General Tikhonov regularization-based load estimation of bridges considering the computer vision-extracted prior information

Estimating the load distribution of a bridge structure enables to evaluate the in-service state and predict the structural responses. This paper develops an iterative strategy to inversely estimate the traffic load distribution of a bridge from limited measurements. The computer vision technologies, including the YOLO network-based object detection and a pixel coordinate-based positioning approach, are used to locate the vehicle positions on the bridge deck and form a prior information vector of the input positions. Then, a generalized Tikhonov regularization method is proposed to estimate the load distribution using the bridge response and prior information. The regularization parameter is determined by the L-curve method. The fusion of computer vision and regularization can improve the load identification accuracy and reduce the overfitting effect. The developed approach is applied to numerical and experimental examples under various load conditions. The load can be accurately identified in all cases, and the full-field responses of the structures can be reconstructed with minor errors.