Virtual Testing of Buckling-Restrained Braces via Nonlinear Autoregressive Exogenous Neural Networks

Abstract

An artificial intelligence model utilizing feedforward back-propagation (FFBP) and nonlinear autoregressive exogenous (NARX) artificial neural networks (ANNs) is presented to model the nonlinear behavior of buckling-restrained braces (BRBs). The NARX ANN is developed using normalized time-delayed inputs and outputs to predict normalized brace forces during load reversals. The values of brace forces are denormalized via an auxiliary FFBP ANN. The training and testing of the proposed model (i.e., the NARX and FFBP ANNs) are performed using experimental data from BRB specimens tested at the Pacific Earthquake Engineering Research (PEER) Center. Experimental data from one specimen is used in the model developing (training) stage. In addition, three sets of data are used to test the model’s learning and generalizing abilities. Brace deformations are used as the network input to estimate the resulting brace forces. The network performance with different parameters is evaluated to arrive at an optimized architecture that best models the phenomenon. The nonlinear hysteretic behavior predicted by the ANN model shows excellent agreement with the experimental results for the training sample. The generalization and prediction capability of the proposed model is further demonstrated by predicting the hysteretic behavior of the testing samples with noticeable accuracy. The presented model is a powerful tool for virtually testing BRB specimens. Such a tool supplements the traditionally available experimental tools for BRB performance investigation. The developed modeling technique facilitates the BRB design and performance investigation processes by minimizing the need for, and extent of, experimental testing.