Practical Use of Computation Fluid Dynamics to Industrial Combustion Applications

Paper from the AFRC 2013 conference titled Practical Use of Computation Fluid Dynamics to Industrial Combustion Applications by Kevin Anderson

An overview of the Computational Fluid Dynamics (CFD) process as applied to full sized industrial combustion applications will be presented which illustrates how CFD is currently used to predict the performance for these cases. The use of non-reacting flow modeling for optimization of burner air distribution, reacting flow modeling for predication of flame characteristics and prediction of pollutants such as carbon monoxide, and post processing CFD results for predication of NOx emissions will be discussed. Simplifications such as reduced chemistry sets will be presented. Example results will be presented including the relevant modeling techniques and field results of several full sized combustion equipment applications.

PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Practical Use of Computational Fluid Dynamics to Industrial Combustion Applications Presented at American Flame Research Committee Combustion Symposium Sheraton Kauai, Hawaii September 22-­‐25, 2013 By Kevin Anderson and Steve Londerville John Zink Company LLC | Coen Division Abstract An overview of the Computational Fluid Dynamics (CFD) process as applied to full sized industrial combustion applications will be presented which illustrates how CFD is currently used to predict the performance for these cases. The use of non-­‐reacting flow modeling for optimization of burner air distribution, reacting flow modeling for prediction of flame characteristics and prediction of pollutants such as carbon monoxide, and post processing CFD results for prediction of NOx emissions will be discussed. Simplifications such as reduced chemistry sets will be presented. Example results will be presented including the relevant modeling techniques and field results of several full sized combustion equipment applications. Page 2 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Introduction CFD software has evolved from a tool to evaluate isothermal flow fields in simple geometries to a powerful and robust tool which can model reacting flow fields in complex geometries. The use of Computational Fluid Dynamics (CFD) in the design of equipment for various industries has become prevalent as the capabilities of the CFD codes have improved. A good general description of CFD applied to combustion is presented by Londerville and Baukal [1]. CFD has proven to be a useful tool specifically for industrial combustion applications, and it is routinely applied in the design of combustion equipment. For example, CFD analysis is used in the design and optimization of air ducts and plenums to ensure airflow is properly distributed to and within combustion equipment. CFD analysis is also useful for optimization of mixing devices such as air/flue gas recirculation (FGR) mixers. Finally, CFD analysis is useful for predicting combustion behavior including evaluation of pollutant emissions. Combustion Airflow Modeling Physical testing of scale models have historically been used for the evaluation of combustion airflow. There are issues that arise with scaling using these methods. For instance, dynamic similitude as described by Streeter and Wylie [2] (i.e. Reynolds number for airflow modeling) is generally not maintained if air is used for the flow media. It can also be difficult to properly simulate some of the equipment on a reduced scale. For example, it is common for fans to be in close proximity to the burners such that flow effects from the fans are significant to the model solution. However, it can be quite difficult to properly model the flow coming from the fan using scale models. Furthermore, it is difficult and sometimes impossible to obtain accurate measurements from scale models. It is relatively simple to eliminate these concerns using CFD modeling. First, there are no issues with similitude, as modeled geometry and fluid properties are not scaled. Second, effects of equipment such as fans in close proximity to burners can be modeled directly as shown in Figure 1. For these cases, the model domain would include the fan scroll (impeller) and fan wheel geometry. A fan wheel discharge profile (radial and tangential velocity components leaving the fan wheel) can be approximated with reasonable accuracy using methods presented by Jorgensen [3]. Page 3 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 1 -­‐ Assumed Fan Wheel Exit Profile (Left), Resultant Fan Impeller and Windbox Pathlines (Right) Geometry simplification can be used to reduce grid size. Perforated plate can be modeled with FLUENT (a commercially available CFD code) as porous jump surfaces using correlations from Idelchik [4]. Burner components with complex geometry such as flame stabilizers can also be modeled with FLUENT as porous jump surfaces if the underlying flow characteristics of the component are understood. The CFD model is used to evaluate combustion air flow distribution in burners. For swirl stabilized burners, secondary air zones (and tertiary air zones, if applicable) are sub divided into multiple zones so that air distribution can be evaluated. Typically, these results are plotted as a function of angle from the air inlet. Size and location of flow correcting baffles can then be adjusted until satisfactory air distribution is observed in the model results. The results can easily be plotted as shown in Figure 2. Page 4 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 2 - Predicted Improvement in Burner Air Distribution due to Engineered Flow Baffle Design (Typical) The CFD model is also used to evaluate combustion air flow distribution between burners. The typical design process would entail building geometry for burners, windbox, ducting, and furnace and evaluating a flow solution without any additional means for flow correction. Proposed changes to the design such as turning vanes and flow correcting baffles would then be tested with the intention of improving the flow distribution between burners while maintaining acceptable flow distribution within individual burners. Actual results from such an approach are shown in Figures 3 and 4. Figure 3 -­‐ Flow in a Six Burner Windbox, Without Baffles (Left), With Baffles (Right) - Pathlines Colored by Velocity, ft/s Page 5 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 4 - Predicted Improvement in Flow Distribution between Burners for a Six Burner Windbox For modeling combustion airflow in ducting and windboxes, turbulence can be modeled using Reynolds-­‐ averaged Navier-­‐Stokes (RANS) models such as k-­‐epsilon. First order discretization of all variables is usually sufficient. Assuming constant density flow is usually a good simplifying assumption for typical cases. Mixing Applications Flue gas recirculation (FGR) is a common and effective method for reducing NOx emissions for combustion applications. Many applications are designed such that a portion of the relatively cool combustion products are conveyed via a FGR fan and mixed with fresh air which has been compressed via a forced draft fan. For these cases, mixing of air and FGR often occurs close to the combustion equipment, and careful design steps must be used to ensure that air and FGR are well distributed in the burner equipment. Poorly distributed FGR and air has deleterious effects on combustion performance. For individual burner applications, mal-­‐distributed FGR tends to affect the local stoichiometry in the burner. For multi-­‐ burner applications, this effect is magnified as it also can result in uneven FGR rates to individual burners which affects the stoichiometry between burners as well as affecting the local stoichiometry in the individual burners. Actual results for an FGR/air mixing application are shown in Figure 5. Page 6 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 5 - Predicted Oxygen Levels Downstream from an FGR/Air Mixer (Colored by Mole Fraction O2) For modeling mixing applications, RANS type turbulence models are usually recommended. Second order discretization of turbulence and species is typically used to reduce the effects of numerical diffusion on the solution. Assuming incompressible ideal gas flow is usually a good simplifying assumption for typical cases. Combustion Modeling Geometry can often be simplified to reduce grid size. Complex components such as swirlers can be greatly simplified if the flow parameters of the device are well understood. Axial, radial, and tangential velocity components for curved blade swirlers can be approximated using techniques from the work of Anderson [5] or similar methods. It can sometimes be advantageous to generate non reacting models of complex components to determine approximations for the flow field to use as inputs for the combustion model. Volumetric reactions with simplified chemistry sets are generally a practical approach for modeling gaseous fuel combustion. Multi-­‐species fuels can be represented by a hydrocarbon analog with properties (ultimate analysis, heat of formation,…etc.) that are set such that combustion products and reaction temperatures match those of the actual fuel. Dissociation cannot be modeled accurately using the hybrid eddy-­‐dissipation/finite rate model. One consequence of this is that these models tend to over-­‐predict temperatures in the highest temperature zones. Another obvious consequence is that radical species concentrations (i.e. [O], [OH], [N], …etc.) Page 7 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. are usually not directly computed in the model. Both of these limitations should be addressed if NOx pollutant formation is to be evaluated. The effects of dissociation on temperature can be approximated by modifying the thermal properties of the main species using the approach suggested by Rose and Cooper [6]. Radical species concentrations (primarily [O]) can be computed using a post processor. Many register type burners are swirl stabilized. Still, the bulk flow in the furnace for these types of applications is not generally swirl dominated flow. RANS type turbulence models give reasonable accuracy for these applications. For swirl dominated furnaces such as tangentially fired boiler applications, the Reynolds Stress model gives more realistic results. Figure 6 shows combustion model results for a tangentially fired boiler. Figure 6 -­‐ Iso Surface of 5,000 ppmv CO for a Tangentially Fired Boiler (Colored by Temperature, deg F) For many applications, reaction temperatures are high enough such that chemical kinetics are very fast compared to species mixing rates. For these mixing limited applications, the eddy-­‐dissipation model based on the work of Magnussen and Hjertager [7] is suitable for combustion modeling with reasonably accurate prediction of combustion including CO pollutant emissions. For most combustion applications, radiation is the primary mode of heat transfer from the flame to other surfaces. Consequently, accurate radiation modeling is critical for accurate solution prediction. The Weighted Sum of Gray Gases model (WSGGM) provides a means to approximate the average radiation properties for a mixture of gases. Radiation calculations themselves can be modeled using the discrete ordinates (DO) model. Soot formation is an important mechanism for heat transfer for most combustion applications, as radiation from the soot can be significant. It is recommended that at least a one-­‐step soot model with radiation be used. Figure 7 depicts a CFD simulation of a gas fired burner including the contribution of soot to the flame radiation. Page 8 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 7 -­‐ Predicted Gas Temperature Contour for an COENTM Burner Installation, Degrees K For modeling combustion, RANS type turbulence models are usually sufficient; however, RSM type turbulence models are sometimes preferable for swirl dominated systems such as tangentially fired boilers. Second order discretization of all variables is typically used to reduce the effects of numerical diffusion on the solution. Predicting Pollutant Emissions For low temperature combustion applications, CO kinetics become an important consideration. For these cases, most of the reactions still occur in a mixing limited flame zone; however, there will be zones in the domain where unburned CO must react with oxygen at low temperature. The hybrid eddy-­‐ dissipation/finite rate model can be used with good success for these applications. Suitable low temperature CO kinetics such as rate formulations reported by Barnes and Barrett [8] should be used for these cases. Exit CO emissions can be predicted with reasonable accuracy directly from the CFD solution results. For example, see Figure 8 where CFD was used to compute furnace exit CO emissions. Page 9 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 8 -­‐ Predicted Furnace Exit CO Profile for a Tangentially Fired Boiler, mole fraction Thermal NOx emissions can be computed based using the extended Zeldovich mechanism. The significant reactions are as follows: + ! ⇌ + Equation 1 + ! ⇌ + Equation 2 + ⇌ + Equation 3 Some simplifications can be made to the thermal NOx calculation. First, Equation 3 is significant only for fuel rich combustion. It can be neglected for lean combustion applications. Second, the quasi-­‐steady assumption for [N] can be used which allows for equations 1 and 2 to be combined. Finally, reverse reactions can be neglected for most applications since resultant NO concentrations are generally very low in all computational cells. Equilibrium [O] concentration can be computed via post processor based on predicted [O2] concentration and predicted temperature. Equilibrium [O] can be approximated using the correlation proposed by Westenberg [9]. Another approach is to compute Equilibrium [O] concentration by interpolation of JANAF thermochemical tables which are available from NIST [10]. Suitable kinetic rate constants such as those proposed by Bowman [11] or by Baulch et al [12] for Equation 1 can be used to compute predicted thermal NOx . Another pathway for NOx formation is called Prompt NOx. Prompt NOx formation generally increases with fuel rich combustion compared to lean combustion. The chemistry of Prompt NOx formation is complex and the species and kinetics driving Prompt NOx formation are not included in simplified combustion simulations as described previously. However, Prompt NOx formation can be approximated for simplified combustion simulations using the model proposed by De Soete [13]. Page 10 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. A third pathway for NOx formation is called fuel-­‐bound NOx. Fuel bound NOx comes from nitrogen in fuel molecules. Generally, conversion of fuel bound nitrogen to NOx decreases with fuel bound nitrogen concentration. Fuel-­‐bound NOx is generally associated with solid and liquid fuel firing. It is not as common for gas fired applications unless the gas contains non-­‐molecular nitrogen compounds (such as HCN or NH3). If reverse reactions are neglected and steady state flow is assumed, then NOx emissions can be predicted in a CFD finite volume domain by using an application of the Reynolds Transport theorem as follows: !" = ∙ !" Equation 4 Where is the molar production rate of NOx is the differential volume is the molar flux of NOx is the differential surface area Essentially, the total predicted NOx emissions through the control surface can be determined by integrating the rate of NOx formation over the control volume. It is a simple extrapolation to extend this concept to a finite number of volumes that constitutes a complete domain as follows. Δ ! !!! = !"# − !" Equation 5 Where is the total number of computational cells in the complete domain Δ is the volume of the individual computational cell !"# is the total moles per unit time of NOx leaving the domain !" is the total moles per unit time of NOx entering the domain The effects of turbulence on variables controlling NOx formation must be addressed. For steady state models, these effects can be included using a beta probability function (PDF) as described by Missaghi et al [14] to approximate the variations in temperature and species. For most cases, it is sufficient to consider only fluctuations in temperature due to turbulence in the PDF. Fluctuations in species can generally be neglected. NOx post processing can be done using the NOx pollutant model included with Fluent or other commercial CFD packages. Alternatively, NOx post processing can be done using stand-­‐alone software which can read output from CFD cases. There are advantages in using a stand-­‐alone post processor. First, there is more control of kinetic rates for [O] and [NO] formation. Second, it is relatively easy to set up zonal conditions for multi-­‐burner firing where peak reaction temperatures, species composition, and Page 11 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. stoichiometry are not uniform throughout all of the burners. An example of post processing NOx correlation to field data for a natural gas fired application is shown in Figure 9. Figure 9 -­‐ Comparison of Field Data vs. CFD Post NOx Emissions for a COENTM Burner Installation Reasonable correlation between field data and predicted NOx emissions (+/-­‐15% error) has been observed for gas firing cases using a NOx post processor as described above for cases firing with fresh air including preheated air. In industry, it is common to vitiate combustion air by dilution with flue gas recirculation (FGR), steam injection, or other means to reduce NOx emissions. For these cases, experience to date has shown that the post processor tends to under predict the NOx compared to actual field data. An empirical correlation is often used to predict the effects of vitiated air flow. Alternatively, a correction can be applied to the NOx formulation as a function of the peak flame temperature. The application of post processing to predict NOx compared to actual field data and empirical methods has shown excellent correlation as shown in Figure 10. Further, this technique can be used to further calibrate empirical models providing rapid and accurate NOx emission predictions potentially eliminating expensive and time consuming field data correlation. Page 12 PROPRIETARY DOCUMENT © 2013 by JOHN ZINK COMPANY LLC. This document is proprietary. It is to be maintained in confidence. Use of, or copying in whole or part is prohibited and may only be granted by written permission of John Zink Company. Figure 10 -­‐ Predicted NOx Emissions for a Natural Gas Fired COENTM Burner Including Low Temperature NOx Correction Summary Effectively applying CFD to the design of industrial combustion equipment requires a thorough understanding of the underlying physics of the combustion system. In a typical industrial combustion application, the science involved is multi-­‐disciplinary and complex, involving fluid flow, heat transfer, and reaction kinetics. By its nature, use of CFD software requires that many simplifications of the underlying physics must be made. The ability to extract realistic results from a CFD simulation can seem daunting. Fortunately, realistic results can be derived from CFD simulations of industrial combustion applications, even with many simplifications made to the CFD simulation. CFD analysis has proven useful in predicting air distribution to and within burners. CFD analysis has also proven useful for optimization of mixing devices such as air/flue gas recirculation (FGR) mixers. Finally, CFD analysis has proven useful for predicting combustion behavior including accurate evaluation of pollutant emissions. Page 13 Bibliography [1] S. B. Londerville Editor and C. E. Baukal Editor, "CFD-­‐Based Combustion Modeling," in The Coen & Hamworthy Combustion Handbook, Boca Raton, CRC Press, 2013, pp. 183-­‐210. [2] V. L. Streeter and E. B. Wylie, "Dimensional Analysis and Dynamic Similitude," in Fluid Mechanics, Eighth Edition, New York, McGraw-­‐Hill, Inc., 1985, pp. 160-­‐182. [3] R. Jorgensen, "Centrifugal Fans," in Fan Engineering, Eighth Edition, Buffalo, Buffalo Forge Company, 1983, pp. 10-­‐1-­‐10-­‐17. [4] I. E. Idelchik, "Resistance to Flow Through Barriers Uniformly Distributed Over the Channel Cross Section," in Handbook of Hydraulic Resistance, 3rd Edition, New York, Begell House, Inc., 1996, p. 516. [5] K. Anderson, "Predicting Performance of curved blade axi-­‐symmetric swirlers," MS Thesis, Deptartment of Mechanical Engineering, CSUS, Sacramento, CA, 2005. [6] J. W. Rose and J. R. Cooper, Technical Data on Fuels, Edinburgh: Scottish Academic Press, 1977. [7] B. F. Magnussen and B. H. Hjertager, "On mathematical models of turbulent combustion with special emphasis on soot formation and combustion," in 16th Symposium on Combustion, Cambridge, 1976. [8] R. H. Barnes and R. E. Barrett, "Chemical Aspects of Afterburner Systems," National Technical Information Service, Springfield, 1979. [9] A. E. Westenberg, "Turbulence Modeling for CFD," Combustion Science and Technology, vol. 4, pp. 59-­‐67, 1971. [10] NIST, "NIST-­‐JANAF Thermochemical Tables," National Institute of Standards and Technology, September 1982. [Online]. Available: http://kinetics.nist.gov/janaf/html/O-­‐001.html. [Accessed 1 May 2013]. [11] C. T. Bowman, "Kinetics of Pollution Formation and Destruction in Combustion," Progress in Energy and Combustion Science, vol. 1, pp. 33-­‐45, 1975. [12] D. L. Baulch, C. J. Cobos, R. A. Cox, P. Frank, G. Hayman, T. Just, J. A. Kerr, T. Murrells, M. J. Pilling, J. Troe, R. W. Walker and J. Warnatz, "Summary Table of Evaluated Kinetic Data for Combustion Modeling: Supplement 1," Combustion and Flame, no. 98, pp. 59-­‐79, 1994. [13] G. G. De Soete, "Overall Reaction of NO and N2 Formation from Fuel Nitrogen," in 15th Symposium on Combustion, 1975. [14] M. Missaghi, M. Pourkashanian, A. Williams and T. L. Yap, "Prediction of NOx Emissions from Page 14 Oxygen Enriched Burners," in International Conference on Environmental Control of Combustion Processes, Honolulu, 1991.