Scholarlink

<p>This study develops and compares machine learning methods to classify pore type and predict porosity characteristics in laser powder bed fused Al–10Si–Mg from computed tomography data, with the goal of supporting process-parameter optimization and fatigue-relevant assessment. Features describing pore size, shape, and morphology were extracted from computed tomography scans and used to train and compare a range of classification and regression models, including conventional, deep learning, and ensemble approaches. The models were used to distinguish lack-of-fusion from keyhole pores and to predict porosity percentage, pore size, and pore morphology. Ensemble models performed best overall, where Gradient Boosting achieved 97% accuracy for pore classification, and Random Forest and Gradient Boosting yielded the strongest porosity predictions with a mean absolute error of 2.02 and 2.03 for porosity percentage. The resulting models enable mapping of porosity metrics and fatigue-relevant defect characteristics across the additive manufacturing process window. A framework was also established to estimate fatigue limit trends by predicting process-parameter-sensitive extreme value distribution parameters and combining them with fracture mechanics concepts. The proposed computed-tomography-based machine learning framework provides an effective route for linking process parameters to pore classification, porosity evolution, and fatigue-critical defect metrics, thereby supporting selection of additive manufacturing conditions that improve structural integrity.</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

A machine learning methodology for porosity classification and prediction from computed tomography data and its application to processing-properties optimization in additive manufacturing

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Mehrnoush Alizade,
  • Akshatha C. Dixith,
  • Anthony G. Spangenberger,
  • Rodica Neamtu,
  • Diana A. Lados

摘要

This study develops and compares machine learning methods to classify pore type and predict porosity characteristics in laser powder bed fused Al–10Si–Mg from computed tomography data, with the goal of supporting process-parameter optimization and fatigue-relevant assessment. Features describing pore size, shape, and morphology were extracted from computed tomography scans and used to train and compare a range of classification and regression models, including conventional, deep learning, and ensemble approaches. The models were used to distinguish lack-of-fusion from keyhole pores and to predict porosity percentage, pore size, and pore morphology. Ensemble models performed best overall, where Gradient Boosting achieved 97% accuracy for pore classification, and Random Forest and Gradient Boosting yielded the strongest porosity predictions with a mean absolute error of 2.02 and 2.03 for porosity percentage. The resulting models enable mapping of porosity metrics and fatigue-relevant defect characteristics across the additive manufacturing process window. A framework was also established to estimate fatigue limit trends by predicting process-parameter-sensitive extreme value distribution parameters and combining them with fracture mechanics concepts. The proposed computed-tomography-based machine learning framework provides an effective route for linking process parameters to pore classification, porosity evolution, and fatigue-critical defect metrics, thereby supporting selection of additive manufacturing conditions that improve structural integrity.