Defect Detection of Grinded and Polished Workpieces Using Faster R-CNN

Polishing and grinding are crucial in the fabrication processes of industrial and commercial products. While the fabrication process can be automated using robots or specialized machines, experienced workers are still needed for the subsequent quality inspection. Here, we report the development of an automatic defect detection system, which is capable of detecting the defects of grinded and polished faucets due to its Faster Region-based Convolutional Neural Networks (Faster RCNN) architecture. The images of the workpieces were taken using a manipulator with a preset trajectory to cover all the surfaces of the workpieces. After labeling, the data were augmented to the trainable level. Three pretrained CNN-based models were utilized and evaluated. The hyperparameters were analyzed to validate their effect on the performance of the model. The mean average precision, using the tuned hyperparameters, was 80.26%.