Interpretable machine learning for asphalt compaction monitoring via spatiotemporal data fusion
Traditional asphalt pavement quality control relies on post-construction spot-testing, which is labour-intensive and spatially fragmented. This study proposes an intelligent compaction (IC) monitoring framework integrating paving and rolling telemetry. Construction data from pavers and rollers were spatially synchronised using KD-Tree-based nearest-neighbor matching and grid-based aggregation to construct a unified spatiotemporal dataset. To address the long-tailed distribution of Compaction Meter Value (CMV), a log-transformed Random Forest (Log-RF) model was developed, achieving a coefficient of determination (R2) of 0.6924 on the field dataset. SHapley Additive exPlanations (SHAP) were used to enhance interpretability. Vibration frequency was the dominant predictive variable, while rolling temperature, rolling speed, and other construction variables provided additional predictive information. The framework provides continuous CMV predictions and spatially distributed information for construction-process monitoring, supporting timely operational assessment during asphalt pavement compaction.
Authors
- Shuai Wang (ORCID: https://orcid.org/0000-0002-7973-7392)
- Jincheng Wei (ORCID: https://orcid.org/0009-0003-4614-7569)
- Lei Niu (ORCID: https://orcid.org/0009-0000-6389-5276)
- Feifei Fan
- Ya Ma
- Dongqi Yan
Institutions
- Shandong Transportation Research Institute (CN)
Publication Details
- Journal
- Road Materials and Pavement Design
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1080/14680629.2026.2759798
- Primary Topic
- Asphalt Pavement Performance Evaluation
- Type
- article
- Field-Weighted Citation Impact
- 0.00