In this study, we developed a fault classification model that combines a coupled hidden Markov model based on multi-channel information fusion with a minimum intra-class distance algorithm. This model relies on statistical features in the current time domain, which are the easiest features to extract for clustering. First, an algorithm is used to select and sequence the statistical features with the minimum intra-class distance in order to form feature vectors, which in turn enhance inter-class discrimination and feature reduction. Following reduction, the coupled hidden Markov model is used to perform classification. The coupled hidden Markov model was shown to reflect the coupling relationships between and among channels. We evaluated the efficacy of the proposed scheme by applying it to the diagnosis of faults in a gyro motor in three groups of experiments. Our results were compared with those obtained using a single-chain hidden Markov model and other intelligent fault diagnosis methods. The proposed scheme outperformed the other methods in terms of correct diagnosis rate, fluctuations in correct diagnosis rate, and excellent robustness against the effects of interference.
