A game theory enhanced domain adaptation network for mechanical fault diagnosis

Abstract

Transfer learning technology has been successfully applied to address the domain adaptation (DA) problem in machinery fault diagnosis. However, the partial DA problem is more suitable for industrial applications, where the target data only covers a subset of the source classes, which makes it difficult to know where to transfer the target data. To overcome this problem, a novel game theory enhanced DA network (GT-DAN) is proposed in this paper. Based on different metrics, including the maximum mean discrepancy, Jensen–Shannon divergence and Wasserstein distance, three attention matrices are constructed to describe the distribution discrepancies between the source domain and the target domain. The optimal coordination between these attention matrices is achieved by a combined weighting based on game theory to generate the optimal probability weights, which can act as a guide to filter out the irrelevant source examples in DA. Two experiments show that the proposed GT-DAN is superior to existing methods in partial DA diagnosis performance.