Abstract:Intelligent fault diagnosis encounters the challenges of varying working conditions and sample class imbalance individually, but very few approaches address both challenges simultaneously. This article proposes an improvement network model named ICDAN-F, which can deal with fault diagnosis scenarios with class imbalance and working condition variations in an integrated way. First, Focal Loss, which was originally designed for target detection, is introduced to alleviate the sample class imbalance problem of fault diagnosis and emphasize the key features. Second, the domain discriminator is improved by the default ReLU activation function being replaced with Tanh so that useful negative value information can help extract transferable fault features. Extensive transfer experiments dealing with varying working conditions are conducted on two bearing fault datasets with the effect of class imbalance. The results show that the fault diagnosis performance of ICDAN-F outperforms several other widely used domain adaptation methods, achieving 99.76% and 96.76% fault diagnosis accuracies in Case 1 and Case 2, respectively, which predicts that ICDAN-F can handle both challenges in a cohesive manner.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are two challenges encountered in fault diagnosis of rotating mechanical equipment under different working conditions: data distribution differences (domain shift) caused by changes in working conditions and sample class imbalance. Specifically: 1. **Data distribution differences caused by changes in working conditions**: Machines are often in different working conditions (such as speed, temperature, load, etc.) during actual operation. This will lead to the domain shift phenomenon in fault diagnosis methods based on deep learning, that is, the distribution between training data and data to be diagnosed is inconsistent, which significantly reduces the diagnostic performance. 2. **Sample class imbalance**: Most similar studies assume that the occurrence probabilities of different data samples are balanced, but in actual industrial operations, there are usually more data in the normal state. Class imbalance will cause the model to be over - trained for classes with a large amount of data, while classes with a small amount of data are squeezed by the decision boundary, thus affecting the diagnostic effect. To solve these two problems, the paper proposes an improved conditional domain - adversarial network model (ICDAN - F), which can simultaneously handle class imbalance and changes in working conditions in fault diagnosis scenarios. The main innovations include: - Introducing Focal Loss as a constraint in the training stage, which effectively solves the problems of class imbalance and inconsistent feature attention, and has low computational cost and is easy to be applied to the network to optimize the fault diagnosis model. - Replacing the ReLU activation function of the domain discriminator in the conditional domain - adversarial network with Tanh, so that negative - value information can participate in the domain adaptation process and better extract metastable fault features that are not affected by changes in working conditions. The effectiveness and feasibility of ICDAN - F have been verified through a large number of bearing fault experiments. The results show that ICDAN - F can achieve high - precision fault diagnosis in the case of intertwined class imbalance and changes in working conditions.

Improved Conditional Domain Adversarial Networks for Intelligent Transfer Fault Diagnosis

Double-level Adversarial Domain Adaptation Network for Intelligent Fault Diagnosis.

Intelligent Fault Diagnosis With Deep Adversarial Domain Adaptation

Intelligent Fault Diagnosis under Varying Working Conditions Based on Domain Adaptive Convolutional Neural Networks

Bearing fault diagnosis based on partial domain adaptation adversarial network

Domain Adaptive Transfer Learning for Fault Diagnosis

Domain Adaptation Network with Double Adversarial Mechanism for Intelligent Fault Diagnosis

Domain Adversarial Transfer Learning for Bearing Fault Diagnosis Based on Data Fusion and Attention

A Balanced Adversarial Domain Adaptation Method for Partial Transfer Intelligent Fault Diagnosis

Collaborative and Conditional Deep Adversarial Network for Intelligent Bearing Fault Diagnosis

DFSA-DAN: Dynamic Fusion of Statistical Metric and Adversarial Learning for Domain Adaptation Network Based Intelligent Fault Diagnosis

Duplex Adversarial Domain Discriminative Network for Cross-Domain Partial Transfer Fault Diagnosis

Fault Diagnosis in Industrial Control Networks Using Transferability-measured Adversarial Adaptation Network

A Novel Attention-Based Domain Adaptation Model for Intelligent Bearing Fault Diagnosis under Variable Working Conditions

Cross-domain Transfer Fault Diagnosis by Class-Imbalanced Deep Subdomain Adaptive Network

Deep Adaptive Adversarial Network-Based Method for Mechanical Fault Diagnosis under Different Working Conditions

Deep Adversarial Subdomain Adaptation Network for Intelligent Fault Diagnosis

Domain Adaptation for Intelligent Fault Diagnosis under Different Working Conditions

A Mixed Adversarial Adaptation Network for Intelligent Fault Diagnosis

Intelligent Fault Diagnosis for Machinery Based on Enhanced Transfer Convolutional Neural Network

Self-supervised Adversarial Network for Intelligent Fault Diagnosis