Bayesian-optimized deep learning model for real-time spatial distribution identification of vehicle axle load

Abstract

Real-time identification of the spatial distribution of vehicle axle loads on the bridge is vital for bridge health monitoring and maintenance. This paper proposes an identification method for the vehicle axle load position with known wheelbase based on a deep learning method. Firstly, the vehicle kinematics equation is established based on the bicycle model, and the axle load distribution identification problem is converted into a state estimation problem of three degree-of-freedom rigid body. Then, the geometric shape of the vehicle is reasonably simplified, and an indirect calculation method for the axle load point position is proposed, based on which a data set is established. Next, a Bayesian-optimized deep learning model is constructed to predict the coordinates of axle load rectangle center. Finally, the optimal trajectory of the predicted axle load rectangle center is estimated by the Kalman filter, and the real-time identification of the axle load spatial distribution is realized by combining the optimized trajectory and the vehicle kinematics equation. The effectiveness and accuracy of the proposed method are tested through the field test.