New Mathematical Formulation of Nonlinear Unsteady Wind Loads on Long-Span Bridge Decks under Nonstationary Winds Using Time-Delay Neural Network

Abstract

This paper presents a novel mathematical formulation of unsteady wind loads on bridge decks by using the neural network technique while incorporating the concurrent effects of nonstationary winds and aerodynamic nonlinearity. For that, a time-delay neural network (TDNN) is developed by recognizing the inputs and target outputs, wherein the inputs entail the wind speed fluctuating components and self-excited motion components, whereas the target outputs entail the buffeting load components. A typical sigmoidal function provided by the hyperbolic function is utilized to simulate the nonlinear features of the wind–bridge interaction (WBI) system. Finally, an elegant formulation for the nonlinear unsteady aerodynamic wind loads considering the nonstationary wind effects is developed in terms of synaptic weights of neurons and biases. The proposed formulation of winds loads has also been applied to a full-scale long-span suspension bridge under real typhoon winds. The buffeting analysis results are also compared with the measured displacement data, which shows the efficacy of the proposed wind load model for real-life bridge structures.