Challenges and advances in computational materials design

Over the last ten years, new methods have emerged that have led to significant acceleration in the field of alloy development. In particular, using high-throughput calculations from the early stages of development, new digital alloy design methodologies enable the development of alloys with multiple optimised characteristics in terms of thermomechanical properties (strength, creep, etc), chemical properties (oxidation) or other properties such as density, cost, etc. The approach used is that of "theoretical combinatorial metallurgy".  It is based on models that can predict the characteristics of interest as a function of the composition. Hence, extensive use of models such as the CALPHAD thermodynamic models, but also Machine-Learning-type regression models, enables to assess the looked-for properties by exploring the full theoretical composition domain. One of the problems to deal with is the gigantic number of possible compositions: for example, with a dozen alloying elements, like in most complex industrial alloys, and around fifty possible concentration levels for each of them, the number of potential alloys is 5012, i.e. more than 2.1020 alloys!  It is therefore impossible, even by calculations, to explore systematically the entire compositional range. Reducing the composition domain explored (concentration ranges, number of elements, number of levels) to authorize brute-force calculations, or using optimization algorithms, such as genetic algorithms are the two strategic solutions. In fact, optimization algorithms enable to perform an "intelligent exploration" of composition space, gradually converging, through iterative processes, towards the Pareto front, i.e. the only areas of the composition domain that are "interesting" in terms of the defined objectives.