Towards autonomous sensor and data analytics systems for bridge structural health monitoring

Abstract

Adoption of effective bridge structural health monitoring (BSHM) systems offers a route to assist in the safe and economic operation of bridges. A number of key limiting factors affect the practical implementation of any BSHM approach, and must be overcome for their wider deployment, particularly in low-resource environments. Most notably, these include the energy consumption of the instrumentation and embedded processing, as well as coping with the data production, transmission, and storage requirements. This work seeks to overcome the constraints around energy consumption and data production, by creating efficient approaches leading to the development of BSHM systems for low-resource environments with lower costs, fewer compute resources, and limited data communications.

A direct damage detection method is proposed and offers enhanced monitoring performance with both lower computational and instrumentation costs. Long-term monitoring is enabled by a signal processing approach, which allows maximum rotation amplitudes to be extracted from DC acceleration measurements under ambient/live traffic loading. The earth movers' distance is shown to provide a low-computational cost means of quantifying the post-damage shift in the statistical distributions of maximum rotation amplitudes. This method is used to produce the first results in rotation-based BSHM to show that grouping rotation signals by axle count allows the same clarity of results as compared to looking at the full population of trucks, thereby reducing the generated data. A pilot study of the approach on an in-service short-span bridge shows, for the first time, the feasibility of extracting rotation signals from direct measurements under live traffic loading for long-term monitoring.

An event detection method was then created to allow event detection capability to be built into existing bridge rotation monitoring instrumentation. This low-cost method is based on the generalised variance over a sliding window of tri-axial accelerometer signals. It can be implemented using available instrumentation and low-computation resources, thereby avoiding the additional cost and energy usage of dedicated traffic-detection hardware. The method is shown to be robust to the effects of sensor noise and rotation signal variability using detailed numerical simulations, which varied sensor location, vehicle speeds and vehicle type. The approach was successfully applied to two in-service highway bridges using signals collected by end-of-span accelerometers.

Finally, to allow the more efficient use of energy in a low-resource sensing environment, a number of sampling methodologies for long-term BSHM have been proposed and evaluated, including random sampling and adaptive sampling of traffic-volume and air temperature information respectively. Extensive numerical simulation results allow us to compare their relative performance using our earth movers' distance--based damage detection method and indicate that sampling methods that give larger numbers of vehicle events provide better damage detection capability. These methodologies have been evaluated for robustness to confounding effects, i.e. varying temperatures and vehicle loads. As this work specifically focuses on long-term BSHM approaches, bridge rotation data was simulated including: (1) implementation of a more realistic sensor noise model based on experimental DC accelerometer noise measurements; (2) creation of a stochastic model for vehicle loading based on a weigh-in-motion dataset to incorporate effect of varying vehicle loads; (3) development of an empirically tuned, temperature capable finite element model based on long-term measurements of air temperatures and bridge natural frequencies.

Date of Award	Dec 2023
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Engineering and Physical Sciences Research Council
Supervisor	Roger Woods (Supervisor) & David Hester (Supervisor)