Condition monitoring and state classification of gearboxes using stochastic resonance and hidden Markov models

This paper introduces a new fault detection and classification system based on the integration of stochastic resonance and the hidden Markov modelling (HMM) of vibration data tested by simulated and real life gearbox vibration data. Stochastic resonance uses noise to amplify weak impulses while HMM approach models the observation of a system as a probabilistic function of the system hidden states. The scheme developed in this paper combines stochastic resonance and hidden Markov modelling to produce a more robust diagnostic system. In addition, the proposed scheme employs other powerful feature extraction techniques such as angular resampling and auto regressive modelling. Features extracted are based on different performance indicators like root mean square (RMS) value and kurtosis. The extracted features are used to train HMMs and a Bayesian control scheme is developed for fault detection while the Viterbi algorithm is used to obtain the system latent states for classification purpose. It is shown that the developed scheme performs quite well with high detection and classification accuracy