Alloy design: from composition to through process models

Abstract

New alloy classes tend to have arisen through chance discoveries or undirected experiments, with improvements then made through directed experimental variations. Improvements directed by theoretical predictions tend to follow a little later, whether by spotting the empirical trends or understanding some of the relevant physics. Artificial intelligence techniques are being explored to aid the spotting of empirical trends within large databases, while improved understanding of theory coupled with dramatic increases in computational ability is now leading to models which can predict the basic physical properties of new compounds and alloys in advance of experimental data. Further models are beginning to make reasonable predictions of actual microstructures, key to the material's mechanical properties. For established alloy systems in mass production, it is unlikely that wholly new phases are lurking undiscovered, or that many new types of microstructure still remain to be seen, and the sheer complexity of a commercial steel's composition and microstructure does not make it the initial target for the application of novel materials design tools. The electronic properties of some exotic inorganic compound never yet made anywhere on earth can be predicted with such confidence that some would ask if an experimental check is even necessary, but a robust model for the mechanical properties of a typical lump of rolled steel still lies beyond current capabilities. For many steels, the rolling history can have the dominant effect on the resultant mechanical properties. This provides scope to design out some of the plethora of steel composition variants, through appropriate processing. The improved facility for quantitative design of finished alloy– microstructure–property combinations is a target of a major modelling programme in British Steel. A variety of model types is used across the business, and a consistent framework is sought for their appropriate, multilayer linking into a ‘through process model’.