Noise-aware machine learning accelerates development of wide-hysteresis NiTi-based shape memory alloys

Abstract

NiTi-based shape memory alloys (SMAs) with wide thermal hysteresis show great potential in engineering applications such as pipe couplings. However, traditional trial-and-error methods are costly and time-consuming, hindering the development of wide-hysteresis alloys. Although machine learning enables efficient exploration of NiTi-SMA compositions, most studies overlook experimental noise. To address this, we propose a noise-aware Kriging model that achieves high predictive accuracy (R2 = 0.91, RMSE = 6.02) for rapidly screening alloys with wide hysteresis. Using this approach, we designed novel NiTiNbTa alloys tailored to specific processing and storage requirements. Among them, Ni49.5Ti44.5Nb4.5Ta1.5 and Ni49.5Ti44.5Nb5.5Ta0.5 can be directly processed after low-temperature storage, while Ni49Ti45.5Nb4Ta1.5 remains stable under ambient conditions. All three exhibit a thermal hysteresis over 70 K without post-processing and a shape memory recovery rate above 90% under 400–900 MPa stress. This work offers a valuable strategy for designing high-performance wide-hysteresis SMAs.