Work

MAJATNet: A Lightweight Multi-Scale Attention Joint Adaptive Adversarial Transfer Network for Bearing Unsupervised Cross-Domain Fault Diagnosis

Year: 2025

Type: article

Abstract: Rolling bearings are essential for modern mechanical equipment and serve in various operational environments. This paper addresses the challenge of vibration data discrepancies in bearings across different operating conditions, which often results in inaccurate fault diagnosis. To tackle this related limitation, a novel lightweight multi-scale attention-based joint adaptive adversarial transfer network, termed MAJATNet, is developed. The proposed network integrates a feature extraction network innovation module with an improved loss function, namely IJA loss. The feature extraction module employs a one-dimensional multi-scale attention residual structure to derive characteristics from monitoring data of source and target domains. IJA loss evaluates the joint distribution discrepancy of high-dimensional features and labels between these domains. IJA loss integrates a joint maximum mean discrepancy (JMMD) loss with a domain adversarial learning loss, which directs the model's focus toward categorical features while minimizing domain-specific features. The performance and advantages of MAJATNet are demonstrated through cross-domain fault diagnosis experiments using bearing datasets. Experimental results show that the proposed method can significantly improve the accuracy of cross-domain fault diagnosis for bearings.

Source: Entropy

DOI: 10.3390/e27101011

Authors: Lin Song, Yanlin Zhao, Junjie He, Simin Wang, Boyang Zhong, Fei Wang

Institutions: Chengdu Technological University, Southwest Jiaotong University, Panzhihua University

Language: English

Topic: Machine Fault Diagnosis Techniques

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $3,137