Fault Detection of Multimodal Chemical Process Based on CLTSA

Abstract

For modern chemical process often has multiple operational modes and monitoring data in the process are often nonlinear and high dimensional, a new fault detection method based on improved local tangent space alignment (LTSA) is proposed in this paper. Firstly, aiming at the characteristics of multi-modality of the chemical process, an improved LTSA algorithm is proposed in the paper, called correlation tangent space arrangement (CLTSA). In CLTSA, the variable $r$ is constructed and used to describe the relationship between the multivariate variables and reconstruct the global coordinates of monitoring data. Then, the incremental learning mechanism is introduced in CLTSA. For newly collected data, only some elements of the transition matrix need to be updated. And the matrix similarity statistics is established to maintain the size of the transition matrix, which has improved the efficiency of the algorithm. Finally, nonlinear principal elements in monitoring data are extracted through CLTSA and statistics $T^{2}$ and $SPE$ are used to monitor the change of the principal elements. When the monitored amount exceeds the threshold, it is determined that a fault has occurred in the chemical process. The simulation results of TE process show that the method proposed in the paper has a high fault detection rate and provides a new way for fault detection of multi-modal nonlinear chemical processes.