Transforming the Open Set Into a Pseudo-Closed Set: A Regularized GAN for Domain Adaptation in Open-Set Fault Diagnosis
Yang Liu, Aidong Deng, Minqiang Deng, Yaowei Shi, Jing Li
Abstract
The field of fault diagnosis has benefited from the development of closed-set domain adaptation (CSDA), a method that can narrow the domain gap caused by variable operating conditions. However, the application of CSDA is limited in practical industrial scenarios where the source label space is a subset of the target label space, i.e., the open set. To address this challenge, we propose a method that transforms the open set into a pseudo-closed set, making CSDA suitable for it. Our approach involves introducing a regularization module to the traditional Generative Adversarial Network (GAN) to generate open data that is similar to the distribution of the entire source domain but significantly different from the distribution of each category. The open data can be used as a new source unknown category, which converts the open set into a pseudo-closed set with an equal number of categories in source and target domains. Furthermore, we introduce a sample-level weighting mechanism into the CSDA algorithm to suppress the negative transfer induced by different unknown categories in the source and target domains. Experimental results on three datasets demonstrate the effectiveness and superiority of our proposed method in open set fault diagnosis.