Dual adversarial network for cross-domain open set fault diagnosis

Recently, cross-domain fault diagnosis methods have been successfully developed and applied. Among them, the ones exhibiting the best performance rely on the common assumption that the training and testing data share an identical label space, implying that fault modes are the same in different engineering scenarios. However, fault modes in the testing phase are unpredictable and new fault modes usually occur, posing challenges for existing cross-domain methods regarding their effectiveness. To address such challenges, a novel open set domain adaptation network based on dual adversarial learning is proposed in this study. An auxiliary domain discriminator assigns similarity weights for individual target samples to distinguish between known and unknown fault modes, and weighted adversarial learning is employed to selectively adapt domain distributions. In separated adversarial learning, the feature generator and the extended classifier are set against each other to construct more accurate hyperplanes between known and unknown fault modes. Comprehensive experimental results for three test rigs demonstrate that the proposed method achieves a promising performance and outperforms existing state-of-the-art open set domain adaptation methods.

Automated summary

In this study, a novel open set domain adaptation network based on dual adversarial learning is proposed to address the challenges in cross-domain fault diagnosis. The method utilizes an auxiliary domain discriminator to assign similarity weights for individual target samples and employs weighted adversarial learning to selectively adapt domain distributions. Experimental results demonstrate the promising performance of the proposed method, outperforming existing state-of-the-art open set domain adaptation methods.