Abstract: Fault diagnosis technology plays a crucial role in ensuring equipment reliability and safety while reducing maintenance time and costs. With the advent of Industry 4.0, equipment systems are trending toward intelligence, autonomy, and unmanned operation, making precise fault diagnosis increasingly essential. Consequently, deep learning-based fault diagnosis methods have emerged. Graph neural networks represent an emerging deep learning approach. Through their unique message-passing and aggregation mechanisms, they enable effective processing of topological data, providing a natural paradigm for modeling complex internal relationships within equipment. In recent years, they have garnered increasing attention, gradually becoming a research hotspot in the field of fault diagnosis of industrial equipment. This paper systematically reviews the research progress of graph neural networks in industrial fault diagnosis over the past five years. It first introduces the fundamental structures of graph neural networks and existing mainstream frameworks, including graph convolutional networks, graph attention networks, graph autoencoders, and graph recurrent neural networks. The paper then analyzes the principles and advantages of using graph neural networks for fault diagnosis, further exploring their application in fault diagnosis technologies for two major categories of objects: industrial components and industrial systems. Finally, it summarizes the practical challenges of applying graph neural networks to fault diagnosis and outlines their future development prospects. This paper aims to provide a systematic review for researchers, promote the deep integration of graph neural networks and fault diagnosis, and offer theoretical support for constructing intelligent equipment fault diagnosis and health management systems.