A Prestress Loss Inverse Analysis Method for Long-Span PC Girder Bridges Combining Finite Element Analysis and Machine Learning

Abstract This study addresses the persistent difficulty of quantifying prestress losses in long-span prestressed concrete girder bridges, where the combined effects of anchorage slip, friction, elastic shortening, relaxation, creep and shrinkage interact in a strongly time-dependent and spatially heterogeneous manner. A hybrid inverse analysis framework that couples high-fidelity finite element simulation with data-driven machine learning is proposed, with the objective of recovering the in-service prestress state from limited and noisy monitoring observations such as deflection, strain and natural frequency. A parametric finite element model that incorporates geometric nonlinearity, time-varying material behavior and discretized tendon profiles is established and used to generate a physically consistent training corpus over the admissible loss parameter space.