Deep domain adaptation based on adversarial training has achieved promising performance in cross-domain fault diagnosis of rotating components.However,the current work mainly focuses on reducing the marginal distribution difference and ignores the mining of category distribution information,which leads to low accuracy in complex scenarios.Aiming at this problem,this paper proposes a sample adaptive conditional adversarial network.The abstract feature decomposition and sample confidence evaluation are used to find the class distribution features to improve the domain adaption ability of the adversarial training and the performance of cross-domain fault diagnosis.The gearbox fault diagnosis experiment verifies the effectiveness and superiority of the proposed method in practical application.
维普
万方数据