Adaptive fault diagnosis method for rotating machinery with unknown faults under multiple working conditions. (April 2022)

• Record Type: Journal Article
• Authors: Ge, Yang; Zhang, Fusheng; Ren, Yong
• Is Part Of: Journal of manufacturing systems. Volume 63(2022)
• Page Start: 177
• Page End: 184
• Publication Date: 2022-04
• Subjects: Diagnosis of unknown faults -- Adaptive -- Multiple working conditions -- Multiscale; Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2022.03.009
• Languages: English
• ISSNs: 0278-6125

Abstract

Abstract: Fault diagnosis is an important part of the health management of many pieces of equipment. It is an effective means to reduce equipment failure rate and shutdown loss. In engineering practice, equipment often has one or more new fault types that have not been discovered before. The classical fault recognition method cannot solve the problem of unknown fault type recognition. An adaptive fault diagnosis network framework is proposed in this paper, which can solve the equipment fault diagnosis problems with new fault types under multiple working conditions. The network framework consists of a multi-scale feature extractor, an adaptive fault discriminator, and a new fault cluster. A loss function is established to identify new fault types and known fault types from the mixed fault data. The new fault cluster divides the new faults into different types at last. Two experiments show that the proposed method can effectively solve the problem of fault diagnosis with new types, and has a high recognition rate and universality. Highlights: An adaptive fault diagnosis network framework is proposed for the equipment with unseen fault types under multiple working conditions. A loss function is established to identify new fault types and known fault types from the mixed fault data. Experiments show that the proposed method has high recognition rate and generalization.