Description
The service life of infrastructure is directly related to the traffic load. In the past, this traffic load has often been underestimated, resulting in significant damage to structures, especially steel bridges. In order to establish the paradigm of predictive maintenance in bridge engineering, prediction models for the state of damage and service life are necessary. In particular, the proportion of heavy trucks leads to a significant decrease in remaining service life. Therefore, knowledge of the composition and frequency of traffic is important. New methods from communications technology based on recurrent neural networks allow rapid identification of specific signals in monitoring data.
A bidirectional Long Short-Term Memory Network (BiLSTM) is used to identify different up to six vehicle types from the drive-by vehicle signals. The method was applied to monitoring data of drive-by data from a German wide span steel bridge with an orthotropic deck (Rhine bridge Neuenkamp). Validation was performed based on weight-in-motion data in combination with a finite element analysis. The extracted data allows a detailed analysis of the classification of vehicle types, sequence effects and clustering of vehicles. The gained data can later be used in digital twin applications.
