Machine Learning Modeling for Predicting Austenite Transformation Temperature and the Latent Heat in CuAl-Based Shape Memory Alloys

Solid-state cooling technologies are gaining increasing attention due to their environmentally friendly characteristics. Among them, elastocaloric cooling, where shape memory alloys (SMAs) are used as the refrigerant, is one of the promising caloric technologies. Through harvesting the latent heat from the reversible stress-induced phase transformation between martensite and austenite under adiabatic loading conditions, the material undergoes temperature changes. While most of the prototypes are designed based on NiTi due to its availability, their high cost and significant work input pose challenges for commercialization. Cu-based SMAs is a competitive alternative compared to NiTi-based SMAs for its lower cost, lower hysteresis, and reduced critical transformation stress characteristics. The performance of these materials is largely dependent on their austenite transformation temperature (A_f), which determines the operational temperature range, and their latent heat, which directly determines the amount of cooling the material can generate during the phase transformation. To accelerate the discovery of optimal SMAs compositions without extensive experimental synthesis, we developed a machine learning model to predict A_f and the latent heat in CuAl-based SMAs. Using 149 experimental data points, composition-related features were generated with Matminer to train the model. The model achieved an R² of 0.90 for A_f and 0.64 for the latent heat. The strong A_f prediction indicates the model's effectiveness, while the lower performance for latent heat is likely due to the absence of key microstructural information, such as lattice constant and grain size, which significantly influence the latent heat. By providing accurate predictions for A_f and reasonable estimates for the latent heat, this model can guide the design of materials, reducing the need for exhaustive synthesis and enabling the development of CuAl-based SMAs with optimized properties for elastocaloric cooling systems.