Fault diagnosis of rotating machinery based on graph weighted reinforcement networks under small samples and strong noise. (1st March 2023)

• Record Type: Journal Article
• Authors: Yu, Xiaoxia; Tang, Baoping; Deng, Lei
• Is Part Of: Mechanical systems and signal processing. Volume 186(2023)
• Publication Date: 2023-03-01
• Subjects: Rotating machinery -- Fault diagnosis -- Small samples -- Strong noise -- Graph weighted reinforcement networks; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2022.109848
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: The graph weighted reinforcement network (GWRNet) is proposed to accurately diagnose the fault of rotating machines under small samples and strong noise. Two highlights of this study can be summarized as follows. The time and frequency domain characteristics of the vibration signal are extracted, and the adjacency matrix is constructed based on the Euclidean distance of these characteristics to achieve node pre-classification. The graph-weighting enhanced mechanism is used to aggregate the node features in the graph, suppress the background noise interference during feature extraction, and realize rotating machinery fault diagnosis under strong noise conditions. Abstract: Available fault vibration signals of large rotating machines are usually limited and consist of strong noise. Existing deep learning methods do not sufficiently extract the correlation relationship between samples, and the process of extracting features is easily disturbed by noise. Therefore, the accuracy of rotating machinery fault diagnosis needs further improvement. A graph-weighted reinforcement network (GWRNet) is proposed to accurately diagnose the faults of rotating machines under small samples and strong noise. First, an adjacency matrix was constructed by measuring the Euclidean distance of the time- and frequency-domain characteristics of small samples to achieve the pre-classification of nodes. Second, the node feature aggregation strategy was designed by dynamically enhancing the largest weight of the multiheaded attention matrix to suppress strong noise interference. Finally, the effectiveness of the proposed method was verified using datasets from the drivetrain diagnostics simulator (DDS) test rig and wind turbine gearboxes.