Scholarlink

<p>Advanced manufacturing processes such as laser additive manufacturing-based selective laser melting (SLM) technique have attracted a lot of attention due to the rising need for high-performance materials in aerospace, automotive, and biomedical applications for fabrication of complex and intricate profiles in short runs. Although, due to&#xa0;the intricate relationships between several parameters including laser power, scanning speed, hatch spacing, and layer thickness, &#xa0;parametric&#xa0;optimization of SLM is a tedious&#xa0;and time consuming task. To resolve this issue, three boosting &#xa0;machine learning (ML) based algorithms such as Gradient Boosting, Ada boosting and XG boost Regressor are used&#xa0;in this work for prediction of surface roughness (<i>R</i><sub><i>a</i></sub>)&#xa0;and parametric optimization&#xa0;of SLM&#xa0;for improved part quality and&#xa0;functionality. The efficacy of the ML models was evaluated in terms of prediction accuracy and computational efficiency after training and testing to predict optimal process parameters for minimum <i>R</i><sub><i>a</i></sub>. The statistical measures on the testing data revealed values of 0.73, 1.27, and 1.21 for MAE, MSE, and RMSE, respectively&#xa0;for XG boost model. The average error of XG boost model is 0.016% and 1.26%,&#xa0;for the training and testing data&#xa0;sets respectively, which is significantly lower as compared to gradient boosting and Ada boosting methods. Therefore, XG boosting showed better accuracy in prediction of <i>R</i><sub><i>a</i></sub> values as compared to gradient boosting and Ada boosting methods. This is because of its better data handling capacity and efficient capturing of complex data sets. A 32.5% reduction in <i>R</i><sub><i>a</i></sub> value is achieved at optimum parameter settings as compared to worst setting. Image J software is used for porosity mapping under different process settings to identify regions of lack of fusion and key holes in fabricated samples. The maximum and minimum porosity in fabricated samples is found be 1.135% and 0.2342%, respectively. This work will be useful in implementation of ML in SLM for better process control, reduction in trial-and-error, and improve the quality of finished product.</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 Parametric Optimization and Porosity Mapping in Selective Laser Melting of SS316L Alloy

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Amit Sharma,
  • Tauseef Uddin Siddiqui,
  • Manoj Kumar Singh,
  • Arshad Noor Siddiquee,
  • Tarun Bhardwaj,
  • Aftab Ansari,
  • Arbab Jamil

摘要

Advanced manufacturing processes such as laser additive manufacturing-based selective laser melting (SLM) technique have attracted a lot of attention due to the rising need for high-performance materials in aerospace, automotive, and biomedical applications for fabrication of complex and intricate profiles in short runs. Although, due to the intricate relationships between several parameters including laser power, scanning speed, hatch spacing, and layer thickness,  parametric optimization of SLM is a tedious and time consuming task. To resolve this issue, three boosting  machine learning (ML) based algorithms such as Gradient Boosting, Ada boosting and XG boost Regressor are used in this work for prediction of surface roughness (Ra) and parametric optimization of SLM for improved part quality and functionality. The efficacy of the ML models was evaluated in terms of prediction accuracy and computational efficiency after training and testing to predict optimal process parameters for minimum Ra. The statistical measures on the testing data revealed values of 0.73, 1.27, and 1.21 for MAE, MSE, and RMSE, respectively for XG boost model. The average error of XG boost model is 0.016% and 1.26%, for the training and testing data sets respectively, which is significantly lower as compared to gradient boosting and Ada boosting methods. Therefore, XG boosting showed better accuracy in prediction of Ra values as compared to gradient boosting and Ada boosting methods. This is because of its better data handling capacity and efficient capturing of complex data sets. A 32.5% reduction in Ra value is achieved at optimum parameter settings as compared to worst setting. Image J software is used for porosity mapping under different process settings to identify regions of lack of fusion and key holes in fabricated samples. The maximum and minimum porosity in fabricated samples is found be 1.135% and 0.2342%, respectively. This work will be useful in implementation of ML in SLM for better process control, reduction in trial-and-error, and improve the quality of finished product.