Even the closest ties between relatives can get strained in a family business environment. Similarly, different metals don’t always work well together in the context of alloys. As a result, finding the perfect blend of metals for specific engineering applications has been an arduous undertaking for materials scientists.
Complex concentrated alloys, or CCAs, are among the trickiest to formulate— the vast possibilities in their melting temperature and chemical composition make optimisation significantly more complicated compared to simpler high entropy alloys (HEAs).
“CCAs typically incorporate a larger variety of elements and more intricate compositions with varying element ratios compared to HEAs,” explained Kewu Bai, a Senior Scientist at A*STAR’s Institute of High Performance Computing (IHPC), adding that CCAs also exhibit multiple phases and involve more complicated manufacturing processes.
A key part of the alloy manufacturer’s design toolkit is CALPHAD (Computer Coupling of Phase Diagrams and Thermochemistry), a thermodynamic modelling technique that helps them to predict how different CCA compositions would behave. “However, the method is time-consuming and may not explore all possible combinations when looking for the best CCA composition,” said Bai.
An alternative CCA-design approach using machine learning (ML) boasts speed and efficiency, but isn’t always practical as it requires large, high-quality datasets to generate accurate predictions.
In response, Bai and colleagues developed a best-of-both-worlds approach to CCA optimisation. Merging both CALPHAD and ML, they started with a small thermodynamic dataset and used reinforcement learning (an ML-training strategy that uses trial and error) to iteratively expand the capabilities of their new platform.
The researchers leveraged the novel CALPHAD/ML platform to predict and assess alloy structures in 20 different CCA families that contained aluminium. Their CCA predictions were found to line up well with the results of real-world testing, boasting over 92 percent accuracy.
“The synergy of CALPHAD and ML thus offers a more comprehensive and efficient approach to the design and optimisation of CCAs compared to using each method individually,” concluded Bai, who added that the team also identified two CCA compositions (AlCoCrFeNi and AlCrFeMnNi) with promising structures and properties for engineering applications.
Next up, the researchers are using a similar computational tactic to tackle the design of eutectic HEAs, alloys featuring unique microstructures with potentially useful material properties, as well as pursuing other rare and powerful metal combinations to advance the world of metallurgy.
The A*STAR-affiliated researchers contributing to this research are from the Institute of High Performance Computing (IHPC).
