부산대학교 도서관

The machine learning-based intelligent fault diagnosis method has the merits of fast response speed and automation, but requires many fault samples which are difficult to obtain. For rolling bearings in engineering, the normal samples collected are sufficient, while the fault samples are scarce. Because the operating time of the equipment in a normal state is much longer than the fault time. This paper proposes a two-stage rolling bearing fault diagnosis method that combines the advantages of machine learning and signal processing. In the first stage, the support vector data description optimized by a multi-objective grasshopper optimization algorithm is used to construct a fault detection model to quickly detect anomaly samples. In the second stage, multi-symplectic geometry mode decomposition is used to analyze anomaly state signals to determine the fault type. The analysis of the bearing dataset shows that the proposed fault diagnosis method can accurately detect early faults and identify fault types, which is robust to the number of normal samples participating in training. Compared with existing detection methods, our method can identify the fault point earlier. The above results demonstrate that the proposed method is expected to practical rolling bearing fault diagnosis.