The Use of Machine Learning in Creating Novel Fe-SMA's

The high cost of Nitinol (Ni-Ti) motivates the exploration of cost-effective iron-based shape memory alloys (Fe-SMAs). This research introduces a novel method for accelerating materials discovery by utilizing an AI Language Model (LLM) trained on published papers to hypothesize new Fe-SMA compositions capable of pseudoelasticity. From the AI-generated list of compositions, a specific Fe-Mn-Al-Si-Ni-C alloy was selected and modified for experimental validation to be compared with the current researched Fe-SMA.

The alloy was subjected to extensive thermomechanical processing, including hot rolling, multi-pass cold drawing with intermediate annealing in controlled atmospheres, and various final heat treatments aimed at achieving abnormal grain growth (AGG). Multiple processing routes were explored to optimize the microstructure and functional properties.

This study revealed significant challenges in achieving the superelastic Fe-based alloys on an industrial scale. This study suggests that while the AI-hypothesized composition has potential, its practical and scalable manufacturing presents considerable hurdles with conventional equipment.