Identifying method of rigid pavement support deterioration based on local CPSD-XGBoost fusion

In response to insufficient robustness in traditional dynamic indicators and poor physical interpretability or over-fitting in purely data-driven models for identifying the deterioration of airport rigid pavement slab support, a data-physics fusion identification method was proposed. The correlation between cross power spectral density (CPSD) and the pavement support state was clarified through theoretical derivation. For distributed vibration sensing, local CPSD was proposed. A data-physics fusion method was developed by combining local CPSD with XGBoost to identify slabs with deteriorated support and subsequently locate the specific regions. Its effectiveness was validated by comparing the performance of the physical indicator method, the principal component analysis-support vector machine (PCA-SVM) data-driven method, and the proposed fusion method across two experimental scenarios. Analysis results show that local CPSD comprehensively reflects the influence of structural mode shapes, frequencies, and damping ratios on the pavement support state. Furthermore, the sensitivity of loading points to support anomalies is significantly higher than that of the CPSD calculation points. Regarding identification performance, the physical indicator method has the lowest average accuracy (83.35%) and recall (50%). Although the data-driven method achieves an accuracy of 90.75%, the actual data exhibit weak separability and feature redundancy (7-14 features). The data-physics fusion method performs the best, with the accuracy rising to 91.65%, the recall reaching 77.5%, and the feature dimension significantly reducing to 3-5. These features correspond to the sensitive frequency bands of the local CPSD for support states, providing strong physical interpretability. A method is provided to combine physical information with data features for identifying support deterioration in rigid pavements. While improving existing physical indicators, it also offers a possible path for applying artificial intelligence methods to identify anomalies in pavement structures.