RANS vs LES CFD for Gas-Fired Combustion Equipment

Paper from the AFRC 2017 conference titled RANS vs LES CFD for Gas-Fired Combustion Equipment

High temperature gas-fired furnaces and gas flares are widely used in the chemical and petrochemical industries. The US EPA estimates that approximately 3,200 process heaters are used in the U.S. petroleum refining industry and 1,400 fired heaters are used in the U.S. chemical process industry [1]. NASA/NOAA estimated that in 2012 there were approximately 7,500 gas flares worldwide burning approximately 143 billion cubic meters (BCM) natural gas [2]. Past developments in combustion diagnostics, computational resources and software development have greatly expanded the application of computational fluid dynamics (CFD) tools for combustion analysis in the process industries. Effective use of CFD can help improve combustion efficiency, enhance heat transfer efficiency, reduce pollutant formation, and establish better safety practices. CFD has been used to develop insights into unique combustion phenomena (i.e., turbulent mixing effect on reacting flows), evaluate new and modified equipment design (i.e., improved low-NOx burners), assess operations (i.e., turn-down conditions), guide experimental work (i.e., reduce costs of full scale testing), and calculate pollutant control impacts (i.e., flare burner tip spacing on soot formation). However, not all CFD tools equally simulate all combustion processes accurately. Successful application of CFD requires an understanding of which tools work best for which applications.; This paper will compare Large Eddy Simulations (LES) to Reynolds-averaged Navier-Stokes (RANS) based CFD applied to several combustion applications to illustrate issues to consider when applying CFD to solve industrial problems. This paper will review the basis, strengths and limitations of these two approaches and provide recommendations on when each is most applicable. An LES based CFD tool simulates turbulent reaction chemistry coupled with radiative transport in buoyancy driven flames (i.e., gas flares) and the impact large flames have on surrounding objects (i.e., wind fence, process equipment, etc.). Validation work comparing LES based CFD simulations to RANS based simulations of a multi-point ground flare test will be shown to illustrate each approach for this application. Recent LES based CFD analysis of transient burner operation will also be discussed. Results of this work will be used to review and discuss how CFD codes may help assess various risk scenarios including wind, % flame coverage, and thermal fatigue for a given geometry. Several examples will be used to illustrate applications where each technology has worked and failed.

AFRC 2017: Industrial Combustion Symposium Hyatt Regency Houston, Houston, Texas - September 17-20, 20017 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis Joseph D. Smith, Ph.D., Laufer Endowed Energy Chair Missouri University of Science and Technology, Rolla, MO Bradley R. Adams, Ph.D., Associate Professor, Brigham Young University, Department of Mechanical Engineering, Provo, UT Robert Jackson, Zachary Smith and Ahti Suo-Antilla, Ph.D. Elevated Analytics, Inc., Catoosa, OK Scot Smith and Doug Allen Zeeco, Inc., Broken Arrow, OK Abstract High temperature gas-fired furnaces and gas flares are widely used in the chemical and petrochemical industries. The US EPA estimates that approximately 3,200 process heaters are used in the U.S. petroleum refining industry and 1,400 fired heaters are used in the U.S. chemical process industry [1]. NASA/NOAA estimated that in 2012 there were approximately 7,500 gas flares worldwide burning approximately 143 billion cubic meters (BCM) natural gas [2]. Past developments in combustion diagnostics, computational resources and software development have greatly expanded the application of computational fluid dynamics (CFD) tools for combustion analysis in the process industries. Effective use of CFD can help improve combustion efficiency, enhance heat transfer efficiency, reduce pollutant formation, and establish better safety practices. CFD has been used to develop insights into unique combustion phenomena (i.e., turbulent mixing effect on reacting flows), evaluate new and modified equipment design (i.e., improved low-NOx burners), assess operations (i.e., turn-down conditions), guide experimental work (i.e., reduce costs of full scale testing), and calculate pollutant control impacts (i.e., flare burner tip spacing on soot formation). However, not all CFD tools equally simulate all combustion processes accurately. Successful application of CFD requires an understanding of which tools work best for which applications. This paper will compare Large Eddy Simulations (LES) to Reynolds-averaged Navier-Stokes (RANS) based CFD applied to several combustion applications to illustrate issues to consider when applying CFD to solve industrial problems. This paper will review the basis, strengths and limitations of these two approaches and provide recommendations on when each is most applicable. An LES based CFD tool simulates turbulent reaction chemistry coupled with radiative transport in buoyancy driven flames (i.e., gas flares) and the impact large flames have on surrounding objects (i.e., wind fence, process equipment, etc.). Validation work comparing LES based CFD simulations to RANS based simulations of a multi-point ground flare test will be shown to illustrate each approach for this application. Recent LES based CFD analysis of transient burner operation will also be discussed. Results of this work will be used to review and discuss how CFD codes may help assess various risk scenarios including wind, % flame coverage, and RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium thermal fatigue for a given geometry. Several examples will be used to illustrate applications where each technology has worked and failed. 1. Introduction and Background Computational Fluid Dynamics has become an accepted engineering tool to help design and optimize aerodynamic equipment since the 1970s. Over the past twenty years, CFD has also successfully been applied to analyze and optimize combustion equipment, including industrial burners, process heaters and gas flares. Current trends in environmental and operational costs are making it increasingly necessary to explore operating conditions, which are beyond the current empirical database of industrial burners. Previous work by Smith, et.al. [3] and Henneke, et.al., [4] have demonstrated the basic fact that CFD based tools hold great promise for economically exploring new combustion equipment designs. The CFD tool used most frequently in these industrial applications was based on Reynolds-averaged Navier-Stokes (RANS) models, while Large Eddy Simulations (LES) models have gained in popularity due to their ability to resolve the transient fluctuations inherent in most combustion problems and industrial access to powerful computers required to perform transient analyses. Understanding which CFD tool to apply is critical to successfully solving the problem and is the focus of this paper. In general, RANS provides time-averaged solutions of the governing equations, whereas LES solves the time-dependent Navier-Stokes equations with direct simulation of large scale motion. Thus, LES simulations yield time-dependent results not possible using standard RANS based models. LES can provide insight into applications where flow instabilities and turbulent fluctuations are important, specifically the impact fluctuations may have on the scalar mixing process [5]. This paper provides a comparison between LES and RANS based CFD analyses using several combustion applications to illustrate key features to consider when using CFD to solve industrial problems. This paper reviews the basis, strengths and limitations of these two approaches and provides recommendations on when each is most applicable. 2. Reynolds-Averaged Navier-Stokes (RANS) Models 2.1. Basis In the Reynolds-Averaged Navier-Stokes (RANS) approach, the Navier-Stokes equations for the instantaneous velocity and pressure fields are decomposed into a mean value and a fluctuating component. The resulting equations for the mean quantities are essentially identical to the original equations, except that an additional term now appears in the momentum transport equation, the Reynolds stress tensor. The Reynolds stress tensor may be modeled in terms of the mean flow quantities using two basic approaches: turbulent-viscosity models and Reynolds-stress models. These models use the concept Page 2 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium of a turbulent viscosity (μt) to model the Reynolds stress tensor as a function of mean flow quantities. The most widely used turbulence model in combustion simulations is the k -  model, which involves transport equations for the turbulent kinetic energy, k, and kinetic energy dissipation rate, . It is a semi-empirical model in that the modelled transport equation for dissipation used in the model depends on phenomenological considerations and empiricism. Variations to the standard k -  model include the following: • Standard k - two-layer model. Combines the standard k - model with the two-layer approach, which allows the k - model to be applied in the viscous sublayer of the wall boundary layer. • Realizable k - model. Proposed by Shih et al. [6] in 1994, this model differs from the standard k - model in that it contains a new formulation for the turbulent viscosity and has a new transport equation for the dissipation rate which is derived from an exact equation for the transport of the mean-square vorticity fluctuation. The term realizable means that the model satisfies certain mathematical constraints on the Reynolds stresses. In addition to k - models, additional two-equation models have been proposed, most of which use the kinetic energy, k, as one of the variables with a variety of choices for the second variable. Perhaps the most common of these is the k -  model proposed in 1988 by Wilcox [7]. Here the specific dissipation rate, = /k, is used as the second variable. A variation of the k -  model is the SST k -  model. This model was developed by Menter [8] and it takes advantage of accurate formulation of the k -  model in the near-wall region with the free-stream independence of the k -  model in the far field. 2.2. Limitations Limitations in the RANS approach are of two general types. First, the k -  and other two equation models are unable to predict anisotropic Reynolds stress tensors, relaxation effects, and nonlocal effects due to turbulent diffusion. More specifically, the standard k - model provides mediocre results for complex flows with severe pressure gradients, strong streamline curvature, swirl and rotation. They also tend to over predict the spread of round jets. The realizable k - model provides improved predictions for jet impingement, separating flows, swirling flows, secondary flows and round jet spread, but is still unable to accurately represent anisotropic stress tensors. For these problems, it is necessary to model the evolution of the full Reynolds stress tensor. The Reynolds stress model introduces 6 new equations (instead of 2 for the k -  model). Although the models include considerably more physics and allow for anisotropy in the Reynolds stress tensor, these models have yet to be optimized to the point that they consistently give superior results. This is particularly true for applications which have a variety of geometric shapes and flow conditions (e.g., swirling burner in one region, axial jets in another, and later flow through tube banks). Because of the minimal improvement in accuracy for 2-3 times higher computational cost, Reynolds stress models are seldom used for practical problems. Page 3 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium Second, RANS simulations cannot represent the transient nature of velocity and turbulent fluctuations in the flow. Since these fluctuations affect the predicted temperatures in combustion applications, the temperatures calculated from averaged fluctuations may not capture the peak temperatures produced during combustion. This can in turn cause the radiant heat transfer from a flame to be underrepresented. Further, flow instabilities cannot be accurately represented with RANS models. 2.3. Strengths RANS models are robust, economical and reasonably accurate. Uncertainties in specifying inlet and boundary conditions for combustion problems are often as significant as uncertainties in the turbulence model. Combustion applications often include a range of geometric and flow scales, making it difficult to apply more specialized turbulence models which are applicable only to a small region of the computational domain. Further, combustion simulations must also account for chemical reactions and radiative heat transfer, which require additional modeling assumptions. Over-emphasizing the accuracy of one model in the context of other less accurate models is not computational efficient. RANS provides an excellent compromise of accuracy, scalability (over different geometry and velocity scales), and computational requirements. Finally, there is considerable experience over the past three decades using RANS models for a variety of applications, including combustion systems. This experience base has allowed the general user base to identify the suitability of RANS to specific applications and identify situations where specialized two-equation models or Reynolds stress models should be used. 2.4. Applications RANS models traditionally have been the most widely used approach for modeling steady-state, industrial combustion applications due to the excellent trade-off between representation of physics and computational tractability. Gas-phase combustion simulations require interdependent calculations of turbulent velocity, fuel-air mixing, reaction chemistry, and radiant heat transfer. Solid-phase combustion is further complicated by the inclusion of heterogeneous particle reactions, gas-solid interactions, and particle radiation. The complex nature and computational resources required for these calculations suggests a turbulence model that compromises between accuracy and computational time is desirable. RANS models have been used successfully to predict a variety of combustion applications in the power generation, chemical process, smelting and waste-to-energy industries. Common uses for RANS modeling include: • • • • • Development and evaluation of new combustion techniques Evaluation of new burner and furnace designs Evaluation of pollutant control technologies Optimization of furnace operating conditions Guidance for combustion experiments Page 4 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium A small sampling of RANS-based CFD modeling of combustion systems includes applications for CO emissions [9], NOx control [10], [11], [12], [13], [14], [15], [16], syngas co-firing [17], [18], flue gas desulfurization and SO2 control [19], [20], [21], airborne mercury control [22], furnace design [23], [24], [25], oxy-combustion technologies [26], [27], [28], and copper flash smelting [29]. These systems include complex interactions between two-phase turbulent mixing and combustion reactions, convective and radiative heat transfer, and pollutant species formation, all with different fuel types, burner configurations and surface conditions. RANS simulations provide the ability to represent all of these processes in a computationally tractable time frame, particularly for industrial applications. 3. Large-Eddy Simulation (LES) Model LES was first proposed by Smagorinsky in 1963 to simulate atmospheric air currents [30]. It was later extended for general flow modeling in 1970 by Deardorff [31]. LES has since been applied to many engineering applications, including combustion [5]. Simulating turbulent flow in a RANS-based CFD model involves a very wide range of time and length scales, which all affect the general flow. Direct Numerical Simulation (DNS) can resolve all scales but requires significant CPU resources to analyze the simplest flow systems and makes simulating complex systems impossible. LES reduces the computational burden by approximating small length scales using subgrid models with low-pass filtering (similar to time-averaging). Small scale information provided by subgrid scale models allows simulation of practical problems where this information is vital (i.e., near-wall flows, reacting flows, and multiphase flows). LES filtering focuses attention on scales ranging from the specific feature size to the preselected minimum filter size. 3.1. LES Filters LES filtering equations can be implicit or explicit. Implicit filtering assumes the smallest scale (turbulent eddy size) dissipates like many other numerical schemes do so the grid size becomes the low-pass filter. Although this approach uses the grid resolution to eliminate the additional computational work to calculate a subfilter scale term, determining the shape of the LES filter associated with some numerical issues is difficult plus truncation error can also be an issue. Explicit filters apply a pre-defined filter shape to the discretized Navier-Stokes equations which reduces truncation error. Explicit filters require finer resolution than implicit filters so computational costs are higher. 3.2. Subgrid Modeling Modeling unresolved scales starts with classifying unresolved scales as either resolved sub-filter scales or sub-grid scales. Resolved sub-filter scales have effects larger than the cutoff scale but whose effects are dampened by the filter. These scales only exist when special types of filters are used such as the Gaussian filter: Page 5 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium These resolved sub-filter scales are modeled using filter reconstruction. Conversely, sub-grid scales include scales smaller than the cutoff filter width with the specific form of the sub-grid scale model depending on the filter implementation. If an implicit LES method is used, no sub-grid scale model is required with the numerical effects of discretization assumed to mimic unresolved turbulent motion. Without a full description of turbulence, empirical information is utilized when constructing and applying sub-grid scale models. Two model classes exist including: functional models and structural models (some categorized as both). Functional models are simpler than structural models since they focus on physically correct energy dissipation based on the artificial eddy viscosity approach (turbulence effects represented by turbulent viscosity). This model assumes equilibrium between energy production and dissipation of small scales. 3.3. Limitations Whereas LES can simulate time-dependent phenomena the computational time required to do so is much greater than that required for RANS based analyses. In addition, the relative approximation of sub-grid scales is dependent on the grid refinement which varies for each application so finding a reasonable approximation of measured reacting turbulent flow systems can be cumbersome and time intensive and relies on the expertise and familiarity of the modeler with the specific system being modeled. Lastly, since LES simulations are transient they must have a good initial condition to start and sufficient time must be allowed for the simulation to come to some quasi "steady-state" condition. Given the relative difficulty of correctly using LES and the time required to derive value from this approach, application of LES should be considered on a case by case basis. Care should be taken to insure grid-independent results which means running the full calculations to the same degree of "steady" operation on finer and finer meshes and then comparing pre-selected metrics to quantify the correct mesh resolution has been used. However, given the increased speed and reduce cost of multi-processor high performance computers LES is becoming more common place in industrial applications. 3.4. Strengths As shown in many applications, LES can provide insight into transient behavior not possible with RANS simulations. Investigating burner stability and radiant flux fluctuations on process tubes are two examples where LES shows great promise. Predicting ignition phenomena is another area where LES has been successfully applied to reduce safety risk to equipment and plant personnel. As pointed out above, given increased access to affordable HPC resources LES can now be used effectively to solve industrial problems not possible with RANS based CFD codes or even experimental investigation due to cost and safety issues. 4. Sample Applications The following examples illustrate where LES has successfully been used to solve industrial problems. These results have previously been published in open literature or presented at earlier Page 6 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium AFRC meetings as referenced for each case. Specific examples consider process heaters and gas flares to show comparison between RANS and LES results with test data provided were available. 4.1. Gas-Fired Furnaces LES simulations of full combustion furnaces are relatively rare [32] due to the significant computational expense to perform these types of analyses. Accordingly, LES simulations are most commonly performed by academic research groups. Much of the academic work has been applied to laboratory-scale or pilot-scale systems, most of the key physical phenomena important in fullscale operation are included which allows these models to be developed, tested, and validated with laboratory data. This also facilitates direct comparison between RANS and LES based models of gas-fired process heaters to illustrate potential differences and advantages of the each approach. In particular, three studies have been identified which examine RANS and LES predictions for gasfired combustion or heat transfer applications. Case 1 - Flow and Heat Transfer Inside Chemical Process Tubes [33] This study reviewed flow inside of a process tube simulated with RANS and LES models. For a tube with enhanced longitudinal fins, results showed the RANS model did not capture the secondary flows induced by the anisotropic turbulence. This led to inaccurate predictions of the local shear stress while global quantities were seen to be in excellent agreement with the LES results. For a separate tube design that produced swirling flow, RANS simulations were found to over predict the turbulence induced by the rotating flow, leading to a shift in the location of peak wall temperatures. However, comparison of global characteristics with experimental data showed good agreement for both RANS and LES modeling approaches with errors within 5% for the relative heat transfer and friction enhancement factors. These results suggest that while LES produced improved local properties, RANS and LES produced similar results for global characteristics. Case 2 - Wall Heat Transfer in Gas-fired Furnaces [34] This study reviewed two studies of swirling diffusion flames and the practicability of prediction methods. Simulation results of a 150-kW swirl burner with unsteady Reynolds-Averaged NavierStokes (U-RANS) model based on second-order moment turbulence closure, as well as by LargeEddy Simulation (LES), showed that the U-RANS model could capture unsteady phenomena qualitatively and "in parts" quantitatively. A second case showed that a traditional RANS firstorder moment closure model with "close attention" to boundary conditions and convergence strategy lead to correct results. A summary of the study concluded, "Recent progress achieved using large-eddy simulations coupled with advanced chemistry models is on the one hand very promising but on the other hand it is still far from being applicable to industrial problems due to excessive computational requirements. Real-world problems are tractable only using supercomputing facilities due to large dimensions of the combustors (on the order of 10 m) and the need to resolve fine features like gas nozzles with diameters on the order of 1 mm. Thus, the only viable alternative for practicable predictions of industrial fired heaters presently and into the Page 7 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium near future will depend on RANS or U-RANS models." This study suggested that regardless of LES improvements, application on the industrial scale will remain prohibitively computationally expensive, implying that LES vs RANS comparisons of industrial furnaces will remain uncommon. Case 3 Modeling of Methane Air Diffusion Flame [35] In this study, RANS and LES simulations were compared for two burner geometries, a piloted (jet) methane-air diffusion flame (Sandia D flame) and a bluff-body burner. Differences between RANS and LES for the Sandia D flame were found to be minimal. For the bluff-body geometry, LES was found to be advantageous over RANS in predicting finite-rate chemistry behavior at the end of the recirculation zone behind the bluff body, where endothermic regions were represented in fuel-rich zones. RANS modeling was not able to represent the endothermic regions with the statistical averages of heat release. This study suggests that localized behavior in swirling or recirculating flames is best modeled by LES, particularly where such behavior can impact combustion or pollutant formation. Conversely, for simple jet flames and overall combustion and heat transfer conditions, RANS results appear to be like LES predictions. 4.2. Flares Three case studies of the use of LES to analyze flares are provided. Each of the studies consisted of the analysis of multipoint flares. The first study deals with an elevated multipoint flare and highlights the LES capability of analyzing transient phenomena such as overpressure associated with ignition delay due to improperly designed pilot systems. The second and third case studies are dealing with large multipoint ground flares (MPGF). These MPGF studies also looked at ignition characteristics such as cross-light performance and ignition delay scenarios which is not possible with RANS, but also include some comparisons of RANS and LES predictions of flame height and shape, as well as flare radiation. Each of the case studies used a proprietary flare modeling tool called C3d which is based on an earlier tool called ISIS-3D [36], [37], [38]. This code was initially validated for predicting the thermal performance of pool fires [39]. C3d has successfully been used to model elevated multipoint flares, large multipoint ground flares (MPGF), air-assisted flares, and utility flares [40]. Several new combustion models for various flare gas compositions including methane, ethylene, propane, benzene, xylene and mixtures of each have been developed and tested. Case 4 - Elevated Multi-Point Flare Elevated multipoint flares represent a special class of flares capable of processing significant quantities of flare gas. A detailed CFD model of an elevated multipoint flare has been developed C3d. This tool has been used to simulate the ignition phenomena for this flare for gas flow rates between 200 to 350 tons per hour (TPH). Simulation results have been directly compared to operating test data for this flare. Results demonstrate the ability of C3d to replicate the measured flame spread rate and reproduce the measured pressure wave generated during the ignition event. Page 8 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium Based on this validation, the tool has been used to conduct over sixty separate simulations to investigate the ignition behavior for this flare. Results from these simulations clearly show the critical effect of ignition delay on the magnitude of the pressure wave generated on ignition. The main conclusion drawn from this analysis is that the ignition system's reliability to quickly ignite the flare gas above the flare tip is critical to safe operation. Predictions show that a 0.6 second ignition delay results in a significant pressure wave generated during flare ignition. Simulations at maximum flow rate (1350 TPH) exhibit explosive tendencies with pressure waves greater than one atmosphere. This confirms the conclusion that the flare must be operated with a continuous pilot to avoid and type of ignition delay. These results underscore the importance of the API recommended practice of continuous pilot operation for all large-scale gas flares. No RANS modeling was included in this case study. Case 5 - MPGF Burner Test A detailed model of a Zeeco flare test using 3 typical MPGF burners, as tested in Zeeco's flare test facility, was developed using C3d and this model was run with conditions from the Zeeco testing. The purpose of the calculations was to predict air demand under various conditions. In addition, the thermal radiation profile around the flare was also determined. The combustion and radiation models were compared to flame size, shape, and radiation measurements, measured during single-burner and multi-burner tests under no-wind and low-wind ambient conditions. The CFD model included various details depending upon the case that was run. For a single burner case, a computational domain of 6 m x 6 m x 30 m was selected. For a multi-burner case, a domain size of 35 m x 35 m x 25 m was selected. For Figure 1 - Direct comparison of C3d predictions (front the full field case, the computational image) to actual flare flame (back image) at Zeeco test domain extended 10m beyond the facility (see https://youtu.be/mCWKT6wAaUE) wind fence surrounding the entire flare field. Testing was done with both propane and ethylene as the flare gas. The shape and height of Page 9 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium the flame and smoking potential were compared to images from the test flares and a comparison is provided in Figure 1. Radiation from the flares was measured and the predicted results from C3d were compared to these measurements for 12 different data points as shown in Table 1. The data included two different flare tip sizes with each operating at three different pressures. Measurements were made at two distances: 15 meters and 30 meters. The results indicate generally good agreement of predictions with radiation test data. The largest discrepancies tended to occur at the low flare pressures. While the study did not identify the specific reason for greater variance at low pressures, it may be related to the wind effects, which as noted below, have been shown found to be important [41]. Table 1 - Comparison of radiation predictions from C3d with test results This study found several important factors when predicting the radiation coming from the flares. The radiation coming from the heated ground was found to be an important factor. The wind speed and direction were also shown to me important factors. The wind tends to not only bend the flames over but also shorten the flame and reduce the radiation emitted as shown in Figure 2. No RANS modeling was included in this case study so comparison to RANS predictions was not possible. Page 10 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium Figure 2 - Effect of wind speed upon radiative heat flux from a triple burner, ethylene flare at a distance of 15 m [41] Case 6 - MPGF Burner Test with 1 and 3 Burners MPGFs are large and difficult to model due to the large meshes needed to account for as many as 400 different flare burner tips in a flare that can be on the order of 80m x 100m. C3d has been used successfully for nearly a decade to model these challenging MPGFs; however, test data is lacking for full MPGF applications. This case study is similar in nature to Flare Case Study 2 in that tests with a few of the flare tips was done for a limited number of tips, using commercial tips designed for the full MPGF. Modeling was done with both LES (using C3d) and RANS (using Fluent). Some of the results of this study were previously presented at the November 2016 API conference. The results included radiation test data and flame shape/height data from the observed flame in the form of photographic documentation of the flame with subsequent post processing by Zeeco test engineers. Comparison of the flame height for both single-burner and three-burner tests is provided in Table 2. The test data indicated a slight increase in height when using multiple burners which is likely due to the flame interaction and resulting decrease in available oxygen available such that the flame gets slightly taller with multiple burners. This tendency was also predicted by both CFD codes. However, where the LES code could predict very accurately the same relative change as the actual observed flames the RANS prediction was much less accurate. The RANS code considerably underpredicted the single burner flame height and somewhat over predicted the three-burner flame height. As can be seen in Figure 3, the flame shape predicted by the LES code was also much closer to the observed flame than the RANS simulation. Page 11 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium Table 2 - Flame height comparison Number of Burners Used During Test Test data reported flame Height LES (C3d) Resulting Time Averaged Flame Height* RANS (Fluent) Predicted Flame Height* 1 19.5 ft 19.8 ft 10 ft 3 21.0 ft 21.1 28 ft * CO mass fraction of 0.0025 used as flame envelope indicator for CFD simulations Figure 3 - Comparison of flame height observed during testing with predictions from RANS and LES CFD Test data of radiation from the flare flame was taken during the single-burner testing. A comparison of this measured data with predictions from both RANS and LES is provided in Table 3. The results indicate a considerable performance improvement for the LES code over the RANS code. At distance from the flare of 100 feet or less the LES code tended to slightly under predict the radiation levels but were generally still within 20% of measured values. The RANS simulation considerably over-predicted the radiation levels for all positions. The LES codes ability to better capture the flame dynamics of air entrainment and wind effects on the flame seems to translate into more accurate radiation predictions. Table 3- Comparison of measured radiation heat flux data with predictions from LES and RANS CFD Wind Speed 12 MPH Distance flare (ft) from Distance from grade = 5ft Radiation heat flux in (BTU/hr◦ft2) Distance from grade = 20ft Radiation heat flux in (BTU/hr◦ft2) Measured LES RANS Measured LES RANS 75 171.25 159 1012 205.5 166 843 100 102.75 85 744 102.75 87 703 150 34.25 37 618 34.25 37 608 Page 12 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium 5. Conclusions and Recommendations Based on the previous discussions comparing RANS to LES based combustion modeling, the following conclusions and recommendations are made: • Current LES simulation results for combustion in enclosed furnaces shows improved predictions in the near-burner region of multiple test furnaces. However, the benefits of LES relative to RANS modeling for a full-scale furnace or large chemical process furnace still have not been fully demonstrated. In cases where transient phenomena cannot be properly resolved using a RANS approximation, and the problem is tied to poor operation, LES can provide more in depth understanding which has led to problem resolution. In some cases, model differences in the near burner region may be less significant relative to combustion and heat transfer in the overall furnace. Conversely, for cases where flame instabilities significantly impact combustion performance, LES can accurately capture this behavior. • LES appears best suited to examine operational safety issues related to ignition and cross lighting of gas flares. Also, LES can capture the dynamic interaction between combustion and cross wind known to be important in smokeless operation of gas flares. • To a lesser degree, LES based CFD has been applied to analyze the transient nature of combustion inside a furnace. However, based on application to gas flares, it appears there are many questions related to burner ignition and "flame-out" inside a process heater that have not been effectively analyzed using RANS based CFD analysis. For this reason, it is recommended that LES based CFD be applied to resolve unstable process burner behavior that can lead to unsafe operation. In summary, RANS is routinely used to analyze internal combustion system (i.e., process heaters) since steady operation is more likely given the system operates in a controlled environment. Conversely, LES has been shown most applicable to external combustion systems (i.e., gas flares) since the combustion process can be significantly impacted by ambient conditions such as cross winds. Overall, the specific flavor of CFD used to solve a problem relies on experience and insight into the problem being solved. Efficient use of CFD to solve operational problems and minimize safety risks relies on experienced engineering. 6. References [1] USEPA, "Alternative Control Techniques Document-NOx Emissions from Process Heaters (Revised)," EPA-453/R-93-034, Research Triangle Park, North Carolina 27711, September 1993. Page 13 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium [2] C. Elvidge, M. Zhizhin, K. Baugh, F. Hsu and T. Ghosh, "Methods for Global Survey of Natural Gas Flaring from Visible Infrared Imaging Radiometer Suite Data," Energies, vol. 9, no. 14, 2016. [3] J. D. Smith and T. Webster, "Using CFD to Solve Challenges of Ultra-low NOx Burner Retrofit in Refinery Process Heaters," Petroleum Technology Quarterly, July (2002). [4] M. Henneke, J. D. Smith, J. D. Jayakaran and M. Lorra, "Computational Fluid Dynamics (CFD) Based Combustion Modeling," in Chapter 9, The John Zink Combustion Handbook, CRC Press, 2001. [5] H. Pitsch, "Large-Eddy Simulation of Turbulent Combustion," Annual Review of Fluid Mechanics, vol. 38, p. 453-482, 2006. [6] T. Shih, W. Liou, A. Shabbir, Z. Yang and J. Zhu, "A new k-epsilon eddy viscosity model for high Reynolds number turbulent flows: Model development and validation.," NASA STI/Recon Technical Report N, vol. 95, p. 11442, 1994. [7] D. Wilcox, "Reassessment of the scale determining equation for advanced turbulence models," AIAA journal, vol. 19, no. 2, p. 248-251, 1988. [8] F. R. Menter, "Two-equation eddy-viscosity turbulence models for engineering applications," AIAA journal, vol. 32 , no. 8, p. 1598-1605, 1994. [9] B. Adams, M. Cremer and D. Wang, "Modeling non-equilibrium CO oxidation in combustion systems," Proceedings of the ASME Heat Transfer Division HTD, pp. 29-34, 2000. [10] M. Cremer, J. Valentine, H. Shim, K. Davis, B. Adams, L. J. and S. Vierstra, "CFD-based development, design, and installation of cost-effective NOx control strategies for coal-fired boilers.," in The Mega Symposium: EPRI-DOE-EPA Combined Utility Air Pollutant Control Symposium, Washington, DC, 2003. [11] B. Adams, D. Wang, M. Cremer, K. Frizzell and S. Conn, "Modeling NOx reduction from fuel lean gas reburning and selective non-catalytic reduction combined with overfire air at OMU's Smith Unit 1," in US EPA/DOE/EPRI Combined Power Plant Air Pollutant Control Symposium: The MEGA Symposium, Chicago, 2001. [12] M. Cremer, B. Adams, D. O'Connor, B. V.N. and R. Broderick, "Design and demonstration of rich reagent injection (RRI) for NOx reduction at Conectiv's B.L. England Station," in US EPA/DOE/EPRI Combined Power Plant Air Pollutant Control Symposium: The MEGA Symposium, Chicago, 2001. [13] Q. Tang, M. Denison, B. Adams and D. Brown, "Towards Comprehensive Computational Fluid Dynamics Modeling of Pyrolysis Furnaces With Next Generation Low NOx Burners Using Finite-rate Chemistry," Proceedings of Combustion Institute, vol. 32, no. 2, pp. 26492657, 2009. Page 14 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium [14] B. Adams and N. Harding, "Reburning Using Biomass for NOx Control," Fuel Processing Technology, vol. 54, no. 1-3, pp. 249-263, 1998. [15] B. Adams and Q. Tang, "Modeling Flame Behavior And Low NOx Emissions In Cracking Furnaces," in 15th International Flame Research Foundation Member's Conference, Pisa, Italy, 2007. [16] Q. Tang, B. Adams, M. Bockelie, M. Cremer, M. Denison, C. Montgomery, A. Sarofim and D. Brown, "Advanced CFD Tools for Modeling Lean Premixed Combustion in Ultra-Low NOx Burners in Process Heaters," in AFRC-JFRC Joint International Symposium, Maui, Hawaii, 2004. [17] B. Adams, "CFD Modeling of Syngas Reburning in the EPA MPCRF Pilot-Scale Coal Combustor," Final Report U.S. EPA, C_EP08C000261_0_0_RCI, 2009. [18] K.-T. Wu, H. Lee, C. Juch, H. Wan, H. Shim, B. Adams and S. Chen, "Study of Syngas CoFiring and Reburning in a Coal Fired Boiler," Fuel, vol. 83, pp. 1991-2000, 2004. [19] I. F. Maroccoa L, "Multiphase Euler-Lagrange CFD simulation applied to wet flue gas desulphurisation technology," International Journal of Multiphase Flow, vol. 35, no. 2, p. 185-194, 2009. [20] M. A. Marocco L, "CFD modeling of the Dry-Sorbent-Injection process for flue gas desulfurization using hydrated lime," Separation and Purification Technology, vol. 108, p. 205-214, 2013. [21] L. Shi, G. Liu, B. Higgins and L. Benson, "Computational modeling of furnace sorbent injection for SO2 removal from coal-fired utility boilers.," Fuel Processing Technology, vol. 92, no. 3, p. 372-378, 2011. [22] M. Cremer, C. Senior, A. Chiodo, D. Wang and J. Valentine, "CFD modeling of activated carbon injection for mercury control in coal fired power plants," in Joint EPRI DOE EPA Combined Utility Air Pollution Control Symposium, The Mega Symposium, Washington, D.C., 2004. [23] B. Adams, M. Cremer and J. Murphey, "Use of CFD in Evaluating Pyrolysis Furnace Design," in Spring Meeting & 11th Global Congress on Process Safety, AIChE, ISBN 9780-8169-1089-2, 2015. [24] M. Cremer, B. Adams, J. Valentine, J. Letcavits and S. Vierstra, "Use of CFD Modeling to Guide Design and Implementation of Overfire Air for NOx Control in Coal-fired Boilers," in Proceedings of Nineteenth Annual International Pittsburgh Coal Conference, Pittsburgh, PA, 2002. [25] B. Adams, M. Heap and S. Chen, "Use of Reacting CFD to Optimize Process Heater Performance," Computational Technologies for Fluid/Thermal/Structural/Chemical Systems With Industrial Applications, ASME, vol. 2, pp. 17-26, 1999. Page 15 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium [26] B. Adams and A. Fry, "CFD Modeling of Pilot-scale Oxycombustion Experiments," in The 36th International Technical Conference on Clean Coal & Fuel Systems, Clearwater, Florida, 2011. [27] A. Fry and B. Adams, "Characterization and Prediction of Oxy-combustion Impacts in Existing Coal-fired Boilers," in 34th International Technical Conference on Clean Coal and Fuel Systems, Clearwater, Florida, 2009. [28] M. Cremer, H. Wang, Z. Chen, K. Davis, B. Adams, L. Bool, H. Kobayashi and D. Thompson, "CFD Evaluation of Oxygen Enhanced Combustion in Coal Fired Boilers: Impacts on NOx, Carbon in Ash and Waterwall Corrosion," in Electric Utilities Environmental Conference, 6th Annual Conference on Air Quality and Global Climate Change, Tucson, AZ, 2003. [29] B. Adams, K. Davis, M. Heap and A. Sarofim, "Application of a Reacting CFD Model to Drop Tube Kinetics and Smelter Simulations," Fluid Flow Phenomena in Metals Processing, pp. 93-100, 1999. [30] J. Smagorinsky, "General Circulation Experiments with the Primitive Equations," Monthly Weather Review, vol. 91, no. (3), p. 99-164, March 1963. [31] J. Deardorff, "A numerical study of three-dimensional turbulent channel flow at large Reynolds numbers," Journal of Fluid Mechanics, vol. 41, no. (2), p. 453-480, 1970. [32] P. Smith, J. Thornock, S. Smith, M. Hradisky, D. Smith, P. Emmett, K. Daines and B. Harris, "Part II: Deployment of a Continuous Monitoring of Tube Metal Temperature at the Holly Refinery Depropanizer Reboiler," in AFRC Annual Meeting , Koloa Kauai, Hawai, 2016. [33] D. Van Cauwenberge, C. Schietekat, J. Flore, V. Van Geem and G. Marin, "CFD-based design of 3D pyrolysis reactors: RANS vs. LES," Chemical Engineering Journal, vol. 282, p. 66-76, 2015. [34] J. Vondal and J. Hajek, "Wall heat transfer in gas-fired furnaces: Effect of radiation modelling," Applied and Computational Mechanics, vol. 9, p. 67-78, 2015. [35] P. Siwaborworn and S. Navarro-Martinez, "Large-eddy Simulation," Research project description at http://www.itv.uni-stuttgart.de/forschung/projekt2/index.en.html. [36] A. Suo-Anttila, K. Wagner and M. Greiner, "Analysis of Enclosure Fires Using the Isis3DTM CFD Engineering Analysis Code," in Proceedings of ICONE12, 12th International Conference on Nuclear Engineering, Arlington, VA, April 25-29, 2004. [37] M. Greiner and A. Suo-Anttila, "Validation of the ISIS Computer Code for Simulating Large Pool Fires Under a Varity of Wind Conditions," in ASME J. Pressure Vessel Technology, 2004. Page 16 of 17 RANS vs LES CFD for Gas-Fired Combustion Equipment Analysis AFRC 2017: Industrial Combustion Symposium [38] M. Greiner and A. Suo-Anttila, "Fast Running Pool Fire Computer Code for Risk Assessment Calculations," in ASME International Mechanical Engineering Congress and Exhibition, Washington, DC., November 15-21, 2003. [39] M. Greiner, N. Are, C. Lopez and A. Suo-Anttila, "Effect of Small Long-Duration Fires on a Spent Nuclear Fuel Transport Package," in Institute of Nuclear Materials Management 45th Annual Meeting, Orlando, FL, July 18-22, 2004. [40] A. Suo-Anttila and J. D. Smith, "Application of ISIS Computer Code to Gas Flares Under Varying Wind Conditions," in American Flame Research Committee International Symposium, Houston, TX, October 16-18 (2006). [41] J. D. Smith, A. Suo-Ahttila, S. K. Smith and J. Modi, "Evaluation of the Air-Demand, Flame Height, and Radiation Load on the Wind Fence of a Low-Profile Flare Using ISIS-3D," in AFRC-JFRC 2007 Joint International Combustion Symposium, Marriott Waikoloa Beach Resort, Hawaii, October 21-24 (2007). Page 17 of 17