Multivariate statistical analysis of a multi-step industrial processes

S.P. Reinikainen, Agnar Høskuldsson

    Research output: Contribution to journalJournal articleResearchpeer-review

    Abstract

    Monitoring and quality control of industrial processes often produce information on how the data have been obtained. In batch processes, for instance, the process is carried out in stages; some process or control parameters are set at each stage. However, the obtained data might not be utilized efficiently, even if this information may reveal significant knowledge about process dynamics or ongoing phenomena. When studying the process data, it may be important to analyse the data in the light of the physical or time-wise development of each process step. In this paper, a unified approach to analyse multivariate multi-step processes, where results from each step are used to evaluate future results, is presented. The methods presented are based on Priority PLS Regression. The basic idea is to compute the weights in the regression analysis for given steps, but adjust all data by the resulting score vectors. This approach will show how the process develops from a data point of view. The procedure is illustrated on a relatively simple industrial batch process, but it is also applicable in a general context, where knowledge about the variables is available.
    Original languageEnglish
    JournalAnalytica Chimica Acta
    Volume595
    Issue number1-2
    Pages (from-to)248-256
    ISSN0003-2670
    DOIs
    Publication statusPublished - 2007

    Keywords

    • monitoring
    • priority regression
    • CovProc
    • partial least squares (PLS)

    Cite this