Accelerated Learning and Co-Optimization of Elastocaloric Effect and Stress Hysteresis of Elastocaloric Alloys

Shi-Yu He,Fei Xiao,Rui-Hang Hou,Shun-Gui Zuo,Ying Zhou,Xiao-Rong Cai,Zhu Li,Yan-Ming Wang,Aysu Catal-Isik,Enrique Galindo-Nava,Xue-Jun Jin

DOI: https://doi.org/10.1007/s12598-024-02827-1

IF: 6.318

2024-01-01

Rare Metals

Abstract:The development of high-performance advanced elastocaloric alloys is crucial for the implementation of elastocaloric refrigeration. Here, we present a machine learning (ML) framework to accelerate the development of novel elastocaloric alloys with large adiabatic temperature change (ΔTad) and low stress hysteresis (Δσhy). The comprehensive framework comprises database construction, feature selection, model construction, alloy design and validation, and model interpretation. Features are selected according to the physical attributes they represent. Properties that may reflect the compatibility between parent and product phases, lattice distortion and the free energy in the alloy are considered in the model. Among them, the key features are selected by recursive feature elimination and exhaustive search methods. The trained models in combination with the Bayesian optimization method are exploited to achieve multi-objective optimization. According to the results, a newly designed elastocaloric alloy shows a large adiabatic temperature change of 15.2 K and low average stress hysteresis of 70.3 MPa at room temperature, which is consistent with our predictions. The predictions of our ML model are interpreted by the Shapley additive exPlainations (SHAP) approach, which explicitly quantifies the effects of each feature in our model on the adiabatic temperature change and stress hysteresis. Additionally, we employ the Sure-independence screening and sparsifying operator (SISSO) method in conjunction with the key features to formulate explicit model.