Rusal has developed an electrolysis control technology using artificial intelligence

2024: AI tests in electrolysis control

RUSAL has completed industrial tests of artificial intelligence in electrolysis management. A specially trained neural network model calculates the chemical composition of the electrolyte in the electrolyzer and corrects the technological process. The project will increase the productivity of aluminum smelting. The company announced this on September 26, 2024.

The technology of using neural networks in the production of aluminum was developed by the RUSAL Engineering and Technology Center (RUSAL ITC), it is based on a neural network model trained on a large array of historical data of the RUSAL aluminum division. Model performs forecast calculation of parameters and controls electrolysis progress based on obtained result.

The chemical composition of the electrolyte in the electrolytic cell is determined by taking a sample and analyzing in the laboratory, measurement for each electrolytic cell, taking into account their quantity, is possible no more than once every two to three days. Based on the measurement, a conclusion is made about the progress of the technological process and the need to supply additional alumina or fluoride salts to the electrolytic cell, timely supply, in turn, makes it possible to increase the rate of aluminum smelting and make the process more efficient. With the introduction of the new technology, the neural network will determine the chemical composition, it will allow supplementing rare laboratory measurements with predictive values, will be able to make calculations as often as possible and will be much more efficient in correcting the electrolysis process, "said RUSAL Technical Director Viktor Mann.

The neural network calculates the chemical composition in the electrolyzer based on dozens of available technological parameters: electrolyte temperature, current, voltage, initial chemical composition, etc.

The neural network was trained at tens of thousands of technological parameters and laboratory measurements. Tests and pilot operation of the technology, which took place for about a year at the Sayanogorsk Aluminum Plant at the latest generation of RA-550 electrolyzers, confirmed its effectiveness. At the beginning of the tests, the model only calculated and gave a predicted value of the chemical composition of the electrolyte for the technologist, who independently, based on his knowledge, made a decision. When we made sure that the model was not wrong, it was introduced into the control loop and began to control the process on its own. In the future, we plan to integrate the model for electrolysis management at all RUSAL aluminum plants, "said Mikhail Grinishin, Director of Production Automation at RUSAL ITC.

RA-550 is a model of high-voltage electrolyzers from those developed by RUSAL. Among its most important advantages are reduced electricity consumption during aluminum smelting and environmental friendliness.

Thanks to the high level of automation of RA-550 electrolyzers, we have accumulated a huge amount of data on their work, the analysis of such an array by engineers is difficult. The neural network allows you to convert this data into digital models and use it to more effectively control electrolysis, while minimizing human participation. The main feature of the approach we applied to training the neural network was to attract the knowledge and skills of process engineers that they acquired over the years of work in production, "said Ilya Puzanov, Director of the RUSAL ITC High-Voltage Technology Improvement Project.