Scholarlink

<p>This research focuses on optimizing the 3D printing process by adjusting key printing parameters, including temperature, speed, and layer thickness, to achieve desired mechanical properties such as yield strain, Young’s modulus, and peak load. A variant of the autoencoder (AE) network is applied to capture the relationship between these printing parameters and the mechanical properties. The proposed model demonstrates accurate bidirectional prediction capabilities, allowing for the estimation of mechanical properties based on printing parameters and vice versa. In particular, by incorporating a guided encoder-decoder training approach, the model prediction accuracy improved significantly, with the <i>R</i><sup>2</sup> score increasing from 90.87 to 93.56%. Furthermore, this study addresses challenges in inverse problems, particularly cases where multiple configurations of mechanical properties correspond to the same set of printing parameters. By utilizing a guided model, ambiguity in parameter prediction is effectively reduced, enhancing the reliability of machine learning-based optimization. These findings contribute to the advancement of additive manufacturing by providing a robust framework for improving the precision and efficiency of 3D printing processes.</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

Optimizing 3D printing through enhancements in inverse autoencoder models and guided decoders for predicting mechanical properties

  • 期刊论文

  • 发表日期：

    2025年5月20日

查看全文
  • Phuong Dong Nguyen,
  • Tien-Dat Hoang,
  • Thanh Q. Nguyen

摘要

This research focuses on optimizing the 3D printing process by adjusting key printing parameters, including temperature, speed, and layer thickness, to achieve desired mechanical properties such as yield strain, Young’s modulus, and peak load. A variant of the autoencoder (AE) network is applied to capture the relationship between these printing parameters and the mechanical properties. The proposed model demonstrates accurate bidirectional prediction capabilities, allowing for the estimation of mechanical properties based on printing parameters and vice versa. In particular, by incorporating a guided encoder-decoder training approach, the model prediction accuracy improved significantly, with the R2 score increasing from 90.87 to 93.56%. Furthermore, this study addresses challenges in inverse problems, particularly cases where multiple configurations of mechanical properties correspond to the same set of printing parameters. By utilizing a guided model, ambiguity in parameter prediction is effectively reduced, enhancing the reliability of machine learning-based optimization. These findings contribute to the advancement of additive manufacturing by providing a robust framework for improving the precision and efficiency of 3D printing processes.