(22a) Root Cause Analysis of Key Process Variable Deviation for Rare Events

Rare events are low-frequency high-impact consequences of abnormal events caused by process disturbances. To reduce the severity of consequences, identification of a root cause of the rare event is of great importance to provide an efficient troubleshooting advice. Root cause analysis of rare events in process industry deals with challenges of data scarcity and uncertainty in both of the rare events data and the abnormal events data. Difficulty presented by sparse rare events data cannot be dealt with existing feature selection techniques as they require vast amounts of data [1]. On the other hand, uncertainty caused by sparse abnormal events data in root cause analysis can be overcome by applying Bayesian model (BM) in conjunction with a fault tree [2]. BM overcomes the unidirectional nature of a fault tree connecting process disturbances to an abnormal event, thus, making root cause analysis possible while accounting for uncertainty. However, the BM approach assumes that the collected abnormal events data come from strictly consistent operating conditions. In other words, it is assumed that different sources of process disturbances have same tendencies to induce disturbances and an abnormal event, which is not true for complex processes. Consequently, it results in unaccountability of uncertainty associated with the source variability of process disturbances. To address this issue, hierarchical Bayesian model (HBM) was proposed which incorporates a stage for process disturbances and their sources, and has been shown to effectively deal with uncertainty caused by source variability of data sources in other fields [3]. However, non-utilization of process knowledge in the application of HBM results in overestimation of uncertainty. In this work we present a new framework to first address the challenge of sparse rare events data by utilizing a combination of existing feature selection techniques to circumvent their respective limitations; and then address source variability of abnormal events data using HBM by utilizing process knowledge for a reliable uncertainty estimation.

The framework presented in this work implements the root cause analysis of rare events in two steps. The first step utilizes a new feature selection technique combining relative information gain and Pearson correlation co-efficient to identify a key process variable significant to the rare event. This technique utilizes the respective advantages of the statistical quantities for capturing overall relationships and working well with smaller datasets. The second step determines the root cause of deviations in the key process variable using HBM. A fault tree is constructed for process disturbances leading to a key process variable deviation. The knowledge from process disturbance data is borrowed in the form of informative prior to be utilized with HBM to avoid uncertainty overestimation. The proposed framework has been effectively demonstrated with a case study of the Tennessee Eastman process. Hence, this work presents a novel methodology for root cause analysis of rare events accounting for source variability of process disturbances and using process knowledge to determine root cause reliably.

Keywords: Hierarchical Bayesian modeling, fault tree, rare events, informative prior

References:

1. Venkatasubramanian, V.; Rengaswamy, R.; Kavuri, S. N.; Yin, K. A review of process fault detection and diagnosis: Part III: Process history based methods. Computers & Chemical Engineering 2003, 27, 327–346.
2. Khakzad, N.; Khan, F.; Amyotte P. Dynamic safety analysis of process systems by mapping bow-tie into Bayesian network. Process Safety and Environmental Protection, 2013; 91(1):46–53.
3. Yu, H.; Khan, F.; Veitch, B. A Flexible Hierarchical Bayesian Modeling Technique for Risk Analysis of Major Accidents. Risk Analysis 2017, 37, 1668–1682.