Development of a moving vehicle identification framework using structural vibration response and deep learning algorithms

Abstract

The accurate identification of vehicle position, velocity, and mass on bridges is crucial in the field of structural health monitoring (SHM). Existing methods for moving vehicle identification primarily rely on staged approaches, which may not adequately address multiple vehicle scenarios. This study is the first to propose a framework that simultaneously considers vehicle velocity, mass, and location as supervised learning problems using structural vibration response. The framework comprises three phases: data generation and preparation, network building and training, and prediction. In Phase 1, a labeling strategy is introduced to convert arbitrary moving vehicle scenarios into a unified, trainable format. Phase 2 presents a novel loss function specifically designed to measure the discrepancy between the network output and the label. Lastly, Phase 3 introduces a data post-processing workflow that converts the network output into moving vehicle information. To demonstrate the effectiveness of the proposed method, the identification accuracy is quantified and analyzed using five different indexes under different vehicle numbers, velocity intervals, and mass intervals. Furthermore, the influences of some key variables, the modeling error and model simplification are discussed in detail. The proposed approach surpasses conventional methods by overcoming staged identification limitations and accommodating multiple vehicles, while efficiently identifying their velocity, mass, and moving range on the bridge without requiring initial conditions.