A concise review of transfer learning and generative learning for autonomous and robotic systems fault detection and diagnosis

In autonomous and robotic systems, the importance of Fault Detection and Diagnosis (FDD)
technologies has increasingly grown as these systems find widespread application across various sectors. There are a wide range of data-driven machine-learning approaches for FDD. These methods often require large amounts of normal and fault data for model training. In scenarios with limited data availability, transfer learning boosts learning efficiency by transferring prior knowledge. In contrast, generative learning, such as Generative Adversarial Networks (GANs), enhances the accuracy and generalization capabilities of fault diagnosis models by generating simulated fault data. While several reviews focus on component-specific fault analysis via transfer learning and generative learning, such as bearings and gearboxes, research on system-level fault diagnosis is comparatively scarce. In this context, the aim of this paper is to review the application of transfer learning and generative learning in FDD technologies for autonomous and robotic systems. The article expands the research scope of FDD from individual components to entire systems. It explores the potential value and challenges of two learning strategies in advancing FDD technology. Finally, it presents new perspectives and directions for future research in this field.