Scholarlink

<p>Selective Laser Melting (SLM) has emerged as a powerful technique for manufacturing Ti-6Al-4V components, widely adopted in biomedical, aerospace, and other high-performance industries. Despite its advantages, the SLM process is challenged by a multitude of interdependent parameters, making it difficult to consistently produce components with optimal tensile strength (TS) and hardness. Moreover, the high cost, time consumption, and expertise required further complicate the process. To address these issues and to enhance prediction performance, this study presents a novel machine learning (ML) approach enhanced by Limited-memory Broyden–Fletcher–Goldfarb–Shanno with Box constraints (L-BFGS-B) optimization to improve prediction accuracy and predict the TS of Ti-6Al-4V parts fabricated through SLM. Among five optimized ML models tested, ensemble models (Random Forest (RF) and Gradient Boosting (GB)) showed the highest performance, with the GB model achieving &lt; 10&#xa0;MPa mean absolute error (MAE), less than 1% of predicted strength. Experimental fabrication with optimized parameters resulted in a favourable distribution of fine grain structures and controlled volume fractions of <i>α</i>, <i>α</i>′ (martensitic), and minor amount of β phases, all resulting from the higher cooling rate of SLM. Microstructural and mechanical characterization revealed that improvement in TS and hardness of SLMed samples compared to conventionally made is due to the predominant finer <i>α</i>′ martensite structure generated by the rapid melting and solidification of SLM process. These findings highlight the effectiveness of the optimization strategy in enhancing model accuracy and robustness, especially within ensemble learning frameworks.</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

A novel L-BFGS-B optimized machine learning model for predicting tensile strength of SLM-manufactured Ti-6Al-4V alloys

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • B. S Arun,
  • V. N Ajukumar,
  • R Anand Sekhar

摘要

Selective Laser Melting (SLM) has emerged as a powerful technique for manufacturing Ti-6Al-4V components, widely adopted in biomedical, aerospace, and other high-performance industries. Despite its advantages, the SLM process is challenged by a multitude of interdependent parameters, making it difficult to consistently produce components with optimal tensile strength (TS) and hardness. Moreover, the high cost, time consumption, and expertise required further complicate the process. To address these issues and to enhance prediction performance, this study presents a novel machine learning (ML) approach enhanced by Limited-memory Broyden–Fletcher–Goldfarb–Shanno with Box constraints (L-BFGS-B) optimization to improve prediction accuracy and predict the TS of Ti-6Al-4V parts fabricated through SLM. Among five optimized ML models tested, ensemble models (Random Forest (RF) and Gradient Boosting (GB)) showed the highest performance, with the GB model achieving < 10 MPa mean absolute error (MAE), less than 1% of predicted strength. Experimental fabrication with optimized parameters resulted in a favourable distribution of fine grain structures and controlled volume fractions of α, α′ (martensitic), and minor amount of β phases, all resulting from the higher cooling rate of SLM. Microstructural and mechanical characterization revealed that improvement in TS and hardness of SLMed samples compared to conventionally made is due to the predominant finer α′ martensite structure generated by the rapid melting and solidification of SLM process. These findings highlight the effectiveness of the optimization strategy in enhancing model accuracy and robustness, especially within ensemble learning frameworks.

Graphical abstract