International researchers investigate more complex matters in digital fabrication, detailing their latest study in the recently published ‘Learning to Accelerate Decomposition for Multi-Directional 3D Printing .’ Delving into a subject of great interest for most users interested in creating complex geometries, the authors explain more about recent work designed to use a beam-guided search algorithm to find an optimized sequence of plane-clipping resulting in the need for ‘tremendously less supports.’ In some cases, no supports may be required at all.
Planar layers with fixed 3D printing direction usually require supports to prevent collapse; however, this added material can be a source of major hassle for even the most experienced users. To solve the problem in requiring supports, the research team created a new algorithm meant to decompose models into sub-components being printed separately in different directions for different components.
“The benefit of the proposed search algorithm is that it can avoid being stuck in local minimum compared to greedily searching the best result. Beam width b = 10 is empirically used to balance the trade-off between computational efficiency and searching effectiveness,” explained the authors. “Though conducting a parallel implementation running on a computer with Intel(R) Core(TM) i7 CPU (4 cores), the method still results in an average computing time of 6 minutes.”
A learning-to-accelerate framework ranks ‘candidate planes’ for the best results, and a new method converts trajectories to ‘pairwise comparisons for training.’ As a sidenote, the authors also mention that the efficiency offered in this work is actually ‘much better’ than previous work. Prior related work by other researchers has involved segmentation-based methods, multi-directional and multi-axis fabrication, and accelerated searches.
“The proposed method utilizes a learning-based method to train a decision-tree-based ensemble that can score the candidates of clipping,” explained the researchers.
Accessibility is key in this new system, with the model and the source codes all available to the public. A high-performance server is used with two Intel E5- 2698 v3 CPUs and 128 GB RAM, with all other tests employing an Intel Core i7 4790 CPU, NVIDIA Geforce GTX 980 Ti GPU and 24 GB RAM.
“We trained our model on the Thingi10k dataset repaired by Hu et al. Instead of training and evaluating on the whole dataset, we extract a subset of the dataset (2061 models) that satisfies every model in the selected dataset should have a few risky faces that can be processed by our plane-based cutting algorithm,” concluded the researchers.
“The computing time is reduced to 1/2 while keeping the results with similar quality. The experimental results demonstrate the effectiveness of our proposed method. We provide an easy-to-use python package and make the source code publicly accessible.”
Requirements for supports tend to be a source of consternation for users, leaving scientists to investigate the use of robotics to help eliminate support materials, exploring materials like resin , and improving ongoing work with material like water-soluble supports . What do you think of this news? Let us know your thoughts; join the discussion of this and other 3D printing topics at 3DPrintBoard.com .
[Source / Images: ‘Learning to Accelerate Decomposition for Multi-Directional 3D Printing