Computational Development of Polycrystalline Alloys Using Automated Importance Sampling

Increasingly stringent demands are being placed on alloys used for high temperature structural applications. For such applications, the successful development of new alloys requires them to exceed numerous property targets. Combinatorial alloy design tools to achieve this are now available. However, efficient selection of optimal alloy compositions requires automated optimisation algorithms. In this work, an approach utilising importance sampling has been used to design a new Nibased polycrystalline alloy. Neural network models have been built to describe various mechanical and thermodynamic properties for Ni-based superalloys. Minimum acceptable values for these properties were then specified, according to design criteria. The boundary of those compositions which satisfy these minimum requirements was then established by the automated importance sampling technique. On a standard desktop computer this technique can output up to ~100000 sets of compositions and design variables per second, which satisfy the design specifications.