Improving prediction of cabin noise with BPANN technique

Existing methods for predicting cabin noise appear to have two shortcomings: (1) Application frequency range is not wide enough; (2) parameters required are unnecessarily too many. Having had several years of research experience in predicting cabin noise, I formed gradually the idea of overcoming these shortcomings with BPANN (back propagation artificial neural network). Instead of several tens of parameters needed by existing methods, BPANN needs only twelve input variables, two hidden variables and one output variable. The data used in training the BPANN algorithm come from two sources: (1) measurements; (2) results predicted by ray-tracing method. Simulation results show preliminarily that BPANN algorithm can predict correctly the sound pressure levels in enclosed spaces and its precision for high frequency band is about the same as that for low frequency band.