From the design of superalloys for making turbine blades that can withstand higher temperatures, to the development of light yet resistant alloys that reduce the weight of vehicles, to the use of high-entropy alloys to make efficient catalysts, advanced metallurgy can bring substantial contributions to alleviate the energy crisis and support the transition to a sustainable development model.
Atomistic modeling can provide mechanistic insights and improved design principles, but it is limited by the complexity of modern alloys that involve up to a dozen carefully tuned components.
We recently proposed a possible solution to deal with the staggering chemical complexity, developing a machine-learning model that can accurately describe alloys containing up to 25 d-block transition metals, with very promising accuracy and transferability.
In this project, we propose to extend our model to disordered and inhomogeneous structures and create a machine learning potential covering all metallic and semi-metallic elements.
The two main outputs of this project will be a large dataset of structure with metallic elements that anyone can use to train machine learning models for advanced metallurgy, and an improved version of our model able to describe a large portion of the periodic table while incorporating surfaces and extended defects.
École Polytechnique Fédérale de Lausanne (EPFL), Switzerland