스마트폰용 미세형상 사출성형부품의 품질검사 자동화를 위한 양불판정 학습시스템 개발 (Development of a pass/fail learning system for automated quality inspection of micro-shaped injection molded parts for smartphones)

서지정보

· 발행기관 : 한국금형공학회
· 수록지 정보 : Design & Manufacturing / 19권 / 2호 / 67 ~ 73페이지
· 저자명 : 박정연, 이준한, 윤선진, 윤길상

초록

This study was conducted to develop an AI-based learning system to automate the quality inspection of micro-shaped injection molded parts such as smartphone cameras. Existing machine vision-based quality inspection systems have limitations in that they cannot sufficiently recognize various shapes and types of defects and have limited inspection areas. In particular, micro-shaped parts such as smartphone camera modules contain metal lead frames with a thickness of 0.1 mm or less, so existing manual inspection methods have limitations in maintaining high productivity and quality. In this study, a deep neural network based on CNN (Convolutional Neural Network) was applied to automate high-precision quality inspection, and a dataset using high-resolution images of 11.94 million pixels was constructed. During the data collection process, 400 good products and 3,300 defective products were secured, and various types of defects were prepared to be learned through data augmentation techniques. As a result of model learning, it was confirmed that the accuracy was 95.2% and the F1-Score was 94.2%, which improved the performance by about 15% and shortened the inspection time by 30% compared to the existing manual inspection method. The AI-based quality inspection system developed in this study can be effectively applied to the mass production environment of micro-shaped injection molded parts, and it is expected to complement the limitations of the existing quality inspection method. Future research will include additional expansion of defect types, improvement of real-time inspection speed, and review of applicability to various materials and parts.

영어초록

This study was conducted to develop an AI-based learning system to automate the quality inspection of micro-shaped injection molded parts such as smartphone cameras. Existing machine vision-based quality inspection systems have limitations in that they cannot sufficiently recognize various shapes and types of defects and have limited inspection areas. In particular, micro-shaped parts such as smartphone camera modules contain metal lead frames with a thickness of 0.1 mm or less, so existing manual inspection methods have limitations in maintaining high productivity and quality. In this study, a deep neural network based on CNN (Convolutional Neural Network) was applied to automate high-precision quality inspection, and a dataset using high-resolution images of 11.94 million pixels was constructed. During the data collection process, 400 good products and 3,300 defective products were secured, and various types of defects were prepared to be learned through data augmentation techniques. As a result of model learning, it was confirmed that the accuracy was 95.2% and the F1-Score was 94.2%, which improved the performance by about 15% and shortened the inspection time by 30% compared to the existing manual inspection method. The AI-based quality inspection system developed in this study can be effectively applied to the mass production environment of micro-shaped injection molded parts, and it is expected to complement the limitations of the existing quality inspection method. Future research will include additional expansion of defect types, improvement of real-time inspection speed, and review of applicability to various materials and parts.

참고자료

· 없음