Early-Stage Development of an Advanced Industrial Scale Multipoint Flare Burner

Paper from the AFRC 2013 conference titled Early-Stage Development of an Advanced Industrial Scale Multipoint Flare Burner by Matthew Martin.

The early stage development process and results for a new multipoint ground flare are presented. A rigorous design process was used whereby established tools allow selection and optimization of a robust flare burner design which in turn enables the optimization of the flare system as a whole in a later exercise. Designed experiments were used in conjunction with computational fluid dynamics to gain insight into the operation of a candidate design over a range of operating conditions in order to select the designs which meet or exceed the desired operating to specifications over the widest range of flare gas flows and pressures. Additional information was gathered which gave insight into the trade-offs between competing constraints in the design space. A method of observation is presented that has been used during the physical testing of flare burners which enables the visualization of flame even under adverse conditions. This visualization data can be directly linked to output variables of a CFD model for use in validation exercises. Screening experiments were simulated using a wide variety of different design concepts. The extent of the virtual testing would have been cost prohibitive if performed physically but due to the efficient use of commercially available CFD software a large variety of disparate design concepts are evaluated. A short discussion on some common design modifications is made in the context of the newly available simulation data. A discussion on the potential merits of a final candidate design is presented with a table of results comparing different designs.

© 2013 Callidus Technologies LLC, All Rights Reserved 1 Early-Stage Development of an Advanced Industrial Scale Multipoint Flare Burner Matthew Martin Callidus Technologies LLC UOP LLC, A Honeywell Company Tulsa, OK USA Email: [e-mail] ABSTRACT The early stage development process and results for a new multipoint ground flare are presented. A rigorous design process was used whereby established tools allow selection and optimization of a robust flare burner design which in turn enables the optimization of the flare system as a whole in a later exercise. Designed experiments were used in conjunction with computational fluid dynamics to gain insight into the operation of a candidate design over a range of operating conditions in order to select the designs which meet or exceed the desired operating to specifications over the widest range of flare gas flows and pressures. Additional information was gathered which gave insight into the trade-offs between competing constraints in the design space. A method of observation is presented that has been used during the physical testing of flare burners which enables the visualization of flame even under adverse conditions. This visualization data can be directly linked to output variables of a CFD model for use in validation exercises. Screening experiments were simulated using a wide variety of different design concepts. The extent of the virtual testing would have been cost prohibitive if performed physically but due to the efficient use of commercially available CFD software a large variety of disparate design concepts are evaluated. A short discussion on some common design modifications is made in the context of the newly available simulation data. A discussion on the potential merits of a final candidate design is presented with a table of results comparing different designs. Keywords Flare, Pipe Flare, Pressure-Relieving and Depressurizing Systems, Computational Fluid Dynamics 1. INTRODUCTION Multipoint ground flares (MPGF) are used to provide smokeless operation over a wide range of flow rates and are suitable for disposing large flow rates of process gas. The process gas is distributed by a manifold to several ‘runners', or smaller pipes. Each of the flare runners may be staged on or off via a controller depending on the flow rate of gas arriving at the flare. Along the length of each runner burner heads are distributed which perform the final entrainment of air and required mixing for smokeless combustion of the flare gas to occur. The burners of an MPGF are typically mounted approximately 10 feet from grade in order to provide low visibility of the flare system to the surrounding community. Due to the proximity of the potentially large source of thermal radiation of the flare at full release the burners are surrounded by a fence. This fence serves the dual purpose of also protecting the burner flames from distortion and potentially poor mixing caused by interaction of the fuel-gas laden jet plumes with cross-winds. Minimization of the flame volume relative to a fixed heat release from each burner head is desirable in order to enable optimal capital cost reduction either by reducing the size of the fence, the number of stages, number of burners, or the required plot space. A rigorous design methodology was employed in order to develop a new multipoint flare burner. This methodology enables the creation of a flare burner which not only meets the requirements for the service but also meets those requirements reliably and repeatedly. Computational fluid dynamics was first used to first screen the candidate designs. The sensitivities of various design variables are exposed though proper experimental design. A large number of designs are quickly screened using low fidelity simulations. The most promising designs can later be optimized for further improvement using higher fidelity simulations. Several of the most promising designs will be advanced to physical testing. During this phase of the development program characteristics of the burner performance which are currently only roughly estimated from CFD such as flame stability and soot production at low pressures will be evaluated. Additionally a new use of flame imaging in the industrial scale will be used to validate the CFD simulations. © 2013 Callidus Technologies LLC, All Rights Reserved 2 2. PRELIMINARIES AND REVIEW 2.1 Flare and Flow Description The flare design to be improved consists of a flare gas distribution manifold which feeds multiple burner heads surrounded by a fence. The fence is designed to allow air flow into the interior of the flare field while shielding the exterior of the flare field from thermal radiation. The extent of the flare field may vary depending on the application and is controlled by the specific site requirements including specified flow rates, flare gas composition and available plot space. Figure 1 shows a picture of a typical MGPF and the burner heads inside the flare field. Each of the burner heads is spaced a critical distance such that when the first burner in a row of burners is lit the flame will propagate along the entire row of burners. If the burners are too close to each other relative to the individual burner flame diameter there will not be adequate air flow for combustion and the flames will become long and produce large volumes of soot. Since one of the primary benefits of a MPGF is low visibility the projection of flames and soot above the fence is generally not acceptable. Therefore and additional design constraint is that the burner flame are not too wide relatively to the desired burner spacing. A nominal flow rate of 5,000 lb/hr of propane with 12 psig supply pressure to each burner head is used for initial screening part this design exercise. Figure 1 - (Top) A typical multipoint ground flare. (Bottom) The burners inside a typical multipoint ground flare. 2.2 Key Performance Measures The key performance measures of the flare burner are the fraction of heat released as radiation, the flame length, the flame diameter, and the visible flame volume. The soot production in the flame front has a strong influence on both the fraction of heat radiated and the visible flame volume and is also compared. The fraction of heat radiated by the flame should also be minimized; if the combustion intensity is increased in order to reduce the flame volume in such a way that the thermal radiation from the flame increases the size of the flare field might have to be increased in order maintain a safe operating temperature for the flare components. The design of the burner head must also be robust. The temperatures inside a MGPF can be high enough to damage poorly designed burner heads; the final design must be resistant to thermal stresses resulting from the high thermal radiation inside the fence. 2.3 Computational Fluid Dynamics Simulation The software package ‘Fluent' was used to perform the computational fluid dynamic calculations. This software has been used to simulate a wide variety of industrial flows with reasonable accuracy. The simulations performed are steady Reynolds averaged Navier-Stokes calculations. The number of simulations to be performed in the screening phase makes time-dependant calculations unfeasible. The fluid was modeled as an ideal gas because the flare gas orifices typically operate with choked flow. Pressure boundaries were used for the atmospheric inlets and outlets and a mass flow inlet was used for the flare gas (fuel) inlet. Turbulence was modeled using the ‘realizable k-epsilon' (RKE) model. The other standard models can become unrealistic in high strain flows, and the RKE model is reported to be more accurate in predicting the spreading rate of round jets [1]. Due to initial convergence difficulties the k-omega model was also used and found to have superior convergence qualities; however agreement between the CFD model the physical test data from the base case was not as good as that of the RKE model. Chemical reactions were modeled using a two-step oxidation model, specifically the ‘eddy-dissipation' model available in Fluent. The chemical reaction rate is limited by the mixing rate of the fuel and its surroundings [3]. For these particular simulations there is good reason to believe that partially premixed effects may be important in calculating the flame and so more a more advanced treatment of turbulence-chemistry interaction should be used that takes into account the chemical kinetics as well as the turbulent mixing rate. For the initial screening simulations ignition in the CFD model proved problematic for the Finite Rate/Eddy Dissipation model. More exacting chemistry-turbulence interaction was therefore abandoned until the final optimization of the new design. © 2013 Callidus Technologies LLC, All Rights Reserved 3 Figure 2 - (Left) Grid dimensions and boundaries. (Right) A single burner head atop a flare gas riser. The Discrete-Ordinates (DO) radiation model was be used. The DO model calculates the propagation of thermal irradiation via a number of control angles, and so a sufficient number of control angles must be used. Additionally, an appropriate mean beam length must be used in the calculation of the absorption coefficient. Soot production was modeled using the ‘Two-Step' model available in Fluent. Preliminary comparisons by this author with data from other fired equipment have shown qualitative agreement between the high soot regions and the most luminous portions of the flame. Fluid properties were defined using Lennard Jones potentials. No attempt was made to account for dissociation either through direct calculation or modification of the fluid specific heats, which might also negatively impact the radiation solution. 3. APPROACH 3.1 Pugh Concept Selection A Pugh concept selection matrix was used to screen eleven conceptual designs against ten different criteria. None of the specific designs selected at this stage actually produced a substantial improvement in the performance of the burner, but some of the concepts introduced were carried forward into subsequent CFD calculations. 3.2 Sensitivity Analysis For the initial burners a sensitivity analysis was performed for the critical dimensions of each design. This step is necessary to quantitatively determine the impact of design changes on the performance of the burner. Process variables are allowed to vary and robust operation is sough in the case of changing process conditions. The sensitivities are exposed through a factorial designed experiment. A CFD simulation is performed each change in input variable as prescribed by a factorial design experiment. 3.3 Screening CFD The screening models only simulate a single burner head with the optimization of the entire flare field being left to a later exercise. The computational grid for each design averaged approximately 4 million cells in size. Each individual fuel gas orifice and a portion of the internal pipe flow leading to the burner head were also included in the simulation. The meshes used were intentionally coarse to facilitate fast calculation but not so coarse that the flame dimensions from the current (base) design could not be reproduced. Figure 2 shows the extent of the computational domain as well as a typical burner head mesh. A succession of substantially different designs was then virtually tested using these CFD screening models. Initially the current best design and six alternate designs determined by the Pugh concept selection process were simulated. Subsequent to the initial designs a series of alternate designs were simulated. 3.4 Physical Testing and Validation At the time of the writing of this paper the physical testing of the new design has not been completed. For this test program it is planned that a infrared camera with a specialized band-pass filter be used to visualize the flame and provide additional information whereby the CFD simulations can be validated. Figure 3 shows the results from using this camera to visualize a UOP Callidus flare burner. The flame is smokeless and essentially invisible in daylight in the left-hand view. In the right hand view one can clearly see the same flame against a dark background. A filtered and averaged image may supply data than can be compared directly to CFD simulations. 4. RESULTS 4.1 Flame Visualization As explained by the author in ‘Comparison of Empirically Based Calculation Methods for Pipe Flares to Computational Fluid Dynamics' [2], a weighted iso-surface of carbon monoxide is used to estimate the visible flame. The use of a carbon monoxide threshold to represent the visible flame surface may be less accurate for the yellow flames of most hydrocarbon fuels and is arguably overly conservative; carbon monoxide could conceivably be present in a non-sooting flame which may be invisible in the open atmosphere. Despite the potential shortcomings of this visualization method it has nonetheless been able to quantitatively reproduce the observed flame length from physical testing to within 10% of those observed by transit. Figure 3 - (Left) Image of a clear, soot-free flame firing in the open atmosphere. (Right) The same flame imaged with an appropriately filtered camera. © 2013 Callidus Technologies LLC, All Rights Reserved 4 4.2 Radiation Results The fraction of heat radiated from the burner was tabulated by integrating the radiation incident on the boundaries of the simulation domain and dividing by the total heat released. This calculation results in a value for the fraction of heat released as radiation in the same manner in which it is used in the API ‘Clause 6' stack-height calculation method of ‘API Standard 521' [3]. The resultant values from the simulations are similar to but lower than those listed in the GPSA Engineering Data Book [4] as might be expected from an advanced flare burner design. 4.3 Results of the Sensitivity Analysis The results of the sensitivity analysis were both trivial for those well-studied in the behavior of flare flames and rewarding insofar as they confirmed and quantified established observations Figure 4 shows example results of the sensitivity analysis for one of the designs. Statistically significance with an alpha value of 0.5 in the Pareto chart can be determined whether the bar associated with the input variable crosses the vertical red line. In the upper-left hand chart one can see that flare gas flow affects the total thermal radiation from the burner just as one would intuit; pressure has no statistically significant affect on the total radiation emitted by the burner head compared to that of flow over the range of flows and pressures simulated. This observation seems reasonable: As the flare gas flow rate to the burner increases the amount of heat radiated from the flame increases. As the design of the burner is changed to operate at a different pressure for a given flow it also affects the total thermal radiation, but not nearly so much as a change in flare gas flow rate. The upper-right hand corner of Figure 4 shows that neither pressure nor flow have a statistically significant affect on the fraction of heat released as radiation over the range tested. This may also be explained by the conjecture that the basic design of the burner head combined with the flare gas being fired are the primary determining factors in the fraction of heat and not the flow rate or the supply pressure. The lower-left hand corner of Figure 4 demonstrates that both the flare gas flow and the supply pressure have a statistically significant affect on the flame length. The lower-right hand corner of Figure 4 shows the normal plot of standardized effects which not only shows the magnitude and significance of the input variables but also shows the sign: Increased flow increases flame length while increased pressure decreases flame length. These affects are well known but the useful piece of information gained is the relative strength of the effect. In other words, one now knows how much the supply pressure must be increased to maintain a given flame length if the flow rate is also increased. . Figure 4 - Clockwise from upper left: Pareto chart of the affect of flow and pressure on the total heat radiated from the flare; Pareto chart of the affect of flow and pressure on the fraction of heat released as radiation; Normal plot of the affect of flow and pressure on flame length; Pareto chart of the affect of flow and pressure on flame length. © 2013 Callidus Technologies LLC, All Rights Reserved 5 Figure 5 - (Left) A basic design (BASE) and a large diameter design (LD). The sensitivity analysis exposed differences between the designs that may be important for operation that are not immediately apparent. Several basic design modifications were included in the sensitivity analysis and the screening analysis. One of the modifications tested was to simply reduce the number of arms on the burner head and increase the length of each remaining arm. This straightforward modification allows for increased inter-port spacing for the plurality of flare gas orifices. Figure 5 shows a comparison of a basic (BASE) design to a large-diameter (LD) design. The simulations revealed that when the flare heads were operated at nominal pressures and flow than the large diameter design achieved calculated a 12% reduction in flame length. However, this reduction in flame length also resulted in an 11% increase in the fraction of heat released as radiation at the same nominal operating point. When the analysis is expanded to explore sensitivities further insight is gained. Figure 6 shows the normal plot of standardized effects for the LD design. When compared to the normal plot of standardized effects for the BASE design shown in the lower-right hand corner of Figure 5 one can see that the affect of increased flow on flame length for the large diameter design is much greater for the LD design when compared to the BASE design. At nominal flow rates the flame length is shorter for this particular large diameter design, but if one attempts to capitalize on this advantage in flame length by increasing flare gas flow rate the flames actually become longer than those of the BASE design. An analysis of the relative fraction of heat released as radiation from the two different designs is also instructive. Figure 7 shows a box-plot of the fraction of heat released as radiation. The data is normalized to the mean flow rate and pressure performance of the BASE design. The boxplot shows that not only is the average fraction of heat radiated by the BASE design lower, but that also the variation in the fraction of heat radiated in the face of differing flow rates is less. This results in more predictable and reliable performance of the burner design regardless of the input flow rates. Another design explored during the sensitivity analysis in comparison to the BASE design was the same basic design pattern with a flare gas orifice disposition change such that shorter flame lengths would result. Figure 6 - Normal plot of standardized effects for the LD design. An increase in flow rate greatly increases the flame length compared to the BASE design. Figure 7 - Boxplot of the fraction of heat released as radiation from the BASE design and the LD design normalized to the mean flow rate results of the BASE design. The BASE design displays more reliable, repeatable radiation performance over a wide range of flow rates and supply pressures. © 2013 Callidus Technologies LLC, All Rights Reserved 6 Figure 8 - On the left a volumetric rendering of the visible flame for the base case [BASE] compared the right hand view of a design with altered flare gas orifice disposition [ALT01]. Although the flame length is shorter it is also too wide for a conventional flare field layout; the flames will merge, become longer and likely produce smoke. Figure 8 shows the result of this change with a volumetric rendering of the estimated visible flame. It is possible to reduce the height of the flame but the total volume of combustion remains roughly constant. This results in flames that are too wide for a conventional flare field layout. When fired to maximum duty the flames will likely merge, become substantially longer, and produce smoke due to inadequate combustion air at the flame surface. This effect is likely not evident in a single burner test or even in a multiple burner firing scenario until high flow rates of flare gas are achieved. 4.4 Results from the Screening CFD Study Table 1 shows the results of the screening CFD study normalized against the performance of the baseline design. All of the configurations in the table are designed to utilize the same flare gas pressure at 5,000 lb/hr of flow. The images appended to this paper show the volumetric rendering of the flames for each of the designs. Of the designs that reduced the flame length by 20% or more there was an average flame length reduction of 58%. However, the radiation from these flames increased by an average of 19% and the diameter or width of the flame increased by 165%. Design 29 substantially reduced the flame length without increasing the flame diameter to an unmanageable size. It has been selected for further optimization. The reduction in flame length did not come without an increase in combustion intensity and therefore the fraction of heat radiated has increased over the BASE case. In its current form this design has the most reduction in flame length combined with the fewest other negative side effects. The increase in radiation can likely be managed either through changes to the tip design itself or though changes within the flare field. Values Calculated from Screening CFD Normalized to the Base Design Design Fraction Heat Radiated Flame Length Flame Diameter/ Width Flame Volume (A) Flame Length / Flow Rate (B) Flame Diameter / Flow (A+B)/2 Base 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1 1.11 0.88 0.96 0.60 0.89 0.98 0.93 2 1.01 1.02 0.97 0.98 1.03 0.98 1.00 3 1.11 0.72 2.70 1.20 0.49 1.81 1.15 4 1.19 0.35 3.64 0.57 0.34 3.67 2.00 5 0.98 1.01 1.00 0.99 1.03 1.00 1.01 6 1.04 1.35 0.96 1.26 1.37 0.98 1.17 7 1.06 0.41 2.26 0.50 0.40 2.29 1.34 8 1.15 1.04 1.14 1.40 1.06 1.14 1.10 9 1.12 0.73 1.12 1.41 1.03 1.14 1.09 10 1.08 0.95 1.08 1.26 0.97 1.10 1.03 11 1.61 0.37 4.71 0.52 0.37 4.76 2.57 12 0.90 0.42 3.05 0.72 0.43 3.07 1.75 13 1.13 0.68 3.16 0.63 0.69 3.19 1.94 14 0.97 1.01 1.00 0.99 1.03 1.00 1.01 15 1.04 1.10 1.06 1.00 1.11 1.07 1.09 16 1.04 0.93 1.03 1.13 0.94 1.05 1.00 17 1.16 0.71 2.15 0.29 0.71 2.15 1.43 18 1.09 1.32 1.24 1.76 1.32 1.24 1.28 19 1.17 0.89 3.60 0.65 0.89 3.60 2.25 20 1.12 1.00 1.09 0.99 1.00 1.09 1.04 21 1.15 1.01 3.16 1.30 1.01 3.16 2.09 22 1.02 1.05 2.48 1.51 1.05 2.48 1.76 23 1.00 1.25 1.07 1.36 1.25 1.07 1.16 24 1.08 0.92 1.40 1.04 0.92 1.40 1.16 25 1.33 1.35 0.93 1.49 1.35 0.93 1.14 26 1.62 1.23 1.06 0.92 1.23 1.06 1.14 27 1.39 1.26 0.88 1.10 1.26 0.88 1.07 28 1.38 1.34 0.89 1.04 1.34 0.89 1.11 29 1.45 0.80 1.16 0.90 0.80 1.16 0.98 Table 1 - Selected normalized measures resulting from the CFD screening study. © 2013 Callidus Technologies LLC, All Rights Reserved 7 5. CONCLUSIONS Well established optimization techniques combined with CFD simulation have been successfully used to investigate the operation of candidate flare burner designs over a range of operating conditions. Investigation of the sensitivity flame length and fraction of heat released as radiation to flow rate and pressure revealed that for some designs performance at nominal pressures and flow rates was not replicated across a likely range of operating conditions. A large number of disparate designs were screened using CFD simulation resulting in a candidate design for further optimization that meets the specified criteria for improved flare design. Subsequent physical testing combined with advanced visualization techniques will provide further validation for the CFD studies as well as further evidence of desirable burner performance. 6. REFERENCES [1] Documentation for Fluent 14.5. www.ansys.com , as of August 2013 [2] Martin, M. ‘Comparison of Empirically Based Calculation Methods for Pipe Flares to Computational Fluid Dynamics'. Presented at the 2007 Joint JFRC-AFRC, 2007 [3] Pressure-relieving and Depressuring Systems. API Standard 521, Fifth Edition, January 2007 [4] GPSA Engineering Data Book, 12th Edition, 2004 © 2013 Callidus Technologies LLC, All Rights Reserved 8 © 2013 Callidus Technologies LLC, All Rights Reserved 9 © 2013 Callidus Technologies LLC, All Rights Reserved 10 © 2013 Callidus Technologies LLC, All Rights Reserved 11 Note: Due to a software bug in the post-processing software a volumetric rendering could not be made in the same manner as the other cases for Design 29. Instead an iso-surface representing the flame surface is shown. The burner head portions of the image are redacted.