| With high overall rigidity,excellent wind resistance and unique advantages in economics and aesthetics,cable-stayed bridges have become the preferred bridge type for medium and large span bridges gradually.Stay cables are the main load-bearing members of cable-stayed bridges and play a vital role in the overall stability and safety of the bridge.During long-term service of a stay cable,on the one hand,the plastic envelope cracks under external environments such as rainwater and sunshine,which causing electrochemical corrosion between rainwater and steel wire of cables;On the other hand,under sudden incidents such as ship collision and damper drop,the cable force will undergo a sudden change;Under the action of vehicle,wind and other dynamic loads,the cable force is relatively stable,but fatigue accumulation is prone to occur under long-term conditions.These potential factors seriously threaten the service safety of the cable.Surface defects and time-varying cable forces are the main causes of corrosion damage,sudden change in cable force and fatigue accumulation for cables;Besides,they are also important reference indicators for the management and maintenance of cable-stayed bridges.Therefore,this paper aims at studying the method of identifying the apparent defects and time-varying cable force of cable-stayed bridges.The main contents of this paper include:Based on the ensemble learning method in deep learning,an improved classification network for surface defects of cables,plastic envelope was proposed which establish a deep-level convolutional neural network to extract the deep characteristics of the surface defects from a large number of surface defect images of cables,plastic envelope to reach automatic classification of the defects.The HRNetw32,EfficientNet b-4 network and SCSE ResNeXt101 network as the three final classification networks.Input raw images into HRNetw32,EfficientNet b-4 network and SCSE ResNeXt101 network for training,and save the optimal model parameters.In the model inference stage,use three models to predict if the image has defecs and generate the final result by voting or weighting to improve the accuracy and robustness of the network.Based on the object detection technology in computer vision,an improved cable surface defects location method called CMR(Cascade Mask RCNN)model is proposed.Three cascaded IoU thresholds are set to solve training overfitting and mis-match problems resulting from the inappropriate IoU setting by increasing the quality of the samples gradually.Deformable convolution is introduced to increase the model’s receptive field to improve the model’s ability to detect defects with large differences in shape distribution.At the same time,the centerline extraction algorithm and the neighborhood search algorithm are combined to extract the pixel features of the defect region identified by the CMR model.Based on the modal decomposition theory,a novel time-varying cable force identification method is proposed which decomposes the original signal into a finite eigenmode function IMF at different time scales,so as to obtain the change of the signal at different time scales.A fast Fourier transform is performed on the vibration acceleration signal of the cable to obtain the fundamental frequency.Based on the fundamental frequency,the signal is decomposed into signals with different modes by using a variational mode decomposition algorithm.Calculate the instantaneous modal frequency and bring it into the cable force calculation formula to get the time-varying cable force value.The scale and full scale models of stay cables are established respectively to validate the time-varying cable force identification method in the catastrophe phase and the stationary phase.Finally,two real bridge tests were carried out to further verify the feasibility of the proposed cable surface defects identification method and time-varying cable force identification method in practical scenarios.Firstly,combining the Cable Inspection Robot with the proposed method for identifying surface defects of the cable’s envelope,defects segmentation,pixel information extraction of defects,and rough estimation of the defects’physical size are cascaded together to form a complete detection and analysis system for surface defects inspection of cable’s envelope.At the same time,a real cable-stayed bridge inspection is carried out;The surface images of the cable’s envelope are obtained by the robot,and transmitted to the server deployed with the surface defects identification algorithm for identification,which output identification results of surface defects.Then,the cable force identification method is used as an auxiliary monitoring method in the cable-stayed bridge bridge replacement proj ect,and compared with the historical cable force value,further verifying the feasibility of the cable force identification method. |
