Transfer Learning-Based Optimization of the Ore Grinding Process Using Artificial Neural Networks

This study proposes a novel and comprehensive mathematical modeling approach designed for the multi-criteria optimization of ore-grinding processes within industrial electric drive mills. In the mining industry, the development of high-fidelity predictive models is frequently impeded by the scarcity of high-quality experimental data—a limitation that traditionally undermines the reliability of process control systems. To address this challenge, this research implements a sophisticated data augmentation strategy integrated with advanced transfer learning methodologies. The proposed framework begins with the artificial synthesis of datasets to enrich the information base, which is subsequently utilized to pre-train, evaluate, and refine various deep neural network architectures. By systematically identifying the most effective architecture for this specific technological context, the study fine-tunes the selected models using domain-specific parameters to ensure maximum operational relevance. The resulting robust neural network framework successfully solves complex optimization problems related to energy efficiency and throughput maximization. Quantitative results demonstrate that the proposed transfer learning paradigm accurately captures the non-linear dynamics of the grinding process, providing a physics-informed surrogate neural control model that serves as a cost-effective alternative to expensive and logistically demanding physical experiments. Ultimately, this research offers a scalable and computationally efficient framework for enhancing the operational performance and sustainability of raw ore-grinding systems in modern mineral processing plants.