The Bayesian Network Method for Handling Cyclic Loops Based on Information Flow

Accurately identifying fault sources in industrial systems is essential for ensuring system safety and optimizing performance. Due to cyclic loops, traditional Bayesian networks often encounter difficulties when dealing with complex systems, which can lead to misjudgments, especially in scenarios involving causal feedback. This paper proposes an improved method based on information flow, utilizing the Liang-Kleeman information flow theory to handle C-R structure cyclic loops within Bayesian networks. By quantifying the strength of causal relationships in cyclic loops, the method effectively identifies and eliminates the weakest causal link, transforming it into a temporal relationship and simplifying the structure into an acyclic form. Experiments were conducted on a C-R structure feedback system to validate the new method’s performance under high-noise conditions. The results demonstrate that the process improves the accuracy of causal relationship identification and exhibits superior robustness in noisy environments, significantly enhancing fault diagnosis accuracy and real-time capabilities. This study provides novel solutions and tools for fault diagnosis in complex industrial systems.