AFRC 2014 Industrial Combustion Symposium Use of CFD in Evaluating Pyrolysis Furnace Design Bradley Adams, Marc Cremer Reaction Engineering International adams@reaction-eng.com 801-364-6925 x18 ABSTRACT This paper examines the usefulness of CFD modeling in evaluating burner and furnace performance in new or retrofit pyrolysis furnaces. A CFD model of a new furnace is often the first evaluation of how the furnace will perform when integrated with burners and process coils. Model results allow burner and furnace vendors to review performance and identify any sub-optimal design elements before construction or retrofits. The paper includes a brief review of the CFD model capabilities required for accurate furnace simulation, followed by examples drawn from a decade of modeling experience demonstrating how predictions have been used to identify and improve sub-optimal designs. Examples will include impacts of burner-end wall spacing, ultra-low NOx premixed burner implementation, and burner design trade-offs for flame shape, heat flux profiles, and NOx emissions. INTRODUCTION Pyrolysis furnaces such as those used for ethylene cracking are widely used in the chemical process industry to transform hydrocarbon feedstocks into more useful chemicals. The radiant furnace in an ethylene cracking plant uses energy from gas-fired flames to heat and chemically crack a hydrocarbon feedstock flowing in process coils (tubes) within the furnace. The rate and profile of heat transfer to the process coils can significantly impact the process yield and availability. The heat transfer behavior in a cracking furnace is primarily controlled by the fuel burners, which are tasked with providing appropriate flame shapes and heat release profiles, and low NO emissions. New and improved burners are constantly under development to meet increasingly strict performance requirements and NO limits. Burners are routinely evaluated in test furnaces prior to installation in full-scale cracking furnaces; however, this testing is necessarily limited in furnace size, geometry, boundary surface conditions and number of burners. Computational fluid dynamics (CFD) modeling provides a cost-effective method for evaluating furnace performance and NO emissions for new furnaces and/or new burner technologies, thus minimizing furnace start-ups times and unforeseen performance impacts. In the case of new furnaces, the CFD modeling is often the first evaluation of the integrated burner-furnace system. To provide useful information, CFD models must be capable of accurately representing the complex geometries, turbulent mixing, combustion, heat transfer and finite rate chemical reactions occurring in the radiant furnace. This is particularly true of the current generation of ultra low-NOx burners. CFD models of cracking furnaces should include: • 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-coil heat exchange; • flow and chemistry sub-models to account for turbulence-chemistry interactions; and • finite-rate chemical kinetics sub-models to account for ppm-level NOx reactions. AFRC 2014 Industrial Combustion Symposium Reaction Engineering International (REI) uses a CFD code called ADAPT that is specifically designed to model combustion and heat transfer associated with ultra-low NOx burners in ethylene cracking furnaces. The characteristics of this software have been previously described [Adams et al., 2007, Tang et al., 2009]. ADAPT capabilities include: • localized mesh refinement to capture burner geometry details and near-burner mixing; • reduced chemical kinetic mechanisms with in-situ adaptive tabulation (ISAT) to compute finite rate chemistry; • discrete ordinates method with SLW gas property model for solving radiative transfer equation; • a combination of conventional Eulerian turbulence modeling with Monte-Carlo PDF methods to model turbulent reactions and mixing; and • a matrix-free Newton-Krylov method to reduce solver computational time and improve robustness. Over the past eight years REI has modeled more than twenty ethylene cracking furnaces to evaluate the performance of different burner designs. A CFD model of a new furnace is often the first evaluation of how the furnace will perform when integrated with burners and process coils. Model results allow burner and furnace vendors to review performance and identify any sub-optimal design elements before construction or retrofits. The following sections summarize three examples which illustrate how CFD predictions have been used to identify and improve sub-optimal designs. EXAMPLE 1 - Burner Arrangement for New Burner This Shaw (now Technip Stone & Webster) furnace utilized John Zink SOLAR NOx Reduction technology that included a combination of floor burners and wall burners. Figure 1 illustrates the arrangement of one set of floor burners and wall burners in a section of the furnace. The full furnace combined 20 LPM Floor (LPMF) burners (10 burners on each side of the furnace; a total of 80 premixed venturis, 40 per side) with five rows of wall-fired burners. The LPMF burner consisted of venturi with the lean premixed air and fuel, one tertiary staged fuel tip per pair of venturi, and staged tips along the furnace floor. The burners in the top row were LPM Wall with Staged Fuel (LPMW-SF) (10 per side). The second and third rows each consisted of 20 LPMW burners without staged tips (10 per side). The fourth and fifth rows each included 20 Remote Fuel Staged (RFS) injectors (10 per side). These injectors are, in effect, the staged tips associated with the LPMW burners in the two rows above. All burners were provided by John Zink. Figure 1. Schematic of Shaw furnace burner arrangement. AFRC 2014 Industrial Combustion Symposium Figure 2 illustrates the layout of the Lean Premixed Flat Flame (LPMF) burner. This burner utilizes lean premixed combustion (small amount of fuel mixed with incoming air), quasi-flameless combustion (oxidant and fuel introduced separately into hot flue gases, each designed to mix with flue gases before mixing with each other), and traditional fuel staging (fuel introduced near air inlet, designed to mix quickly with air for flame stability) to achieve the desired heat flux and emissions performance characteristics. Because of the relative complexity of the SOLAR system, CFD modeling was an essential step in selecting the location of the various combustion elements, particularly the floor burners. In this project, REI used its state-of-the-art reacting CFD software ADAPT to evaluate furnace performance for specific burner configurations and firing conditions used. Full-scale Furnace Modeling The key modeling objective for this project was to determine furnace performance for different locations of the combustion elements and thereby determine the final design (optimal combustion element locations) of the SOLAR system. Furnace performance was evaluated based on some of the following parameters: • Flame shape as defined by CO concentrations and gas temperatures, • Furnace CO and NOx emissions, • Furnace exit temperature (hence radiant efficiency), • Heat transfer to process coils, • Heat flux profiles to process coils, • Process outlet temperature (coil outlet temperature, a.k.a. COT). Several different burner configurations were modeled for this project in order to identify improved design options. Most changes focused on the venturi and staged fuel tip configuration along the furnace floor. Each of the configurations (or cases) will be discussed in turn below. All cases used the same furnace operating conditions, inlet fuel and air properties, burner tip design, furnace boundary conditions, and process fluid conditions. Fuel GasAirEductor(Venturi)PrimaryUltra LeanPremixFlameStaged RiserTertiaryRiserFuel RiserTertiaryRiser Figure 2. Schematic of John Zink LPMF burner. AFRC 2014 Industrial Combustion Symposium Case 1 The initial configuration modeled (Case 1) was based on the floor burner arrangement shown in Figure 3. The most significant result of Case 1 was the biased flow profiles. The predicted temperature and velocity profiles shown in Figure 4 illustrate this flow bias. Figure 4 plots the gas temperature and velocity profiles at three locations near the burners - adjacent to the wall, at the centerline of the venturis, and at the centerline of the first row of staged tips. The burner flames were pushed toward the end wall and a large recirculation zone formed at the left side of the burner (left edge in the plots). After internal review, it was decided that the spacing between burner blocks (group of ten venturis) may have been too large. Burner TileStaged Fuel TipsTertiary Fuel Tips6-Venturi Section4-Venturi SectionEnd WallBurner Wall Figure 3. Case 1 floor burner configuration. GasTemp.3000 °F500 °FAdjacent to wallVenturicenterline1strow staged tip centerlineAdjacent to wallVenturicenterline1strow staged tip centerlineVelocity 70 ft/s0GasTemp.FGasTemp.s0Velocity s0 Figure 4. Case 1 temperature (left) and velocity (right) contours. AFRC 2014 Industrial Combustion Symposium Case 2 After reviewing the Case 1 results, the decision was made to modify the burner spacing in an attempt to reduce the recirculation zone predicted between burners. Figure 5 shows the modified burner design used in Case 2. Here the 10-venturi burners were divided into a 4-venturi and 6-venturi blocks in order to distribute the flames more evenly along the length of the furnace. All wall burners and fuel firing conditions remained unchanged from Case 1. Results indicated that the modified spacing of the burners did, indeed, result in more uniform flame distribution in the furnace. Figure 6 shows gas temperature and velocity plots comparable to Case 1 (Figure 4). The profiles were clearly more uniform with higher temperatures and velocities focused in the center of the burner group where the 6-venturi clusters were. The higher temperatures were caused by the additional fuel release in this region from the second row of staged tips. Velocity 70 ft/s0GasTemp.3000 °F500 °FAdjacentto wallVenturicenterline1ststaged tip centerlineAdjacentto wallVenturicenterline1ststaged tip centerlineVelocity s0Velocity FGasTemp.centerline Figure 6. Case 2 predicted gas temperature (left) and velocity (right) contours. Burner TilesStaged Fuel TipsTertiary Fuel Tips6-Venturi Burner4-Venturi BurnerEnd WallBurner Wall Figure 5. Case 2 floor burner configuration. AFRC 2014 Industrial Combustion Symposium Case 3 After reviewing Case 1 and 2 results, the decision was made to redistribute some of the fuel introduced by the second row of staged tips in a more uniform manner. The modified distribution of secondary row staged tips improved the gas temperature profiles in the furnace. The area above the 6-venturi burner blocks was still higher in temperature than the area above the 4-venturi burner blocks, but the difference was less than in Case 2. Figure 7 again plots the gas temperature and velocity profiles at three locations near the burners. Compared with Figure 6, the profiles were even more uniform. The peak gas temperatures away from the burners also decreased slightly relative to Case 2. Case 4 The objective of Case 4 was to determine if complete elimination of the second row of staged tips might improve even further the uniformity of heating of the process coils. The elimination of secondary row staged tips did not produce any major changes to the furnace performance. There were only minor changes in the flow patterns, gas temperature profiles, species concentration profiles, and heat flux values. All of the characteristics from Case 3 remained the same. Figure 8 plots the incident radiative flux arriving at the process coils for Cases 2, 3 and 4. The plotted values were an average of the normal incident flux for a group of five coils. The profiles can be divided into two groups - coils near the 6-venturi burners (coils 6-10 and 11-15) and coils near the 4-venturi burners (coils 1-5 and 16-20). The flux profiles reflected the higher flame temperatures above the 6-venturi burners. The overall average is also shown (dark blue). The difference between the lowest peak incident flux (coils 16-20, nearest the furnace end wall) and highest peak incident flux (coils 6-10, opposite the 6-venturi burner block farthest from the end wall) was 5568 Btu/hr-ft2 for Case 2, a relatively large value suggesting that the uniformity of heat flux distribution to the process coils should be improved. For the Case 3 burner configuration, the general profile trends were the same as Case 2, but the elevation for the peak fluxes across all coils was more uniform, supporting the observation that the temperature profiles in the furnace were more uniform. Similarly, the difference between the lowest and highest peak flux averages was only 3114 Btu/hr-ft2; just over half the corresponding value in Case 2. One outcome of this more uniform heat flux was that Velocity centerlineVelocity 70 ft/s0Velocity s0GasTemp.3000 °F500 °FGasTemp.FAdjacentto wallVenturicenterline1ststaged tip centerlineAdjacentto wallVenturicenterline1ststaged tip centerline Figure 7. Case 3 predicted gas temperature (left) and velocity (right) contours. AFRC 2014 Industrial Combustion Symposium the average heat flux was slightly higher in Case 3, but the highest peak flux was lower for the coils adjacent to the 6-venturi burner block. This led to lower peak metal temperatures for these coils as well. For the Case 4 configuration, the incident flux profiles were similar to Case 3, but the elevation for the peak fluxes across all coils was even more uniform and the difference between the lowest and highest peak flux averages was reduced an additional ten percent from 3114 Btu/hr-ft2 to 2748 Btu/hr-ft2. The small change in incident heat flux values was consistent with the minor changes in gas temperature profiles and was also consistent with the minor changes in heat absorbed by the process coils. Model Validation - Data from Test Furnace In accordance with normal practice, detailed burner tests were made in John Zink's Test Center in Tulsa, Oklahoma. The largest test furnace was used and the components in the test included both types of LPMF burners and five rows of RFS, LPMW and LPMW-SF burners, two per row, laid out in the same way as in the process furnace. The predicted heat flux profile (for the process furnace) is compared with the measured flux profile (in the test furnace) in Figure 9. Agreement is very good and reinforces the soundness of the underlying design basis of the furnace. CFD results predicted NOx levels at the furnace exit of 15-16 ppm (dry 3% O2) for Cases 3 and 4 whereas test measurements showed NOx levels of 17-21 ppm (dry 3% O2). Predicted levels of CO exiting the box of the process furnace were ~1 ppm. Exit CO levels measured during the test were non-detectable. 8000090000 5568 Btu/hr-ft28000090000 3114 Btu/hr-ft28000090000 ft2 8000090000 ft28000090000 2748 Btu/hr-ft2 Case 2Case 3Case 4 Figure 8. Incident heat flux profiles. AFRC 2014 Industrial Combustion Symposium EXAMPLE 2 - End Wall Burner Spacing An ethylene cracking furnace was simulated to evaluate burner-burner interactions and burner-end wall interactions. The floor burners were fired with a slight swirl, creating a tendency for the flames near the furnace end walls to lean towards or away from the walls. The baseline furnace geometry was modeled as shown in Figure 10. This approach utilized the two burners adjacent to the furnace end walls from each side of the furnace to achieve a total of eight burners (see Figure 11). This "partial" furnace model had the advantages of simulating burner-end wall interactions (previously seen as a design concern), burner-to-burner interactions in the center of the furnace, potential flame rollover, and heat fluxes to full sets of coil sections, while still completing detailed chemistry calculations in a reasonable time frame. For this case, extending the model to include more burners would have required significant increases in computer run times without providing a comparable improvement in furnace performance insight or accuracy. 70%75%80%85%90%95%100%051015202530354045Height Above Hearth (ft)Normalized Incident Flux at Coil Plane (%)REI CFD SimulationsMeasured in John Zink Test Furnace Figure 9. Comparison of predicted and measured normalized heat flux profiles. Figure 10. Baseline furnace geometry. AFRC 2014 Industrial Combustion Symposium Calculations for the baseline furnace were first completed with equilibrium chemistry, which provides a good estimate of the aerodynamic and heat release behavior of the burners in the furnace, but does not provide any detailed chemistry information such as NOx concentrations. These results showed that flame profiles were very poor (see Figure 12) in the baseline design, mostly due to recirculating flow regions between the furnace end walls and the adjacent burners. If these recirculation zones grow too large, they can impact the development of the flames emanating from the burners (see Figure 13). This behavior was seen in previous projects and was attributed to burner spacing. This led to the recommendation to modify the burner spacing in the furnace. For the new design, the burners nearest the end walls were moved 10% closer to the end walls. Use of this revised burner spacing eliminated the large recirculation zones that were distorting the flames (see Figures 14 and 15). Once the benefit of the revised burner spacing on flame behavior was established, the revised baseline model was run to completion with full finite-rate chemistry calculations. The revised baseline results showed good flame behavior and predicted heat absorption and NOx emissions within the design specifications. CO concentrations predicted with detailed chemistry indicated some minor interaction between flames and furnace end walls, but in general flame behavior was acceptable. The flame appeared to develop two distinct lobes on each side of each burner due to the location of the staged Figure 11. Baseline furnace 8-burner layout. Figure 12. Iso-surfaces of CO concentration (1500 ppmv). Velocity100 ft/s0Plane APlane B Figure 13. Velocity profiles at burner inlets. AFRC 2014 Industrial Combustion Symposium fuel tips and the design of the injector tips. There was no evidence of tendencies for flame rollover. The effectiveness of heat transfer to the process coils (44.44%) was in the expected range of 45% thermal efficiency (percent of fired duty absorbed by coils). The predicted NOx concentration (47 ppmvd) and CO concentration (<1 ppm) at the furnace exit were also in the expected range. EXAMPLE 3 - Burner Retrofit Study For this project, a large ethylene cracking furnace was to be retrofit with new floor burners in order to improve flame quality and maintain low NOx emissions. Furnace simulations were done for the original (existing) burners and three new burner options. Burner locations in the furnace were the same for all cases. Figure 16 shows the section of the furnace modeled, including the layout of the floor burners and the process coils. A ¼-furnace model was utilized to capture effects of burner-burner interactions and tube heat flux profiles, while maximizing burner resolution. The key metrics for measuring the performance of the three burner retrofit options were: • Flame quality (no propensity for flame rollover) • Low NOx and CO emissions • Acceptable heat flux profile Figure 17 shows flame iso-surfaces as defined by CO concentrations of 5000 ppm for the original burners and the three retrofit options. These iso-surfaces are helpful in visualizing flame behavior and flame quality. Figure 18 shows the predicted wall temperatures for each of Figure 15. Velocity profiles through burners for revised burner spacing. Velocity100 ft/s0Plane APlane B Figure 14. Flame iso-surfaces at 1500 ppm CO for revised burner spacing. Figure 16. Example 3 cracking furnace burner and coil layout. AFRC 2014 Industrial Combustion Symposium the burner configurations. These plots provide insight into the heat release profiles of the flames. Hotter wall temperatures correlate with higher flame temperatures and heat release. Figure 19 plots the NOx concentrations for the different burners at a vertical plane near the burner centerlines. The NOx concentrations show the cumulative NOx amounts throughout the furnace (as opposed to NOx formation rates which are localized in the flames). Original Burner Option 1 Option 2 Option 3 Figure 17. Flame iso-surfaces based on CO concentrations of 5000 ppm for four burner options. Figure 18. Furnace wall temperatures for four burner options. AFRC 2014 Industrial Combustion Symposium The original burners showed excellent flame attachment, but had some excess fuel between the burners resulting in higher CO levels at the floor of the furnace between burner tiles. It appears that some fuel becomes trapped between the burners. The flame is not as cohesive as desired, but does remain attached to the furnace wall. The flame attachment caused the heat flux profile to peak at a lower elevation than desired in the furnace, as seen in the wall temperature plots. These burners injected more fuel through primary tips than the other options, which may have contributed to less distributed fuel and oxidant mixing for these burners. The more rapid mixing and attached flame resulted in higher NOx emissions for these burners. The Option 1 burners showed a more cohesive flame (e.g., no fuel between burners), but did show some propensity for flame rollover (CO iso-surfaces moved more toward the center of the furnace). This "bushier" flame did not appear to attach to the wall as well as flames from the other burners. This was likely due to the secondary fuel jets being angled too close to vertical. As shown in Figures 17 and 18, this flame was the least attached to the burner tile. The peak heat flux for this flame was about 0.5 m higher than the original burner flames. This burner did produce the lowest NOx emissions of all the burner options. The Option 2 burners showed a cohesive flame that was well-attached to the furnace wall. The flame was reasonably well attached to the burner tile, but was also limited in height, which produced a peak heat flux elevation about 0.3 m higher than that for the original burner. NOx emissions for this burner were between original and Option 1 burner emissions. The Option 3 burners appeared to produce the best flame shape. Flames were well-attached to the furnace wall, and extended well above the burner. This produced a slightly more uniform heat flux profile with a peak heat flux approximately 0.8 m higher than that for the original burner. The overall heat transfer to process tubes was similar for all burner options, but varied locally between tubes for different burners. Although not shown here, the longer flame from this burner resulted in peak tube metal temperatures about 4°C lower than with the Option 2 burners. This burner design provided a good combination of flame quality and heat release distribution. NOx emissions for Option 3 burner were about 3 ppmv higher than Option 2, but still lower than emissions for the original burners. Original burner Option 1 Option 2 Option 3 NOx (ppmvd) Figure 19. NOx concentration at burner plane for four burner options. AFRC 2014 Industrial Combustion Symposium The greatest differences in burner designs were the distribution of fuel between primary and secondary tips and the injection angle of the secondary fuel tips. This caused differences in flame attachment, flame shape, rollover propensity, and heat release profile. From this perspective, burner Option 2 and Option 3 appeared to provide the best performance. All burners met the specification for NOx emissions and total heat absorption to the process tubes. CONCLUSIONS REI has utilized the ADAPT CFD software to evaluate proposed burner design and furnace performance, including NOx predictions, for a number of pyrolysis furnaces over the past eight years. Results showed CFD tools can be useful for predicting behavior in pyrolysis furnaces if: 1) the software contains appropriate sub-models for representing key mixing, combustion, finite-rate reaction, and heat transfer processes; 2) furnace geometry and operating conditions are accurately represented. Three examples shown here illustrate how CFD has been a useful tool for: • The design review process, where the CFD model is often the first analysis of the integrated burner-furnace system, and can identify potential problems with burner location, spacing, or firing. • Performance assessment of flame shape/rollover, heat flux profiles, process coil heat absorption and NOx emissions. • Quantitative evaluation of different technologies for furnace revamps or burner retrofits. Although not shown here, REI has also used CFD models to troubleshoot existing operations such process tube local hot spots and decoking particulate flows/burnout. It should be noted that CFD results are nearly always modeled for steady state operation, therefore evaluation of dynamic burner behavior, such as start-up and turn down, is best accomplished with physical burner tests. ACKNOWLEDGMENTS Technip Stone & Webster (formerly Shaw Energy & Chemicals) provided project support for the examples shown. John Zink provided the LPMF burner schematics and burner test measurements. CFD graphics were created with Fieldview software by Intelligent Light (http://www.ilight.com). REFERENCES Adams, B., Tang, Q., Ma, J., Brown, D., "Modeling Combustion in Pyrolysis Furnaces with Next Generation Low NOx Burners," AFRC-JFRC International Symposium 2007 Tang, Q., Denison, M., Adams, B., and Brown, D., "Towards Comprehensive Computational Fluid Dynamics Modeling of Pyrolysis Furnaces With Next Generation Low NOx Burners Using Finite-rate Chemistry," Proceedings of Combustion Institute, Volume 32, Issue 2, 2009, Pages 2649-2657.