Fault Recognition Method Based on Independent Component Analysis and Hidden Markov Model

Abstract

Independent component analysis (ICA) is a powerful tool for analyzing non-Gaussian data. Through the use of ICA, invariable features embedded in multi-channel vibration measurements made in different operating modes can be extracted. The hidden Markov model (HMM) is a statistical model of the time series, and has a strong capability in pattern classification, especially for signals with abundant information quantity, non-stationary natures and poor repeatability and reproducibility. A new approach to fault recognition is proposed in this article, in which ICA is used for feature extraction, and the HMM as a classifier. Fault recognition in the speed-up and speed-down processes of rotating machinery has been successfully completed. The proposed approach is compared with another recognition approach, in which principal component analysis (PCA) is used for feature extraction, and HMM as a classifier, and is shown to be very effective.