In this study, new methodologies are introduced to analyze combustion instability in a lab-scale swirled combustor. First, with the help of radial basis function neural network (RBFNN), the flame describing function (FDF) is effectively modeled from a limited number of experimental data. This neural-network-based FDF method is able to generate more refined FDF data in an extended range. In addition, instead of a perforated plate with round holes, a slotted plate is utilized as a stabilization device. In this approach, the acoustic impedance of a slotted plate is modeled by the Dowling approach, and the dimensions of a slotted plate are optimized by simulated annealing (SA) algorithm to get the highest average absorption coefficient in a given frequency range. The present RBFNN-based FDF approach yields the reasonably good agreements with the measurements in terms of the limit-cycle velocity perturbation ratio and resonant frequency. It is also found that a slotted plate optimized by SA algorithm is quite effective to attenuate combustion instability. Numerical results obtained in this study confirm that these new methodologies are quite reliable and widely applicable for the analysis of combustion instability encountered in practical combustion systems.
Physics-aware neural network flame closure for combustion instability modeling in a single-injector engine
