Algorithmic Advancements in Drive-by Inspection Methods Towards Intelligent Bridge Monitoring

Since conventional vibration-based structural health monitoring (SHM) methods typically require the installation of numerous sensors directly onto the bridge, the substantial costs associated with on-site sensor installation and maintenance have long rendered such technology an expensive option. Recently, an alternative approach referred to as the "drive-by bridge inspection method" has attracted scholarly attention. This method necessitates no instrumentation on the bridge; instead, it uses a few sensors positioned on the vehicle traversing the bridge, with the vehicle functioning as both an exciter and a receiver. It offers the possibility to efficiently monitor groups of bridges (particularly small and medium-sized ones), presenting an economically efficient solution to bridge health monitoring problems. While previous research on the drive-by method has yielded promising results, there are still some challenging problems and ample room for improvement before its engineering application. This thesis aims to propose algorithmic solutions to address the challenges faced by the current drive-by methods, successfully extracting bridge modal parameters and identifying damage. In terms of bridge modal parameters, two algorithms are proposed. Algorithm 1 extracts the bridge frequency as a common signal component in the responses of multiple sensors mounted on the vehicle. This method does not require specially designed vehicles or multiple cars, as in earlier studies, but instead utilizes a single ordinary vehicle, providing a practical solution. Algorithm 2 employs the Coherence-PPI (Prominent Peak Identification) method, which extracts the bridge frequency from the common vibrational components of multiple passes of the same vehicle. Rather than seeking to minimize differences, it encourages variability in drive-by measurements (e.g., varying vehicle parameters, avoiding traversing the same road surface) to filter bridge frequencies. It is particularly suitable for vehicles passing the same bridge multiple times (such as buses). In terms of damage identification, a data-driven algorithm based on an optimized AdaBoost-linear SVM is proposed (Algorithm 3), which accurately indicates bridge damage using only the raw vibration signals received from vehicles passing the bridge. To further improve damage identification accuracy and address certain limitations of drive-by measurements, such as noise and data redundancy, a time-domain signal processing algorithm for the raw vehicle accelerations is also proposed (Algorithm 4). The proposed algorithms have been validated through numerical simulations, laboratory experiments, and field tests on simply supported and continuous beam bridges using truck and bus models. The ultimate goal is to achieve a practical and intelligent bridge health monitoring system.