Scholarlink

Screw – the small but important elements used in various industry. Its presence plays a significant role as it securely holds the product in place, preventing loosening or collision with the case. Such occurrences could lead to the displacement of small components or compartments, resulting in product failure. The advent of Industry 4.0 has contributed to reducing labor costs and human errors. This research aims to develop a robust classification model for machine vision detection, specifically for identifying the absence or presence of a screw. The model can be integrated into relevant robotics applications. To collect the customized dataset, a 6-degree-of-freedom (DOF) robot. The collected dataset was then categorized into two groups: absent and present. For the training process, a pretrained dataset called ImageNet was employed to facilitate the training process. Transfer learning techniques were used to extract the features required for different machine learning models. Each machine learning model underwent hyperparameter tuning to achieve the highest classification accuracy. The data was divided into training, validation, and testing sets using a sampling ratio of 60:20:20, respectively, before being fed into the various machine learning models.

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

Utilizing Transfer Learning Models to Classify Absence Defects on Aluminum Plates Using Feature-Based Approaches

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kiran Pandian,
  • Lim Weng Zhen,
  • Anwar P. P. Abdul Majeed,
  • Sze-Hong Teh,
  • Koon Yin Goon,
  • Mohd Azraai Mohd Razman

摘要

Screw – the small but important elements used in various industry. Its presence plays a significant role as it securely holds the product in place, preventing loosening or collision with the case. Such occurrences could lead to the displacement of small components or compartments, resulting in product failure. The advent of Industry 4.0 has contributed to reducing labor costs and human errors. This research aims to develop a robust classification model for machine vision detection, specifically for identifying the absence or presence of a screw. The model can be integrated into relevant robotics applications. To collect the customized dataset, a 6-degree-of-freedom (DOF) robot. The collected dataset was then categorized into two groups: absent and present. For the training process, a pretrained dataset called ImageNet was employed to facilitate the training process. Transfer learning techniques were used to extract the features required for different machine learning models. Each machine learning model underwent hyperparameter tuning to achieve the highest classification accuracy. The data was divided into training, validation, and testing sets using a sampling ratio of 60:20:20, respectively, before being fed into the various machine learning models.