Sample-Driven Design of Multi-porous Structures

In order to design porous models with specific functions, good internal connectivity and natural structures, this paper proposes a new learning-based porous structure modeling method.Firstly, the porosity and connectivity are selected as the evaluation indices.Then a structure with 6-adjacent voxel is designed as the parametric design unit, which is used to establish a manually labeled training database.Random decision forest(RDF) is utilized to learn the correlation model between sample structures and design targets.This correlation model is finally integrated into a scalable porous structure modeling framework.Experimental results show that the generalization capability of the RDF is able to give correctly judgments for those structures beyond the training data, which makes it possible to generate more natural porous structure to satisfy certain design goals.