Fault diagnosis of nonlinear process based on KCPLS reconstruction

In this paper, a new kernel concurrent projection to latent structure (KCPLS) reconstruction method for process monitoring is proposed. The main contributions of the proposed approach are as follows: (1) the KCPLS method provides a complete monitoring of faults that happen in the predictable output subspace and the unpredictable output-residual subspace; (2) after the fault reconstruction approach is proposed, the fault-relevant direction is determined; (3) the fault is effectively diagnosed compared to the conventional KPLS method. The proposed method is applied to penicillin fermentation process and is compared to the KPLS method. Experiment results show that the KCPLS can more accurately detect the fault compared to the KPLS method. In addition, the fault-relevant direction is identified more effectively by using the KCPLS reconstruction algorithm compared to the KPLS reconstruction approach. (C) 2014 Elsevier B.V. All rights reserved.