Intelligent compaction quality assessment of vibratory roller using deep neural network

Intelligent compaction technology uses the acceleration signal features of the roller’s vibration drum to assess the compaction quality of the subgrade. Currently, most methods for assessing compaction quality rely on traditional compaction indicators. The existing compaction quality evaluation methods are basically based on intelligent compaction measurement values, however, the compaction quality of subgrade is affected by the interaction of many factors and it is not enough to describe the complex mapping between the drum acceleration signal and the compaction quality by using only a few indicators. This limits the development of intelligent compaction technology. This paper proposes a novel method for assessing subgrade compaction quality based on Gaussian process. A deep neural network is used to extract features from acceleration signals. Based on the features extracted by the network, the Gaussian process is used to accurately evaluate and predict the compaction quality. The results indicate that compared to traditional compaction quality assessment methods, the proposed method significantly improves prediction accuracy, demonstrating clear engineering value for the realization of intelligent subgrade compaction.