Computer-aided design of formulated products: A bridge design of experiments for ingredient selection

• Published in: Computers & chemical engineering Vol. 169; p. 108083
• Main Authors: Cao, Liwei, Russo, Danilo, Matthews, Emily, Lapkin, Alexei, Woods, David
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2023
• Subjects: Bayesian optimization; Design of Experiments (DoE); Gaussian processes; Machine Learning (ML); Product design; Design of Experiments (DoE); Product design; Machine Learning (ML); Bayesian optimization; Gaussian processes
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2022.108083

•A bridge-design of experiments was developed for formulations optimization.•The methodology allowed to simultaneously optimize discrete and continuous variables.•The methodology allowed to choose a subset of ingredients from n available ones.•Formulations were optimized with an automated platform in 17 working days. Formulations are ubiquitous in many industries. As formulations are being modified and re-developed to include more renewable and recyclable ingredients, the speed of formulations development becomes important. This study expands on the previous work demonstrating successful application of multi-objective Bayesian optimization to design of formulations within a restricted set of the available ingredients. Here we develop an approach that resolves the un-solved to date problem in algorithmic formulations development, when a subset of ingredients should be chosen from a larger available pool of suitable ingredients. The new DoE algorithm was demonstrated in a workflow making use of a 'make and test' formulation robots. The developed new DoE procedure demonstrated an efficient selection of a subset of ingredients from a larger number of the available ones, optimizing their concentration and allowing assignment of differential priorities to the optimization objectives.