OCR Text |
Show Results and Discussion The velocity fields obtained from both the eddy-breakup and the conserved scalar models are displayed along with experimental measurements as radial profiles at three different stations, 27 mm downstream of quarl, 109 mm downstream of quarl and 343 mm downstream of quarl (Figures 5-7). The formation of the internal and external recirculation zones can be clearly seen. Predictions obtained from the structured (FLUENT) and the unstructured (FLUENTfUNS) solvers are seen to be in excellent agreement. The unstructured solver is seen to perform just as well as t.he structured solver for predicting flow velocities indicating that it may be used the with the same degree of confidence as the structured solver. Comparing the different modeling schemes used in the unstructured solver (FL UENT fUNS) to the experimental data, both the eddy breakup and the pdf models are seen to overpredict the strengt.h of the reverse flow velocities near the centerline. It is believed that this is because of the 2d axisymmetric represent.ation of the 24 injection holes by an axisymmetric slit. The 2d assumption causes delayed mixing between the fuel and air, hence, the heat release in the near burner zone is smaller. Thereby the effective swirl number is not reduced [2] to the same degree as it would be in the 3D case, where the heat release would be higher. The peak velocities are also overpredicted. Further, the peak velocities do not decay as fast as indicated by the experimental dat.a. Tangential velocity profiles are shown in the next set of figures (Figures 8-10). The comparisons reveal that at 27 mm downst.ream of quarl, both the structured and the unstructured solvers capture the double peak structure revealed in the experiments, although the magnitude of the inner peak is underpredicted. Comparing different modeling schemes, t.he pdf model predicts the swirl velocities better than the eddy breakup model (single step chemist.ry). At. the station 109 mm downstream of the quarl, all schemes predict a higher decay of the tangential velocit.y as compared to experiments. This is a well known shortcoming of the k - t model that has been used. Temperat.ure profiles shown in Figures 11-13 show that predictions obtained from both solvers are very similar. However, the predictions indicate a longer and thinner flame than experimentally observed (Figure 14). Again, this is because of t.he axisymmetric assumption, since the gas is assumed to issue from an annular slit rather than 24 discret.e holes. Peak temperatures are predicted to occur further downstream in the combustor as compared to experiment.al dat.a. The pdf model shows the presence of a sharp spike, which is quite possibly an inherent limit.at.ion of t.he model resulting from ignoring finite chemistry effects that are important in this region. The eddy breakup model, on the other hand, does not show the presence of such a peak. Further downstream, the same behaviour is seen with the temperature being underpredicted by approximately 130 K in all models in the outer regions of the combustor, while near the centerline, the temperature is higher than indicated by experiment. Radial profiles of O2 and CO2 (dry volume per cent) are shown at the three different stations (Figures 15-17). The unst.ruct.ured solver (FLUENT fUNS) predictions are in good agreement with the predictions obtained from the st.ruct.ured solver, FLUENT. This indicates that the unstructured solver can be used with the same degree of confidence as t.he structured solver. Figures 18 and 19 show comparisons of O2 and CO2 (dry mole fraction) concent.rat.ions t.o experimental data at the station 27 mm downstream of the quarl. Profiles of O2 in t.he near burner region (rich flame) show predictions of mole fraction close to zero, a consequence of t.he equilibrium chemistry assumption inherent in the pdf model. Finite rate chemistry effects, that are important in t.he rich flame region are thought to be responsible for the errors in prediction of species concentrations. Errors in t.he predict.ion of t.urbulent mixing in the Magnussen model owing to variance in the mixing constant A for different flames can be responsible for some of the differences observed. Errors also occur in the PDF model because of the modeling of t.he t.ransport equation for the mixture fraction and its variance. Flow field modeling errors can arise due t.o t.reat.ment. of the 3-dimensional furnace geometry as a 2-dimensional, axisymmetric configuration. Conclusions and Future Efforts Validat.ions for a swirling natural gas flame have been carried out for the unstructured solver FLUENT fUNS against a st.ruct.ured solver (FLUENT). Comparisons to experimental data have also been shown. Two combustion models were used the eddy breakup model with both the single step and two-step reaction schemes, and the partial 5 |