Manufacturing State Recognition of Machine Center Based on Revised WPT and PCA

Abstract

The abnormal manufacturing state recognition of machine center is of great significance to reduce downtime and ensure quality of productions. A revised wavelet packet transform and principal component analysis method has been developed to extract features from vibration signals of machine center. The revised wavelet packet transform employs local discriminant bases method to select optimal wavelet packet nodes. The principal component analysis is conducted for features dimensionality reduction to obtain the final features. A BP neural network is used to classify manufacturing states based on final features of three different manufacturing processes. The comparison result indicates that the revised WPT and PCA method is an efficient feature extraction method for manufacturing stage recognition of machine tools.