APPLICATION OF 3D CONVOLUTIONAL NEURAL NETWORK IN PREDICTING SHRINKAGE STRESSES AND DISPLACEMENTS IN MONOLITHIC CONCRETE SLABS ON BASE

Abstract

The aim of this study is to demonstrate the capabilities of artificial convolutional neural networks (CNNs) in problems related to mechanics, in particular, in the design of monolithic slabs on a base. In this paper, for the first time, an approach based on the use of a voxel description of the object under study is proposed. In a number of cases at the design stage, the presence of technological holes of various shapes is envisaged, the slab surface may have a complex geometric shape. Determination of the stress-strain condition in closed form in such cases is very labor-intensive or not achievable at all. This paper presents an alternative approach based on the application of 3D CNNs with U-Net architecture, which allows to obtain reasonably accurate predictions of shrinkage stresses and displacements in slabs in a simpler way compared to finite element methods. The work highlights the promising potential of U-Net architecture neural networks.