CFD Prediction of Visible Flame Height for Pressure-assisted Flares

Paper from the AFRC 2014 conference titled CFD Prediction of Visible Flame Height for Pressure-assisted Flares by M. Henneke.

Multi-point ground flares are often the solution of choice to smokelessly flare large quantities of gas. These flare systems are staged so that smokeless operation is possible over a very wide range of gas flows. In most cases, a fence is required around the perimeter of the flare array so that the flare flames are concealed. Because of the requirement to conceal the flames, visible flame height is a key performance metric in the application of multi-point ground flares. Visible flame height is known to vary with burner design, relief gas flow rate, relief gas composition, and relief gas supply pressure. The spacing between adjacent burners is also known to impact flame height. Computational Fluid Dynamics (CFD) modeling is a simulation methodology that can provide a prediction of flow, mixing, and combustion in these flare flames. However, developing a CFD methodology that can accurately predict the visible flame height can be challenging. In this paper we will discuss the impact of mesh fidelity and turbulence-chemistry interaction on CFD predictions of flame height. We will show that certain commonly used modeling methods fail to capture the important combustion physics in multi-point ground flares. We will further illustrate the impact of mesh fidelity or resolution on capturing the flame shape.

AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 1 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM CFD Prediction of Visible Flame Length for Pressure-­‐Assisted Flares Mike Henneke Shirley Chen Kevin Leary Abstract Multi-point ground flares are often the solution of choice to flare large flow rates of gas without visible smoke. These flare systems are staged so that smokeless operation is possible over a very wide range of gas flows. In most cases, a fence is required around the perimeter of the flare array so that the flare flames are concealed. Because of the requirement to conceal the flames, visible flame length is a key performance metric in the application of multi-point ground flares. Visible flame length is known to vary with burner design, relief gas flow rate, relief gas composition, and relief gas supply pressure. The spacing between adjacent burners is also known to impact flame length. Computational Fluid Dynamics (CFD) modeling is a simulation methodology that can provide a prediction of flow, mixing, and combustion in these flare flames. However, developing a CFD methodology that can accurately predict the visible flame length is challenging. In this paper we will discuss the impact of mesh fidelity and turbulence-chemistry interaction on CFD predictions of flame length. We will show that certain commonly used modeling methods fail to capture the important combustion physics in multi-point ground flares. We will further illustrate the impact of mesh fidelity or resolution on capturing the flame shape. Introduction Modern multi-point ground flares often employ pressure-assisted flare burners. This paper discusses the class of pressure-assisted burner designs that have multiple drilled orifices. At typical high firing pressures, these orifices are choked such that the gas flow leaving the burners is sonic. The high velocity of the individual gas jets would suggest that the resulting flames are momentum dominated (Hawthorne, 1949). However, in practice, large flame structures are present in the upper flame, suggesting that the flames transition from being momentum-dominated in the lower portion to being buoyancy dominated (like a fire) in the upper portion. The lower flame is often lifted above the burner at operating pressures. This lift-off creates a partially-premixed combustion process that can be challenging for CFD analysis. Traditional methods for combustion modeling CFD such as the hybrid eddy dissipation concept can readily be adjusted so that the flame length prediction matches at a single point. However, single point validation does not ensure that the model has the correct sensitivity to pressure, burner design, and fuel. As we show later, even a highly detailed CFD model using chemistry validated for in-furnace applications does not show the correct sensitivity to the important inputs. AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 2 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM This paper presents an overview of the challenging aspects of ground flare CFD analysis. Experience from the authors' company is provided when relevant. In particular, we note that single-point model validation should not be considered sufficient when assessing the suitability of a particular modeling methodology. Issues to consider for pressure-­‐assisted flares The following sections discuss several issues related to the quality of CFD models for pressure-assisted flares. This discussion is not all-encompassing but provides information on some of the key considerations. High-fidelity vs. low-fidelity CFD As described in the introduction, typical pressure-assisted flare burners have a moderate level of geometric complexity due to burner geometry and large number of fuel ports. For CFD analysis, the challenge that the small fuel ports pose is the disparity in length scales. The drilled holes produce gas jets that typically have a length scale of 1 cm or smaller. The length of the flare flame might be 10 m or more. Previous testing work indicates that the details of the drilling pattern are often key parameters in the performance of a particular design. For this reason accurately capturing the effects of these small holes is key to the utility of the CFD model - however, capturing these jets is computationally expensive. For example, in previous CFD modeling work at John Zink Hamworthy Combustion, relatively high resolution meshes with more than 5 million cells per burner have been required for accurate flame length predictions. In simulations of large arrays, this mesh resolution would lead to very large cell counts so engineers at John Zink Hamworthy Combustion have developed mapping strategies to reduce the cell count for ground flare array simulations. The term ‘mesh resolution' is used to describe the ability of a particular mesh to capture the important physics in a CFD simulation. Note that resolution requirements are different for the different types of physics models that can be used in CFD. For example, as discussed below, the resolution requirements of a Large Eddy Simulation are often significantly more stringent than for the equivalent Reynolds' Averaged Navier-Stokes simulation. Figure 1 shows a section slice through a portion of a typical burner mesh. The highlight arrows call out the mesh included in the individual fuel ports. AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 3 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM Figure 1: Section cut showing mesh resolution through individual ports of a typical burner design. A rendering of the burner is shown in the inset If the mesh resolution is insufficient to capture the behavior of the flow leaving the burners, then the combustion zone behavior cannot be correctly predicted by the model. Types of CFD analysis In this section, we describe the attributes of Reynolds' Averaged Navier-Stokes (RANS) CFD solutions and contrast these with Large Eddy Simulations (LES). To understand the differences in these models it is necessary to understand some principles of turbulent fluid dynamics. Turbulence is a result of instabilities that develop in fluid flows above a certain Reynolds' number. The instabilities in the flow are rotational and vortex structures of many sizes are an important characteristic of turbulence. Flow quantities such as velocity appear to be random - however the fluctuations will have a range of length and time scales. Turbulence creates additional challenges in combusting flows because of the interaction of the turbulence with the chemistry. Reynolds' Averaged Navier-Stokes (RANS) RANS solutions are often described as steady-state solutions. The aim of these simulations is to produce a time-averaged solution. Note mesh resolution in the individual burner jets AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 4 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM In a RANS solution, turbulent fluctuations are not explicitly included in the CFD simulation, but instead are modeled using a turbulence model such as the k-ε model. The k-ε model is often called a two-equation turbulence model and it has been widely evaluated for many different internal and external flows - most of the evaluations have been for non-reacting, single-phase flows. Turbulent fluctuations cause enhanced transport of mass, momentum, and energy. In a k-ε model the transport due to turbulent fluctuations is modeled as a diffusive process where the diffusivity depends on the strength of the turbulent fluctuations. Large Eddy Simulation (LES) The turbulence assumptions used by RANS simulations described previously work reasonably well for many engineering flows. However, RANS requires that much of the important turbulent phenomena be simplified and captured by time-averaged quantities. Large eddy simulation, in contrast, directly simulates much more of the important turbulent phenomena. In a large eddy simulation, the large and important turbulent length scales are directly simulated by the numerical solution while only the small scale turbulent behavior is approximated. Knowledge of the important turbulent length scales is key to successful LES. For example, the often-used Pope criterion (Pope, 2004) requires that Large Eddy Simulation capture 80% of the turbulent kinetic energy in the flow. While it is not possible to provide exact guidance, work done by John Zink Hamworthy has suggested that near-burner mesh resolution of 1 mm or smaller will be required. Combustion Physics and Turbulence Chemistry Interaction Hydrocarbon flames are thin - in the range of 0.5 mm in thickness. Industrial CFD models do not capture the detailed flame behavior for this reason. The models only capture quantities like average temperature or average concentrations. The underlying combustion physics are very sensitive to fluctuations in these quantities. Laminar chemistry is a frequently applied approximation but it usually performs poorly (Poinsot, 2012). Reaction rates are highly non-linear yet in the laminar chemistry approach the reaction rate is approximated as: (! are molecular concentrations) , ! = (, !) This type of approximation can lead to very large errors (Bakker 2006). The laminar chemistry assumption is often applied as a hybrid eddy dissipation model where the reaction rate is computed by the eddy dissipation approach and the Arrhenius approximation shown above and the simulation uses the smaller rate expression. In many situations of interest, the errors introduced by the laminar chemistry approximation may not be significant because the Arrhenius rates are not being used in the simulation. The classical Burke-Schumann flame (Williams, 1994) assumes infinitely fast chemistry. That work is the basis for a many non-premixed combustion models. This work and the work that follows has shown that in many situations of interest, the fast chemistry assumption can provide reasonable and useful results. However, the fast chemistry approach (and by extension, the so-called hybrid eddy dissipation models) has poor predictive capability in cases where the combustion process is not fast. Lifted AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 5 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM flames are a good example of flames where the combustion chemistry is too slow for a fast chemistry model to be appropriate. Experience with these sorts of fast chemistry approximations and the simplistic adjustments provided by the hybrid eddy dissipation approach is that these models do not capture flame length with sufficient accuracy to be useful. While the models can be tuned to predict flame length, the hybrid eddy dissipation approach is unable to predict the effect of pressure and fuel type on flame length. Similarly, this approach is not able to predict the effect of burner design on flame length. These are key capabilities that are needed for pressure-assisted flares. Figure 2 shows the results of previous work at John Zink Hamworthy Combustion in this area. The figure illustrates that the results obtained using a Reynolds' averaged Navier-Stokes CFD simulation along with a hybrid eddy dissipation concept combustion model fail to capture the pressure dependence of flame length for a propane flame. The effort was not a simplistic effort - the chemistry used had been experimentally validated in other industrial combustion applications at the JZHC R&D test center and the CFD model used had over 8.5 million cells in a single burner simulation. In this case, the CFD model was calibrated with the lowest pressure point to assess if a calibrated model could be used in the product development and engineering process. The results of the analyses indicate that the errors in the CFD model are too large to be useful. Figure 2: Comparison of flame length predictions using CFD with experimental flame observations This particular example clearly reveals the problem with a single point validation effort. Meaningful validation must consider a wide range of operating scenarios. This means that 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Normalized flame length Supplied gas pressure (normalized) Experimentally observed flame length CFD-predicted chemical flame length Calibration point AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 6 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM experimental data must be collected over a range of fuel pressures, fuel blends, and burner types in order to properly validate a CFD methodology. This type of validation effort is very compute-intensive. Visible and Chemical Flame Length In many industrial applications, the chemical flame length is the parameter of interest. For example, it is typical to measure flame boundaries using a pre-determined concentration of carbon monoxide such as 1,500 ppmvd. However, for ground flaring applications, chemical markers are rarely measured - instead it is the visible flame length that is of interest commercially and technically. Wade and Gore (1996) measured visible and chemical flame length in acetylene/air diffusion flames. These flames produce significant amounts of dark soot at low heat release rates. Since the soot particles are luminous, the visible flame length is largely determined by the soot particle concentration and temperature throughout the flame. The results presented showed that in some cases the visible flame length may be 1.5 times the chemical flame length or even longer. Soot formation in the flame For many ground flare applications, the visible flame is due to luminous soot particles. These are small carbonaceous particles that are formed in the fuel-rich portions of the flame. In the high temperature regions of the flame, these particles emit bright yellow or white light as well as significant thermal radiation. At high fuel pressures, these soot particles are usually completely oxidized in the flame and unburned soot (smoke) is not visible above the flame. However, as the pressure is reduced, eventually unburned soot will start to escape the flame and produce visible smoke. This is one of the reasons that large multipoint ground flares are staged - prevention of visible smoke requires a minimum operating pressure. This pressure is sometimes called the de-staging pressure. Soot formation in the flame is relevant to the CFD analysis because the visible luminosity in these hydrocarbon flames is largely due to soot. As discussed in the previous section, inferring flame length from chemical markers alone cannot be expected to match observations of the visible flame length. Accurate CFD predictions of soot concentration and morphology in a flame are areas of current research (Kennedy,1997). Conclusions CFD analysis of multipoint ground flare systems could offer important insights into system performance that are difficult and costly to obtain by full-scale testing. For this reason, the use of CFD during the engineering phase of these projects is expected to grow. This challenge of accurate flame length assessment using CFD has been discussed in this paper. Traditional methods used in CFD analysis for industrial combustion systems do not yield flame length predictions of sufficient accuracy for use in design. Single-point validation, where the CFD model performance is evaluated at a single operating point, is not sufficient to develop a robust methodology. John Zink Hamworthy is heavily invested in bridging the gaps identified in this paper. Overcoming these challenges requires extensive test work to generate validation data as well as AFRC Copyright 2014, John Zink Company, LLC. All rights reserved 7 AFRC 2014 INDUSTRIAL COMBUSTION SYMPOSIUM developing significant CFD expertise in advanced combustion modeling, large eddy simulation, meshing methodologies, and high-performance computing. References 1. Hawthorne WR, Weddell DS, Hottel HC. Mixing and combustion in turbulent gas jets . 3rd Symp. (Int.) Combust, Williams and Wilkins, Baltimore, 1949, p266-288. 2. Poinsot, T and Veynante, D, Theoretical and Numerical Combustion, 3rd edition, 2012 3. Pope, SB, Ten questions concerning the large-eddy simulation of turbulent flows, New Journal of Physics 6 (2004) 35 4. Williams, FA, Combustion Theory, Westview Press, 1994 5. Bakker, A, Modeling Chemical Reactions with CFD, http://www.bakker.org/dartmouth06/engs199/10-react.pdf, 2006 6. Wade, R. W.; Gore, J. P., Visible and Chemical Flame Lengths of Acetylene/Air Jet Diffusion. NISTIR 5904; October 1996. 7. Kennedy, I. M., Models of Soot Formation and Oxidation, Progress in Energy and Combustion Science, 23:95-132, 1997