To improve the performance and interpretability of fault detection in complex mechatronic systems,a graph convolutional network based on causal path(GCN-CP)is proposed in this paper.Firstly,hybrid causal discovery,combining empirical constraints with data-driven causal discovery,is employed to construct causal graph,which describes the causal relationships among monitoring variables.Since faults propagate along the cause-and-effect relationships,the causal paths are extracted as the receptive field for the proposed GCN-CP.The output of the proposed model is the system fault detection result based on information extracted from causal paths.A real case study concerning fault detection in high-speed train braking systems is carried out,to verify the effectiveness of the proposed method.Comparative experiments are carried out against several benchmarks,and the proposed method exhibits over 5％improvement with respect to two evaluation metrics.Additionally,the proposed method also demonstrates desirable interpretability and generalizability,indicating its potential for application in fault detection of complex mechatronic systems with high-dimensional monitoring variables.
Fault DetectionGraph Convolutional NetworkMechatronic SystemsCausal DiscoveryHigh-speed Train Braking System
NETL
NSTL
万方数据
