Abstract:In the conditioning process of fine-grained minerals, the quantitative relationship between the microscopic-scale dispersion of oil-based water-insoluble reagents and macro-regulation parameters is a critical theoretical basis for guiding the design of stirring equipment. Therefore, this study proposes a dynamic online testing method for microscopic-scale reagent oil droplets by integrating high-speed microscopic imaging technology with unsupervised learning-based image processing. A high-precision microscopic flow field testing platform was constructed to capture microscopic flow field characteristics, and unsupervised learning image processing was employed to identify reagent oil droplet sizes, achieving boundary recognition of droplets in dynamic, complex multi-droplet microscopic scenarios. Through multi-condition tests at varying stirrer impeller speeds, it was found that stirring intensity exhibits a nonlinear correlation with reagent dispersion efficiency: as the impeller speed increases, the proportion of small-scale droplets (< 40μm) rises, while the proportion of large-scale droplets (> 80μm) decreases. Further increases in stirring speed show limited impact on the proportion of large droplets, and the growth rate of small droplet proportion slows down. This research realizes dynamic online recognition of reagent dispersion scales during the stirring process, providing theoretical support for optimizing flow field parameters and impeller design.