Identification of railway bridge modal properties based solely on acceleration data from traversing trains

Railway bridges are vital components of rail infrastructure, yet many are aging and require effective monitoring. Traditional Structural Health Monitoring (SHM) methods, while accurate, are often costly and impractical for large-scale deployment. This study proposes a novel indirect monitoring approach to identify bridge eigenfrequencies using only acceleration data from trains. Leveraging vehicle–bridge interaction, the method proposes a two-step identification framework. First, a Bayesian filter is employed to estimate the unknown vehicle states and contact forces, removing the need for direct force and vehicle states measurements, which is commonly assumed as a prerequisite in previous works. Second, the estimated states and contact forces are used to define the inputs and outputs in the identification of the bridge modal properties. To this end, this study examines two identification approaches: an input/output, covariance-driven subspace identification (IO-COV) algorithm and an Auto-Regressive Moving Average with eXogenous input (ARMAX) method. The considered IO-COV approach additionally includes an uncertainty quantification step, used to filter out spurious frequency estimates. The approach is validated through numerical simulations. More importantly, the proposed methodology is verified on data collected via the Swiss Federal Railways diagnostic vehicle from an operational bridge structure, namely the Aarebrücke bridge in Uttigen, Switzerland. The identified frequencies align well with reference values, confirming the feasibility of indirect methods for accurate and cost-efficient bridge monitoring. The study addresses key challenges such as unknown inputs and high vehicle speeds, while also outlining remaining limitations and future directions. • Indirect estimation of railway bridge modal properties using acceleration data collected from traversing trains. • Bayesian Dual Kalman Filter framework to infer vehicle states and contact forces without direct sensing. • Identification of bridge eigenfrequencies via covariance-driven subspace identification (IO-COV) and ARMAX methods. • Validation on the Aarebrücke railway bridge (Uttigen, Switzerland) using real SBB diagnostic vehicle data.