Scholarlink

<p>This paper presents a novel methodology for determining the relative permittivity (ε<sub>r</sub>) of 3D-printed materials by leveraging Rectangular Microstrip Patch Antenna (RMPA) design equations and machine learning techniques, specifically a 1D Convolutional Neural Network (1D-CNN). Unlike conventional methods, such as the ring resonator, which require extensive geometry creation and complex calculations, the proposed method predicts ε<sub>r</sub> by analysing the shift in the S<sub>11</sub> response of an inset feed patch antenna. Initially, a RMPA is designed for a 2.4&#xa0;GHz frequency, assuming a relative permittivity of 2.6, and simulated 50 times with varying ε<sub>r</sub> values using Ansys HFSS. The resulting S<sub>11</sub> data is used to train the CNN model, which is then validated using a 3D-printed Polylactic Acid (PLA) substrate. Experimental results show good agreement between the CNN predictions and the values obtained using the ring resonator method, with the CNN achieving a Mean Square Error (MSE) of 0.03 for unknown samples. This approach provides a simple alternative for material characterization, particularly in antenna and sensor design, and has potential for further improvements in accuracy by expanding the dataset and incorporating the loss tangent. The synergy between antenna design theory and machine learning is demonstrated, offering significant potential for advancements in material engineering and characterization.</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

New Approach of Determination of Relative Permittivity of 3D Printed Materials Using RMPA and 1D-CNN

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Atul M. Kulkarni,
  • Garima Saini,
  • Shyam S. Pattnaik

摘要

This paper presents a novel methodology for determining the relative permittivity (εr) of 3D-printed materials by leveraging Rectangular Microstrip Patch Antenna (RMPA) design equations and machine learning techniques, specifically a 1D Convolutional Neural Network (1D-CNN). Unlike conventional methods, such as the ring resonator, which require extensive geometry creation and complex calculations, the proposed method predicts εr by analysing the shift in the S11 response of an inset feed patch antenna. Initially, a RMPA is designed for a 2.4 GHz frequency, assuming a relative permittivity of 2.6, and simulated 50 times with varying εr values using Ansys HFSS. The resulting S11 data is used to train the CNN model, which is then validated using a 3D-printed Polylactic Acid (PLA) substrate. Experimental results show good agreement between the CNN predictions and the values obtained using the ring resonator method, with the CNN achieving a Mean Square Error (MSE) of 0.03 for unknown samples. This approach provides a simple alternative for material characterization, particularly in antenna and sensor design, and has potential for further improvements in accuracy by expanding the dataset and incorporating the loss tangent. The synergy between antenna design theory and machine learning is demonstrated, offering significant potential for advancements in material engineering and characterization.