Advancing Machine Learning Interpretability in Materials Science Through Algorithm Frameworks and High-Throughput Microscopy

Research question

How can we use microstructure to establish explainable machine learning (ML) for engineering materials? In this project, we will start from realising high throughput microscopy enabled by ML, and then design explainable ML frameworks to unlock the processing-microstructure-property relationship in materials. We will use recycled aluminum alloys to showcase the developed methodologies. We expect that the high throughput experimental and microscopy workflows and ML frameworks developed in this project to be applied to a wide spectrum of recycled metals, enabling quick determination of the composition and processing window with optimal properties and highest possible recycled content, thus address the challenges of data-driven materials science for sustainability.

Sustainability aspects

Aluminium alloys play a critical role in the green transition. Thanks to their recyclability, 75% of all aluminium ever produced is still on the market today. Recycled aluminium is projected to meet 50% of the total aluminium demand in 2050. Using recycled aluminium significantly enhances sustainability by saving nearly 95% of the energy compared to using primary aluminium extracted from minerals. However, due to the impurities accumulated in recycling processes, recycled aluminium alloys often have a chemical composition that falls outside the standard range of conventional aluminium alloy grades. Consequently, these recycled aluminium alloys are often blended with primary aluminium alloys to achieve the specified nominal composition or redirected to less demanding and lower-value applications, a process known as downcycling. This project will leverage data-driven methods to mitigate future downcycling of aluminium alloys.