Key points are not available for this paper at this time.
Real-time assessment of subgrade compaction quality poses a significant challenge in the implementation of intelligent compaction (IC). Current compaction evaluation models are confined to specific scenarios and lack robustness. This study proposes a subgrade compaction strategy that utilizes a heterogeneous dataset to estimate compaction quality across diverse scenarios while maintaining model accuracy. Field compaction tests are conducted in four distinct scenarios, considering various construction parameters. Compaction models are developed using several machine learning algorithms. The datasets are thoroughly assessed in terms of quality, diversity and similarity. The proposed model exhibits good performance in new scenarios by incorporating an additional 5% to 8% of new data for retraining. The model’s generalization capability is enhanced by conducting a limited number of field tests, which are labor-saving and time-efficient. The model’s accuracy consistently improves across diverse scenarios and optimal algorithms. The proposed compaction strategy adopts a physics-and-data dual-driven approach, aimed at practical engineering applications and guiding the compaction procedure.
Wang et al. (Sun,) studied this question.