-low NOx burners in process furnaces and boilers was presented. The tool combines state-of-the-art techniques for comprehensive modeling of practical turbulent combustion problems, including: a reduced chemical kinetic mechanism for describing finite-rate chemical kinetics; an in-situ adaptive tabulation algorithm for incorporating realistic combustion chemistry accurately and efficiently; and an adaptive mesh refinement finite volume flow solver for treating complex geometries of AFRC 2014 Industrial Combustion Symposium - 15 - REACTION ENGINEERING INTERNATIONAL practical combustion equipments. Other models such as a radiation heat transfer model based on a discrete ordinate method were also developed to be used in the simulations. ADAPT has been successfully applied to over 25 ethylene furnaces as well as gas fired power boilers and other industrial furnaces. Particular focus has been devoted to the development and implementation of reduced mechanisms based on the steady state species approach including NOx emissions. The model has proven to be a reliable tool for assessment not only of performance related issues in boilers and furnaces, but also with regards to NOx emissions in these systems. Several improvements to the new modeling tool are in-progress. The new capabilities are intended to decrease the computational cost of using the new modeling tool by improving the scalability using domain decomposition. Future plans also call for extension of the model to simulate solid and liquid fuel combustion. Acknowledgements CFD graphics in the paper are produced by Fieldview of Intelligent Light.