Scalability of Learning Tasks on 3D CAE Models Using Point Cloud Autoencoders

Abstract

Geometric Deep Learning (GDL) methods have recently gained interest as powerful, high-dimensional models for approaching various geometry processing tasks. However, training deep neural network models on geometric input requires considerable computational effort, even more so if one considers typical problem sizes found in application domains such as engineering tasks, where geometric data are often orders of magnitude larger than the inputs currently considered in GDL literature. Hence, an assessment of the scalability of the training task is necessary, where model and data set parameters can be mapped to the computational demand during training. The present paper therefore studies the effects of data set size and the number of free model parameters on the computational effort of training a Point Cloud Autoencoder (PC-AE). We further review pre-processing techniques to obtain efficient representations of high-dimensional inputs to the PC-AE and investigate the effects of these techniques on the information abstracted by the trained model. We perform these experiments on synthetic geometric data inspired by engineering applications using computing hardware with particularly recent graphics processing units (GPUs) with high memory specifications. The present study thus provides a comprehensive evaluation of how to scale geometric deep learning architectures to high-dimensional inputs to allow for an application of state-of-the-art deep learning methods in realworld tasks.

CarHoods10k: An Industry-Grade Data Set for Representation Learning and Design Optimization in Engineering Applications

“…Examples are partial occlusion or scans, and noise, which often occur in point clouds obtained from 3D scanning in robotics or autonomous driving. More recent work has therefore explored the applicability of GDL for point clouds specifically in engineering applications [12], [23], [24].…”

“…Autoencoders have been proposed also for point clouds by Achlioptas et al [11], where a particular challenge was to make the architecture invariant to permutations in the ordering of the input [21]. Building on the work by Achlioptas, Rios et al adapted the point cloud autoencoder (PC-AE) for application in engineering contexts [23]. The authors showed that the learned compact representations could be successfully used in engineering tasks, such as optimization or building metamodels, [12], [23], [24].…”

“…Building on the work by Achlioptas, Rios et al adapted the point cloud autoencoder (PC-AE) for application in engineering contexts [23]. The authors showed that the learned compact representations could be successfully used in engineering tasks, such as optimization or building metamodels, [12], [23], [24]. Here, due to the compact representation of the geometry, global optimization algorithms require less function evaluations and unsupervised representation learning potentially leads to "problem-agnostic" representations that "capture more relevant design features than a human designer would suggest" [12].…”

“…As one approach to representation learning for design optimization and other engineering tasks, we used a deep autoencoder network for point cloud representations of 3D shape data (PC-AE) 1 was introduced by Rios et al [23] and is based on a model originally proposed in [11]. The PC-AE accepts a point cloud as input, that is an unordered set of points in 3D space, P = {p i }, i ∈ {1, .…”

Approximating the Steady-state Temperature of 3D Electronic Systems with Convolutional Neural Networks

“…One way to reduce memory requirements would be to use a smaller representation of the geometry instead of images [21]. Although some alternatives, like point clouds based on CAD geometries [22][23][24] or octrees were proposed [25], most available NN architectures are still developed for images.…”