Answers are grounded in this paper and cite their sources.
Deep learning-based moving load identification (MLI) methods can circumvent the time-consuming inverse operation of matrices by directly learning the mapping relationship between vehicle-induced bridge responses and vehicle loads. However, these pure data-based approaches often inadequately integrate physical prior information, and the complex spatiotemporal features of bridge responses are incompletely extracted in existing models. Therefore, a novel MLI framework named physics-aware loss and multi-scale spatiotemporal fusion network (PhyMS-Net) is proposed in this study. Firstly, a vehicle-bridge coupling system (VBCS) is established to obtain the dataset containing vehicle loads and vehicle-induced bridge responses, followed by a global extreme value proportional scaling to retain the physical prior information on the amplitude ratio among multiple axle loads. Based on the prior information on moving vehicle loads, a joint error penalty function is defined for decoupling the static and dynamic load components. Secondly, the proposed method is validated by numerical simulations. The effects of road surface roughness, loss function weighting coefficients, training data, etc., are also studied. Finally, similarity law for the VBCS is conducted in experimental verifications to construct the training set, allowing for a reasonable identification of vehicle axle weights. The results indicate that the MLI accuracy has been enhanced by the proposed method, and the similarity law employed in experiments is reasonable. The loss function weighting coefficient set as α = 0.7 can lead to a higher MLI accuracy and reduce the computation cost in both numerical simulations and experimental verifications.
No takes yet. Share an insight, caveat, or question.
Ran et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: