The PCA Fault Diagnosis Method Based on Causal Relationships

Fault diagnosis is a critical and central issue in industrial engineering, essential for addressing problems encountered in engineering systems. In recent years, data-driven fault diagnosis methods have gained extensive attention and application. Traditional fault diagnosis techniques primarily focus on analyzing data from individual fault nodes, making it challenging to capture the causal relationships between fault nodes accurately. This paper proposes a fault diagnosis method based on the combination of principal component analysis (PCA) and Liang-Kleeman Information Flow (LKIF). This method quantifies causal interactions between nodes by introducing information flow, thereby enhancing diagnostic efficiency. To validate the effectiveness of this approach, we conducted simulation experiments using the Tennessee Eastman Process (TEP) as a testbed. Specifically, we analyzed two faults in the TE platform and designed comparative experiments by varying two input variables. The results demonstrate that the LKIF-PCA fault diagnosis method achieves high accuracy, indicating its potential for practical applications.