Adaptive models of process flow charts for mining and processing plants using machine learning technology

A software application has been developed to create adaptive models for mineral processing flow charts and processes, leveraging machine learning technology. The application enables generation of synaptic weights in a neural network that establish precise correlations between the input and output parameters of industrial processes, facilitating the creation of adaptive models. Its algorithm is based on the architecture of feedforward neural networks, incorporating bias neurons. This tool is designed for direct integration into automated process control systems, enabling real-time optimization of enrichment processes, identification of optimal input and output values, and accurate forecasting to support rapid and effective decision-making. The application was tested on data from the iron ore enrichment section of a crushing and beneficiation plant, which led to the development of a block-based adaptive model representing the qualitative and quantitative process flow charts of the section. The model comprises interconnected adaptive submodels for processes such as grinding, wet magnetic and magnetic-gravity separation, classification, and desliming. These submodels, represented by trained neural networks, are linked according to the operational flow chart, allowing comprehensive validation of the overall model and analysis of the impact of changes in any single process on the entire section. The ability to update training datasets ensures that the neural network adapts to operational changes in the beneficiation plant. These changes may include variations in initial ore batch composition, useful component content in the products, process performance, slurry solid content, water consumption, circulation load, and concentrate grade.