Speaker
Description
The surface quality of a finished product on a galvanizing line is a very important criteria, especially for automotive and home appliance exposed products. To control it, automatic surface inspection systems (ASIS) have been developed since many years and are installed on all galvanizing lines dealing with high product quality. A highly performant ASIS from Primetals Technologies has been developed for many years and is especially protected under the worldwide known Trademark SIAS®.
In the last ten years, ASIS performance improvements are mostly due to the integration of new developments in lighting and camera technologies. Deep-learning and artificial intelligence are fast developing technologies. Convolutional neural networks have proved to be the most efficient tool to address many image processing problems like image retrieval and classification. It remains still very challenging to use these technologies industrially for real-time applications on large video streams.
A new online real time defect detection/classification system using full convolutional network has been developed and the prototype of this system is installed in Liège on the EUROGAL galvanizing line (ArcelorMittal Belgium). The added value is expected to be threefold:
Improved defect detection/classification performances (for textured product in particular).
Provide an easier tuning on multi-camera systems integrating several lighting conditions.
Decrease the sensitivity to the tuning parameters and provide a more generic detection configuration, easier to transfer from one product or one line to another.
This paper presents the principle, the first results of the prototype which is installed at EUROGAL. We will present also the vision of the future of surface inspection system, dealing with training the neural network on multiple sites, and the SIAS Fleet Management System.
Keywords
Automatic surface inspection, Artificial Intelligence, Deep Learning, Convolutional Neural Network.