RUSAL, one of the world's largest aluminium producers, has completed industrial artificial intelligence tests in electrolysis control. A specially trained neural network model calculates the chemical composition of the electrolyte in the electrolyser and adjusts the technological process accordingly. The project aims to increase productivity in aluminium smelting.
The new neural network technology for aluminium production is a proprietary development by RUSAL's engineering department. It is based on a neural network model trained on a large dataset of historical information from RUSAL's aluminium division. The model performs predictive parameter calculations and, based on these results, manages the electrolysis process.
Currently, the chemical composition of the electrolyte in the electrolyser is determined through sampling and lab analysis, with measurements possible only once every two to three days per electrolyser. This technique helps monitor the process and determine the need for additional alumina or fluoride salts to improve efficiency. With the new technology, a neural network will supplement lab measurements with predicted values, allowing for more frequent calculations and quicker adjustments to the electrolysis process.
The neural network calculates the chemical composition in the electrolyser based on several available technological parameters, like electrolyte temperature, current, voltage, initial chemical composition, etc., and its effectiveness was confirmed through testing and pilot operations at the Sayanogorsk Aluminium Smelter. Initially, the model only predicted the electrolyte's chemical composition, which technologists used for decision-making. After confirming the model's accuracy, it was integrated into the control loop to manage the process autonomously. There are future plans to implement this model for electrolysis management across all RUSAL aluminium smelters.
RA-550 is RUSAL's latest and most advanced high-current electrolyser model. Its key advantages are reduced energy consumption during aluminium smelting and enhanced environmental friendliness.
Automated RA-550 electrolysers have generated massive operational data, making analysis challenging for engineers. Utilising a neural network can transform this data into digital models for more efficient electrolysis management with minimal human input. RUSAL leveraged the expertise of process engineers to train the neural network, using their years of experience in production.
Responses
