Research On The Elasticaloric Properties Of TiNi-based Shape Memory Alloys Based On Machine Learning

TiNi shape memory alloys are considered to be one of the most promising materials in the engineering application of elastocaloric refrigeration due to its excellent superelastic properties and good machining properties.How to further develop higher-performance elastocaloric materials of TiNi-based alloys has become a key issue in elastocaloric refrigeration research.In recent years,developing high-performance elastocaloric materials of TiNi-based alloy with the experimental main method of trial-error,but its long cycle and high cost greatly limit the development of TiNi-based alloys.As the highly development of computing power,machine learning has been gradually applied to the field of material design and has achieved great success.In this thesis,we propose a method for predicting the elastocaloric properties(including transformation temperature and adiabatic temperature change)of TiNi-based alloys by machine learning combined with highly relevant physical characteristics,and respectively reveal the alloy composition,properties and energy difference(?E_A-M)for transformation temperature modeling prediction and the influence of alloy composition,properties and crystal cell volume change(?V)effect on the adiabatic temperature change modeling prediction.The machine learning model of TiNi-based alloys transformation temperature and adiabatic temperature change is change is built successfully,which brings a new way and notion for the design of TiNi-based alloy elastocaloric materials.Calculating the lattice parameters and?E_A-M data of the austenite and martensite phases of TiNi-based alloys by the first-principles method,and the quantitative relationship between the TiNi-based alloys composition and the?E_A-M and lattice parameters is established by linear regression.Through the prediction equation,the lattice parameters and?E_A-M of a large number of unknown systems of TiNi-based alloys can be quickly predicted,which provides a physical basis for feature selection for subsequent machine learning modeling to predict transformation temperature and adiabatic temperature change.According to the transformation temperature data of TiNi-based alloys,the composition,properties and?E_A-M of the alloy were obtained as the basic characteristic dataset by combining the prediction equation of composition and?E_A-M.Through feature selection and XGBoost machine learning,the correlation of determination(R~2)is 0.934 on the test set.Root mean squared error(RMSE)is 9.846,and mean absolute error(MAE)is 7.490.The results show that the XGBoost model can efficiently predict the transformation temperature of TiNi-based alloys,and provide valuable reference for its application in the field of elasticaloric refrigeration.Based on the adiabatic temperature change data of TiNi-based alloys and the prediction equation of composition and lattice parameters,the?V of TiNi-based alloys was further calculated according to the calculation results,and the basic data set was established to characterize the composition,properties of the alloy and the relationship between the?V and adiabatic temperature change.Feature dimensionality reduction was applied to evaluate the prediction ability of different algorithms,and the decision tree regression(DTR)training model was used to obtain R~2?0.9(MAE and RMSE were 1.109 and 1.413,respectively)on the test set.The outcome shows that the model can predict the adiabatic temperature change of TiNi-based alloys,and provide reference for the study of elasticaloric properties of TiNi-based alloys.The above research shows that the elastocaloric properties of TiNi-based alloys can be successfully modeled and predicted by using machine learning methods with highly correlated physical characteristics,which is useful to figure out the influencing factors of the elastocaloric properties of TiNi-based alloys and provides strategies for the design of other new elastocaloric materials.