Dynamic Load Identification on Prefabricated Girder Bridges Based on a CNN and Dynamic Strain Data

Abstract

The vehicle load on a bridge is a critical and dynamic variable. It adversely affects bridges, especially when overloading occurs. Bridges are prone to fatigue damage or collapse. Therefore, identifying the size and type of dynamic vehicle loads on bridges is critical for theoretical studies and practical applications, such as health monitoring, daily maintenance, safety assessment, and traffic planning. The paper proposes a method for identifying the dynamic load parameters based on a convolutional neural network (CNN) and dynamic strain data. The model is implemented in MATLAB. An initial finite-element model of a three-span precast beam bridge is established in the software ABAQUS and modified by combining the modal and experimental data derived from a segmental girder bridge. The dynamic strain response of the bridge under a moving vehicle load is simulated under different working conditions. The results are used as the training data of the CNN to identify the vehicle’s position, speed, and load on the bridge. The high prediction accuracy indicates the proposed model’s suitability for identifying the dynamic load parameters.