A reliable feature-assisted contrastive generalization net for intelligent fault diagnosis under unseen machines and working conditions. (1st April 2023)

• Record Type: Journal Article
• Authors: Shi, Zhen; Chen, Jinglong; Zhang, Xinwei; Zi, Yanyang; Li, Chen; Chen, Jin
• Is Part Of: Mechanical systems and signal processing. Volume 188(2023)
• Publication Date: 2023-04-01
• Subjects: Contrastive learning -- Confidence assessment -- Domain generalization -- Feature-assisted multi-task learning -- Intelligent fault diagnosis; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2022.110011
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: A novel domain generalization network is proposed in this paper, which can not only achieve intelligent fault diagnosis of unseen machines, but also achieve intelligent fault diagnosis under unknown working conditions, especially varying speeds. A multi-source contrastive generalization framework is introduced in this article to learn machine-independent knowledge. A feature-assisted multi-branch module is presented in the paper, which could extract fault-specific features through guiding the network to focus on fault characteristic frequency while ignoring operation conditions information. A confidence evaluation index is suggested to understand the decision-making behavior and assess the confidence level of diagnosis results. Abstract: Intelligent fault diagnosis has made significant progress in recent years. However, due to the following two difficulties, these solutions are still difficult to implement: 1) The majority of intelligent fault diagnosis methods assume that there is enough target data to train the model. The procedure will fail if the training data becomes scarce. 2) Current intelligent diagnosis methods could only provide a conclusion, but cannot explain why the conclusion is reached or how confident the conclusion is reached. Therefore, a reliable feature-assisted contrastive generalization net (RFACGN) is proposed. Firstly, a contrastive framework is presented for eliminating domain-specific knowledge by minimizing the difference between the same faults in multiple domains while maximizing the difference between distinct faults in different domains. Then, a feature-assisted multi-branch module is introduced to guide the model to learn fault-related features while ignoring the interference of operating condition information such as speed and load. Additionally, a confidence metric is utilized to understand the net and to assess the believability of the results. The proposed method is validated on four cases and compared to some state-of-the-art methods. The results demonstrate the effectiveness of RFACGN under unknown machines and working conditions, particularly under varying speeds.