A hybrid domain adaptation approach for estimation of prestressed forces in prestressed concrete bridges under moving vehicles

Ensuring the accurate and reliable estimation of prestressed forces (PFs) in prestressed concrete bridges is vital for operational performance and public safety. Traditional vibration-based methods often face challenges when applied to real-world scenarios. In this study, a novel hybrid domain adaptation approach is proposed to predict the prestressed force of prestressed concrete bridges under moving vehicles. The finite element model for Vehicle-Bridge Interaction (VBI) systems is established and validated using the experimental results. This validated model is then used to generate a synthetic training dataset of the source domain. Continuous Wavelet Transform (CWT) is employed for feature extraction from VBI acceleration responses, capturing their time-frequency properties. Preliminary feature extraction is enhanced through the use of a pre-trained AlexNet network. Following this initial step, a novel hybrid domain adaptation (DA) approach is applied to close the gap between the synthetic and real-world data. Specifically, Maximum Mean Discrepancy (MMD) and adversarial DA techniques are synergistically combined. Three types of loss functions are incorporated: Regression Loss for precise force prediction, MMD Loss to align the synthetic and real-world data domains, and Adversarial Loss to ensure domain invariance. The effects of uncertainties in VBI system, such as errors of the bridge length, flexural rigidity, density, damping, boundary conditions, errors of the vehicle model and moving speed and road surface roughness, have been discussed. The results indicate that the integrated approach effectively mitigates the challenges posed by domain shift, enabling robust and reliable predictions of PFs in actual bridge structures. The finding of this study offers a comprehensive, data-driven solution with significant implications for the future of structural health monitoring and bridge condition assessment.