AI-based production optimization system for flotation control tasks: Talnakh Concentrator`s experience

Digitalization of industrial sites at the current stage of development includes not only equipping them with sensors and a single control interface via SCADA systems, but also unification of processes (according to the methodology of Data Governance — data quality management). The desire to increase the volumes and quality of finished products, reduce costs inevitably leads to experimental modes of technological processes that were not included at the design stage. The accumulated statistics of changes in physicochemical properties in such conditions is an invaluable source for comparing control algorithms. Generalization of the results by means of machine learning allows to identify and programmatically implement the most effective of them, taking into account multi-purpose, multi-criteria optimization problems — to obtain automatic control services, operators` digital twins. The methodological and practical experience of creating a production optimization system that ensures an increase in metal extraction and product quality at the Talnakh Concentrator of the Polar branch of PJSC MMC Norilsk Nickel is considered. An approach to formalizing management in a plant environment is presented: classification of observed physical and chemical parameters for identifying conditions (Condition-Based Maintenance), a hybrid modeling format combining the results of simplified physical and mathematical models and machine learning, control based on forecast data (Model Predictive Control), division of control scales into dispatcher and operator levels, development and updating of services in a subject-oriented paradigm (Domain-Driven Design) and taking into account the flow of requests from technologists who carry out the actual enrichment process using automatic control services. The result of this work by the end of 2024 is an increase in the through extraction of nickel (by 0.36% (rel.)) and copper (by 0.16% (rel.)) into collective concentrate, and stabilization of flotation processes in conditions of ore variability. A technical effect was also achieved due to a more optimal distribution of metals in profile concentrates: an increase in Ni extraction in nickel concentrates by 0.5% (rel.) and a decrease in Ni extraction in copper concentrate by 0.5% (rel.) with a fixed through extraction (an effect due to the absence of losses at the Copper Plant).