Review on trustworthy process monitoring and fault diagnosis with epistemic uncertainty

Significant epistemic uncertainty is common due to inadequate fault samples, environmental changes, and other factors, which affects the reliability of results. Although related research has increased about fault diagnosis and uncertainty quantification, most existing work has focused on specific methodological designs. There remains little systematic review for epistemic uncertainty. To address this issue, this paper reviews research on epistemic uncertainty of data-driven trustworthy process monitoring and fault diagnosis. Firstly, the primary sources and root causes are analyzed for epistemic uncertainty based on uniform fault diagnosis model. Secondly, this paper provides a systematic overview of existing uncertainties quantification methods. The principles, advantages, limitations, and the specific applications of these methods are compared from the perspectives of probability, belief, likelihood, and uncertainty measures. Furthermore, uncertainty support vector machine (UT-SVM) is described to illustrate the approach and implementation of epistemic uncertainty modeling in trustworthy fault diagnosis. Finally, it summarizes the main challenges and future directions of current studies. This paper aims to provide a comprehensive reference for trustworthy process monitoring and fault diagnosis methods under epistemic uncertainty.