Use of Convolutional Neural Networks for Quality Control in Automotive Industry

The automotive industry is composed of multiple manufacturing processes of primary parts for assembly of assemblies. These parts are generally built on massive manufacturing processes, based on stamping and simple assemblies, through nailing, welding, grinding, adding components, etc. Defects in elements that affect the process can lead to the generation of whole batches of defective parts. The inspections are carried out by artificial vision cameras based on patterns recognitions algorithms. The efficiency to recognize patterns depends on many variables such as the position of the pieces, changes in the brightness, appearance of dust, sensor deterioration, etc. The use of convolutional neural networks has proven to be an effective method to perform this type of recognition in images. Furthermore, automotive industry requires more information and data on what is happening at each stage of the process. This means that it is no longer useful to send a few digital signals to make the process work, but that the massive sending of data that has been processed apart from the process is requited to adjust it constantly and in real time, at the stage it is at or before or after, based on the information received, which is achievable through the use of the Internet of Things. In this work a new approach based on algorithm YOLO v3 and Internet of Things is presented. A comparison of the performance with other neural networks architectures is performed. The accuracy is studied with different conditions to check the performance of the classification for real-time applications.