Experimental Design of Formulations Utilizing High Dimensional Model Representation.

Area of Science:

• Chemistry
• Chemical Engineering
• Materials Science

Background:

• Mixture formulation often requires selecting optimal compositions from numerous potential components.
• Traditional design of experiments (DoE), like simplex-lattice sampling, becomes impractical for large numbers of components due to the high experimental cost.
• Efficiently modeling complex mixtures with many components is a significant challenge in formulation science.

Purpose of the Study:

• To develop a novel approach for constructing a single predictive model for mixtures with numerous components.
• To reduce the number of experiments required for mixture screening and optimization.
• To enable accurate performance prediction for full, binary, and ternary component mixtures.

Main Methods:

• Utilized biased random sampling to select a modest number of experimental points.
• Employed high dimensional model representation (HDMR) to build a comprehensive model from the sampled data.
• Replaced traditional simplex-lattice design in design of experiments (DoE) with the proposed sampling and modeling strategy.

Main Results:

• Successfully constructed a single model capable of predicting mixture performance using significantly fewer experiments than DoE.
• Demonstrated the model's efficacy in predicting performance for full, binary, and ternary component mixtures.
• Achieved substantial reduction in experimental effort, particularly for systems with a large number of potential components.

Conclusions:

• The combination of biased random sampling and HDMR offers a highly efficient alternative to traditional DoE for mixture modeling.
• This approach is particularly advantageous for complex formulations with many permissible components.
• The method was successfully illustrated through a solvent mixture solubility model, showcasing its practical applicability in mixture screening.