Okba Taouali

Moving window KPCA with reduced complexity for nonlinear dynamic process monitoring.

This paper proposes an improved Reduced Kernel Principal Component Analysis (RKPCA) for handling nonlinear dynamic systems. The proposed method is entitled Moving Window Reduced Kernel Principal Component Analysis (MW-RKPCA). It consists firstly in approximating the principal components (PCs) of the KPCA model by a reduced data set that approaches properly the system behavior in the order to elaborate an RKPCA model. Secondly, the proposed MW-RKPCA consists on updating the RKPCA model using a moving window. The relevance of the proposed MW-RKPCA technique is illustrated on a Tennessee Eastman process.

Fault detection and isolation in nonlinear systems with partial Reduced Kernel Principal Component Analysis method

In this article, we suggest an extension of our proposed method in fault detection called Reduced Kernel Principal Component Analysis (RKPCA) (Taouali et al., 2015) to fault isolation. To this end, a set of structured residues is generated by using a partial RKPCA model. Furthermore, each partial RKPCA model was performed on a subset of variables to generate structured residues according to a properly designed incidence matrix. The relevance of the proposed algorithm is revealed on Continuous Stirred Tank Reactor.

Online process monitoring based on kernel method

This paper discusses the monitoring of dynamic process. In recent years, Kernel Principal component analysis (KPCA) has gained significant attention as a monitoring method of nonlinear systems. However, the fixed KPCA model limit its application for dynamic systems. For this purpose a new Variable Moving Window Kernel PCA (VMWKPCA) method is introduced to update the KPCA model. The basic idea of this technique is to vary the size of the moving window depending on the normal change of the process. Then the VMWKPCA method is performed for monitoring a Chemical reactor (CSTR). The simulation results proved that the new method is effective.

Fault detection localization and reconstruction in nonlinear system using RKPCA method and RBC

In this paper we exploit the use of the proposed RKPCA method ([1], [2], [3]) for sensor fault detection, localisation and reconstruction. To this end, a set of structured residues is generated by using partial RKPCA technique. Also to identify fault, the Reconstruction Based Contribution RBC approach [4] was used. The relevance of the evaluated techniques partial RKPCA and RBC is revealed on Continuous Stirred Tank Reactor (CSTR).

Identification of nonlinear systems with kernel methods

This paper present a nonlinear system identification based kernel methods, such as regularization networks, support vector regression and kernel principal component analysis. In this case, black-box models are used in a particular space named reproducing kernel Hilbert space (RKHS) which only considered the input/output signals of the nonlinear system. In this particular space, the model is a linear combination of kernel functions applied to transform the observed data from the input space to a high dimensional feature space of vectors, this idea known as the kernel trick. To prove the performances of the kernel methods, identification examples are illustrated with three single-input single-output (SISO) benchmark models.

A new online kernel method identification on RKHS space

This paper proposes a new kernel method for online identification of nonlinear system. The proposed Support Vector Regression-Regularized Network (SVR-RN) method uses the technique SVR in an offline phase to reduce the parameters number of the RKHS. Then the RN method is used to update theses reduced parameters.