From a pool of 8 million potential candidates, mechanical engineers at the University of Wisconsin-Madison have swiftly identified a number of promising high-performance polymers.
These polymers, known as polyimides, are used in a wide range of applications in the aerospace, automotive, and electronics sectors because they have superior mechanical and thermal characteristics, such as strength, stiffness, and heat resistance.
Because designing polyimides is expensive and time-consuming, there are now just a few available.
The UW-Madison engineers, however, use their data-driven design framework to accelerate the identification of novel polyimides with even superior attributes by utilizing molecular dynamics simulations and machine learning predictions.
According to the research’s principal investigator, Ying Li, an associate professor of mechanical engineering at the University of Wisconsin-Madison, their findings have significant ramifications for the area of materials science and will spur additional study on the development of sophisticated data-driven methodologies for materials discovery. Their design approach is significantly more effective than the traditional trial-and-error method and is also applicable to the molecular design of other polymeric materials.
Diamine/diisocyanate and dianhydride molecules react by condensation to form polyimides. The engineers initially gathered open-source information on the chemical compositions of all known dianhydride and diamine/diisocyanate molecules for their investigation, and then they used that information to create an extensive database of 8 million fictitious polyimides.
Li compares it to constructing something out of LEGOs. Several various dianhydride and diamine/diisocyanate molecules are the fundamental building blocks that you have. Additionally, you might try to manually construct each of the potential structures, but it would take a very long time due to how numerous the possible combinations are.
As a result, Li and his colleagues placed the building blocks together using a computer, which allowed them to compile all potential combinations into a sizable database.
Using a database and experimentally reported results, the researchers developed a number of machine learning models for the thermal and mechanical characteristics of polyimides. The researchers identified chemical substructures that are most crucial for defining individual attributes using a range of machine learning techniques.
Our machine learning model isn’t a black box since they used methodologies that, in essence, describe how it operates, claims Li. They have created a transparent box that enables subject-matter experts to instantly see the reasoning behind a particular choice made by the machine learning model.
The researchers were able to anticipate the characteristics of the 8 million hypothetical polyimides using their well-trained machine learning models. Then, after screening the entire dataset, the three best hypothetical polyimides were found, whose combined attributes were superior to those of existing polyimides.
The researchers created all-atom models for their top three possibilities and used molecular dynamics simulations to determine an important thermal parameter. They also double-checked their results.
They have confidence that their forecasts are quite accurate because the molecular dynamics simulations and the predictions from the machine learning models agreed well, according to Li. Additionally, the simulations demonstrated how simple it would be to create these novel polyimides.
The scientists created one of the novel polyimides as a final confirmation procedure, and studies showed the material to have outstanding heat resistance. In line with their machine learning predictions, their experimental findings revealed that the new polyimide could endure temperatures of up to 1,022 degrees Fahrenheit before it began to degrade.
Existing polyimides, however, could only withstand temperatures between 392 and 572 degrees Fahrenheit. The researchers also developed a web-based programme that enables users to interactively visualize the new, high-performing polyimides.
