Abstract: To accurately invert the dynamic resilient modulus of in-service subgrades from pavement vibration responses under impact loading, overcoming the limitations of existing methods in simulating the dynamic characteristics of layered pavement systems during testing, an inversion method integrating dynamic modeling and artificial intelligence is proposed. First, fully considering the layered nature of pavement systems and the unsaturated characteristics of subgrades, a three-layer pavement dynamics model is established. The solution for the model’s dynamic response is derived using the Laplace-Hankel double transform and its inverse transform. A comprehensive database correlating pavement surface vibration responses with the dynamic resilient moduli of layered pavement systems is constructed through extensive computations. Subsequently, an artificial neural network (ANN) model is developed based on this database. Bayesian optimization is employed to adjust its hyperparameter combinations, yielding an intelligent rapid inversion framework for subgrade dynamic resilient modulus. To validate the method, measured data from 19 typical test sections in the U.S. long-term pavement performance program (LTPP) database are utilized for comparative verification. Results demonstrate that the dynamic resilient moduli obtained via the intelligent inversion method exhibit strong correlation with laboratory-measured values (R2=0.815 0), significantly outperforming conventional inversion methods relying on static mechanics- based models.

Key words: artificial intelligence, subgrade, resilient modulus inversion, dynamic response