Dynamic transfer entropy diagram for fault root cause diagnosis in multi-mode industrial process

The complexity of multi-mode industrial processes poses challenges for traditional fault diagnosis methods in achieving satisfactory results. This challenge arises from the non-stationarity caused by changes in modes, which often overlaps with temporal variations in fault generation processes. To address this issue, we propose a fault diagnosis method based on multi-mode categorization and dynamic transfer entropy graph analysis. This method can uncover the causal information changes resulting from both mode switches and fault occurrence. Firstly, modes are categorized using the criterion of maximum intra-segment similarity and minimum inter-segment similarity, and a static causal graph is constructed for each mode. Secondly, a framework for dynamic transfer entropy graph is established to extract short-term causal information transfer using transfer entropy and compare it with the static causal graph of normal modes. Time points with significant changes in causal relationships are selected as fault time points for further analysis. Finally, an anomaly score is designed to identify critical fault nodes by selecting nodes with significant causal changes before the fault time point, thereby accurately locating the fault root nodes. To validate the performance of the proposed method, a case study using the Tennessee Eastman process is provided. Simulation results demonstrate the effectiveness and feasibility of the proposed method.