Computational Study of Particle Behavior in the Blast Furnace Charging Process

The blast furnace is the dominant ironmaking route, operating as a large-scale multiphase reactor with heat and mass exchange and chemical reactions between the ascending gases and the descending solid burden. Within this system, burden distribution plays a critical role in determining productivity and product quality by regulating gas flow. This underscores the fundamental importance of the charging process, as it dictates the initial burden distribution at the furnace throat. The raw materials, such as coke, sinter and pellet, are granular in nature, serving as the fundamental basis for studying burden behavior. A fundamental understanding of particle behavior is therefore a prerequisite for elucidating the mechanisms governing burden distribution.
However, due to the complexity of industrial operations and intricate of granular behavior, a comprehensive understanding of granular dynamics, bulk properties and the charging mechanism has yet to be fully elucidated. To bridge this gap, this thesis investigates the particle and bulk properties of raw materials, as well as their dynamics during the charging process, by integrating image processing, deep learning techniques, and numerical simulations.
A coke particle detector based on the YOLOv3 algorithm was developed to identify coke particles and calculate their particle size distribution (PSD). This method enables the rapid acquisition of accurate PSD data for the coke to be charged, providing a robust foundation for future applications in online monitoring and intelligent control of blast furnace burden quality. In parallel, to overcome the limitations of particle shape simplification in burden studies, a three-dimensional (3D) model of a realistic irregular sinter particle was constructed. Using this geometric template, 3D-printed particles were fabricated for physical experiments, while multi-sphere particles were generated for simulations. Taking advantage of Discrete Element Method (DEM) simulations, both controlled-parameter pile modeling and particle-scale analysis of granular behavior were enabled.
The non-spherical particles enabled the simulation to capture essential mechanical interactions during pile formation, such as particle interlocking and rolling resistance. The angle of repose was found to be jointly influenced by particle shape, static friction, and rolling friction, increasing with greater resistance to motion. The effect of particle shape on porosity was non-monotonic, with non-spherical particles exhibiting higher sensitivity to frictional variations. Packing porosity in a pile remained relatively uniform along the axial direction but increased radially from the center to the edge. In piles of triple-sized particles, strong size segregation was observed in both axial and radial directions, with radial particle mobility being more sensitive to particle shape. Finally, the charging process was studied in a scaled-down model of a bell-less blast furnace charging system. The behavior of sinter particles as they passed through the charging system was examined both temporally and spatially, revealing distinct motion characteristics across different components. The burden distribution in the throat was also evaluated radially and axially, revealing the particle size distribution patterns within the sinter layer and the mixing behavior of the sinter with the coke layer. Comprehensive analysis of particle behavior during the charging process indicated that variations in particle dynamics across different components specifically influence the distribution of the burden.
This work advances the understanding of fundamental burden properties in the blast furnace, particularly the influence of irregular particle shapes and the improved depiction of particle dynamics during charging. Nevertheless, several limitations remain, such as the far more complex particle geometries and size distributions of raw materials encountered in industrial practice, as well as the large scale of commercial blast furnaces. Overcoming these difficulties will require concurrent progress in experimental techniques, numerical modeling approaches, and computational capabilities.