Data-driven fault detection and diagnosis in industrial process systems: A systematic review and perspective

Industrial process systems, given their complexity and high risk, can cause catastrophic accidents in the case of failure, leading to casualties, environmental pollution, and economic losses. In the era of Industry 4.0, the rapid progress of intelligent production turns data-driven fault detection and diagnosis (FDD) into a crucial means to ensure the safety and reliability of representative systems, particularly across chemical, petrochemical, and energy industries where benchmark platforms such as the Tennessee Eastman Process (TEP), continuous stirred-tank reactor (CSTR), and Cranfield multiphase-flow facility exemplify practical applications. This paper presents the first systematic review of data-driven FDD research in industrial processes using the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) methodology, which focuses on the core objectives of process fault diagnosis, prognosis, predictive maintenance, and condition monitoring. With a structured approach to identify, appraise, and analyze screening literature, this paper thoroughly comprehends the complete pipelines from data benchmarking to modeling and diagnosis. The application of large language model (LLM), multi-source digital twin (DT), and explainable artificial intelligence (XAI) techniques in complex fault pattern recognition is further emphasized, providing practical guidance for the enhancement of system safety and sustainability in the context of smart industry.