Deep learning based 3D defect detection system using photometric stereo illumination

Part inspection machines of industrial manufacturing systems are being newly evolved as intelligent machines with the technology innovation of artificial intelligence. Especially, the automation of defect detection systems in the field of casting industry has been widely studied, applying deep learning based inspection algorithms due to its inspection difficulties with 2D and 3D detects. In this paper, we proposed an automatic defect detection system based on deep learning technology in artificial intelligence using fused illumination images to get 2D and 3D information of target objects simultaneously. Due to the characteristics of the cast product surfaces, the success rate of a conventional 2D detect detection system is easily affected by illumination location and angle for uneven surfaces and small defects. To solve this problem, a photometric stereo system to generate its reflectance, roughness and slope information is used to generate feature fusion data. This dataset is used to improve the defect detection performance of automatic inspection machines for casting products with a deep learning model. Experiment results show that the detection accuracy of the proposed method is 62.58%. The proposed system is expected to be used as a new technology to improve the detection performance of AOI(Automated Optical Inspection) machines with deep learning.