A fault diagnosis method for rolling bearing sample class imbalance based on the improved ACGAN
WANG Rongzhen
XU Zheng
ZHUO Shuai
YU Xiao
GAO Weishan
Abstract：[Objective]Data-driven fault diagnosis methods frequently encounter the problem of sample imbalance in rolling bearing fault diagnosis.To address this issue,a sample class imbalance augmentation method based on an improved auxiliary classifier generative adversarial network(ACGAN)was proposed,namely the global attention mechanism independent classifier generative adversarial network(GAMICGAN).[Methods]Firstly,global attention mechanisms and octave convolutions were introduced,enabling the generator to fully attend to spatial,channel,and dimensional information as well as high-and low-fre-quency component information in time-frequency images,thereby improving the generation quality of specified class samples.Secondly,to mitigate the decline in diagnostic model accuracy caused by distribution differences in sensor data under varying operating conditions,a domain adaptation fault diagnosis method for bearings based on joint distribution domain adversarial neu-ral networks was proposed.A backbone network structure based on global attention residual networks was designed.Finally,joint maximum mean discrepancy metrics were utilized to quantify marginal and conditional distribution differences between source and target domains,achieving bearing fault diagnosis under varying operating conditions.[Results]The results indicate that the proposed method maintains high diagnostic accuracy in variable operating condition applications with imbalanced samples and demonstrates robust adaptability to complex scenarios.
Keywords：Fault diagnosisACGANSample class imbalanceDomain adaptationVariable working condition
Publication Date：2026-07-15
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：13( 163-174,184 )