Regional graph-based pavement condition data quality assessment in support of trustworthy highway infrastructure digital twins

Abstract

Pavement condition data plays a critical role in highway digital twins (DTs) for intelligent infrastructure management. However, automated condition surveys often contain abnormal measurements that, if left undetected, can distort deterioration modeling and compromise maintenance decision-making. Existing section-level abnormal data identification methods remain limited in accuracy, as they typically rely on statistical thresholds or temporal heuristics and overlook the spatial-temporal dependencies inherent in pavement data. To address this hurdle, this paper introduces a regional graph-based method for section-level data quality assessment. Specifically, it first groups neighboring road sections into homogeneous clusters based on their historical condition patterns. Subsequently, a Graph-Mamba Attention Network (GMAN) is utilized to capture spatial-temporal dependencies and identify abnormal data points within each cluster. A case study on the UK highway network showed that the proposed method significantly outperforms the existing methods, thus signifying its potential to enhance the trustworthiness of pavement data in highway DTs.