Scholarlink

<p>Microscopic images captured from various cross-sections were used to detect and identify defects in additively manufactured parts. Image processing was performed using object detection models trained on microscopic images from multiple sections. Subsequently, the relationship between process parameters and defect behavior was analyzed in the second stage. Three classes of defects —cracks, lack of fusion (LOF), and porosity— were labelled. The proposed YOLOv8-based model achieved a mean average precision (mAP) of 77% at 50% intersection over union (IoU) and a recall of 78%, and was selected for quantifying defect areas. In the second stage, 94 new samples were fabricated with varying process parameters, including laser power, scan speed, and hatch distance. The outputs from the image processing stage were combined with the corresponding input parameters to construct a dataset for machine learning. Among the evaluated models, the Bagging (Decision Trees) model was selected for predicting defect areas. The results demonstrated that process parameters have a significant influence on defect formation. According to the predictions of the model, the combination of 230&#xa0;W laser power, 1200&#xa0;mm/s scan speed, and 0.06&#xa0;mm hatch distance minimized the total defect area in the fabricated parts. </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

Correlation between Process Parameters and Defects in Additively Manufactured Components Using Image Processing and Machine Learning

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • M. Memari,
  • A. Farzadi,
  • A. Karami Fini,
  • M. Aliabadi,
  • A. Shamsipoor

摘要

Microscopic images captured from various cross-sections were used to detect and identify defects in additively manufactured parts. Image processing was performed using object detection models trained on microscopic images from multiple sections. Subsequently, the relationship between process parameters and defect behavior was analyzed in the second stage. Three classes of defects —cracks, lack of fusion (LOF), and porosity— were labelled. The proposed YOLOv8-based model achieved a mean average precision (mAP) of 77% at 50% intersection over union (IoU) and a recall of 78%, and was selected for quantifying defect areas. In the second stage, 94 new samples were fabricated with varying process parameters, including laser power, scan speed, and hatch distance. The outputs from the image processing stage were combined with the corresponding input parameters to construct a dataset for machine learning. Among the evaluated models, the Bagging (Decision Trees) model was selected for predicting defect areas. The results demonstrated that process parameters have a significant influence on defect formation. According to the predictions of the model, the combination of 230 W laser power, 1200 mm/s scan speed, and 0.06 mm hatch distance minimized the total defect area in the fabricated parts.