A Data-Driven Causality Analysis Tool for Fault Diagnosis in Industrial Processes

Data-driven causality analysis is an important step towards fault diagnosis in complex industrial processes. Although many causality analysis tools were developed in different domains, only a few of them are applied in the industry. Accordingly, there is a need to develop a causality analysis tool that serves for fault diagnosis in large-scale chemical plants. This paper develops a decision-support tool to perform causality analysis by extracting useful information from the process historical data. The aim is to help the process operator to understand the underlying systems conditions with minimal efforts and to take appropriate actions in a short response time. The tool is implemented as a graphical user-friendly interface (GUI) that exploits the multivariate time series data and provides the user with stationarity tests and Granger causality analysis. It also offers various visualization charts such as pairwise causality relationships and most importantly the final causal graph. In order to demonstrate the easiness and usability of the developed tool, two different case studies are considered. The first case study is a time varying simulated model and the second one is the Tennessee Eastman Process as a well-known benchmark. The results show that the cause-and-effect information obtained by the developed tool can assist the user to deeply analyze causal variables and diagnose the corresponding fault with minimal involvement. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.