基于样本自适应条件对抗网络的齿轮箱跨域故障诊断研究

Cross-domain Fault Diagnosis of Gearboxes Based on Sample Adaptive Conditional Adversarial Network

基于对抗训练的深度领域适应在旋转部件跨域故障诊断中应用效果良好。然而,现有研究主要致力于降低边缘分布差异而忽略对类别分布信息的挖掘,导致其在复杂场景下诊断准确性不足。针对该问题,提出一种样本自适应条件对抗网络,通过分解抽象特征和评估样本置信度挖掘类别分布特征,增强对抗训练的域适配能力,从而有效提高跨域诊断性能。通过齿轮箱故障诊断实验验证所提方法在实际应用中的有效性和优越性。

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.