Materials Informatics Approaches Towards the Acceleration of Hard Metal Development

High entropy alloys (HEAs) are a relatively new class of alloys comprising multiple principal elements in either equimolar ratios or with elemental concentrations between 5 and 35 at.%. These alloys exhibit exceptional properties, making them a promising alternative to conventional WC-Co cemented carbides, which have dominated metal tooling applications for nearly a century. HEAs offer the potential to meet the design criteria of metal tooling parts, including wear resistance, ductility, and compatibility with additive manufacturing (AM). However, the vast compositional space occupied by HEAs renders trial-and-error experimental approaches impractical due to time and cost constraints. Thus, machine learning (ML) has emerged as a powerful computational tool, capable of accelerating the alloy design process by predicting microstructural and mechanical properties across unexplored regions of compositional space based on existing experimental data. Herein, ML models were developed and trained to explore the HEA compositional space and identify candidate compositions with properties suitable for tooling applications. These ML-guided predictions were complemented by thermodynamic analysis using the calculation of phase diagrams (CALPHAD) method to refine alloy selection. Subsequently, nine novel alloy compositions were experimentally characterised, incorporating carbon reinforcement to enhance their microstructural and mechanical properties. While ML predictions did not perfectly align with experimental results, the combined ML and CALPHAD approach successfully accelerated alloy discovery, identifying several promising HEA compositions for tooling applications. Building on this success, a second iteration of the ML model was proposed, incorporating an expanded HEA database and an improved construction pipeline. However, the anticipated performance enhancements were not realised, highlighting key challenges and future opportunities for ML in the HEA field. This work underscores the critical need for significantly larger and more diverse experimental HEA datasets, along with standardised data reporting practices. Addressing these limitations will enable ML to realise its full potential in exploring uncharted compositional spaces and advancing HEA development for demanding industrial applications