A novel fault diagnosis framework for rotating machinery with hierarchical multiscale symbolic diversity entropy and robust twin hyperdisk-based tensor machine

Tensor learning has the advantage of fully leveraging the rich information in tensor features, and has been successfully applied to intelligent fault diagnosis of rotating machinery. Unfortunately, previous tensor learning-based fault diagnosis frameworks have limitations in both tensor feature extraction and classification. Aiming at these limitations, a new fault diagnosis framework is proposed with hierarchical multiscale symbolic diversity entropy (HMSDE) and robust twin hyperdisk-based tensor machine (RTHDTM). Firstly, HMSDE is presented for tensor feature extraction. HMSDE evaluates the signal complexity at different hierarchical layers and different scales, and thus more comprehensive information can be extracted. Then, the extracted HMSDEs are fed to RTHDTM classifier to recognize health states automatically. The proposed RTHDTM has the following novelties: First, the twin hyperdisk model takes into account the distance to other class samples, and thus gives a more reasonable approximation of the actual class region. Second, based on priori knowledge of tensor samples, RTHDTM implements a confidence weight assignment strategy to enhance the robustness against outliers. Experiment results demonstrate that the proposed framework has excellent feature extraction and fault recognition performance. Compared to the state-of-the-art tensor learning-based diagnosis frameworks, the proposed framework has great advantage in diagnosis accuracy and robustness.