Result: A piezoelectric sensing and machine learning integrated method for modulus inversion and structural strength prediction in asphalt pavements.

This serves as a graphical abstract for this paper, facilitating the reading of reviewers and readers, to quickly grasp the research framework and content of this study. As illustrated in graphical abstract, this study proposes PWPM based on the piezoelectric effect, and develops an intelligent evaluation framework that integrates numerical simulation with data-driven modeling. The framework systematically includes indoor material parameter testing, FE model development, and machine learning-based prediction model construction. The logical relationships among these components are clearly delineated, forming a coherent workflow. Ultimately, this framework is designed to enable in-situ sensing and real-time evaluation of structural health condition of asphalt pavement. By leveraging PWPM, the proposed method provides a novel approach for the rapid inversion of material modulus and non-destructive assessment of pavement structural strength. [Display omitted] • Proposed an in-situ monitoring framework integrating PWPM and machine learning. • Established a nonlinear mapping between PFs and material modulus. • Applied PWPM as a novel DNT technique for PSSI evaluation. • Developed a Python-based automated modeling and sample generation system. • Provided technical support for lifecycle health management of pavement structures. Modulus (M) and Pavement Structural Strength Index (PSSI) are critical indicators for evaluating the health condition and remaining service life of asphalt pavements. To enable in-situ perception and real-time evaluation, this study proposes a Piezoelectric Wave Propagation Method (PWPM) based on the piezoelectric effect and develops an intelligent assessment framework integrating numerical simulation and data-driven modeling. First, Falling Weight Deflectometer (FWD) tests were conducted on selected sections of the Nanjing-Hangzhou Expressway to collect deflection response data. Subsequently, in-situ coring and laboratory tests were performed to obtain material parameters and piezoelectric features (PFs) of different structural layers. Based on this, finite element (FE) models for pavement structural deflection response and piezoelectric ultrasonic active sensing were established. A Python-based automation pipeline was developed for sample generation and parametric modeling, ensuring uniform coverage and representativeness in the high-dimensional parameter space. Furthermore, machine learning algorithms were employed to construct a PF-based Modulus Inversion Model (PF-M Inversion Model) and a PF–PSSI Mapping Model for accurate prediction of M and PSSI. Experimental results indicated a strong positive correlation between PFs and M. The XGBoost-based PF-M Inversion Model achieved high prediction accuracy on both training and validation sets, with R2 values exceeding 0.98, demonstrating excellent adaptability across different layers. Additionally, performance metrics confirmed that the XGBoost model provided robust and accurate mapping between PFs and PSSI. Comparison with measured data showed that model predictions closely matched the in-situ test results, with relative errors (REs) within 10%, validating the feasibility and engineering potential of the proposed PWPM framework for in-situ monitoring and performance evaluation of asphalt pavement structures. [ABSTRACT FROM AUTHOR]