(70c) Utilizing Statistical Doe and Modeling to Accelerate Polyurethane Research

The Dow Chemical Company is a world leading material science company where statistical techniques have been widely used to enable efficient scientific research for industrial innovation. Through a real polyurethane (PU) research example, this presentation introduces how we have applied statistical design of experiments (DOE) and modeling to accelerate new product development with statistical confidence.

The DOE study incorporated three categorical and three continuous factors, each with three levels. A full factorial design would require 729 experiments, which is impractical. A 60-run D-optimal design was developed to screen polyol backbones and additives for formulation optimization. The DOE was evaluated to ensure its suitability for prediction and the investigation of factors’ impacts, with a focus on aspects such as relative prediction variances, correlation between model terms, and variance inflation factors. The DOE significantly reduced experimental work and still allowed the fitting of relevant statistical models with the DOE data collected.

Based on the DOE, statistical models were developed by integrating various sources of data, such as analytical image analysis, visual inspection, rheology test, and dynamic mechanical test. Several statistical modeling techniques including beta regression, logistic regression, classification tree, and random forest were utilized to determine the impacts of critical factors and identify optimal formulations. Simulation was also performed to assess variable importance in explaining the variability of the performance measurement. The statistical analysis successfully screened the polyol backbones and additives for next-stage formulation optimization. This presentation demonstrates the power of statistical DOE coupled with statistical modeling in accelerating PU research and driving data-based business decisions.