Compound fault diagnosis method of rotating machinery using multi-view multi-label feature selection based on label compression and local label correlation

The missing fault labels and the complexity of inter-fault correlations pose a great challenge for compound fault diagnosis of rotating machinery. Therefore, this paper proposes a compound fault diagnosis method using multi-view multi-label feature selection based on label compression and local label correlation (MVML-LCLLC). Firstly, the method develops an adaptive view weight assignment mechanism that dynamically assign weights according to the importance of each view in the fault information representation. Secondly, it achieves effective compression and recovery of labels through low-rank decomposition of sparse label matrix, while local label correlation is introduced to compensate for the lack of global information. Furthermore, to solve the optimization problem in the model, an alternating optimization algorithm is designed to generate sparse feature weight matrix for feature selection. Finally, the top-ranked features from the MVML-LCLLC method are selected and fed into a multi-label k-nearest neighbor (MLKNN) classifier to complete the diagnosis task. By comparing six multi-label classification evaluation metrics and fault classification confusion matrices for three rotating machinery cases, the results show that the proposed method possesses high accuracy and stability.