Prediction of the transverse elasticity modulus of unidirectional composites through artificial neural network analysis using Rprop algorithm
Giorgio André Brito Oliveira1; Raimundo Carlos Silverio Freire Jr2; Eduardo César Bezerra Câmara2
1 Univerdade Federal do Rio Grande do Norte; 2 Universidade Federal do Rio Grande do Norte
doi:10.20906/CPS/CILAMCE2017-0711
Resumo
The artificial neural networks (ANN) is a new computational tool which has been used in several areas of Engineering. In Mechanical Engineering, among its newest applications is the evaluation of mechanical properties of composite materials. This work aims to create an ANN capable of predicting the transverse elasticity modulus (E2) of unidirectional composites, comparing two types of training algorithms of ANN, to verify which one has the best mathematical model. To achieve that, two groups of data will be analyzed using cross validation. Thus, two kinds of ANN architecture will be analyzed, one of them with three inputs, and a mixed model that combines a three input model with equations developed by Halpin-Tsai. These sorts of architecture were already evaluated using the backpropagation algorithm, on this work, another type was ascertained, called resilient backpropagation (Rprop). After the training, the results were compared and the ANN with Rprop training algorithm showed a high efficiency and relevance. The Halpin-Tsai model was utilized as a comparison parameter, thereby, higher correlation coefficients were observed as well as lower root mean square values.
Palavras-chave: Neural Network; Mechanical Properties; Rprop Algorithm