Graph attention network with granger causality map for fault detection and root cause diagnosis

• Published in: Computers & chemical engineering 2024-01, Vol.180, p.108453, Article 108453
• Main Authors: Liu, Yingxiang, Jafarpour, Behnam
• Format: Article
• Language: English
• Subjects: Conditional Granger causality analysis; Fault detection; Graph attention network; Root cause diagnosis
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2023.108453

Unsupervised data-driven methods are widely used for fault detection and diagnosis in modern industrial processes. However, previous studies have limitations in accurately distinguishing faults from normal feedback control system adjustments and promptly identifying root causes. To address these limitations, we propose a neural network model consisting of one-dimensional convolutional neural networks and a graph attention network (CNN-GAT) that uses a causal map derived from fault-free data using conditional Granger causality analysis. The CNN-GAT model produces a monitoring index that accurately reflects the operating conditions of the process and distinguishes faults from normal control adjustments. Using the causal map and prediction results from the CNN-GAT model, the root cause diagnosis can be performed promptly after faults are detected, allowing operators more time to deal with the fault. We demonstrate the performance of the proposed framework using the benchmark Tennessee Eastman process case studies and comparison with other fault detection methods. •The proposed CNN-GAT model leverages one-dimensional Convolutional Neural Networks and a Graph Attention Network to predict multivariate time series data based on causal relationships rather than correlations among variables.•The CNN-GAT model effectively handles nonstationarity and shifts in data distribution, resulting in a fault detection index that accurately reflects actual operating conditions and distinguishes faults from normal control adjustments.•The fault diagnosis method based on the CNN-GAT model enables quick identification of the root cause without relying on a large amount of faulty data collected after faults.