Neural network architecture optimization by analyse the mechanical characteristics of civil engineering materials

Our proposal for this paper is to diversify the architecture of neural networks in order to optimize it and to obtain the best performing configurations that minimize errors of predictive mechanical properties of polymeric concrete. In this paper different architectures of artificial neural networks will be used for investigating the flexural strength of polymer concrete with fly ash and fibres. In the present study the epoxy resin was used for binding the aggregates. In the composition were introduced near the fly ash, used as filler, the cellulose fibres for improving the properties. The characteristics of these artificial neural networks architectures will be presented and analysed in order to choose the one that minimizes the prediction errors of the mechanical characteristics of polymer concrete and presents an optimal configuration that allows a high working speed that can adapt to this type of approaching the problem with a strong nonlinear character. By using this modern predictive methods, it was attempted to highlight its basic character - learning by examples specific to the human brain but much more efficient due to the mathematical models of the activation functions and the interconnection between the layers of neurons that exponentially increase their ability to adapt to strong nonlinear phenomena. Thus one can say that such a prediction helps to reduce the number of real experiences and can greatly contribute to obtaining the optimal configuration of parameters necessary to obtain a desired mechanical characteristic of the analysed concrete.