Using CFD to Reduce Unburned Carbon during Installation of Low NO x Burners Lawrence D. Berg 1, John Goldring 2, Lyle Woodard 3, and Joseph D. Smith, Ph.D. 4 32nd International Technical Conference on Coal Utilization & Fuel Systems Agenda Clearwater, Florida, USA June 10-15, 2007 INTRODUCTION To meet mandated NO x levels, coal plants are using a variety of reduction methods: low NO x burners, Over Fire Air (OFA), Selective Non-Catalytic Reduction (SNCR), Advanced Reburn, Selective Catalytic Reduction (SCR), etc. In general, low NO x burners and OFA systems are installed as a first step. Combustion modifications of this sort reduce NO x by decreasing the amount of air near the primary combustion region, resulting in conditions favorable to fuel nitrogen being converted to diatomic nitrogen (N 2 ). Unfortunately, these conditions also tend to increase the amount of unburned carbon in the flyash. In many locations, elimination of flyash with high carbon content can become a significant economic liability. European plants, in particular, are interested in maintaining a low Carbon-in-Ash (CIA) level as this allows them to sell the ash, as opposed to having to pay for removal and disposal. During the design phase for the low NO x equipment retrofit of a tangentially fired (T-fired) pulverized coal (PC) boiler in the UK, the client (RJM Corporation) requested Computational Fluid Dynamics (CFD) be used to verify the CIA guarantees. Previous CIA modeling attempts with FLUENT (versions 6.1 and earlier) had produced unsatisfactory results. The char burnout model that had been implemented only allowed one specie (either CO 2 or CO) to be evolved from the char oxidation. If CO only was employed, CIA levels were unrealistically low (~ 0.01% CIA predicted compared to ~ 6.2% actual) and predicted CO from the furnace was too high. If CO 2 was employed as the only product of char oxidation, the predicted CIA levels were too high (~ 25% CIA compared to ~6.2% actual) and the predicted CO levels were too low. Clearly a balance was needed. Recent experience with coal gasification modeling indicated that the new multi-char reaction option in FLUENT 6.2 provided better estimates of CIA. This option allows for multi-path and multi-specie reactions with the char. As a commercial project with tight submission deadlines, prompt execution time was of the essence. This paper reports on a unique CFD methodology that was developed using FLUENT 6.2 for accurately predicting trends in flyash carbon. As will be seen, the model was able to predict CIA with remarkable accuracy. CIA MODEL Computational Fluid Dynamic modeling has been utilized for years [1, 2] on utility PC boilers to understand combustion dynamics and to predict NO x and CO trends. For example, Goldring and 1 Correspondence Author, Alion Science and Technology, Inc. Owasso, OK, USA lberg@alionscience.com 2 RJM Corporation, Ltd. 3 AES Kilroot Power Ltd. 4 Alion Science and Technology, Inc. Owasso, OK, USA jdsmith@alionscience.com
Berg present some of RJM s previous successful application of CFD to various projects and applications [3]. The CFD models (continuity, momentum, turbulence, species and reactions, etc.) that were utilized have been sufficiently documented elsewhere [4] and will not be addressed in this paper. What is of interest, however, is the approach to coal modeling that was employed / developed for the current project. For a T-fired PC coal furnace, experience has shown that the k-ε turbulence model, with either the RNG or Realizable modification reasonably simulates the turbulent flow inside the furnace. Even though there is a strong tangential furnace circulation, experience shows that the full RSM turbulence model is not required. The combustion model includes species conservation and uses the eddy-dissipation model. A single species is devolved from the coal, and spontaneously decomposes to form a combination of methane, CO, NO, SO 3 and water. The specie split is calculated to maintain a reasonable mass balance of the C, H, O, N, S and H 2 O from the proximate and ultimate analyses. Boundary conditions (air mass flow rates, temperatures, coal particle size distribution, etc.) are determined from field measurements and data. The two furnace configurations that were analyzed will be discussed in greater detail in the next section. The discrete phase model is used to model particle flow in the furnace. This model is based on a Lagrangian particle tracking technique, which traces a particle trajectory through the phases of coal combustion. Figure 1 shows a cartoon that illustrates the steps which occur during general coal combustion. For the computer code utilized in the present study, these steps are sequential, and proceeds as follows: 1. Particle begins to heat up - particle achieves the boiling temperature of water, the temperature does not change until enough heat has been absorbed to boil off all of the inherent water in the coal particle. Since specie conservation is used, the mass of water evolved from the coal is transferred to the gas phase. 2. Devolatilization - begins once water has been driven off coal particle. The particle heats up rapidly due to radiant interaction with the flame. Kobayshi et. al. [5] discusses volatile yield as a function of the conditions along the particle trajectory. It is important that a path dependent devolatilization method is employed. The one used for this study is based on the Kobayashi model which has been modified to allow for different volatile yields, depending on path conditions. 3. Char Burn-out - begins once devolatilization is complete. The multiple char reaction model was used. Bartok and Sarofim [6] discuss various competing reactions on page 695 of their book. They provide applicability guidance and list references for specific rate data. The interested reader is highly encouraged to review this information prior to attempted implementation. From previous work in coal gasification, the following reaction set was employed: C + O 2 => CO 2 Reaction 1 C + CO 2 => 2CO Reaction 2 C + H 2 O => CO + H 2 Reaction 3 Reaction 1 is exothermic, and Reactions 2 & 3 are endothermic.
4. Leaving Domain - freezes unburned carbon in particle. In addition to leaving the domain, the Lagrangian tracking methodology require a maximum residence time. If the particle trajectory time exceeds this maximum time, the particle is dropped out of the calculation. A good practice is to have a residence time long enough so that any carbon dropped out in this manner is at least one order of magnitude less than the total carbon leaving the furnace exit. MODELING OF EXISTING FURNACE OPERATIONS Figure 2 shows a wireframe of the furnace geometry with coal injection points identified. The furnace has four injection levels; A through D with D being the lowest injection level. Firing coal the furnace has a nominal capacity of 220 MW gross power production. The furnace is typically operated with three of the four levels in service, with the fourth level kept in reserve. This allows the plant to maintain a high level of on-line availability. A previous modification to the furnace included installation of offset secondary air buckets and Separated Over-Fire Air (SOFA) ports. The geometry of the furnace, existing SOFA ports, and burner corners were developed from drawings supplied by the client. As the plant performance is critical, AES-Kilroot Power Limited performed as series of baseline tests to help establish the existing performance of the plant. The furnace operates on both coal and oil, so baseline testing of various configurations for each fuel type was accomplished over about a 5-day period. Of specific interest to this work were two 220 MW configurations firing pulverized coal through: 1) A-C level burners, and 2) B-D levels burners. During each test, continuous measurements of nearly two hundred set points collected process information over approximately a four hour period. The information gathered during the baseline testing was used to set the flow conditions (velocity and temperature) for each air or fuel injection location. Modest adjustments were accomplished to ensure excess O 2 matched measured values. After the CFD model had converged, comparison of furnace exit values to measured data was accomplished to ensure that the CFD model reasonably reproduced field data. Table 1 compares the CFD model to actual data for CO, NO x and Furnace Exit Temperature (FEGT) for upper mills (A-C) in service with excess O 2 of 4.3%. Table 1 - Comparison of Data to CFD Upper Mills Operation Data CFD CO 11 ppm 2200 ppm NO x 641 mg/nm 3 * 649 mg/nm 3 * FEGT 0 C 1120 0 C 1158 0 C Carbon in Ash (CIA) 6.2% 22% * To convert mg/nm 3 to lbs/mmbtu, multiply by 6.655E-04 Iso-surfaces of the predicted CO, NO x, and furnace temperatures are provided as Figures 3 through 5. Since the CO was measured after the economizer, continued CO oxidation is expected in the post furnace region (i.e., economizer, etc.). While it was not possible to directly compare the CO numbers, the predicted values were typical of coal furnaces. NO x and FEGT match reasonably well. As discussed earlier in the paper the CIA prediction was so far off in
the initial CFD modeling that predicted trends were not expected to be meaningful. Thus, what was needed was a more accurate CIA model. Using RJM s combustion expertise and building upon experience gained from recent work modeling coal gasification, the three char reaction model discussed earlier was applied to the current CFD simulation. Starting reaction rate parameters were taken from literature sources outlined by Bartok and Sarofim [6]. Not all coals are identical, so it was anticipated that rational adjustments to the kinetic parameters would be required to match the data. Since each rate parameter adjustment required a complete re-convergence of the CFD solution and since the expected affect of the kinetic rate parameters have been shown to be nonlinear on CIA [7], it was not possible to completely match the predicted data. However, after just a few variations, a prediction of 6.48% CIA (compared to measured ~6.2% CIA) was accomplished. This comparison is not particularly remarkable as it was accomplished by adjusting kinetic parameters to achieve a good comparison. The second baseline test more critically tested the models ability to match observed CIA for a completely different operational scenario. Using the same methodology, a baseline CFD model of running the lower mills was executed. Results are presented in Table 2 (excess O 2 of 3.7%): Table 2 - Comparison of Data to CFD Lower Mills Operation Data CFD CO (ppm) 13 1770 NO x (mg/nm 3 ) * 554 560 FEGT ( 0 C) 1246 1185 Carbon in Ash (CIA) 4.4% 5.98% * To convert mg/nm 3 to lbs/mmbtu, multiply by 6.655E-04 Again, the NO x and FEGT predictions are reasonable while the CO is high but anticipated given the relative location of the predictions. In this case, using the adjusted 3 char reaction mechanism accurately predicted the trend (CIA went down for this operating scenario). This was especially encouraging, as the excess O 2 was reduced from 4.3% to 3.7%. In addition to the furnace exit predictions, the new CIA model also has the full range of diagnostic tools possible in a comprehensive CFD code. The following were particularly useful during the equipment design phase: Identification of CIA sources In addition to better accuracy, the new model allows for identification of carbon sources from individual injection locations. Table 3 shows the predicted Upper Mills Baseline CIA, broken down by coal injection point. In this case, the first letter is for the mill (a, b, c, or d mill) and the number is for the particular furnace corner where the coal was injected. Using this nomenclature, injection c1 means Corner 1, C-mill injection. 5 Interestingly, 52% of the predicted baseline Upper Mills CIA comes from the lowest level C mill. 5 Absolute location of each corner with respect to furnace nose and ash pit geometry can not be included due to the proprietary nature of the solution developed for the client.
Identification of C concentrations Visual diagnostics are also available to supplement the quantitative information. Figure 6 gives an example of this type of diagnostic. It shows coal particle trajectories colored by carbon concentration. The B4 injection shown starts off with maximum carbon concentration (red), and as carbon is oxidized, the path color becomes bluer, with dark blue representing nearly 0% carbon. This diagnostic not only shows where coal is going in the furnace, but where it is being oxidized. Table 3 - Upper Mills Baseline CIA by Coal Injection c4 0.03877 18.5% b4 0.017 8.1% c3 0.02169 10.3% b3 0.00411 2.0% c2 0.0344 16.4% b2 0.0185 8.8% c1 0.0142 6.8% b1 0.00587 2.8% a4 0.0296 14.1% a3 0.00649 3.1% a2 0.014 6.7% a1 0.00514 2.5% Particle / SOFA Interaction Using a 10% iso-surface of oxygen, Figure 7 combines the iso-surface with coal path lines colored by carbon concentration. This unique view shows how coal particles interact with surrounding air being supplied though either the offset or SOFA ports. LOW NO X BURNERS AND CIA MINIMIZATION After completion of baseline modeling, the proposed low NO x burners and modifications to the OFA system were modeled. Comparison to the baseline model is shown in Table 4. Table 4 - Comparison of Baseline to Upgrade (CFD) Upper Mills Operation Baseline Upgrade CO (ppm) 2200 4700 Excess O 2 4.3% 4.3% NO x (mg/nm 3 )* 649 430 Carbon in Ash (CIA) 6.45% 14.70% * To convert mg/nm 3 to lbs/mmbtu, multiply by 6.655E-04 As expected, lower NO x leads to higher CO and CIA. The roughly 33% NO x reduction predicted was sufficient to provide confidence in RJM s solution s ability to meet the NO x guarantees. The higher CO number was not as much of a concern. This was due to previous baseline furnace modeling results which indicated sufficient residence time to accomplish CO oxidation. However, the predicted increase in CIA number was troubling since it was critical that the CIA not increase above the value for the existing operation. The following shows some of the diagnostic power of the new modeling tool.
Three methods were employed to reduce CIA and maintain the same NO x reduction: 1. Upgrade rotary classifiers and reduce overall particle size, 2. djust angle and damper positions of SOFA ports to capture high amounts of CIA, and 3. Identify high CIA injections, and selectively add secondary air. Option 1 was simple to incorporate into the model by changing particle size distribution in the model input file. This was accomplished and the predicted CIA value decreased to about 10%. This was encouraging, but still not sufficient. The SOFA ports were installed as part of a previous modification and were installed fairly close (in vertical direction) to the furnace exit. As a result, when the furnace was operated on the upper mills the residence time for carbon burnout could become an issue especially with the lower furnace air level being lowered to obtain a compliant NO x performance. Figure 8 compares the carbon burnout for the base case and the upgrade. Clearly the SOFA ports need to be optimized. Figure 9 exemplifies how the SOFA ports were optimized. It shows a series of three injections that analysis had shown were significant contributors to the CIA problem. As seen, they merge together, in the back left corner. Changing the angle of SOFA air injection on the back right (top) port, virtually eliminated the CIA. This was a significant finding since the existing SOFA ports were fixed angle. The client, as part of the upgrade, now knew that SOFA ports with yaw angle control was required. Figure 10 shows the virtual elimination of CIA with upper mill operation. In this case a remaining SOFA port was angled into the center of the furnace to interact with the high concentration of CIA in center furnace. The final predicted CIA is 4.65% - nearly 28% less than the original base line prediction. Lower mill operation presented a different set of problems. As a practical matter, SOFA yaw angles could not be varied, so the angles that were set for upper mill operation had to be used for lower mill operation as well. This prevented using the SOFA angle optimization just outlined. When the upper mill SOFA settings were used, the CFD prediction for lower mill CIA was 7.5%. A good starting point, but still too high. Table 5 shows the CIA levels for lower mill operation using the optimized SOFA angles from the upper mill study. Table 5 - Upgrade CIA by Injection Source (Lower Mill Operation) Injection Mass Flow % of CIA d4 0.0672 26.6% d3 0.0272 10.8% d2 0.0575 22.8% d1 0.0226 8.9% c4 0.0083 3.3% c3 0.00766 3.0% c2 0.0131 5.2% c1 0.0323 12.8% b4 0.00025 0.1% b3 0.0055 2.2% b2 0.0048 1.9% b1 0.00616 2.4%
Unexpectedly, almost 70% of the CIA came from the lowest injection level (D mill) in this operating scenario. When this row of nozzles were modified to a more optimum angle of injection, the CFD model predicted a CIA of 3.45% - with most of the reduction from the predicted D mill injections. SUMMARY Table 6 shows the final CFD predictions for the upgrade. Table 6 - Comparison of Baseline to Upgrade (CFD) Upper Mills Operation Upper Mill Upgrade (CFD) Lower Mill Upgrade (CFD) CO 2400 ppm 2600 ppm Excess O 2 4.56% 4.3% NO x 430 mg/nm 3 * 416 mg/nm 3 * Carbon in Ash (CIA) 4.65% 3.45% * To convert mg/nm 3 to lbs/mmbtu, multiply by 6.655E-04 Often, these types of adjustments are made at start-up by experienced personnel which can be time consuming, labor intensive and costly. In this case, CFD analysis identified additional equipment modification (yaw angles to the SOFA ports, and modification to the D mill injectors / secondary air) and valuable operational settings prior to installation. REFERENCES 1. Smoot, L. D. and Smith, P. J., Coal Combustion and Gasification, Plenum Press, New York, 1985. 2. Fiveland, W. A., Wessel, R. A., Numerical Model for Predicting Performance of Three- Dimensional Pulverized-Fuel Fired Furnaces, AIAA/ASME Thermophysics and Heat Transfer Conference, Boston MA, ASME paper 86-HT-35, 1986 3. Goldring, J. Berg, L.D., How Experienced use of CFD Analysis allow Compliance with strict LCPD NOx Emissions Requirements, International Power Generation Conference (IPG), Leipzig, Germany, November, 2006. 4. FLUENT User s Guide, Fluent Inc., Centerra Resource Park, 10 Cavendish Court, Lebanon, NH. 5. Kobayshi, H., Howard, J. B., and Sarofim, A. F., Coal Devolatilization at High Temperatures, 18 th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, PA (1977), p. 411. 6. Bartok, W. and Sarofim, A. F., Fossil Fuel Combustion A Source Book, Wiley- Interscience Publication, 1991. 7. Smith, J.D., Smith, P.J., Hill, S.C., "Parametric Sensitivity Study of a CFD-Based Coal Combustion Model," AIChE Journal, Vol. 39, No. 10, October (1993).
Heat Up and drive off H 2 O Devolitization Char Burn-out Leave Domain Or Incomplete CO CO 2 H2O Volatile gas Coal particle O 2 Ash and Unburned Carbon Figure 1 Stages of Coal Combustion a b c d Coal Injection Levels Figure 2 Furnace Wire Frame and Coal Injections
Figure 3 Furnace 15,000 ppm CO Iso-Surface Figure 5 Furnace 1600 0 C Iso-Surface Figure 5 Furnace 1600 0 C Iso-Surface
Blue is 100% char Oxidation Red is no char Oxidation Figure 6 Identification of C Concentrations 10% O 2 Iso-Surface (from SOFA) Interactions Increase Blue of Coal particles Interactions also reduce O 2 Volume Figure 7
Initial Burnout not as complete Base Upgrade Figure 8 Base Case to Upgrade Carbon Burn-out Comparison Coal Inj. C Coal Inj. C & B Coal Inj. C & B & A Figure 9 Three Level Contribution to CIA <C> at Exit
High CIA in Center High CIA at Corners 2 & 4 Eliminated CIA Reduced by Aiming Left Back SOFA (lower) into center of furnace Figure 10 Elimination of Upper Mills CIA