Product design through multivariate statistical analysis of process data

Authors

  • Christiane M. Jaeckle,

    1. Dept. of Chemical Engineering, McMaster University, Hamilton, Ontario, Canada, L8S 4L7
    Search for more papers by this author
  • John F. Macgregor

    Corresponding author
    1. Dept. of Chemical Engineering, McMaster University, Hamilton, Ontario, Canada, L8S 4L7
    • Dept. of Chemical Engineering, McMaster University, Hamilton, Ontario, Canada, L8S 4L7
    Search for more papers by this author

Abstract

A methodology is developed for finding a window of operating conditions within which one should be able to produce a product having a specified set of quality characteristics. The only information assumed to be available is that contained within historical data on the process obtained during the production of a range of existing product grades. Multivariate statistical methods are used to build and to invert either linear or nonlinear empirical latent variable models of the existing plant operations to obtain a window of operating conditions that are capable of yielding the desired product and that are still consistent with past operating procedures and constraints. The methods and concepts are illustrated using a simulated high-pressure tubular reactor process for producing low-density polyethylene.

Ancillary