This work presents a numerical workflow to generate a virtual packed bed made of non-spherical polydisperse particles, and subsequently predict its permeability. Wood chips were taken as an illustration. First, chips are sized before being recreated numerically. Then, using LMGC90, a DEM code, a packed bed made of those chips was generated. Once bed internal had been sampled, CFD tools belonging to the OpenFOAM library were used to mesh the geometry (snappyHexMesh) and compute fluid motion (simpleFoam). Finally, using numerical results, the bed permeability was computed in both Stokes and inertial regimes - turbulence being described by Launder-Reece-Rodi model. In parallel, experimental measurements of the permeability of a packed bed, made of the exact same wood chips, was carried out. These experiments were used as a reference to challenge numerical results. The permeability value delivered by the workflow is 16.0% higher than the experimental value. This value has to be compared with Kozeny-Carman equation estimation which overestimates bed permeability by 115%. Going one step further, this framework was successfully used to compute inertial effects constant of the Forshheimer equation for our packed bed. Throughout this article, a special care has been taken in explaining and evaluating the impact of all the key parameters, namely, number of particles that have to be sized, mesh refinement level, numerical domain dimensions. This workflow opens the door to numerical estimation of bed tortuosity, dispersion coefficients, volumetric heat exchange coefficients, and much more, using the particle size distribution as unique input data.