.png){kind=icon}
.png){kind=icon}
Return
A novel performance-oriented causal graph-based fault detection and root cause diagnosis integrated framework for complex manufacturing processes
.png){kind=icon}
PRE
AI
T
Y
X
X
董
DOI：10.1016/j.psep.2026.109222.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the increasing complexity of industrial manufacturing processes, modern production systems have become highly interconnected and strongly coupled. As a result, performance anomalies may propagate and intensify through intertwined material, energy, and information flows, posing significant challenges to fault detection and root cause diagnosis. To address these challenges, a performance-oriented causal graph-based integrated framework is proposed for fault detection and root cause diagnosis. It characterizes the propagation of performance-related information among process variables with respect to the key performance indicator. First, a performance-oriented gated recurrent unit-based stacked autoencoder (PGRU-SAE) is developed to extract deep dynamic performance-related features. A performance-supervised prediction model is then established by integrating the PGRU-SAE with a multi-layer perceptron. Based on the prediction network, weighted path scores are defined to identify the performance-oriented causal skeleton. Second, an effective model complexity-based causal direction identification method is proposed to infer causal directions from the network’s internal training parameters. In addition, industrial process knowledge is incorporated to prune redundant edges, resulting in an interpretable performance-oriented causal graph. Then, the extracted deep performance-related features are employed for online fault detection, and Bayesian inference is integrated to enhance detection accuracy and reliability. After an anomaly is detected, variable contribution scores are computed to identify candidate faulty variables. These variables are combined with the established performance-oriented causal graph to analyze fault propagation paths, which enables precise tracing of root cause variables. Experiments on a real finishing mill process demonstrate that the proposed method achieves superior monitoring and diagnostic performance compared with some baseline methods.
Keywords:
Performance-oriented causal graph
Gated recurrent-based stacked autoencoder
Process monitoring
Root cause diagnosis
Complex manufacturing process
Journal
IF:
7.8
Papers:
9.4K
Citations:
3.8W
Organization
No organization information available
Cited Papers