Computing and Informatics

Industrial companies are more and more interested in the use of artificial intelligence (AI) in the control and monitoring of their processes. They try to take advantage of the power of this technology in order to increase the level of automation and to build smarter machines with new capabilities of self-adaptation and self-control. Especially, the automotive industry, with their high requirements in productivity and diversity management, are eager to adapt AI concepts to their processes. However, the training of Deep Learning (DL) models requires an important effort of data preparation, providing a dataset of all possible configurations. Indeed, this dataset must be collected and then annotated. Considering the fact that automotive industry deals with a huge number of references and that it often and quickly needs to modify their products, it is very difficult, if not impossible, to gather sufficient datasets for each produced reference and to have the time to train DL models in the plants with the traditional methods. This paper presents an innovative methodology to prepare the dataset by creating virtual images instead of collecting real ones and then automatically annotating them. It will demonstrate that this method will reduce the efforts and the time of the preparation of the dataset significantly. The paper will also present how this method was deployed for the quality control of welding operations in the automotive industry.