Deep Mixed Domain Generalization Network for Intelligent Fault Diagnosis Under Unseen Conditions

Abstract

Emerging intelligent fault diagnosis models based on domain adaptation can resolve domain shift problems produced by different working conditions. However, the prerequisite of obtaining target data in advance limits the application of these models to practical engineering scenarios. To address this challenge, a deep mixed domain generalization network (DMDGN) is proposed for intelligent fault diagnosis. In this novel model, data augmentation is applied to both class and domain spaces, adversarial learning is employed to introduce adversarial perturbations, and a domain-based discrepancy metric is used to balance intra-domain and inter-domain distances. The model can effectively learn more domain-invariant and discriminative features from multiple source domains to perform different generalization tasks for different working loads and machines. The feasibility of the DMDGN model is verified on two public datasets and one private dataset collected from practical production processes. Empirical results show that the DMDGN model outperforms several state-of-the-art models.