Machine learning-based beta transus temperature prediction for titanium alloys

• DOI: 10.1016/j.jmrt.2023.01.019
• Publication Date: 2023-01-06T17:11:33Z
• Authors: Yong Niu; Zhi-qiang Hong; Yao-qi Wang; Yan-chun Zhu
• Keywords: Titanium alloy; Root mean square

Beta transus temperature (βtr) is one of the most crucial features titanium alloys. It typically used as index while designing heat treatment process for The βtr also a significant parameter to optimize processing technology Four machine learning algorithms and empirical formula developed in this study estimate alloys: Artificial Neural Networks (ANN), Gauss Processing Regression (GPR), Super Vector Machine (SVM), Ensemble Trees (ERT). According correlation coefficient (R), Mean Absolute Error (MAE) Root Square (RMSE) verify accuracy models, experimentally measured phase transition Ti600 alloy was generalization ability model. Choosing best model analyze sensitivity elements determine how each component affects βtr. result demonstrated that ANN has highest prediction among five different structures have effects on predicting new data. with 10 neurons accuracy, 8 strongest ability. results analysis proved all compositions input parameters were valid parameters.