A computational approach to partial least squares model inversion in the framework of the process analytical technology and quality by design initiatives

Abstract

In the context of the paradigms founding the Quality by Design and Process Analytical Technology initiatives, the work herein presents a computational approach to support the decision-making process, in particular, about the feasibility of a product defined for some a priori given quality characteristics.

The approach is based on the computation of the pareto-optimal front when simultaneously minimizing the expected differences between the predicted and the desired characteristics. Thus, the feasibility is tackled as an optimization problem with the novelty of doing so simultaneously for all the characteristics, preserving the correlation structure, but by separately handling each individual characteristic.

With data from a low-density polyethylene production process, with fourteen process variables and five measured characteristics of the final polyethylene, solutions are found to define the Design Space for targeted quality characteristics on the product, and without the need of explicitly inverting the PLS (Partial Least Squares) prediction model fitted to the process.