Fault Propagation Path Inference in a Complex Chemical Process Based on Time-delayed Mutual Information Analysis

Abstract Process monitoring plays more and more important role in modern process industry. Early root cause isolation is the most attractive character to process operators. Currently, the signed directed graph (SDG) is a widely used method for fault diagnosis, in which graphs are employed to represent the causality between process variables. In most cases, the SDG model is obtained from expert experience. The challenge of this approach is that it is hard to include all knowledge required in complex chemical process operation, which may not be available to experts and operating professionals, as they can be significantly different with changing control strategies, even to same set of process operation. With the universal application of distributed control system (DCS), operation data have been recorded for a certain long time, which contain comprehensive information regarding the process itself. It can be expected that the logic and time dependence among all variables can be extracted with proper analysis method. Mutual information is a commonly used data-based method for measuring the interaction of two objects. The initiator and responder between a pair of variables can be identified by adding an appropriate time lag. In this paper, the identification of fault propagation path is achieved based on a time-delayed mutual information method. When a fault occurs, the response information among variables will be used to explore the propagation path, which provides a more objective information for fault diagnosis. The methodology is applied to a simulated process and a practical industrial case, an ethylene cracking process. The result illustrates that the proposed data-based method shows a good capability for identifying the fault propagation path.