Application of multivariate statistics to support process control

This paper describes two methods by which multivariate statistics can be exploited to provide benefits to industrial process control. In the first application, the inner structure of a partial least squares model is applied to provide automatic model switching in piecewise linear systems. In the second application, multivariate statistical noise models, together with kernel density techniques, are used to determine the uncertainty in critical plant sensors. Based upon this uncertainty, it is shown that the set-points and constraints used within a model predictive control algorithm can be manipulated to ensure quality constraints are not violated.