ADVANCED MATERIALS & PROCESSES | SEPTEMBER 2023 10 MACHINE LEARNING | AI SUPER SPEEDY POLYMER DISCOVERY Researchers at the University of Wisconsin-Madison are combining machine learning predictions with molecular dynamics simulations to dramatically speed discovery of new polyimides with excellent mechanical and thermal properties. In their study, the engineers first collected opensource data of the chemical structures of all the existing dianhydride and diamine/diisocyanate molecules, then took that data and built a library of eight million hypothetical polyimides. The team used a computer to combine these building blocks together, which allowed them to organize all possible combinations into a huge database. They then created multiple machine learning models for the thermal and mechanical properties of polyimides based on experimentally reported values. Using a variety of machine learning techniques, the researchers identified the chemical substructures that are most important for determining individual properties. Applying their machine learning models, the team obtained predictions for the properties of all eight million hypothetical polyimides. Then they screened the entire dataset and identified the three best hypothetical polyimides with combined properties superior to those of existing polyimides. They checked their work by building all-atom models for their top three candidates and conducted molecular dynamics simulations to calculate a key thermal property. “The molecular dy- namics simulations were in good agreement with the predictions from the machine learning models, so that gives us confidence that our predictions are quite reliable,” says lead researcher Ying Li. “In addition, the simulations showed that these new polyimides would be easy to synthesize.” wisc.edu. MACHINE LEARNING FOR MATERIAL MODELING Scientists from the Center for Advanced Systems Understanding (CASUS) at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) in Germany, and Sandia National Laboratories, Albuquerque, N.M., developed a machine learning-based simulation method that surpasses traditional electronic structure simulation techniques. Their Materials Learning Algorithms (MALA) software stack reportedly enables access to previously unattainable length scales. Lenz Fiedler, MALA’s key developer, explains, “MALA integrates machine learning with physics-based approaches to predict the electronic structure of materials. It employs a hybrid approach, utilizing an established machine learning method called deep learning to accurately predict local quantities, complemented by physics algorithms for computing global quantities of interest.” Machine learning is being used to discover and design new polymers. Courtesy of Xin Zou/UW-Madison. The MALA software stack takes the arrangement of atoms in space as input and generates fingerprints known as bispectrum components, which encode the spatial arrangement of atoms around a Cartesian grid point. The machine learning model in MALA is trained to predict the electronic structure based on this atomic neighborhood. A significant advantage of MALA is its machine learning model’s ability to be independent of the system size, allowing it to be trained on data from small systems and deployed at any scale. The team achieved a speedup of more than 1000 times for smaller system sizes, consisting of up to a few thousand atoms, compared to conventional algorithms. Further, the team demonstrated MALA’s capability to accurately perform electronic structure calculations at a large scale, involving more than 100,000 atoms. “The key breakthrough of MALA lies in its capability to operate on local atomic environments, enabling accurate numerical predictions that are minimally affected by system size. This groundbreaking achievement opens up computational possibilities that were once considered unattainable,” says Attila Cangi of CASUS. www.helmholtz.de/en. Deep learning simulation of more than 10,000 beryllium atoms. Courtesy of HZDR/CASUS.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==