Recent review on deep learning-based Dempster–Shafer decision-making framework for chemical process monitoring

Abstract Chemical process monitoring is vital for ensuring safe, efficient, and reliable operations in the chemical and process industries. However, practical implementation remains challenging due to complex nonlinear dynamics, high-dimensional sensor networks, and inherent uncertainty in industrial environments. Conventional monitoring and decision-making methods often struggle to manage these complexities effectively. This review provides a comprehensive overview of integrated deep learning (DL)–Dempster–Shafer theory (DST) frameworks for chemical process monitoring. DL techniques enable automatic extraction of nonlinear and temporal features from high-dimensional process data, while DST offers a principled mechanism for uncertainty representation, multi-source evidence fusion, and conflict resolution. The review summarizes core components of process monitoring, including sensor-based systems, fault detection and diagnosis, and intelligent decision-making strategies. It then examines recent DL–DST integration approaches, demonstrating improvements in diagnostic robustness, confidence-aware decision-making, and reliability under noisy and uncertain conditions. Benchmark studies, including the Tennessee Eastman process, highlight the advantages of DL–DST frameworks over standalone DL models. Industrial applicability is further analyzed, addressing challenges such as real-time deployment, process drift adaptation, high-dimensional data processing, and the absence of standardized integration guidelines. Future research directions are outlined to support scalable, interpretable, and industrially deployable monitoring solutions.