Graph Neural Networks-Empowered Intelligent Diagnostics for Electro-Hydraulic Systems in Smart Manufacturing

The Electro-Hydrostatic Actuator (EHA) is widely used in aerospace applications due to its high load capacity and control efficiency. However, its performance degrades over time, leading to hidden, diverse, and complex failures. Effective health monitoring and fault diagnosis are therefore crucial but challenged by system nonlinearity, limited fault samples, and fluctuating operating conditions. This chapter focuses on intelligent fault diagnosis for EHA hydraulic systems. We first analyze typical failure modes, such as hydraulic cylinder leakage and pipeline blockage, through both fault test rigs and simulation models. To address diagnostic challenges under varying conditions, we propose an Adaptive Graph Convolutional Manifold Embedded Neural Network (AGCMENet), integrating graph convolution, Locality Sensitive Discriminant Analysis (LSDA), and Particle Swarm Optimization (PSO). To overcome the scarcity of fault data, we further develop a Model-Assisted Multi-source Fusion Hypergraph Convolutional Neural Network (MAMF-HGCN), leveraging simulation models and hypergraph structures. Additionally, a Prognostics and Health Management (PHM) platform is established to support data monitoring, signal processing, model training, and practical deployment. The proposed methods and platform provide a comprehensive and effective solution for real-world EHA fault diagnosis, offering valuable insights for the development of intelligent maintenance systems.