A Two-Tier Causal Network for Analyzing Primary Sources and Tracing Fault Propagation Routes in Intricate Manufacturing Systems

Research and discussion focus on the issue of identifying root causes and propagation paths of faults in complex systems with the primary aim of enhancing system safety. Initially, based on process knowledge, A complicated industrial procedure is segmented into several submodules with the aim of reducing system complexity, followed by the establishment of monitoring models for fault detection in each submodule. Secondly, contribution plots obtained from fault detection are used to identify and select submodules and variables related to the corresponding faults. Subsequently, a causal analysis using partial symbolic transfer entropy (PSTE) is performed on these selected components to construct a two-layer causal diagram, making causal relationships more concise and clearer. A novel fault propagation path recognition method based on a Causal Gated Recurrent Unit (C-GRU) is then designed. This framework simultaneously considers the mechanistic knowledge and data characteristics of industrial systems, thereby enabling more accurate fault diagnosis and path identification. A comprehensive experiment using Tennessee Eastman industrial process data verified the effectiveness of our method and highlighted its practicality in real-world scenarios. Actual data experiments show that this framework can accurately capture the fault source components and their propagation paths.