Jordan Journal of Civil Engineering

Nonlinear Seismic Response Approximation of Steel Moment Frames using Artificial Neural Networks

Authors:

Fatima Noori; Hesam Varaee;

Abstract:

Determining the seismic response of structures has always been one of the most critical structural and earthquake engineering challenges. In recent years, various methods with different performances in terms of accuracy and computational costs have been developed by researchers. Among them, accurate methods such as Nonlinear Time History Dynamic Analysis (NTHDA) require very high computational costs. Therefore, providing predictive models with sufficient accuracy will significantly reduce the computational demand, thus making the seismic analysis of structures more feasible. On the other hand, recently, Artificial Neural Networks (ANN) have been used to estimate various engineering issues and have successfully performed. Therefore, in this study, Multilayer Perceptron (MLP) neural networks with Back Propagation (BP) algorithm have been used to estimate the dynamic response of steel moment frame (SMF) structures subjected to seismic loads. Nonlinear responses of 3, 10, and 24 floor SMF structures are used as different examples for training ANNs. The responses of nonlinear frames are evaluated by the maximum acceleration of the earth in different amplitudes in the trained neural network. The section's modulus of elements is considered the input and the parameters obtained from the NTHDA, such as axial force, bending moment, rotation, hinge states, and hinge position in the structural members different ANNs. Hinge states and hinge location mainly have been studied as an innovative point. Comparison of the neural network estimation results and NTHDA results show that the trained neural networks can estimate the dynamic response of two-dimensional steel moment frames with acceptable accuracy.

Keywords:

Dynamic Response of Structures, Artificial Neural Networks, Steel Moment Frames, Nonlinear Dynamic Analysis, Time History Analysis, Multilayer Perceptron Neural Networks