Formation of the structure and optimal parameters of the neural network for determining the composite plate deflections

Abstract

The paper defines the structure of an artificial neural network (ANN), the required amount of data for training and algorithms for their processing to ensure the ability of the network to recognize dependencies between the structure of the composite material and the quantitative parameters of the deformed state under a given load with acceptable accuracy. To test the network, the problem of determining the deflection of a three-layer (glass-film-glass) laminate plate with a wide range of changes in physical and mechanical characteristics and geometric dimensions under conditions of three-point bending was used. It is shown that the most effective is the ANN consisting of two hidden layers with 12 and 6 nodes in the layers, respectively, when processing data based on the Levenberg-Marquardt algorithm. At the same time, the error of the obtained results lies within 0.5% for data from the range used for training the network and within 5% for data outside this range.