Bearing fault diagnosis based on negative selection algorithm of feature extraction and neural network

As the fault diagnosis based on neural network needs typical features samples, the paper proposes a hybrid fault diagnosis method which integrates the RNSA (real-valued negative selection algorithm) and the radial basis function network. In this method, we choose typical fault samples (generated by Negative selection algorithm) as the inputs of the neural network, which solves the difficulty of obtaining typical samples, then extracting feature extraction from rolling bearing vibration signal with wavelet packet analysis is finished. At last, the fault samples (generated by RSNA) have been used to validate the new algorithm, and the accuracy is up to 97.2%, which verifies validity of the algorithm.