AI-Driven Iterative Design of Shape Memory Alloys for Elastocaloric Refrigeration

The development of efficient, high gravimetric density elastocaloric refrigeration technology necessitates the discovery of elastocaloric materials optimized to achieve maximal caloric effects. Shape memory alloys (SMAs) which possess low transformation hysteresis, high enthalpy, and resistance to functional fatigue are desirable for elastocalorics. In this work, we utilized an AI-guided iterative material design approach, employing multi-objective Batch Bayesian Optimization integrated with Deep Gaussian Process regression models, to identify SMAs optimal for room-temperature elastocaloric refrigeration (-20°C ≤ Af ≤ 50°C). Our design criteria targeted minimal temperature hysteresis, maximal ratio of transformation enthalpy to transformation strain, and maximal ratio of transformation enthalpy to the average transformation temperature. Selected alloy compositions were experimentally synthesized through arc melting, followed by solution heat treatments and thermomechanical processing, in particular hot rolling. Martensitic transformation behaviors were characterized via differential scanning calorimetry (DSC) and isobaric thermal cycling (i.e. actuation loading) under tension and compression conditions. The iterations performed resulted in a set of alloys which exhibit optimized properties based on the established objectives, demonstrating the effectiveness of this AI-driven framework in accelerating the discovery of SMAs tailored to elastocaloric refrigeration.