Digital Twin Modeling and Transformer Prediction-Based Overall Particle Size Distribution Online Estimation

Assessing the total particle size distribution (PSD) on conveyors is vital for overseeing blast furnace production. Nevertheless, challenging environments and restricted detection capabilities restrict current online methods to estimating partial PSD, hindering the prediction of the total PSD. Diverging from segmentation techniques that focus on surface PSD, we propose a novel method combining digital twin modeling and local-global fused prediction networks for estimating the total PSD. Initially, we create a digital twin model of particle accumulation, extracting global field distribution characteristics. Subsequently, we train a particle segmentation model using surface images from detection devices like cameras to acquire surface PSD. Next, we present the Local and Global Feature Fusion-based Transformer Prediction Network to forecast the total PSD. The encoder integrates orthogonal fusion and PointNet to efficiently merge local and global attributes. Finally, we verify the precision and efficiency of our methodology through thorough ablation experiments.