Engineers accelerate polymer discovery with transparent machine learning tools

Using the power of prediction, mechanical engineers at the University of Wisconsin-Madison quickly discovered several promising high-performance polymers among eight million candidates.

In the aerospace, automotive and electronics industries, these polymers, known as polyimides, are used in a wide range of applications due to their excellent mechanical and thermal properties such as strength, stiffness and heat resistance.

Currently, the number of existing polyimides is limited due to the costly and time consuming design process.

But with a data-driven design framework, engineers at Madison, Wisconsin are leveraging machine learning predictions and molecular dynamics simulations to dramatically speed the discovery of new polyimides with even better properties. increase.

The researchers detailed their approach in a paper published this month in the Chemical Engineering Journal.

“Our findings will have far-reaching implications for the field of materials science and will stimulate further research in the development of advanced data-driven techniques for materials discovery,” said the study’s lead researcher, the University of Washington Madison Machinery. Associate Professor of Engineering Ying Li said. “Our design strategy is much more efficient than the traditional trial-and-error process and can be applied to molecular design of other polymeric materials.”

Polyimides are produced by the condensation reaction of dianhydride and diamine/diisocyanate molecules. For the study, the engineers first collected open-source data on the chemical structures of all existing dianhydride and diamine/diisocyanate molecules, then used that data to generate a comprehensive set of 8 million virtual polyimides. I built the library.

“It’s like building something out of Lego bricks,” Lee says. “There are basic building blocks for different dianhydride molecules and diamine/diisocyanate molecules. And while you could try to build every possible structure by hand, the variety of combinations is huge. , it will take forever.”

So Lee and his colleagues were able to use computers to combine the building blocks and organize all possible combinations into a huge database.

Utilizing the database, the team created multiple machine learning models for the thermal and mechanical properties of polyimide based on experimentally reported values. Researchers used a variety of machine learning techniques to identify the most important chemical substructures for determining individual properties.

“The model is not a black box because we have incorporated techniques that essentially explain how the machine learning model works,” Lee says. “We built a transparent box that allows human experts to immediately understand why a machine learning model made a certain decision.”

Researchers applied a well-trained machine learning model to obtain predictions of the properties of 8 million virtual polyimides. Then, by screening that entire data set, he identified the three best hypothetical polyimides that combined better properties than existing polyimides.

They also checked their achievements. The researchers built all-atom models and performed molecular dynamics simulations on the top three candidates to calculate important thermal properties.

“The molecular dynamics simulations were in good agreement with the predictions from the machine learning model, which gave us confidence that our predictions are very reliable,” Lee says. “Furthermore, simulations show that these new polyimides are easy to synthesize.”

As a final validation method, the team created one of their new polyimides and conducted experiments demonstrating the material’s superior heat resistance. Their experimental results showed that the new polyimide could withstand temperatures of about 1,022 degrees Fahrenheit before it began to degrade. This result matched their machine learning predictions. In contrast, existing polyimides can only withstand temperatures in the range of 392 degrees Fahrenheit to 572 degrees Fahrenheit. The researchers also created a web-based application that allows users to explore new high-performance polyimides with interactive visualizations.

Additional authors of the Chemical Engineering Journal paper include Jinlong He of the University of Wisconsin-Madison, Lei Tao of the University of Connecticut, and Nuwayo Eric Munyaneza of Virginia Tech and State Universities as first authors with equal contributions . Vikas Varshney of the Air Force Research Laboratory, Wei Chen of Northwestern University, and Guoliang Liu of Virginia Tech and State University are also authors of the paper.

This work was supported by funding from the Air Force Young Researcher Research Program, the Air Force Research Laboratory, and the Air Force Office of Scientific Research through the National Science Foundation.