Study on the evaluation method for compaction quality of rockfill dams based on seismic wave dispersion characteristics and deep learning

The compaction quality of rockfill dams is a pivotal factor in determining the stability and safety of such structures. However, the conventional on-site test pit flooding method is both costly and limited in its ability to accurately reflect the compaction quality of the entire rolling surface. Consequently, this paper proposes a novel approach to the non-destructive evaluation of compaction quality in rockfill dams. Firstly, the relationship between the dispersion characteristics of seismic waves and the compaction quality of rockfill dams was discussed through the indoor seismic wave test experiment. The porosity was utilised as the fundamental metric for evaluating the compaction quality of rockfill dams, and the compaction quality was categorised into various grades. The dispersion images obtained from the seismic wave test were labelled accordingly. Secondly, four models were utilised to train in different compaction quality classifications of heap rock dams, respectively. The outcomes demonstrated that all four models could achieve the recognition and evaluation of compaction quality classes of rockfill dams, with MobileNetV2 performing the best and having the highest accuracy rate. When 2 classification and 5 classification recognition evaluations were performed, the accuracy reached 93.41% and 91.40% respectively.