Noncontact cable tension estimation using edge recognition technology based on convolutional network

The cable is the main load-bearing component of cable-stayed, suspension, and tied-arch bridges. The tension of the cable is crucial to the bridge's overall safety. Therefore, achieving an accurate and rapid cable tension estimation is of great practical importance. Currently, the method of estimating cable tension based on image processing technology avoids the disadvantages of conventional estimation methods, such as high cost, difficult sensor installation, and potential structural damage. However, for estimating cable tension based on image processing technology, a complex optical background may reduce the accuracy and stability of cable tension estimation. A method for estimating cable tension based on a deep convolutional network is proposed to solve the problem. The convolutional network is initially used to identify cable edges. Second, analyzing the edge pixels' motion identifies the cable's fundamental frequency. Finally, the cable tension is estimated using the correlation between fundamental frequency and cable tension. A field experiment was conducted on cable-stayed and tied-arch bridges in Chengdu to validate the proposed method's applicability and accuracy. The results of the experiments indicate that the edge detection effect of the proposed method is demonstrably superior to that of the Canny algorithm under various optical backgrounds and that the cable tension relative difference between the proposed method and the conventional method based on accelerometer is within 5%.