Scholarlink

<p>Shape Memory Alloys (SMA) are smart materials with widespread applications in the fields of aeronautics, biomedicine, automobiles, and robotics due to their special properties such as the shape memory effect, superelasticity, and high damping capacity. Machine learning (ML) provides an efficient and reliable alternative to conventional experimental approaches for predicting transformation behavior in SMAs. This research work is intended to design and validate a precise PIML models for predicting the martensitic start temperature (Ms) and thermal hysteresis (TH) of Cu–Al–Ni SMAs, thereby reducing experimental cost, time, and material consumption associated with traditional trial-and-error alloy development. Literature-derived compositional data and physics descriptors were used to build and validate the four boosting-based ML models such as Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), and Categorical Boosting (CatBoost). Model performance was enhanced through systematic hyper-parameter tuning using grid search, and robustness was evaluated via 5-fold cross-validation. Among the models, XGBoost exhibited the best predictive performance with R²= 0.967,0.94 for Ms &amp; TH, accompanied by low error metrics (RMSE (10.53, 2.6), MAE (5.12, 1.77), and MSE (112.48, 7.04)) and low average error percentages of 4.6% and 3.01%, respectively. Cross-validation and residual analyses confirmed the superior stability and generalization capability of XGBoost without evidence of overfitting. SHAP analysis showed that aluminium content and lattice distortion are the key factors controlling Ms and TH. XGBoost accurately predicts Ms and TH while reflecting the underlying physical mechanisms, providing valuable guidance for designing the Cu–Al–Ni alloys.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physics-informed boosting-based machine learning prediction of martensitic start temperature and thermal hysteresis in Cu–Al–Ni shape memory alloy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月19日

查看全文
  • Saravanamuthukumar Ponraj,
  • Ahmad Baharuddin Abdullah

摘要

Shape Memory Alloys (SMA) are smart materials with widespread applications in the fields of aeronautics, biomedicine, automobiles, and robotics due to their special properties such as the shape memory effect, superelasticity, and high damping capacity. Machine learning (ML) provides an efficient and reliable alternative to conventional experimental approaches for predicting transformation behavior in SMAs. This research work is intended to design and validate a precise PIML models for predicting the martensitic start temperature (Ms) and thermal hysteresis (TH) of Cu–Al–Ni SMAs, thereby reducing experimental cost, time, and material consumption associated with traditional trial-and-error alloy development. Literature-derived compositional data and physics descriptors were used to build and validate the four boosting-based ML models such as Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), and Categorical Boosting (CatBoost). Model performance was enhanced through systematic hyper-parameter tuning using grid search, and robustness was evaluated via 5-fold cross-validation. Among the models, XGBoost exhibited the best predictive performance with R²= 0.967,0.94 for Ms & TH, accompanied by low error metrics (RMSE (10.53, 2.6), MAE (5.12, 1.77), and MSE (112.48, 7.04)) and low average error percentages of 4.6% and 3.01%, respectively. Cross-validation and residual analyses confirmed the superior stability and generalization capability of XGBoost without evidence of overfitting. SHAP analysis showed that aluminium content and lattice distortion are the key factors controlling Ms and TH. XGBoost accurately predicts Ms and TH while reflecting the underlying physical mechanisms, providing valuable guidance for designing the Cu–Al–Ni alloys.

Graphical Abstract