Scholarlink

<p>This study presents a comprehensive machine-learning (ML)-based surrogate modeling framework to predict multiple, coupled thermo-mechanical responses of laser powder bed fusion additively manufactured (L-PBF-AM) NiTi shape memory alloys (SMAs). Four supervised ML algorithms, Linear Regression (LR), Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM), were systematically trained and benchmarked using curated experimental data. The input features included feedstock chemical composition, L-PBF processing parameters, and loading mode/test temperature, while the target outputs encompassed relative density, phase transformation temperatures, ultimate tensile/compressive strength, elongation strain, total superelastic/shape memory strain, and recovery ratio. Comparative evaluation revealed that RF achieved the best overall predictive accuracy, delivering high coefficients of determination (R²) and low mean absolute errors (MAE) across most property categories. However, ANN outperformed RF in predicting transformation temperatures, highlighting its enhanced sensitivity to nonlinear phase-transformation phenomena and suggesting the complementary potential of hybrid ML schemes. Model validation using independent datasets excluded from training confirmed strong generalization performance, with RF exhibiting less than 15% prediction error across most mechanical and functional properties, while ANN achieved comparable accuracy for transformation-temperature predictions. The proposed surrogate modeling framework demonstrates robust predictive capability under unseen conditions and positions RF and ANN as synergistic data-driven tools for capturing process-structure-property linkages in L-PBF-AM NiTi systems. This approach enables rapid virtual screening and process optimization, providing a practical foundation for accelerated design and property tailoring of NiTi-based SMAs and related metallic materials.</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

Machine-learning framework for predicting process-property relationships in additively manufactured NiTi shape memory alloys

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Sayed Ehsan Saghaian,
  • Milad Hemmati,
  • Syed Mahedi Hasan,
  • Othmane Benafan

摘要

This study presents a comprehensive machine-learning (ML)-based surrogate modeling framework to predict multiple, coupled thermo-mechanical responses of laser powder bed fusion additively manufactured (L-PBF-AM) NiTi shape memory alloys (SMAs). Four supervised ML algorithms, Linear Regression (LR), Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM), were systematically trained and benchmarked using curated experimental data. The input features included feedstock chemical composition, L-PBF processing parameters, and loading mode/test temperature, while the target outputs encompassed relative density, phase transformation temperatures, ultimate tensile/compressive strength, elongation strain, total superelastic/shape memory strain, and recovery ratio. Comparative evaluation revealed that RF achieved the best overall predictive accuracy, delivering high coefficients of determination (R²) and low mean absolute errors (MAE) across most property categories. However, ANN outperformed RF in predicting transformation temperatures, highlighting its enhanced sensitivity to nonlinear phase-transformation phenomena and suggesting the complementary potential of hybrid ML schemes. Model validation using independent datasets excluded from training confirmed strong generalization performance, with RF exhibiting less than 15% prediction error across most mechanical and functional properties, while ANN achieved comparable accuracy for transformation-temperature predictions. The proposed surrogate modeling framework demonstrates robust predictive capability under unseen conditions and positions RF and ANN as synergistic data-driven tools for capturing process-structure-property linkages in L-PBF-AM NiTi systems. This approach enables rapid virtual screening and process optimization, providing a practical foundation for accelerated design and property tailoring of NiTi-based SMAs and related metallic materials.