Dual adversarial and contrastive network for single-source domain generalization in fault diagnosis

TL;DR

This work tackles fault diagnosis under single-source domain generalization, where unseen operating modes must be diagnosed using data from only one seen mode. It introduces the Dual Adversarial and Contrastive Network (DACN), which generates diverse pseudo-sample features via a feature transformer and extracts domain-invariant health-condition representations through a domain-invariant feature extractor guided by supervised contrastive learning and a gradient-reversal-based discriminator. The model is trained in two stages (pre-training on the seen mode and then joint training with pseudo-features) and validated on the Tennessee Eastman process and a continuous stirred-tank reactor, showing robust generalization to unseen modes with a compact parameter count. Ablation studies confirm that pre-training, contrastive learning, and adversarial components each contribute to improved cross-mode performance, highlighting the approach's practical value for reliable fault diagnosis in variable industrial environments.

Abstract

Domain generalization achieves fault diagnosis on unseen modes. In process industrial systems, fault samples are limited, and it is quite common that the available fault data are from a single mode. Extracting domain-invariant features from single-mode data for unseen mode fault diagnosis poses challenges. Existing methods utilize a generator module to simulate samples of unseen modes. However, multi-mode samples contain complex spatiotemporal information, which brings significant difficulties to accurate sample generation. To solve this problem, this paper proposed a dual adversarial and contrastive network (DACN) for single-source domain generalization in fault diagnosis. The main idea of DACN is to generate diverse sample features and extract domain-invariant feature representations. An adversarial pseudo-sample feature generation strategy is developed to create fake unseen mode sample features with sufficient semantic information and diversity, leveraging adversarial learning between the feature transformer and domain-invariant feature extractor. An enhanced domain-invariant feature extraction strategy is designed to capture common feature representations across multi-modes, utilizing contrastive learning and adversarial learning between the domain-invariant feature extractor and the discriminator. Experiments on the Tennessee Eastman process and continuous stirred-tank reactor demonstrate that DACN achieves high classification accuracy on unseen modes while maintaining a small model size.

Figures (14)

• Figure 1: Normal and fault 2 data of material A flow rate after standardization of the Tennessee Eastman process in six modes.
• Figure 2: Overall structure of DACN.
• Figure 3: Forward and backward propagation of DACN.
• Figure 4: Two sets of adversarial learning in DACN.
• Figure 5: P&ID of the revised process model RN237021.