Invited Seminar by Prof. Mohammad AlHamaydeh

The interaction between vehicles traversing a bridge and the bridge’s vibrations is a dynamic phenomenon. An acceleration sensor mounted on the vehicle captures signals emanating from various sources of interaction, including bridge vibrations, road irregularities, vehicle characteristics, and speed. Specific algorithms are essential to discern the vibrational properties of the bridge from this complex signal. This thesis presents a comprehensive framework designed to extract the vibrational characteristics of the bridge using data from multiple vehicle sensors. Additionally, two novel techniques for source separation are introduced.

The framework is structured into two primary stages. The initial stage focuses on source separation by eliminating the effects of vehicle dynamics and road irregularities from the original sensor-recorded signal. Two established approaches, Frequency Response Function (FRF) and Ensemble Empirical Modal Decomposition (EEMD), are considered for removing vehicle dynamics. Additionally, a hybrid algorithm (FRF+EEMD) is introduced and evaluated. Two road roughness profile removal methods are tested: the Second-Order Blind Identification (SOBI) algorithm from the literature and the Signal Subtraction Algorithm (SSA), a newly proposed technique. SSA utilizes the disparity between responses from two vehicles to isolate pure bridge frequencies. After filtering out residual frequencies from the deconvoluted vehicle response, these frequencies yield the pure bridge response. Source separation is conducted for multiple vehicles, generating a sparse observation matrix containing bridge vibration readings across both spatial and temporal coordinates.

In the second stage of the framework, the sparse matrix is completed using the Alternating Least Squares (ALS) algorithm. New signal processing techniques are introduced to compute initial frequency guesses and prepare the data for structured optimization analysis. The hybrid technique (FRF+EEMD) and SSA demonstrate comparable or superior performance to existing algorithms in the literature. Furthermore, the introduced signal processing techniques successfully differentiate between vertical and torsional vibration modes and automatically detect initial guess values closely aligned with the actual values of fundamental frequencies.