We consider a material made of alumina platelets. The platelets shape is unknown, but their size is of the order of a few nanometers. In a first step, we extract information from noisy and blurry 2D images obtained by transmission electron microscopy. These images were acquired on samples with thickness about 300 nm. The platelets can not be distinguished. We measure the average correlation between points on the image s as a function of the distance between the two points. In a second step, we generate numerically 3D microstructures representative of the real material at the nanoscale. Two scales are present in the microstructure model: that of the platelets and that of clusters of platelets, where the particles are aligned with their neighbors. This feature is necessary to recover the measurements determined on transmission image. Varying geometries are simulated depending on the shape of the platelets. Finally, we use porosimetry data and a tomography model to validate our model.