Application of neural networks for localization a concentrated load on the surface of a deformable body based on strain data

In this paper, we consider an inverse problem of deformable solid mechanics related to the search for boundary conditions with known information about the stress-strain state. A specific version of the inverse problem consists in finding the point of application of a concentrated load on the surface of the body with deformation values known at a finite number of points. As an example, we consider a thin rectangular plate rigidly fixed on two opposite sides. A feedforward neural network is used to build an algorithm for solving the problem under consideration. The necessary information about deformations is based on the finite element method. The result of the performed research is a comprehensive analysis of the factors determining the accuracy of solving the problem under consideration: the configurations of the monitoring system, the amount of training data and the initialization phase.