Generation of 3D realistic geological particles using conditional generative adversarial network aided spherical harmonic analysis

The reconstruction of 3D realistic geological particles remains a significant challenge in the field of granular mechanics. Specifically, numerous spherical harmonic (SH) based generation frameworks have been proposed to synthetic new particle shapes retaining majority particle morphology yet having a certain variety. However, given the fact of assuming one or more established distributions or ignoring secondary particle features, the regenerated particles inevitably lose certain diversities. To address this issue, the deep learning method, conditional generative adversarial network (CGAN) was introduced to the SH analysis for particle shape regeneration. Three kinds of sand particles were synthesized and compared with their real mother particle samples concerning the distribution features of SH coefficients and particle shape parameters for validation. Results prove the proposed method has a good reliable and diverse regeneration performance. This approach is promising to facilitate a more reality closer research on 3D particle -related issues in the future.