SVM Optimization based on BFA and its Application in AE Rotor Crack Fault Diagnosis

According to the fact that parameter selection of support vector machine(SVM) for fault diagnosis is difficult, a new method based on bacterial foraging algorithm(BAF) for support vector machine parameter optimization was proposed , then the faster optimization of the parameters C and RBF kernel parameter γ was performed. The crack rotor as the experiment object, firstly, AE signal of rotors with different crack depth was collected, according to the wavelet packet better time-frequency analysis ability, and then time-domain AE signal 4 layers wavelet packet decomposition was implemented by wavelet db10, the nodes (4, 2) and (4, 6) was reconstructed to time signal, the feature vector was constructed by spectral entropy and peak frequency, finally, using the optimized support vector machines to pattern recognition. The experiment showed, bacterial foraging algorithm had best effect in comparison to other optimization algorithm to the rotor crack fault diagnosis and was suitable for support vector machine parameters optimization.