Automated Visual Inspection System of Gear Surface Defects Detection Using Faster RCNN

A company that produces gears is capable of producing 50 gears per hour every day. The gears will be sorted by the QC (Quality Control) division by taking faulty gears based on the surface and rigidity shape. In one production process, it produces 3-6 faulty gears. The QC process has shortcomings such as the sharpness of vision, concentration duration and accuracy of each person varies, so faulty gears can escape the QC stage, which will become a problem if it reaches the distributor's hands as it will affect the level of trust. Therefore, a visual inspection system was created that contains an algorithm that detects image defects using the Faster RCNN deep learning method. The gear detection process in this project uses the Faster RCNN method, which has a higher accuracy than the Frame Per Second speed or detection speed. This method aims to achieve a higher accuracy. The average accuracy produced in the condition of 1 gear and 1 type of defect in 1 frame and the conveyor is not moving is 86%, but the value decreases when detecting 2 classes of defects in one frame and the conveyor is not moving, the average accuracy produced is 76%, and also when on the conveyor that is moving at speeds of 2-8.3 RPM and 1 gear in 1 frame, the average accuracy decreases to 83.32%.