Deep Learning for Inspection at Tata Steel
Tata Steel in IJmuiden produces 6.5 mln tonnes of steel strip products for the automotive, engineering, building, battery and packaging market which require a high surface quality. About 25 years ago human inspectors got support from cameras and later from automatic classification systems.
The current classification techniques are sensitive to change of environment (lighting, steel surface etc.) and continuous learning is missing. Tata Steel has now started to develop a new generation of surface inspection systems using artificial intelligence (deep learning and active learning) with improved classification performance over the current technique.
The presentation shows the development, challenges and the potential of deep learning.
Artificial & Data Intelligence
Digital Transformation & Control
Johan Bernard is a chemical engineer and made a PhD at Delft University of Technology. In 1992 he started his career at Hoogovens in IJmuiden which is now part of Tata Steel. He had a variety of positions in research, production, IT and OT project management and head of process engineering. Since 2017 he is running a part of the Smart Industry/Industry 4.0 programme at Tata Steel in IJmuiden which contains simulation and digital twin, data sharing with customers, new sensor applications and artificial intelligence for surface inspection system in the steel industry.