Review of recent applications and future perspectives on process monitoring approaches in industrial processes

Abstract

Process monitoring is essential in industrial production, as it ensures product quality and production efficiency through real-time data monitoring and analysis during the manufacturing process. Process monitoring generally includes four procedures: fault detection, fault diagnosis, fault isolation and root cause diagnosis. However, few current works present comprehensive review papers covering the four aspects. Thus, this review presents a timely and comprehensive retrospective analysis of process monitoring techniques and provides an in-depth review of research developments in process monitoring across different scopes. Firstly, this review discusses the characteristics and applications of both traditional machine learning-based and deep learning-based process monitoring methods, which offer a comprehensive comparison and evaluation of their respective strengths and limitations. Secondly, the extensions, prospects and challenges in data-driven process monitoring (i.e., adaptive, interpretable approaches as well as contrastive learning and meta-learning-based techniques) are discussed to lay a solid foundation for future research. Thirdly, the application procedures, including fault detection, fault diagnosis, fault isolation and root cause diagnosis, are elaborated to provide valuable research references and insights for both academics and practitioners. Finally, the existing challenges and promising research directions are discussed, which can pave the way for future research and contribute to the advancement in process monitoring.