Generative Inversion for Property-Targeted Materials Design: Application to Shape Memory Alloys

The paper addresses a challenge in designing shape memory alloys (SMAs): achieving both high transformation temperatures and large mechanical work output. The authors propose a data-driven inverse design method using a generative adversarial network (GAN). They combine a pretrained GAN with a property prediction model and then optimize in the GAN’s latent space to generate alloy compositions and processing parameters that match user-specified property targets. They report experimental validation by synthesizing and characterizing five NiTi-based SMAs. One composition, Ni49.8Ti26.4Hf18.6Zr5.2, is reported to reach a transformation temperature of 404 °C, mechanical work output of 9.9 J/cm³, transformation enthalpy of 43 J/g, and thermal hysteresis of 29 °C, outperforming existing NiTi alloys. The improved performance is attributed to transformation volume change and microstructural features involving Ti2Ni-type precipitates, with effects linked to sluggish Zr/Hf diffusion and semi-coherent interfaces with localized strain fields.