Comparative study on constitutive models of a near β titanium alloy TC18 during thermoplastic deformation based on machine learning

The hot deformation behavior of TC18 alloy has been systematically studied at the temperature of 993-1113 K and strain rate of 0.001-1 s-1. Based on the stress-strain data obtained under the above process parameters, machine learning models describing the thermoplastic constitutive relationship of the alloy were established by using random forest (RF), support vector regression (SVR), back propagation artificial neural network (BPANN) and Gaussian process regression (GPR). The hyper-parameters of these models were optimized by adaptive inertial weighted particle swarm optimization (APSO) algorithm, and the models were further evaluated by statistical analysis and cross-validation. The results show that the prediction ability of the developed four machine learning models was ranked as GPR>BPANN>RF>SVR. Then APSO was applied to four models to further enhance their prediction accuracy, and the prediction accuracy of these APSO models was ranked as APSOGPR>APSO-BPANN>APSO-RF>APSO-SVR. The developed APSO-GPR model has the highest R2 (>0.999), as well as the lowest RMSE (<1.77) and MAPE (<0.63 %) on both the training set and the testing set, demonstrating its strong predictive performance. Sixteen cross-validation tests also confirm the APSO-GPR model has high prediction accuracy.