Research on Self-diagnosis Method of Nickel top-Blowing Furnace Sensor Based on Causal Graph Neural Network

As the scale of nickel top-blowing furnace melting systems expands and the complexity of the equipment increases, the maintenance workload for sensors has gradually increased. Manual maintenance approaches consume significant resources and are not suitable for the development of smart sensors. To address this issue, an unsupervised sensor self-diagnosis method is proposed in this paper. Firstly, the Pearson correlation coefficient is introduced to construct a graph network of sensors of the same type, and node embedding is used to form super nodes of sensors. Secondly, transfer entropy is utilized to assess the causal relationships between these super nodes, based on which a super node graph network is constructed. Finally, a Multi-Head Self-Attention Gated Graph Convolutional Network (MCB-GCN) is introduced. This network captures spatial and temporal dependencies of nodes and, with the help of an enhanced fusion module, achieves time-series prediction for sensors. Subsequently, sensor self-diagnosis is achieved through a standardized residual consistency check algorithm. Simulation experiments on the measurement system of a nickel top-blowing furnace demonstrate the effectiveness of the proposed method, in addressing the problem of autonomous sensor diagnostics.