Damage Identification in Steel Girders of Highway Bridges Utilizing Vibration Based Methods and Convolution Neural Network in the Presence of Noise

Abstract

The vibration-based damage identification method utilizes changes in the vibration properties of a structure to detect damages. The presence of noise makes the use of these methods unreliable. Therefore, it is necessary to develop and apply a robust technique in noisy conditions. The main purpose of the proposed method in this study is to investigate the effect of noise on highway bridges and reduce its effects in determining the precise and correct location and severity of damages on these types of bridges. Therefore, a dual-criteria method based on modal flexibility change (MF) and modal strain energy (MSE) damage index is considered as the bases for training convolution neural network (CNN). This method aims to identify more accurate the damage location and intensity with and without the effect of noise. The feasibility of the proposed method is indicated on a validated FE model applied to the portion of the I-40 bridge as a sample of steel girders highway bridge by its application to a range of damage scenarios. The numerical simulation of damage scenarios is utilized to achieve both noise-polluted damage indexes for training CNN. The well-trained CNN is then applied to double-check the location and attain the intensity of unknown single and multiple damages (up to four simultaneous damages) in noisy conditions. The results demonstrate that dual criteria damage indexes along with CNN can practically and accurately identify unspecified location and severity of single and multiple damage scenarios in the presence of noise.