ABSTRACT
Multiple breakthroughs have emerged across various applications due to deep learning. These advancements can also be adapted for Bridge Weigh-In-Motion (BWIM) systems. A deep learning-based algorithm has been developed previously by the authors, which can be integrated into these systems. That algorithm shows promising results when compared to so-called static methods. One of the main contributions of our developments is handling the error of the preprocessing steps more efficiently. A developed axle load estimator is presented in this paper that builds on our previous solution, and which was tested in operation on a cable-stayed bridge. The algorithm handles multi-lane scenarios more efficiently when compared to our previous solution. The comparison of this novel algorithm and Moses’ algorithm is also presented in this paper. Our solution achieves lower error rates both on a synthetic dataset, and a real measurement-based annotated dataset.
