Control of metallurgical processes with indirect measurements and machine learning (MetMaskin)

Important results from the project

The aim is to demonstrate a model for better monitoring and process control. A model that facilitates the operator´s decisions using measured quantities that are not used today, and reduces the importance of individual skills of operators and facilitates the training of personnel. The goal was to be able to detect a stirring intensity of a steel melt by measuring vibrations. The measurement must be input data to a mathematical algorithm that, via machine learning, creates an operator support for operators for a metallurgical process where the stirring of the steel melt is important.

Expected long term effects

The results obtained may form the basis for further decisions about a more permanent installation at a steel plant. Then experienced operators can decide whether the developed method and model description should be used as support for the operators´ decisions such as time lengths of e.g. the step change of gas composition in an AOD converter or time optimization of process steps such as treatment time under vacuum in a vacuum station. These time steps are crucial for an optimized process control and resource-efficient process with optimized quality of produced metal.

Approach and implementation

The project was planned as an initial measurement campaign at two steel companies with a vacuum station and an AOD converter. The initial campaign provided measurement data and experience for a second, longer campaign at each of the companies. The project was followed by other steel companies in a project group with continuous meetings during the time of the project. The collaboration worked well, but the evaluation of measurement data was delayed due to the increased workload during the Covid-19 pandemic then also the second campaign was delayed due to the visit and travel restrictions.