Aiming at realizing the effective fault diagnosis for aviation bearing, a method based on a coupled hidden semi-Markov model (CHMM) is proposed. Firstly, a monitoring network is designed for observing the radial and axial vibration data of bearings according to the aviation dynamic components of transmission structure. Secondly, a nonlinear feature extraction method is applied for obtaining a few key features, and it provides a prerequisite for improving the efficiency and accuracy of fault diagnosis. Finally, the left and right types of the homogenous hidden Markov chains, extending CHMM to multi-channel data fusion fault diagnosis, are utilized for building the reasonable state model of residence time distribution in practical problems. Moreover, the initial state model selection and parameter estimation algorithm of CHMM with two chains are researched on a probability reasoning algorithm. The validation and effectivity of the proposed CHMMis verified by the results of the fault diagnosis on rolling bearing.