Computer power alone was not enough to design successful mixtures, but neither was the brute-force approach of physically testing every possible combination.
Instead, a computer algorithm was linked to robotic system that blends polymers, tests them and passes the results back to the algorithm.
“The researchers originally tried a machine-learning model to predict the performance of new blends, but it was difficult to make accurate predictions across the astronomically large space of possibilities,” said MIT. “Instead, they utilised a genetic algorithm, which uses biologically inspired operations like selection and mutation to find an optimal solution.”
It sent 96 polymer blend recipes at a time to the autonomous mixing and testing platform – in all it is capable of making at testing 700 blends a day, only requiring human intervention to refill chemical reservoirs.
The system autonomously identified hundreds of blends that outperformed their constituent polymers, said MIT, with some of the best-performing blends not including any of the best-performing components, and the best overall blend performing 18% better than any of its components.
“Using a different approach, you could easily overlook the under-performing components that happen to be the important parts of the best blend,” said team scientist Connor Coley.
“We could sometimes blend existing polymers to design new materials that perform even better than individual polymers do,” added fellow researcher, Guangqi Wu, now at the University of Oxford.
The work is published as ‘Autonomous discovery of functional random heteropolymer blends through evolutionary formulation optimization‘ in Matter.
