OCR Text |
Show Advanced Models Although two-step chemistry mechanisms have been a standard approach for many combustion modeling applications, this approach ignores the minor and radical species which are strongly linked to the generation of pollutants such as NOx. NOx predictions for these types of solutions require "global" pollutant models that rely on empirical rates and constants that must be tuned from case to case. These global models have been used successfully for predicting overall furnace emissions (Fiveland and Wessel, 1991), but are much less reliable for burner or local variations. To correctly predict in-flame NOx, a kinetics mechanism with sufficient detail to capture the key NOx reactions must be used. This approach, which replaces the two-step chemistry with an increased number of detailed, elementary kinetics mechanisms, can provide information on minor and radical species included in the mechanism. However, to model hydrocarbon chemistry and nitrogen chemistry to correctly predict nitrogen oxides, a large number of reactions (over 2(0) must be included. The addition of this large number of stiff equations to the solution significantly increases the computational time required for even a modest two-dimensional prediction. To use these models effectively, a reduced mechanism set must be sought that provides sufficient detail for combustion and pollutant formation, while minimizing computer time. This is best accomplished by comparison of reduced mechanisms with complete mechanisms such as GRI-Mech (Frenklach, et ai, 1995). Practical mechanisms may be derived that will consist of between twenty and seventy reactions. As part of an on-going internal program at Babcock & Wilcox, several reduced kinetics mechanisms are being evaluated against combustion solutions completed with the full 279 reactions of GRI-Mech 2.11. Predictions with these mechanisms have been compared with the 2-step EDM predictions and the BERL data. An example of such a comparison is shown in Figure 13, which shows the standard 2-step mechanism (EDM), as well as a revised 2 reaction mechanism (EDC 2-Rx) and a 20 reaction combustion mechanism (EDC 20-Rx) compared with the full 279 reaction mechanism of GRI-Mech (EDC 279-Rx) and the measured data at two traverse locations. This comparison demonstrates significant improvement in the combustion prediction using the 20 reaction mechanism; however, this mechanism is not sufficiently detailed to permit prediction of in-flame NOx as shown for the full GRI-Mech prediction. The use of the BERL data set in this way provides a strong argument for the need for such data when evaluating advanced combustion chemistry models. CONCLUSIONS A set of in-flame and supporting data for an industrial natural gas flame has been successfully collected at the Sandia BERL test facility. The test was designed to provide a complete, well documented data set for validation of numerical combustion models of natural gas burners. The validation document builds on measurements from the GRI sponsored SCALING 400 program, and successfully used combustion modeling as an integral part of the program to guide and focus experimental testing. The collected data provides a baseline for evaluation of differences between idealized 2-D axisymmetric models, typically used for parametric studies, and more detailed 3-D models. The data also provides a vehicle for evaluation of advanced methods such as the implementation of detailed kinetics mechanisms. However, to provide a comprehensive validation data set, additional testing is still required. These measurements must provide quantitative information and visualization of in-flame radicals and further insight into importance of turbulence and improved turbulence models for combustion. In addition, a strong sensitivity case is required to quantify the model's ability to predict changes in performance with changes in operating conditions. Comparisons to date between predictions and data indicate that existing combustion models are capable of replicating many of the experimental trends and predicting main features of the natural gas flames, but that additional work must still be done to improve model performance. Physical models of chemistry and turbulence must be improved. This improvement remains a significant challenge. The implementation of detailed kinetics mechanisms has already shown great promise using the described data set; however, significant work remains to reduce the increased computational times required by these models. This effort must combine identification of reduced reaction chemistry that will provide pollutant predictions, and implementation of the algorithms on parallel or distributed computing systems. 8 |