Characterization of Two Novel NiTiHf Shape Memory Alloys Designed by Machine Learning Utilizing Novel Experimental Techniques

Abstract

This paper details the experimental characterization techniques utilized to establish the shape memory characteristics of two new NiTiHf shape memory alloys (SMAs) designed by machine learning (ML). Specifically, a multi-layer feed-forward neural network (MLFFNN) framework was developed with the aim of finding the optimum NiTiHf shape memory alloy (SMA) composition that exhibits an austenite finish temperature (Af) beyond 400 degrees C with a stable reversible phase transformation behavior. The predicted Ni49.7Ti26.6Hf23.7 and Ni50Ti27Hf23 alloys, with respective predicted Af values of 424 and 401 degrees C, were cast by vacuum arc melting (VAM), and the validation experiments revealed that the Ni49.7Ti26.6Hf23.7 SMA exhibited a 404 degrees C Af. In order to overcome the difficulties associated with the small size of the samples manufactured by VAM, further experiments utilizing a Vickers indenter adopting a heated plate and a 3D laser scanning microscope were carried out, demonstrating that both Ni49.7Ti26.6Hf23.7 and Ni50Ti27Hf23 SMAs exhibited a stable and reversible martensitic phase transformation. Overall, the combined ML-based alloy design and experimental validation effort presented herein opens a venue for exploiting new alloy systems to address challenging materials problems in a timely and cost-efficient manner.