Headlight lens defect inspection using neural networks

The aim was to complement the existing inspection cell with cameras and electronics for evaluation using neural networks. ​

The existing equipment sends an image to a small industrial computer (subject of delivery) where it´s analysed using a neural network based on ResNet-50. All scratches, fingerprints, fibres, etc. are searched for. Subsequently, information is sent back to the inspection equipment to find out the image is OK or NOK.

The actual communication is done using a standard RPC call, so implementation into the control system of an existing cell was easy.

Other Case Studies