Abstract: In view of the low accuracy of mechanical fault diagnosis caused by imbalance data sample distribution and mixed characteristic space, a three-layer hyper-sphere classifier abnormity detection model is proposed, which separates the characteristic space into positive field, negative field and mixed field. In order to decide the labels of samples in mixed field, a definitive method of discrimination hyper-plane based on Bayes is proposed, which uses the samples distribution information of distance and probability density. The samples in mixed field are classified according to Bayes criterion, so as to minimize the classification error probability. The method is applied to the analysis of diesel engine cylinder cover vibration signal, and the result shows that it can improve fault diagnosis accuracy.

Key words: Bayes, Diesel engine, Fault diagnosis, Hyper-sphere classification