Nonlinear AutoRegressive eXogenous Artificial Neural Networks for predicting Buckling restrained braces force

Abstract

This paper proposes a novel approach for modeling the nonlinear dynamic behavior of Buckling-Restrained Braces (BRBs). The proposed approach is based on a combination of two architectures of Artificial Neural Networks (ANN) namely, Nonlinear AutoRegressive eXogenous (NARX) ANN and feed forward back propagation (FFBP) ANN. The proposed model predicts (outputs) the brace force at a certain time from the brace deflection and its history and the history of the brace force. The data used in training and testing of the model is acquired from the experimental testing of four BRB specimens. The proposed model is trained on data from one specimen while tested against the rest to demonstrate its learning and generalization capability. Optimum values for various network parameters are selected empirically to obtain the best network performance. The results show that the prediction error for the peak response (maximum tensile/compressive force) lies within ±5% confidence interval for all cycles.