A Novel Statistical-Based Monitoring Approach for Complex Multivariate Processes

Zhiqiang Ge, Lei Xie, Zhihuan SongIndustrial 2009 · DOI

Abstract

Conventional methods are under the assumption that a process is driven by either non-Gaussian or Gaussian essential variables. However, many complex processes may be simultaneously driven by these two types of essential source. This paper proposes a novel independent component analysis and factor analysis (ICA-FA) method to capture the non-Gaussian and Gaussian essential variables. The non-Gaussian part is first extracted by ICA and support vector data description is utilized to obtain tight confidence limit. A probabilistic approach is subsequently incorporated to separate the residual Gaussian part into latent influential factors and unmodeled uncertainty. By retrieving the underlying process data generating structure, ICA-FA facilitates the diagnosis of process faults that occur in different sources. A further contribution of this paper is the definition of a new similarity factor based on the ICA-FA for fault identification. The efficiency of the proposed method is shown by a case study on the TE benchmark process.
BibTeX
@inproceedings{ge2009novel,
  author = {Zhiqiang Ge and Lei Xie and Zhihuan Song},
  title = {{A Novel Statistical-Based Monitoring Approach for Complex Multivariate Processes}},
  booktitle = {Industrial},
  year = {2009},
  doi = {10.1021/ie800935e},
}

Loading related content...

Figure