Engineers at the University of Wisconsin-Madison have harnessed the power of prediction to quickly identify several high-performance polymers from a field of eight million candidates. These polymers, known as polyimides, possess exceptional mechanical and thermal properties that make them ideal for a wide range of applications in the aerospace, automobile, and electronics industries. The process of designing polyimides is currently costly and time-consuming, but the UW-Madison team used a data-driven design framework to leverage machine learning predictions and molecular dynamics simulations to accelerate the discovery of new and superior polyimides.
The team’s research is detailed in a paper published this month in the Chemical Engineering Journal. Using open-source data of chemical structures, the team built a comprehensive library of eight million hypothetical polyimides, much like building something with LEGO blocks. A computer was then used to combine the building blocks together, creating a vast database of all possible combinations.
The team then created multiple machine learning models based on experimentally reported values, identifying chemical substructures critical to determining individual properties. Their well-trained models predicted the properties of the eight million hypothetical polyimides, and the team screened that dataset to identify the three best hypothetical polyimides with superior properties to existing polyimides.
To validate their predictions, the researchers built all-atom models for their top-three candidates and conducted molecular dynamics simulations. The simulations showed that these new polyimides were easy to synthesize and would provide excellent heat resistance, agreeing with their machine learning predictions.
In contrast, existing polyimides could withstand temperatures only in the range of 392 to 572 degrees F, while the new polyimides could withstand temperatures up to 1,022 degrees Fahrenheit. The researchers also developed a web-based application that allows users to explore the new high-performing polyimides with interactive visualization.
The team’s findings promise to inspire further research in advanced data-driven techniques for materials discovery, with broad implications for the field of materials science. The design strategy is more efficient than the conventional trial-and-error process and can also be applied to the molecular design of other polymeric materials.
