Ensemble deep learning-based fault diagnosis of rotor bearing systems

For rotating machinery, early and accurate diagnosis of rotor and bearing component fault is of great significance. The classic fault diagnosis model includes two key modules, feature extraction and fault classification. In order to enhance the practicability, the deep learning models realize the end-to-end fault diagnosis by integrating this two modules, thus avoids the problems caused by the inadequate adaptability of manual designed features. However, considering the wide application scenario of fault diagnosis technology, the application scope of single deep model may have corresponding limitations. Accordingly, in this paper, an ensemble deep learning diagnosis method based on multi-objective optimization is proposed. The multi-objective optimization algorithm is used as the ensemble strategy in this method, the Convolution Residual Network (CRN), Deep Belief Network (DBN) and Deep Auto-Encoder (DAE) are weighted and integrated to realize the effective diagnosis of rotor and bearing faults for rotating machinery. The experimental results demonstrate the better adaptability of the proposed method compared to other single and ensemble deep models. (C) 2018 Published by Elsevier B.V.