Identification of contributing variables using kernel-based discriminant modeling and reconstruction

Abstract

Fault identification is one of essential operational tasks required for process safety and consistent production of high quality final products. The objective of fault identification is to identify process variables responsible for causing a specific fault in the process. Such an identification of contributing process variables helps process operators or engineers to diagnose a root cause of the fault more effectively. A new nonlinear fault identification method is developed using a nonlinear kernel-based Fisher discriminant analysis (KFDA). The proposed method performs a pair-wise KFDA on normal and fault data. Thus it characterizes the change of each process variable’s contribution relative to normal operating conditions when a specific fault occurs. A case study on the Tennessee Eastman process has shown that the proposed method produces reliable identification results. Moreover, the proposed method outperforms the contribution chart approach based on linear PCA. The use of a nonlinear technique of KFDA in a fault identification task was shown to be a promising tool for determining key process variables of various faults.