Fluid flow through porous media is important in applications such as water treatment, energy recovery, and material processing. Pore-to-pore conductance provides a reduced description of local hydraulic transport, but classical formulations generally rely on simplified throat geometries and a limited set of structural descriptors. In this study, a hybrid mechanistic–data-driven framework is developed for modelling pore conductance in uniform sphere packings with different porosities. The packed beds are generated using the discrete element method (DEM), and fully resolved computational fluid dynamics (CFD) simulations provide pore-scale pressure and velocity fields. Machine learning (ML) is used to predict the CFD-derived pore-to-pore pressure difference from throat-centre velocity and a progressively enriched description of the surrounding pore–throat–pore geometry. A comparative conductance is subsequently calculated from the predicted pressure response and evaluated against representative analytical formulations. Model performance is evaluated using five-fold cross-validation. In a controlled comparison using the same learner, progressively adding the particle-spacing and adjacent-pore descriptors reduced the pressure-difference RMSE from 0.150 to 0.0031 Pa, while the ML-derived comparative conductance showed closer agreement with the CFD-based reference than the selected analytical models over the sampled packings. Across the investigated particle Reynolds-number range, the conductance distributions show modest flow-condition-dependent shifts. The proposed framework therefore combines mechanistic pore-scale definitions with data-driven learning to extend conductance modelling for irregular pore–throat configurations in sphere packings.
