Fault detection for a class of industrial processes based on recursive multiple models

Abstract

Traditional multiple model process monitoring methods usually yield satisfactory results for multi-mode processes under the assumption that the processes are time invariant. However, for some petrochemical processes, such as ethylene cracking furnace, the process is time varying as the coking in the furnace tubes. To solve this problem, this study proposes a multiple model recursive monitoring method. A computational intelligence-based cluster algorithm is employed to separate different operating modes. Then, recursive kernel principal component analysis is used to reduce the dimension of the time-varying process data and extract the nonlinear principal components recursively. Furthermore, support vector data description is utilized to build models because the process data are non-Gaussian. Finally, the corresponding statistics are constructed to detect the process fault. The performance of this method is evaluated through a case study of ethylene cracking in a petrochemical plant.