Novel Multimodule Bayesian Network with Cyclic Structures for Root Cause Analysis: Application to Complex Chemical Processes

Currently, most industrial processes are equipped with complex alarm systems for the purpose of safety. Unfortunately, in the face of the increasing complexity of industrial process scales, alarm floods often occur, resulting in the failure of alarm systems to respond effectively. The Bayesian network can be used to solve alarm floods by finding the root causes of alarms. However, the cyclic relationship of process variables cannot be learnt by the traditional Bayesian network, reducing the accuracy of root cause analysis. To solve this problem, a novel multimodule Bayesian network with cyclic structures is proposed. The development of the proposed method is divided into four steps: decomposing the whole system into submodules, Bayesian network structure learning at the state node level, merging the modular Bayesian networks into a special Bayesian network with cyclic structures at the node level, and finally analyzing the root cause of alarms. In addition, a hybrid greedy firework algorithm is used to avoid overfitting and the event expansion method is utilized to accurately determine the final source of alarms. To verify the performance of the proposed method, a case study using the Tennessee Eastman process is provided. The simulation results prove the effectiveness and feasibility of the proposed multimodule Bayesian network with cyclic structures.