Unsupervised data-driven assessment of internal defect in airport concrete pavement structures

Abstract Deep learning has been widely applied in structural defect identification. However, the results of defect identification usually cannot directly support the decision-making process of intelligent operation and maintenance. Defect assessment is the key link that connects detection data with operation and maintenance strategies. To this end, this paper proposes an internal defect assessment method for airport concrete pavements based on unsupervised data-driven approaches. This method relies on ground-penetration radar detection technology. Based on the measured data from 10 detection areas, 141 detection lines, a total detection area of 13 654.1 m 2 , and a detection length of 11 492.48 m, it extracted defect density, the proportion of defect length, and the proportion of defect area as characteristic indicators to quantitatively assess the internal damage conditions of the pavement from multiple dimensions. In the research, a variety of unsupervised clustering algorithms, including the partitioning method, the hierarchical method and the density-based method, were adopted to conduct cluster analysis on the defect states at different detection lines and regional scales respectively. The evaluation results showed that different algorithms consistently classify the samples into three basic patterns based on defect features, revealing that there is an inherent category law in the structural distribution of pavement damage. Finally, based on the results of cluster analysis and in combination with the current specification requirements, this paper proposes the pavement damage score, providing a new indicator for quantitatively evaluating the degree of internal structural damage of the pavement.