\u017diga Kolar, David Susi\u010d, Martin Kone\u010dnik, Domen Prestor, Tomo Pejanovi\u010d Nosaka, Bajko Kulauzovi\u0107, Jan Kalin and Matja\u017e Gams<\/p>\n
Abstract
Accurate vehicle axle weight estimation is essential for the maintenance and safety of transportation infrastructure. This study evaluates and compares the performance of various algorithms
for axle weight prediction using time-series data. The algorithms assessed include traditional machine learning models (e.g., random forest) and advanced deep learning techniques (e.g., convolutional neural networks). The evaluation utilized datasets comprising time-series data from 10 sensors positioned on a single lane of a bridge, with the goal of predicting each vehicle\u2019s axle weights based on the signals from these sensors. Each algorithm\u2019s performance was measured against the OIML R-134 recommendation, where a prediction was classified as accurate if the error was within \u00b14 percent for two-axle vehicles and \u00b18 percent for vehicles with more than two axles. Tests were conducted on several bridges, with this paper presenting detailed results from the Lopata bridge. Findings indicate that deep learning models, particularly convolutional neural networks, significantly outperform traditional methods in terms of accuracy and their ability to adapt to complex patterns in time-series data. This study provides a
valuable reference for researchers and practitioners aiming to enhance axle weight prediction systems, thereby contributing to more effective infrastructure management and safety monitoring.<\/p>\n
\u00a0<\/p>\n\n\n\n