Assessment of the drag model for nanoparticle agglomerates based on the BP neural network method

Traditional empirical drag coefficient models, although effective under certain conditions, fail to account for the complex interparticle interactions, agglomerate structures, and slip-correction effects of nanoparticles, resulting in inadequate accuracy for predicting the drag coefficient of nanoparticle agglomerates. In this work, the derived semi-empirical drag coefficient model is combined with shape and Cunningham slip-correction factors to formulate the drag coefficient for nanoparticle agglomerates. Relevant agglomerate parameters are extracted via nanoparticle fluidization experiments and image-processing methods to assemble a comprehensive dataset. This dataset is used to train a BP neural network, whose performance is then compared with traditional empirical correlations such as those of Haider and Ganser to evaluate the optimized model's predictive capability. The trained network is subsequently distilled into an explicit mathematical model for drag-coefficient prediction through symbolic regression. Finally, the influence of each input parameter on prediction accuracy and error is analyzed to elucidate the relationships between key features and model performance.