Empirical Formulation of Ferrocement Members Moment Capacity Using Artificial Neural Networks

Abstract

In the two past decades, ferrocement members have been with a wide variety of uses in structural applications because of their unique physical properties (high surface-area-to-volume ratio and possible fabrication in any shape). In this study, two models were presented for a predict of the moment capacity of ferrocement members, one based on a back-propagation multilayer perceptron artificial neural network and the other proposing a new equation based on the multilayer perceptron network trained. These models with five input parameters including volume fraction of wire mesh, tensile strength, cube compressive strength of mortar, and width and the depth of specimens are presented. The results obtained from the two models are compared with experimental data and experimental equations such as plastic analysis, mechanism, and nonlinear regression approaches. Also, these results are compared with the results of the equations that researchers have proposed in recent years with soft computing methods (ANFIS, GEP, or GMDH). The prediction performance of the two models is significantly better than the experimental equations. These models are comparable to that of models provided with different soft computing methods to predict the moment capacity of ferrocement members. The result of this research has proposed a general equation with less mathematical complexity and more explicit.

Keywords

• moment capacity
• ferrocement members
• artificial neural networks
• flexure failure