Approximation Of Stress-Strain Curve Of Rubber-Like Material Using An Artificial Neural Network

Abstract

Neural networks are widely used in various fields, including computer simulation and mechanics. Neural networks allow solving problems in mechanics with parameters that are poorly formalized. Rubber-like materials have a complex deformation curve. Therefore, the approximation of such a curve can be performed using a neural network. The objective of the work is to create a set of neural networks with different input parameters and internal structure. These neural networks are used for interpolation and approximation of experimental data. To do this, it is constructed and trained neural networks of direct distribution. The training is carried out by the method of back error propagation. The data set was generated based on the results of the experiment. For testing, all networks received the same dataset that was not used during the training but was known from the experiment. This allows determining the network error for the loading cycle area and the root mean square deviation. It is described in detail the type of network and its topology, the method of training, and the preparation of the training sample.