Railway traffic characterisation data based on weigh-in-motion and machine learning: A case study in Portugal.

Metallic railway bridges built during the 20th century are ageing, but their continued operation is essential for enhancing the capacity of the network. Structural degradation phenomena are a cause for concern, particularly fatigue, which requires accurate knowledge of service railway loads for structural integrity assessment in order to avoid unnecessary strengthening or premature bridge replacement. Quantifying present traffic scenarios is, therefore, critical and provides a basis for deriving past and future scenarios, given the challenges of implementing permanent weighing systems. Weigh-In-Motion (WIM) approaches are used to periodically collect data on train loads and geometry, generating large datasets containing information such as axle loads, axle spacings, and train speeds. However, identifying train types in these data is often a manual and laborious task, with this information being crucial not only for structural assessment, such as fatigue analysis, but also for broader considerations within the railway sector, including economic and social impacts. This paper presents the outcomes from a WIM system installed on the Alcácer do Sal bridge in Portugal to capture real-time train data, which was then post-processed through an automated Machine Learning (ML) approach for train classification. The resulting information is valuable not only for the national context but also for other countries with comparable traffic characteristics.