Independent component analysis-based non-Gaussian process monitoring with preselecting optimal components and support vector data description

Abstract

Independent component analysis (ICA)-based process monitoring methods have rapidly progressed, but independent components (ICs) selection remains an open question. Subjective ICs selection would lead to useful information dispersion and affect the ICA monitoring performance. A novel ICA-based method integrated with preselecting optimal components and support vector machine data description (SVDD) technique is proposed to improve the non-Gaussian process monitoring performance. The proposed method first concentrates the informative ICs into one subspace for each fault and then the SVDD is employed to examine the variations in all subspaces. Case studies on a simulated process and Tennessee Eastman benchmark process demonstrate the effectiveness of the proposed scheme. The monitoring performances are significantly improved compared with the conventional ICA method.

Keywords:

• process monitoring
• multivariate statistics
• independent component analysis
• optimal component selection
• support vector data description

Funding

The authors gratefully acknowledge the support from the following foundations: 973 project of China [2013CB733605]; National Natural Science Foundation of China [21176073] and the Fundamental Research Funds for the Central Universities.