Design and implementation of a visual recognition system in the PLC-based production line of the Audi e-tron

Nowadays, the industrialization 4.0 and the announcement of the Audi e-tron production in Brussels are leading to an update of the production line at Audi Brussels. The skids, metal structures on which the frames are placed, includes an identification plate for the operators. Information useful for the production of the car (color, roof type, etc.) is also encoded in an RFID cell attached to the skid. The proper management of these skids requires facilitating human-machine interaction. This master’s thesis aims to develop a system for skid recognition, error identification and management. To achieve these goals, several concepts are considered in this paper to replace the QR codes previously used by Audi. The chosen recognition system will use a combination of a deep neural network and visual recognition techniques. Different network architectures, optimizers, data augmentation techniques, standardization or template matching methods will be compared during the analysis of experimental results using the prototype implemented on the production line. Finally, prospects for generalizing the developed system are conceivable, such as multimodal recognition or image rectification. Ultimately, the model as designed achieves an accuracy level of 99.93%, which exceeds the requirements set for this project.