Application of improved RBF network training method in fault diagnosis

RBF Network;  Orthogonal least squares;  Input clustering;  Output-input clustering

At present, several existing RBF network training methods are proposed for complex samples with random noise training speed is too slow and the classification performance is unstable, according to the principle of minimum relative entropy, an improved RBF network training method-output-input clustering method is proposed. This method is used to train the rotating machinery fault samples and compare with other methods, and the results show that the training method takes a short time, has a simple network structure, and is less affected by noise. The created network is applied to fault diagnosis, and the examples show that the network diagnosis results trained by this method are accurate and have good application prospects in fault diagnosis.