Scholarlink

<p>Because of their intricate composition–processing–property relationships and the high experimental cost of alloy synthesis, shape memory alloys (SMAs), especially ternary Ni-Ti-Hf alloy, remain difficult to predict in terms of their behavior. Efficient materials discovery and alloy design optimization for practical uses depend on the precise classification of SMA and non-SMA compositions. This work offers an interpretable, optimization-driven machine learning (ML) framework to tackle this problem. It combines an adaptive hybrid dandelion optimizer (DETDO) and Bayesian hyperparameter tuning (Optuna) with sophisticated classifiers, such as logistic regression, XGBoost, and CatBoost. To increase generalization and reduce class bias, the dataset was balanced using SMOTEENN resampling and improved with engineered compositional and processing features. With a cross-validation F1-score of 0.9846, a test F1-score of 0.9524, and a test accuracy of 0.9167, the DETDO–XGBoost model outperformed the other models in the test, demonstrating its superior predictive stability. The main factors influencing SMA prediction, according to SHAP interpretability analysis, were Ni atomic percentage, delta_mass, and final-aging duration. This study shows that SMA functionality is defined by a combination of thermomechanical and compositional factors. The suggested framework supports useful developments in automotive, aerospace, and renewable energy applications by offering a quick, scalable, and explicable method for intelligent alloy screening.</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

AI-Enhanced Materials Selection Pipeline for High-Performance Actuators in Power and Aerospace Systems: Predicting Thermal Hysteresis in High-Temperature Shape Memory Alloys

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Xuezhe Zhang,
  • Liqing Su,
  • Cuihua Li

摘要

Because of their intricate composition–processing–property relationships and the high experimental cost of alloy synthesis, shape memory alloys (SMAs), especially ternary Ni-Ti-Hf alloy, remain difficult to predict in terms of their behavior. Efficient materials discovery and alloy design optimization for practical uses depend on the precise classification of SMA and non-SMA compositions. This work offers an interpretable, optimization-driven machine learning (ML) framework to tackle this problem. It combines an adaptive hybrid dandelion optimizer (DETDO) and Bayesian hyperparameter tuning (Optuna) with sophisticated classifiers, such as logistic regression, XGBoost, and CatBoost. To increase generalization and reduce class bias, the dataset was balanced using SMOTEENN resampling and improved with engineered compositional and processing features. With a cross-validation F1-score of 0.9846, a test F1-score of 0.9524, and a test accuracy of 0.9167, the DETDO–XGBoost model outperformed the other models in the test, demonstrating its superior predictive stability. The main factors influencing SMA prediction, according to SHAP interpretability analysis, were Ni atomic percentage, delta_mass, and final-aging duration. This study shows that SMA functionality is defined by a combination of thermomechanical and compositional factors. The suggested framework supports useful developments in automotive, aerospace, and renewable energy applications by offering a quick, scalable, and explicable method for intelligent alloy screening.