Application of Machine Learning and Generative Design Strategies for the Design and Optimization of Alloys

Abstract

As technological advancements accelerate, the demand for new materials becomes increasingly imperative, particularly within industries such as aviation, medical, advanced manufacturing and other high-end applications. The development and exploration of novel materials have emerged as an indispensable research topic. The quest for advanced materials, tailored to meet the rigorous demands of modern industries, naturally extends to the realm of alloys, which have long played a crucial role in various industrial applications. In recent years, high-entropy alloys (HEAs), also generally referred to as multi-principal element alloys (MPEAs), composed of five or more principal elements in near-equal proportions, have attracted significant research interest. These alloys benefit from their unique configurational entropy, leading to simple solid-solution phases and outstanding mechanical properties. Given their remarkable strength, hardness, and enhanced corrosion resistance, HEAs hold promise for advanced applications in sectors such as aerospace, automotive, and advanced manufacturing.

The advent of the Material Genome Initiative (MGI) holds substantial potential to revolutionise the field of materials science. This initiative aims to expedite the development and commercialization of new materials, fuelling advancements in various engineering and scientific applications. Computational techniques, such as density functional theory (DFT) and high-throughput screening (HTS), have become prevalent in recent materials explorations, notably in alloy development. These methodologies facilitate simultaneous computations of properties across an extensive array of materials. Furthermore, the infusion of machine learning (ML) in this domain is gaining momentum. While traditional simulation and optimization algorithms often demand substantial computational resources and time, machine learning, through its ability to learn from vast amounts of data, presents a more resource-efficient alternative.

This study presents an innovative integration of machine learning and traditional optimization algorithms, aimed at the systematic investigation of high-performance novel materials, with a primary focus on alloys. This research delves into several facets of applying machine learning in the domain of material science, encompassing feature analysis, property prediction, optimization, and generative design. Additionally, a systematic generative design metaheuristic integrating machine learning with multi-objective optimization algorithms was introduced. Recognizing the depth of existing work in this domain, our methodology provides an alternative lens, focusing on the potential of generative models and data-driven strategies. Such an approach may offer insights beneficial for broader applications in materials science. Show more