Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-15-0038
Surface Scratch Defect Detection of Titanium Spacer Ring in Hard Disk based on Convolutional Neural Network
Department of Mechanical Engineering, National Taiwan University of Science and Technology, 1, Sec. 3, Zhongxiao E. Rd., Taipei 10608, Taiwan
ABSTRACT
In terms of quality management in the metal machining industry today, most of the defects are classified and judged by the human eye. The decision of the human eye will be affected by the external environment and mental fatigue problems, and small defects will also cause misjudgment due to lack of experience. In this research, aimed at the titanium metal spacer ring inside the hard disk inspection, surface scratch defects are identified. Because of the different depths of scratches and placement angles, shallow scratch features cannot be identified. To fully capture the defect features, referring to the distribution concept of bright field and dark field, a set of the lighting system with a single light source and motor control is designed, which can acquire multiangle images at one time. Due to the large difference in the shape and intensity of scratches, it is difficult to detect defects using the traditional digital image processing method. In this study, the deep learning network architecture was used. First, the images were pre-processed to reduce the interference of noise, and the number of image data increased through data augmentation. Then, the FasterRCNN and Unet models are used for training and prediction. The final experimental results showed that the success rate of defect identification was 95% and 82%%, respectively.
Keywords: Defect Detection, Convolutional Neural Network, Multiangle Illuminations, Automated Optical Inspection, Metallic Ring
Department of Mechanical Engineering, National Taiwan University of Science and Technology, 1, Sec. 3, Zhongxiao E. Rd., Taipei 10608, Taiwan
ABSTRACT
In terms of quality management in the metal machining industry today, most of the defects are classified and judged by the human eye. The decision of the human eye will be affected by the external environment and mental fatigue problems, and small defects will also cause misjudgment due to lack of experience. In this research, aimed at the titanium metal spacer ring inside the hard disk inspection, surface scratch defects are identified. Because of the different depths of scratches and placement angles, shallow scratch features cannot be identified. To fully capture the defect features, referring to the distribution concept of bright field and dark field, a set of the lighting system with a single light source and motor control is designed, which can acquire multiangle images at one time. Due to the large difference in the shape and intensity of scratches, it is difficult to detect defects using the traditional digital image processing method. In this study, the deep learning network architecture was used. First, the images were pre-processed to reduce the interference of noise, and the number of image data increased through data augmentation. Then, the FasterRCNN and Unet models are used for training and prediction. The final experimental results showed that the success rate of defect identification was 95% and 82%%, respectively.
Keywords: Defect Detection, Convolutional Neural Network, Multiangle Illuminations, Automated Optical Inspection, Metallic Ring