Abstract
Fault diagnosis of rotating machinery plays a pivotal role in ensuring operational reliability, efficiency, and productivity in modern manufacturing systems. However, limited fault samples and class imbalance issues impede the effective development and deployment of fault diagnosis models. Generative adversarial networks (GANs) and their variants have emerged as promising solutions to these challenges, and studies in this research field are expanding. However, the latest GAN variants specifically applied to vibration data generation for fault diagnosis have not been systematically reviewed. Moreover, the existing literature lacks a comprehensive evaluation of the quality metrics used to assess GAN-generated vibration data and inadequately addresses the practical challenges faced in industrial settings. Therefore, this paper provides a review of the latest studies on the applications of GANs and their variants for generating vibration data in the fault diagnosis of rotating machines. The review begins with an overview of the basic structure and principles of standard GANs and recent advancements, followed by a detailed analysis of their applications in generating synthetic vibration data for fault diagnosis. It then reviews the metrics employed to evaluate the quality and diversity of GAN generated vibration data. Finally, this review provides practical considerations, ongoing challenges, recent advancements, and future directions. The insights shared in this review can guide researchers and practitioners in this rapidly evolving field.
