Scholarlink

<p>Laser powder bed fusion (LPBF) Ti-6Al-4&#xa0;V parts exhibit variable fatigue strength due to inherent processing defects such as high residual stress, rough surface finish, subsurface and internal pores, and a brittle metastable microstructure. Traditional models often fail to predict this fatigue strength accurately due to the complexities involved. This study employs machine learning (ML) to analyze the sensitivity of various input features on fatigue strength predictions for LPBF-produced Ti-6Al-4&#xa0;V parts. Through feature sensitivity analysis and testing of different feature selection methods, critical predictors such as surface roughness, stress intensity factor (Δ<i>K</i><sub>th</sub>), hot isostatic pressure (HIP) and elongation at break were identified. Utilizing a comprehensive dataset derived from 55 studies, a gradient boosting decision tree model was refined through feature selection, demonstrating a prediction accuracy with an <i>R</i><sup>2</sup> of 80%. Moreover, incorporating these sensitive features into ML models significantly enhances prediction reliability and offers insights into the critical factors affecting fatigue strength in LPBF. This research also contributes to the understanding of feature selection in the context of additive manufacturing.</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-Based Prediction of Fatigue Strength in Additively Manufactured Ti-6Al-4V Parts: A Sensitivity Analysis of Input Features

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月23日

查看全文
  • Michael Andrew Hills,
  • Thorsten Hermann Becker

摘要

Laser powder bed fusion (LPBF) Ti-6Al-4 V parts exhibit variable fatigue strength due to inherent processing defects such as high residual stress, rough surface finish, subsurface and internal pores, and a brittle metastable microstructure. Traditional models often fail to predict this fatigue strength accurately due to the complexities involved. This study employs machine learning (ML) to analyze the sensitivity of various input features on fatigue strength predictions for LPBF-produced Ti-6Al-4 V parts. Through feature sensitivity analysis and testing of different feature selection methods, critical predictors such as surface roughness, stress intensity factor (ΔKth), hot isostatic pressure (HIP) and elongation at break were identified. Utilizing a comprehensive dataset derived from 55 studies, a gradient boosting decision tree model was refined through feature selection, demonstrating a prediction accuracy with an R2 of 80%. Moreover, incorporating these sensitive features into ML models significantly enhances prediction reliability and offers insights into the critical factors affecting fatigue strength in LPBF. This research also contributes to the understanding of feature selection in the context of additive manufacturing.