PoreFlow-Net: a 3D c`onvolutional neural network to predict fluid flow through porous media

•A 3D convolutional neural network is able to create a functional relationship between pore morphology and the steady state solution of the Navier-Stokes equation for laminar flow.•Four geometric features extracted from the binary image are needed to make the model robust.•A model trained only with spherepacks is able to perform accurately in different domains including non-consolidated samples, synthetic heterogeneous geometries, and sandstone and carbonate x-ray images.