Principal Component Analysis and Hidden Markov Model

Aiming at accurately and rapidly recognizing bearing fault pattern and performance degradation, a bearing fault detection and diagnosis method based on principal component analysis and hidden Markov model is proposed. The mixed domain fault feature set of bearing vibration signal, which corresponds to different bearing conditions, is extracted with principal component analysis to reduce the dimension of the feature set, then hidden Markov models are trained with part of the reduced feature set. The performances of trained models are verified with the remaining parts in this feature set. The bearing fault patterns are recognized and bearing performance degradation is assessed by comparing the logarithmic likelihood probability value of the hidden Markov models. Experiments under different bearing conditions are carried out, vibration signals are collected, and the correct classification rate of the proposed method reaches 100%. Compared with the compensation distance evaluation based feature dimension reducing technique and hidden Markov model, the classification dispersion of the proposed method is increased by 123.74%. In the bearing performance degradation monitoring, the proposed method exhibits better effectiveness and accuracy for early bearing degradation warning.