Finding best recipe for Magnesium alloys

In recent decades, industries such as automotives and electronics have seen a growing need for new alloys that are both lightweight and strong. Magnesium (Mg) stands out for being the lightest structural metal, and alloys made with it have a diverse range of applications, but their lack of strength and ductility limits their widespread use.

The properties of Mg alloys can be enhanced in many ways, like adding new elements (especially rare earths), but determining precisely which approaches and formulations give the desired properties requires dedicated and lengthy experiments. The research is long and resource-intensive, which causes the final products to be costly.

Sandeep Jain (now at Yeungnam University), with collaborators from Sungkyunkwan University in Korea and Indian Institute of Technology Delhi, developed a computational approach to overcome this bottleneck. “The innovation in this work is its integration of machine learning with traditional materials science to predict alloy behaviour under different thermomechanical processes,” says Jain, the first author of the paper published in Science and Technology of Advanced Materials.

The team’s approach was to test six different machine learning models, each of which had a different strength for data modelling, to predict mechanical properties under diverse processing conditions.

The researchers used a pre-determined set of evaluation parameters to see which model best addressed their goals.

“Instead of relying solely on lengthy experimental testing, our study used data-driven models to capture complex relationships between processing conditions and mechanical properties,” explains Jain.

The researchers’ goal was to evaluate if any of the models could use data to accurately predict three mechanical properties: ultimate tensile strength, yield strength, and elongation. They selected 389 data entries encompassing variations in alloys under different thermomechanical processing parameters, and then ran the six models on this database (using a validation-and-testing approach). The best performer, K-Nearest Neighbors (KNN), was then used to predict the mechanical behaviour of five alloys whose properties are well-characterised in literature. The predictions from the KNN model were more than a 95% match for the known data for all the three parameters.

“It’s interesting that the KNN model outperformed more commonly used machine learning approaches, delivering highly accurate predictions of strength and ductility,” says Jain. “This shows that even relatively simple algorithms, when applied carefully to well-curated data, can provide powerful insights for alloy design.”

Read the paper

Science and Technology of Advanced Materials: https://www.tandfonline.com/doi/full/10.1080/14686996.2025.2449811 

Further information

Dr Sandeep Jain 
[email protected]
Yeungnam University

STAM Inquiries 
[email protected] 
STAM Editorial Office

Download this story

More Asia Research News 2026 stories