Temporal Graph Convolutional Network-Based Fault Detection Method for the Two-Wheeled Self-Balancing Robot

The fault detection method based on graph neural networks has been widely applied. However, few of these methods consider the dynamic characteristics and prior knowledge of control systems, resulting in limited usage for fault detection in control systems. To address this, this article proposes a fault detection method based on temporal graph convolutional neural network (T-GCN) and multi-layer perceptron (MLP), which is applicable to general linear dynamic control systems. The proposed method first constructs a relational graph based on prior knowledge of the control system and utilizes T-GCN to extract spatio-temporal relationships between various state variables and control input signals within the system. Then, a MLP integrates the extracted information to detect whether the control system has experienced faults. Finally, a two-wheeled self-balancing vehicle is utilized to test the effectiveness of the proposed method. The results demonstrate that the proposed approach could achieve high-precision detection of sensor faults within the system.