When assembling motors on the assembly line at Domel, there is a possible incorrect insertion of a spring, which results to the brush being pushed onto the commutator. Such spring position may lead to a shortening of the motor life span. The position of the spring is checked by a machine vision system, which is used to detects any incorrect position of the spring. This machine vision system may malfunction with certain types of products, which leads to an incorrect classification.
The first part of this work describes and verifies our assumptions of the alleged causes of this misclassification. The second part presents our solution to upgrading the machine vision system in order to properly detect incorrect spring positions. This solution is based on finding reference points on segmented areas of the spring and brush guide. For image segmentation, we used a convolution neural network, which was thought to perform the image segmentations.
As a solution to the problem, an additional algorithm is presented, which is used for recognizing images with faulty products, as well as sorting images with correct and incorrect spring positions with respect to the distance between the spring and the brush guide. This additional algorithm also takes into account the preliminary classifications of the existing machine vision system.
|
