A Semantic-Consistent Conditional Generative Adversarial Model for Zero-Shot Learning in Bearing Fault Diagnosis

The scarcity of labeled data remains one of the most significant challenges in industrial fault diagnosis, where the industrial processes lack readily sampled in many fault types. This issue is often called the zero-shot learning problem in industrial fault diagnosis. To address this challenge, this paper proposes a novel fault diagnosis model for rotating machinery, leveraging Generative Adversarial Networks (GANs) and semantic consistent loss. Our approach aims to generate a rich set of samples for fault types that are difficult to collect, thereby enabling the training of an accurate classification model. The proposed model introduces a one-dimensional convolutional neural network to enhance the quality of samples and the accuracy of the model. A semantic consistent discriminator is proposed to ensure that the distribution of generated samples closely aligns with the distribution of actual samples. The performance of the model is validated using the CWRU bearing fault dataset, and the results demonstrate its efficacy in addressing the zero-shot fault diagnosis challenge.