March 31, 2023 (Nanowerk News) Advanced materials are urgently needed in everyday life, be it in high technology, mobility, infrastructure, green electricity or medicine. However, conventional methods for discovering and researching new materials are limited by the complexity of chemical compositions, structures and desired properties. Furthermore, new materials should not only enable novel applications, but also include sustainable ways of making, using and recycling them. Researchers at the Max Planck Institute for Iron Research (MPIE) review the state of physics-based modeling and discuss how combining these approaches with artificial intelligence can open up previously untapped spaces for the design of complex materials. They published their perspective in the journal Nature Computational Science (“Accelerating the design of Compositionally Complex Materials via Physics-informed Artificial Intelligence”). Combining physics-based approaches with artificial intelligence To meet technological and environmental challenges, increasingly sophisticated and diverse material properties must be considered, making alloys more complex in terms of composition, synthesis, processing and recycling. Changes in these parameters lead to changes in their microstructure, which directly affect the material properties. This complexity must be understood in order to be able to predict the structures and properties of materials. Computational materials design approaches play a crucial role here. “Today, our options for designing new materials are based exclusively on physically based simulations and experiments. This approach can encounter limitations when it comes to quantitatively predicting high-dimensional phase equilibria and in particular the resulting non-equilibrium microstructures and properties. In addition, many microstructural and property models use simplified approximations and rely on a large number of variables. However, the question remains whether and how these degrees of freedom are still able to cover the complexity of the material,” explains Professor Dierk Raabe, Director at the MPIE and first author of the publication. The paper compares physics-based simulations such as molecular dynamics and ab initio simulations with descriptor-based modeling and advanced artificial intelligence approaches. While physics-based simulations are often too expensive to predict materials with complex compositions, using artificial intelligence (AI) has several advantages. “AI is able to automatically extract thermodynamic and microstructural features from large data sets obtained from electronic, atomistic and continuum simulations with high predictive power,” says Professor Jörg Neugebauer, Director at the MPIE and co-author of the publication. Improving Machine Learning with Large Datasets Since the predictive power of artificial intelligence depends on the availability of large datasets, ways to overcome this obstacle must be found. One way is to use active learning cycles, where machine learning models are trained with initially small subsets of labeled data. The model’s predictions are then checked by a labeling engine, which feeds high-quality data back into the pool of labeled records, and the machine learning model is run again. This step-by-step approach results in a high-quality final dataset that can be used for accurate predictions. There are still many unanswered questions for the use of artificial intelligence in materials science: How to deal with sparse and noisy data? How to account for interesting outliers or “missfits”? How to implement unwanted intrusion of elements from synthesis or recycling? However, when it comes to the design of complex composition alloys, artificial intelligence will play a more important role in the near future, especially in the development of algorithms and the availability of high-quality material data sets and high-performance computing resources.
