Automated inspection of molten metal using machine learning

During metallurgical operations such as storage, stirring or casting, monitoring of metal properties is a crucial part of quality assurance. Nowadays, this is usually done by a human operator who observes the metal/slag surface distribution in an uncomfortable environment. In this paper, a novel computer vision method for metal/slag characterization is proposed. The goal is to support the ladle operator with automated inspection. The method relies on artificial neural networks-based classification to differentiate between thick slag, thin slag and bare metal. Validation has been done using a cold model where water, oil and coal are mixed in order to mimic the metal-slag interface. The proposed solution is shown to outperform conventional methods.