Abstract

Support generation is one of the crucial steps in 3D printing to make sure the overhang structures can be fabricated. The first step of support generation is to detect which regions need support structures. Normal-based methods can determine the support regions fast but find many unnecessary locations which could be potentially self-supported. Image-based methods conduct a layer-by-layer comparison to find support regions, which could make use of material self-support capability; however, it sacrifices the computational cost and may still fail in some applications due to the loss of topology information when conducting offset and boolean operations based on the image. In order to overcome the difficulties of image-based methods, this paper proposes a surfel convolutional neural network (SCNN)-based approach for support detection. In this method, the sampling point on the surface with normal information, named surfel (surface element), is defined through layered depth-normal image (LDNI) sampling method. A local surfel image which represents the local topology information of the sampling point in the solid model is then constructed. A set of models with ground-truth support regions is used to train the deep neural network. Experimental results show that the proposed method outperforms the normal-based method and image-based method in terms of accuracy, reliability, and computational cost.