SimuStruct: Simulated Structural Plate with Holes Dataset with Machine Learning Applications

Bruno Alves Ribeiro, João Alves Ribeiro, Faez Ahmed, Hugo Penedones, Jorge Belinha, Luís Sarmento, Miguel Anibal Bessa, Sérgio Tavares

Keywords: Plate with Holes, Structural Analysis, Computational Design, Machine Learning, Finite Element Method

TL;DR: Simulated Structural Parts Dataset

Abstract: This paper introduces the SimuStruct dataset, which consists of 2D structural parts along with their respective meshes and the outputs of numerical simulations for various properties such as linear and elastic material, boundary and loading conditions, and different levels of refinement. The dataset includes the classic case of plates with holes, which is a common and analytically resolvable 2D case found in different mechanical design applications. SimuStruct provides a diverse and realistic dataset as it includes multiple cases for different loading and boundary conditions, various material properties, and mesh refinement levels. Furthermore, it is flexible, versatile, and scalable as all algorithms and codes, where each case is solved using standard Finite Element Method (FEM) with the open-source package FEniCS. The primary aim of the SimuStruct dataset is to serve as training and evaluation data for Machine Learning (ML)-based methods in structural analysis and optimal mesh generation, thereby supporting the development of ML-based optimal mechanical design solutions. In this paper, it is also presented an application of SimuStruct to train and test a Graph Neural Network (GNN) model to predict stress-strain fields, demonstrating the dataset's potential for use in structural analysis. The SimuStruct dataset will facilitate the integration of the Mechanical Engineering and Machine Learning communities and enable faster and more efficient research in the computational design field.