Enhancing robustness of cross-machine fault diagnosis via an improved domain adversarial neural network and self-adversarial training

With the diversification and complexity of industrial equipment, cross-machine fault diagnosis (FD) faces increasingly severe challenges. Especially when the data distribution differences between devices are significant, traditional deep learning methods often fail to balance high diagnostic accuracy and robustness against adversarial attacks, which limits their practical applications. To this end, this paper proposes a robust fault diagnosis framework based on an improved domain adversarial neural network (DANN) and multi-module fusion. This framework innovatively combines adversarial training, meta-self-training, and an enhanced DANN architecture to achieve high-precision diagnosis and strong robustness in cross-device scenarios. It not only effectively solves the problem of domain transfer between devices, but also significantly improves the adaptability and robustness of the model under different operating conditions by using adversarial training devices. Experimental verification results based on multiple data sets show that the proposed method has significant advantages in both diagnostic performance and robustness against adversarial attacks, as well as good application prospects in practical engineering.