Numerical Prediction of the Dynamic Behaviour of Turbulent Diffusion Flames

Environmental compatibility requires low emission burners for gas turbine power plants as well as for jet engines. In the past significant progress has been made developing low NOx and CO burners. Unfortunately these burners often have a more pronounced tendency than conventional burner designs to produce combustion driven oscillations The oscillations may be excited to such an extent that pronounced pulsation may possibly occur; this is associated with a risk of engine failure. The stability of a burner system can be investigated by means of a stability analysis under the assumption of acoustical behaviour. The problem with all these algorithms is the transfer function of the flame. A new method is presented here to predict the dynamic flame behaviour by means of a full Navier-Stokes-simulation of the complex combustion process. The first step is to get a steady-state solution of a flame configuration. After that a transient simulation follows with a sudden change in the mass flow rate at the flame inlet. The time-dependent answer of the flame to this disturbance is then transformed into the frequency space by a Laplace Transformation. This leads, in turn, to the frequency response representing the dynamic behaviour of the flame. In principle, this method can be adapted for both diffusion as well as premixed flame systems. However, due to the fact that diffusion flames are more controlled by the mixing process than by the chemical kinetic, the method has first been used for the prediction of the dynamic behaviour of turbulent diffusion flames. The combustion has been modelled by a mixed-is-burnt model. The influence of the turbulence has been taken into account by a modified k-ε-model and the turbulence influences the combustion rate by presumed probability density functions (pdf). The steady-state as well as the transient results have been compared with experimental data for two different diffusion flame configurations. Although the burner configuration is relatively complex, the steady state results collaborate very well with the experiments for velocity, temperature and species distribution. The most important result is that the heat release which drives the oscillations can be modelled sufficiently accurately. The effect of using different pdf-models has been discussed and the best model has been used for the transient calculations of the dynamic flame behaviour. The results for the frequency response of the flame are very encouraging. The principal behaviour of the flame — higher order time element with a delay time — can be predicted with sufficient precision. In addition, the qualitative results collaborate fairly well with the experiments.