Computational alloy design by artificial intelligence and thermodynamics

This article describes a generic method for computational alloy design, which offers gains in time, cost and performance over trial-and-error development methods. The method is based, on the one hand, on artificial intelligence methods such as combinatorial multi-objective optimization (genetic algorithms, etc.) and data mining (artificial neural networks, Gaussian processes, etc.) and, on the other, on predictive thermodynamics using the so-called "Calphad" method (for "CALculation of PHAse Diagrams" in English). These are the main tools used today to embrace the complexity of compositional spaces in modern metallic materials, whether to establish links between composition, processes and properties, or to search for compositions that will lead to optimal combinations of characteristics.

The method potentially concerns the entire metallurgical industry (steels, nickel alloys, aluminum, titanium, etc.), and all fields requiring the use of high-performance metal alloys: transport (automotive, aeronautics, space, rail, naval, etc.), energy (nuclear, gas, oil, etc.), chemical engineering (reactors, etc.), biomedical (prostheses, implants, staples, etc.), etc. The overall context for the development of new alloys is first set out from a threefold historical, combinatorial and complexity perspective. After presenting some of the technical features of the tools used, the method is illustrated using examples of complex alloy design, such as nickel-based superalloys or multi-concentrated "high-entropy" alloys. We then show how the method can be exploited to design materials with environmental considerations in mind (eco-design). Finally, we outline a number of prospects linked to current developments in the method.