Reduced Rank KPCA based on GLRT chart for sensor fault detection in nonlinear chemical process

Abstract

Kernel Principal Components Analysis (KPCA) method it is the frequently used among the other kernel methods due to their easiness and it competence in modeling processes and fault diagnosis with less necessity for prior knowledge on process mechanism. To treat the high calculation costs and the memory storage, in this article, a new defects detection process is developed. The developed detection method, which is called RR-KGLRT. The RR-KPCA approach is utilized to build a reduced reference model and we use Generalized Likelihood Ratio Test (GLRT) for detection purposes. The proposed technique upgrade the computational competence by downsizing the kernel matrix size than the KPCA technique based on GLR test. Furthermore, the proposed method can ameliorate the detection performances compared to RRKPCA method based on SPE index. In order to ameliorate the performances of the proposed method we have also proposed another fault detection method that fuse the benefit of the proposed RR-KGLRT technique with the exponentially weighted moving average (EWMA) filter. The suggested RR-KGLRT based EWMA can reduce significantly the false alarm rate and ameliorate the good detection. The Continuous Stirred tank Reactor (CSTR) and the Tennessee Eastman process (TEP) are used to value the capability of the proposed RR-KGLRT to detect faults.