Spatial-Temporal Dependency Based Automatic Failure Mode Classification for Industrial Processes

With the rapid development of network technology, cyber-physical systems in the chemical industry are facing increasingly severe cybersecurity threats. Cyber attacks can lead to the failure of critical protection layers within these systems, including the basic process control system (BPCS), alarm systems, and the safety instrumented system (SIS), thereby endangering the safety and stability of industrial processes. In order to accurately identify the system state, this study proposes a spatial-temporal dependency based multivariate time series classification method for industrial processes. By preprocessing the process data from cyber-physical systems in the chemical industry and subsequently employing graph attention networks and diagonal state space models for feature extraction, we capture the spatial-temporal dependencies in the process data and obtain low-dimensional representations. This approach enables accurate monitoring of the states of industrial control systems and precise localization of failures. We validated the performance of the proposed method using the Tennessee Eastman Process (TEP) dataset and the Catalytic Cracking Process (CCP) dataset. Experimental results demonstrate that this method achieves high accuracy and reliability, providing strong support for enhancing the security of industrial control systems.