A numerical Investigation into the use of forced vibration due to Vehicular Loads for structural health monitoring of bridges

Structural Health Monitoring (SHM) aims to achieve early damage detection in bridges by characterizing changes in dynamic properties from the structural response acquired via sensors such as accelerometers. Currently, a new generation of SHM technologies is emerging to gather information in a low-cost and energy-efficient manner from the vibrations induced by vehicular loading. More specifically, the developments included in this thesis apply to technologies avoiding the need for a permanent installation on the bridge. They typically record measurements for a short duration of the response, which makes the limited info captured about the bridge highly sensitive to vehicular and road properties. The accelerometers can be placed on: (i) the bridge (i.e., direct measurements gathered through unmanned aerial vehicles), and (ii) the axles of a vehicle crossing the bridge (the so-called indirect measurements, or ‘drive-by’). Therefore, this thesis intends to detect, locate and quantify damage using short data bursts of forced vibration due to vehicular loading. Here, damage is understood as either a stiffness loss (global or local) or a deterioration of the support conditions. This research contributes to technologies based on location (i) by proposing an algorithm that relies on the measurements from an accelerometer during the forced vibration of the bridge while traversed by a specialised vehicle. The algorithm exploits the variation in bridge frequency, not only with the magnitude and location of the moving vehicle but also with the stiffness profile of the bridge. A k-Nearest Neighbours algorithm searches for the patterns of forced eigenfrequencies that are closest to the on-site instantaneous frequencies to determine the location and severity of the damage. The algorithm shows promising results, although it is limited to low vehicle speeds and low road roughness. Furthermore, this thesis contributes to the technology (ii) with three model-based drive-by algorithms, i.e., involving a finite element model of the bridge and a vehicular model, which are necessary to locate and quantify the damage. Three components can be distinguished within a measured vehicular acceleration: bridge, vehicle, and road components. The last two components usually govern the frequency spectrum, which makes it hard to distinguish the targeted bridge information. As a result, ‘drive-by’ solutions are often developed for damage detection purposes only, meaning that they rarely locate or quantify the damage, and when they do so, is usually under strict requirements, i.e., specific vehicle properties, low vehicular speed and smooth road profiles. In order to mitigate the undesired vehicle component, the concept of transfer function is applied to derive the response of the contact point between the vehicle and the bridge from the response of the acceleration of an axle. In order to deal with the road component, each algorithm uses a different strategy as follows: (1) estimation of the energy content in the frequency domain following the removal of the road profile in the spatial frequency domain based on measurements at different speeds. Damage is located and quantified by comparison to a database of responses generated via a simple P-load model; (2) optimisation of the road profiles estimated in the time domain using measurements in one axle at two different speeds. Based on cross-entropy optimization, the stiffness profile of a simply supported bridge model is accurately predicted, even for relatively high speeds and rough roads, except for locations next to the supports; (3) optimisation of the road profiles estimated in the time domain by two axles of the same vehicle. This algorithm addresses the condition of the bridge supports. Finally, drive-by algorithms (2) and (3) yield an accurate representation of the road profile on the bridge, in addition to an assessment of the mechanical properties of the bridge.