BUILDING NEURAL NETWORKS TO PREDICT THE BEHAVIOR OF ELASTIC FLATE STRUCTURES WITHOUT HISTORICAL DATA

Abstract

In this study, we establish a neural network-based computational framework to predict the displacement of elastic plates under mechanical loads. A neural network is constructed by adding physical laws that describe the mechanical behavior of materials, through which the neural network can learn itself without historical data. Several numerical examples are investigated and compared with the finite element method (FEM) to demonstrate the effectiveness of the proposed method.