Scholarlink

<p>Circuits on printed circuit boards (PCBs) must be very reliable and of high quality, especially in the current dynamically changing electronics production line. This paper outlines a novel approach toward using the Faster Region–Convolutional Neural Network (Faster R-CNN) deep learning model for PCB fault identification and categorisation. A modelling process that consists of model construction, training, evaluation, and dataset generation was developed. A well-labelled and well-trained heterogeneous dataset with various types of PCB faults including various flaws such as spurs, open circuits, mouse bites, missing holes, shorts, and spurious copper was used in this study. Data augmentation techniques were employed to improve the data. After being adjusted and assessed, the Faster R-CNN model produced some excellent performance metrics. It had an accuracy of 0.912, precision of 0.905, recall of 0.930, F1 score of 0.917, and mean average precision (mAP) of 0.919. The inference outcomes showcase the capacity of the model in swiftly identifying various faults so that efficient actions in quality control processes can be determined on time. This work shows how complex deep learning algorithms can be applied to support PCB inspection and, in the long run, enhance manufacturing efficiency and product reliability.</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

Real-time machine learning for in situ quality control in hybrid manufacturing: a data-driven approach

  • 期刊论文

  • 发表日期：

    2025年1月14日

查看全文
  • Dinesh Mavaluru,
  • Akanksha Tipparti,
  • Anil Kumar Tipparti,
  • Mohammed Ameenuddin,
  • Jayabrabu Ramakrishnan,
  • Rafath Samrin

摘要

Circuits on printed circuit boards (PCBs) must be very reliable and of high quality, especially in the current dynamically changing electronics production line. This paper outlines a novel approach toward using the Faster Region–Convolutional Neural Network (Faster R-CNN) deep learning model for PCB fault identification and categorisation. A modelling process that consists of model construction, training, evaluation, and dataset generation was developed. A well-labelled and well-trained heterogeneous dataset with various types of PCB faults including various flaws such as spurs, open circuits, mouse bites, missing holes, shorts, and spurious copper was used in this study. Data augmentation techniques were employed to improve the data. After being adjusted and assessed, the Faster R-CNN model produced some excellent performance metrics. It had an accuracy of 0.912, precision of 0.905, recall of 0.930, F1 score of 0.917, and mean average precision (mAP) of 0.919. The inference outcomes showcase the capacity of the model in swiftly identifying various faults so that efficient actions in quality control processes can be determined on time. This work shows how complex deep learning algorithms can be applied to support PCB inspection and, in the long run, enhance manufacturing efficiency and product reliability.