An interactive walkthrough of a vehicle-bridge interaction and Bridge Weigh-in-Motion model.
Live demo: https://bypire.github.io/bridgebeat/
Five sections:
- Weigh. A truck crosses a bridge. The deck deflects by less than a millimetre, and that deflection, measured at a single point, recovers the truck's weight while it is moving (Bridge Weigh-in-Motion). Increasing the speed increases the dynamic amplification and the recovery error.
- Diagnose. A loss of stiffness lowers the bridge's natural frequency. Increasing the damage moves the spectral peak to a lower frequency, which a passing vehicle can detect without sensors on the bridge.
- Real data. The measured natural frequency of the KW51 bridge (Leuven), recorded daily for sixteen months and coloured by deck temperature. A retrofit in 2019 raises the frequency by about two percent.
- Why it matters. Fatigue damage scales with roughly the cube to fourth power of axle load, so a small overloaded minority causes most of the wear, and B-WIM identifies those trucks.
- A viaduct under traffic. A 500 m multi-span viaduct carrying a stream of cars and trucks, with a running tally of load, overloads, and accumulated fatigue cost.
This is the front end for a 2D vehicle-bridge interaction and B-WIM model: an Euler-Bernoulli beam finite-element model (consistent mass and stiffness, Rayleigh damping), a quarter-car coupling, RK4 and Newmark-beta integration, and a Moses, Tikhonov-regularised, and Bayesian inverse for the axle weights. It is verified against closed-form references (Frýba, modal frequencies) and validated against the measured KW51 monitoring dataset. The numpy-only physics core is in a separate repository.
Derived data only; no raw monitoring records. The KW51 tracked modal frequencies are from Maes and Lombaert, Monitoring data for railway bridge KW51, Zenodo 3745914.