Self-adaptive Artificial Neural Network in Numerical Models Calibration

Abstract

The layered neural networks are considered as very general tools for approximation. In the presented contribution, a neural network with a very simple rule for the choice of an appropriate number of hidden neurons is applied to a material parameters’ identification problem. Two identification strategies are compared. In the first one, the neural network is used to approximate the numerical model predicting the response for a given set of material parameters and loading. The second mode employs the neural network for constructing an inverse model, where material parameters are directly predicted for a given response.