Towards Trustworthy Highway Infrastructure Digital Twins: A Regional Graph-Based Method for Section-Level Pavement Data Quality Assessment

Digital twins (DTs) are rapidly gaining attention as a powerful tool for intelligent highway infrastructure management. By integrating real-time sensing, modeling, and data-driven analytics, DTs enable transport agencies to monitor pavement performance and make informed maintenance decisions. However, data quality remains a critical trustworthiness concern, particularly as automated pavement condition surveys increasingly replace manual inspections. This study presents a regional graph-based method for assessing the quality of pavement condition data at the section level. Specifically, the method consists of two key steps: first, a region-based clustering approach using DBSCAN is applied to group neighboring road sections with similar deterioration patterns into homogeneous clusters. Then, a Graph-Mamba Attention Network (GMAN) is introduced to evaluate the reliability of each condition data point within these clusters by capturing spatial and temporal dependencies in the data. The method was evaluated using real-world characteristic skid coefficient (CSC) data from the UK’s Strategic Road Network (SRN), supported by TRL Software’s iROADS digital twin platform. Results showed that the proposed method consistently outperformed existing methods in identifying abnormal data points (38.8% and 7.1% increase in accuracy and recall, respectively), demonstrating its potential to improve the effectiveness and trustworthiness of pavement data quality management within highway DTs.