Scholarlink

<p>Achieving a smooth surface finish is crucial in 3D printing, as it affects both the quality and functionality of the printed parts. In the Fused Deposition Modeling (FDM) process, selecting the optimal process parameters to minimize surface roughness (Ra) is particularly challenging. This often results in wasted material and time, and sometimes even unsuccessful outcomes. To address this issue, the current study aims to optimize FDM printing parameters such as layer height, print speed and nozzle temperature, to minimize surface roughness in PLA parts. Additionally, it focuses on developing machine learning models to accurately predict surface roughness based on these selected parameters. Five different data mining algorithms were used to develop models that are capable of forecasting the Ra value of the parts fabricated through FDM with polylactic acid (PLA) plus filament. The models are generated by altering key process parameters. The significant FDM process variables such as layer height, nozzle temperature, and print speed were chosen with three levels of input parameters; a total of 27 parts were printed. Moreover, a dataset comprising 15 additional instances has been constructed to assess the performance of these models. The mean Ra value was determined using a surface roughness tester with the outer portion of the printed part in the print direction. The Ra values were taken for further machine learning analysis (using WEKA software) like Naïve Bayes, Multilayer Perceptron, J48, Random Forest, and Random Tree. The models generated by random forest and J48 have obtained the best results in the training and testing. The J48 algorithm, also known as C4.5, is capable of generating decision tree models. These models are valuable for identifying the key parameters that affect surface roughness in printing processes. Notably, the most significant influencing parameters are in the order of Print speed (PS), layer height (LH), and Nozzle temperature (NT). These printing parameters improve the quality and suitability of PLA components printed through FDM for a range of engineering applications.</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

Evaluating machine learning methods for predicting surface roughness of FDM printed parts using PLA plus material

  • 期刊论文

  • 发表日期：

    2025年1月9日

查看全文
  • R. Soundararajan,
  • A. Sathishkumar,
  • S. Abdul Aathil,
  • N. Gnana Chandran

摘要

Achieving a smooth surface finish is crucial in 3D printing, as it affects both the quality and functionality of the printed parts. In the Fused Deposition Modeling (FDM) process, selecting the optimal process parameters to minimize surface roughness (Ra) is particularly challenging. This often results in wasted material and time, and sometimes even unsuccessful outcomes. To address this issue, the current study aims to optimize FDM printing parameters such as layer height, print speed and nozzle temperature, to minimize surface roughness in PLA parts. Additionally, it focuses on developing machine learning models to accurately predict surface roughness based on these selected parameters. Five different data mining algorithms were used to develop models that are capable of forecasting the Ra value of the parts fabricated through FDM with polylactic acid (PLA) plus filament. The models are generated by altering key process parameters. The significant FDM process variables such as layer height, nozzle temperature, and print speed were chosen with three levels of input parameters; a total of 27 parts were printed. Moreover, a dataset comprising 15 additional instances has been constructed to assess the performance of these models. The mean Ra value was determined using a surface roughness tester with the outer portion of the printed part in the print direction. The Ra values were taken for further machine learning analysis (using WEKA software) like Naïve Bayes, Multilayer Perceptron, J48, Random Forest, and Random Tree. The models generated by random forest and J48 have obtained the best results in the training and testing. The J48 algorithm, also known as C4.5, is capable of generating decision tree models. These models are valuable for identifying the key parameters that affect surface roughness in printing processes. Notably, the most significant influencing parameters are in the order of Print speed (PS), layer height (LH), and Nozzle temperature (NT). These printing parameters improve the quality and suitability of PLA components printed through FDM for a range of engineering applications.