2019 International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS) | 2019
Fault Detection of Discriminant Improved Local Tangent Space Alignment
Abstract
With the increasing complexity of automation systems in chemical processes, online monitoring and fault diagnosis are effective ways to ensure product quality and reduce equipment failure. Aiming at the characteristics of nonlinear, multi-noise, Gaussian and non-Gaussian mixing of real industrial process data, this paper proposes a chemical process fault detection method based on Discriminant Improved Local Tangent Space Alignment (DILTSA). Firstly, the low-dimensional eigenvectors of high-dimensional matrices are extracted by using the discriminative improvement of DILTSA with high robustness and less adjustment parameters. The dimension reduction is performed to obtain the corresponding optimal mapping matrix. Then, the support vector data description (SVDD) is introduced. The SVDD model is built from the feature space. Finally, the method proposed in this paper is applied to the TE platform. The classical detection method is compared and the experimental simulation results are analyzed. It shows that the method can achieve better fault detection results.