A novel data-driven methodology for fault detection and dynamic risk assessment

This paper presents a novel methodology for dynamic risk analysis, integrating the multivariate data-based process monitoring and logical dynamic failure prediction model. This concept for dynamic risk analysis is comprised of the fault assessment and dynamic failure prognosis modules. A combination of the naive Bayes classifier, Bayesian network, and event tree analysis is utilized to manifest the concept. The naive Bayes classifier is used for fault detection and diagnosis; it also generates a multivariate probability for a fault class in each time-step, which is used for dynamic failure prognosis by different paths a fault can lead a process to failure. The proposed framework has been applied to two process systems: a binary distillation column and the RT 580 experimental setup in four fault scenarios, and it is found the developed technique can effectively monitor the process and predict the failure.