In high-volume production of stamped parts, unwanted deformations occur, which need to be eliminated. This thesis presents the development of a machine vision system for recognizing deformed parts. To develop the system, we created a test station where a camera, aided by custom developed illumination, captures images of the parts. The images are then processed using RoboRealm software. The image processing is based on a shape-matching algorithm, which uses a deep learning model in the background. For training this model, we had to create a training dataset using test parts. The trained model was then applied to images of parts that were not used in model training. The results show that the system reliably detects deformed parts.
|
