An adversarial gradual domain adaptation approach for fault diagnosis via intermediate domain generation

Intelligent fault diagnosis technology, as a fundamental aspect of modern Nondestructive Testing and Evaluation (NDT&E), is critical for ensuring the safe operation of electrical equipment. However, traditional domain adaptation methods are prone to negative transfer when significant distributional disparities exist between source and target domains. To address this, this paper proposes an Adversarial Gradual Domain Adaptation Framework based on Intermediate Domain Generation (GAGA). The framework transforms time-series signals into two-dimensional images using the symmetrized dot pattern (SDP) method to enhance feature representation. Intermediate domain samples are generated along the Wasserstein Distance (WD) distribution based on optimal transport (OT) theory, and a multi-objective optimisation strategy – integrating adversarial training, consistency constraints and entropy regularisation – enables gradual knowledge transfer. The effectiveness of the proposed method is validated through comprehensive experiments on the Paderborn University bearing dataset and the Permanent Magnet Synchronous Motor (PMSM) dataset. Results demonstrate that, across nine severe domain shift scenarios, the average classification accuracy surpasses those of state-of-the-art algorithms and significantly alleviates negative transfer. This study presents a robust and highly generalisable solution for fault diagnosis in electrical equipment, offering substantial value for engineering applications and theoretical advancements.