Optimization of Formulations Using Robotic Experiments Driven by Machine Learning DoE

Highlights

• A commercial formulation is optimized using a Bayesian approach

• Discrete and continuous targets are optimized without any available physical model

• A robotic platform can perform semiautomated sample preparation and characterization

• The adopted iterative approach gives satisfactory results within 15 working days

Summary

Formulated products are complex mixtures of ingredients whose time to market can be difficult to speed due to the lack of general predictable physical models for the desired properties. Here, we report the coupling of a machine learning classification algorithm with the Thompson sampling efficient multiobjective optimization (TSEMO) algorithm for the simultaneous optimization of continuous and discrete outputs. The methodology is successfully applied to the design of a formulated liquid product of commercial interest for which no physical models are available. Experiments are carried out in a semiautomated fashion using robotic platforms triggered by the machine learning algorithms. The procedure allows one to find nine suitable recipes meeting the customer-defined criteria within 15 working days, outperforming human intuition in the target performance of the formulations.