Domain Contrastive Alignment Network for Pipeline Fault Diagnosis

The general domain adaptation (DA) approach focuses on making the source and target domains of the feature space as close as possible, while ignoring the inter and intra-class distances of the domain samples, leading to poor generalization performance of the model. To tackle this issue, a novel Domain Contrastive Alignment Network (DCAN) algorithm is proposed in this paper, which extracts fine-grained information by minimizing the intra-class distance and maximizing the inter-class distance. The proposed DCAN algorithm is applied to the fault diagnosis of oil and gas pipeline networks to verify its performance, and the results of cross-domain experiments show that the DCAN algorithm effectively improves the accuracy of fault diagnosis in comparison with state-of-the-art algorithms.