Nonparametric Wavelet-Based Multivariate Control Chart for Rotating Machinery Condition Monitoring

Abstract

Signal processing of vibration signals from rotating machinery has been an active research area for recent years. Especially, discrete wavelet transform (DWT) is considered as a powerful tool for feature extraction in detecting fault on rotating machinery. However, the number of retained DWT features can be still too large to be used for standard multivariate statistical process control (SPC) techniques although DWT significantly reduces the dimensionality of the data. Even though many feature-based SPC methods have been introduced to tackle this deficiency, most of methods require a parametric distributional assumption that restricts their feasibility to specific problem of process control and thus limits their applications. This study introduced new feature extraction technique to alleviate the high dimensionality problem of implementing multivariate SPC when the quality characteristic is a vibration signal from bearing system. A set of multiscale wavelet scalogram features was generated to reduce the dimensionality of data, and is combined with the bootstrapping technique as nonparametric density estimation to set up an upper control limit of control chart. Our example and numerical simulation of a bearing system demonstrated that the proposed method has satisfactory fault-discriminating ability without any distributional assumption.