A Bridge Vibration Measurement Method by UAVs based on CNNs ‎and Bayesian Optimization‎

Abstract

A bridge vibration measurement method by Unmanned Aerial Vehicles (UAVs) based on a Convolutional Neural Network (CNN) and Bayesian Optimization (BO) is proposed. In the proposed method, the video of the bridge structure is collected by a UAV, then the reference points in the background of the bridge and the target points on the bridge in the video are tracked by the Kanade-Lucas-Tomasi (KLT) optical flow method, so that their coordinates can be obtained. The BO is used to find the optimal hyper-parameter combination of a CNN, and the CNN based on BO is used to correct the bridge displacement signal collected by the UAV. Finally, the natural frequency of the bridge is extracted by processing the corrected displacement signals with Operational Modal Analysis (OMA). Moreover, a steel truss is used as the experimental model. The number of reference points and the shooting time of the UAV with the optimal correction effect of the BO-based CNN are obtained by two groups of comparative experiments, and the influence of the distance between structure and reference points on the correction effect of the BO-based CNN is determined by another group of comparative experiment. The static reference points are not required for the proposed method, which evidently enhances the applicability of UAVs; the conclusion of this paper has great guiding significance for the actual bridge vibration measurement.

Keywords

• bridge vibration
• unmanned aerial vehicles
• correction effect
• convolutional neural networks
• bayesian optimization‎