Prediction of compressive mechanical properties of three-dimensional mesoscopic aluminium foam based on deep learning method

Abstract

To achieve efficient and accurate prediction of the mechanical properties of aluminium foam, this study proposes a deep learning-based mechanical property prediction framework. The aluminium foam modelling plug-in based on the Voronoi model is selected to establish a three-dimensional solid model of aluminium foam, and the voxelization method is utilized to transform the geometric model of aluminium foam into a voxel model that can be recognized by the neural network. The mechanical properties of aluminium foam are calculated by the finite element method, including the stress‒strain response, densification strain and plateau stress. A deep learning sample dataset containing the aluminium foam voxel model and mechanical properties is established. The deep learning model based on a three-dimensional convolutional neural network (3D-CNN) is trained to identify the mesostructure features of aluminium foam, and a high-precision mapping relationship between the mesostructure and the macroscale mechanical properties is established. The research results show that the deep learning model established in this research can efficiently and accurately predict the mechanical properties of aluminium foam. Compared with conventional prediction methods, the prediction method based on deep learning has more advantages in efficiency and accuracy, and is an effective alternative to numerical simulation. In addition, the mechanical property prediction framework is highly scalable, extending the application scope of deep learning in the field of materials.