A Novel Process Monitoring and Root Cause Diagnosis Strategy Based on Knowledge-Data-Integrated Causal Digraph for Complex Industrial Processes

With the integrated and scaled development of modern industrial processes, multiple control units are strongly coupled, forming a complex interconnected network. This leads to the propagation and evolution of faults within the network, which will affect the quality of products and the safety of industrial processes. This article proposes a novel process monitoring and root cause diagnosis strategy for complex industrial processes based on a knowledge-data-integrated causal digraph. Compared with traditional single data-driven methods, this strategy combines process data and knowledge to improve the ability of fault detection and diagnosis. First, an attention-based time convolutional network is performed on process variables to construct a causal digraph. The causal digraph is trimmed and refined using the process knowledge to solve the problem of redundant causality and enhance interpretability. Second, the complex industrial process is decomposed into multiple sub-blocks, and the causal relationship between sub-blocks is obtained. On this basis, a process monitoring model for collaborative analysis of temporal and spatial information is established, where spatial information among sub-blocks is obtained through the interaction of information between them, and the temporal information within sub-blocks is captured by kernel canonical variate analysis (KCVA). Subsequently, a fault diagnosis method based on global and local causal digraphs is designed. Process data and causal digraphs are used to select fault variables and analyze causality relationships respectively, which can infer the fault root cause and propagation path. Finally, the experimental results on the real dataset of the float glass production process demonstrate that our strategy not only achieves significant improvements over other methods but also has a favorable application in industrial processes.