Abstract:The sling is widely used in large-scale structures. And the characteristics of its long span and low stiffness make it prone to significant vibration caused by environmental excitation, thereby threatening the safety and durability of the structures. Aiming at the problem, a non-contact high-precision vibration displacement measurement method integrating high-speed visual perception and deep convolutional neural network (DCNN) is proposed and verified. The core of this method lies in using a high-speed camera to collect a continuous image sequence of sling vibration, and automatically learning the spatiotemporal features in the image sequence through a pre-trained DCNN model directly to map and output the physical displacement. The sling of a long-span continuous beam arch bridge in a high-speed railway is selected as the research object for on-site measurement. The markers are set on the surface of the sling, and the high-speed cameras are used to record its vibration video. After the video data is preprocessed, it is input into the DCNN model based on the improved ResNet-18 architecture for prediction. The pixel displacement output by the model is converted into physical displacement through on-site calibration. The measured results show that this method has successfully obtained the clear vertical vibration displacement time history curves of two slings. The vibration amplitude ranges are -1.6~1.2 mm and -3.2~2.4 mm respectively, and it is identified that its first-order natural frequency is 1.9 ~ 2.0 Hz. Compared with the synchronous measurement results of traditional sensors, the average relative error of this method is 6.15%, which is significantly better than the template matching and optical flow methods of traditional image algorithms, and has broad application prospects.
