Abstract:Domain generalization achieves fault diagnosis on unseen modes. In process industrial systems, fault samples are limited, and only single-mode fault data can be obtained. Extracting domain-invariant fault 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. Therefore, double gradient reversal network (DGRN) is proposed. First, the model is pre-trained to acquire fault knowledge from the single seen mode. Then, pseudo-fault feature generation strategy is designed by Adaptive instance normalization, to simulate fault features of unseen mode. The dual adversarial training strategy is created to enhance the diversity of pseudo-fault features, which models unseen modes with significant distribution differences. Subsequently, domain-invariant feature extraction strategy is constructed by contrastive learning and adversarial learning. This strategy extracts common features of faults and helps multi-mode fault diagnosis. Finally, the experiments were conducted on Tennessee Eastman process and continuous stirred-tank reactor. The experiments demonstrate that DGRN achieves high classification accuracy on unseen modes while maintaining a small model size.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address the single-source domain generalization problem in multi-modal fault diagnosis. Specifically: 1. **Background and Challenges**: - In industrial systems, fault data is limited and usually only available in a single mode. - Extracting domain-invariant fault features from single-mode data for diagnosing faults in unseen modes is challenging. - Existing methods use generator modules to simulate unseen mode samples, but due to the complex spatiotemporal information in multi-modal samples, accurately generating samples becomes difficult. 2. **Proposed Method**: - The paper proposes a Dual Gradient Reversal Network (DGRN) to achieve the goal through the following steps: 1. Pre-train the model to acquire fault knowledge from known single-mode data. 2. Use an Adaptive Instance Normalization (AdaIN) strategy to generate pseudo fault features to simulate fault features of unseen modes. 3. Design a dual adversarial training strategy to enhance the diversity of pseudo fault features. 4. Combine contrastive learning and adversarial learning strategies to extract domain-invariant fault features. 3. **Main Contributions**: - Proposed DGRN to solve the single-domain generalization problem in multi-modal fault diagnosis. - Designed a pseudo fault feature generation strategy based on AdaIN and enhanced the diversity of pseudo fault features through a dual adversarial training strategy. - Established a domain-invariant feature extraction strategy combining contrastive learning and adversarial learning, achieving the goal of extracting common fault features from single seen modes and pseudo fault features. - Validated the effectiveness of DGRN through extensive experiments on the Tennessee Eastman process and continuous stirred tank reactor, showing that DGRN not only achieves strong generalization ability but also maintains a small model size. Through these methods, the paper aims to solve the problem of generalizing single-mode fault data to multiple unseen modes, thereby improving the accuracy of fault diagnosis in industrial systems.

Double Gradient Reversal Network for Single-Source Domain Generalization in Multi-mode Fault Diagnosis

Chemical Fault Diagnosis Network Based on Single Domain Generalization

Adversarial Mutual Information-Guided Single Domain Generalization Network for Intelligent Fault Diagnosis

Gradient Aligned Domain Generalization with a Mutual Teaching Teacher-student Network for Intelligent Fault Diagnosis

Contrast-Assisted Domain-Specificity-Removal Network for Semi-Supervised Generalization Fault Diagnosis

Domain fuzzy generalization networks for semi-supervised intelligent fault diagnosis under unseen working conditions

Multi-scale Style Generative and Adversarial Contrastive Networks for Single Domain Generalization Fault Diagnosis

Adaptive Open Set Domain Generalization Network: Learning to Diagnose Unknown Faults under Unknown Working Conditions

A domain generalization network for imbalanced machinery fault diagnosis

Domain Transferability-Based Deep Domain Generalization Method Towards Actual Fault Diagnosis Scenarios.

Gradient Consistency Strategy Cooperative Meta-Feature Learning for Mixed Domain Generalized Machine Fault Diagnosis

A Hybrid Generalization Network for Intelligent Fault Diagnosis of Rotating Machinery under Unseen Working Conditions.

Imbalanced domain generalization via Semantic-Discriminative augmentation for intelligent fault diagnosis

A New Adversarial Domain Generalization Network Based on Class Boundary Feature Detection for Bearing Fault Diagnosis

Domain Augmentation Generalization Network for Real-Time Fault Diagnosis under Unseen Working Conditions

Causality-inspired Multi-source Domain Generalization Method for Intelligent Fault Diagnosis under Unknown Operating Conditions

Deep Transfer Fault Diagnosis Using Digital Twin and Generative Adversarial Network

A Domain Generalization Network Combing Invariance and Specificity Towards Real-Time Intelligent Fault Diagnosis

A Domain Generalization Network Exploiting Causal Representations and Non-Causal Representations for Three-Phase Converter Fault Diagnosis

A novel domain generalization network with multidomain specific auxiliary classifiers for machinery fault diagnosis under unseen working conditions

Dynamic Balanced Dual Prototypical Domain Generalization for Cross-Machine Fault Diagnosis