Intelligent compaction quality evaluation using Morse wavelet transform and deep neural network

Abstract

The intelligent compaction technique uses the longitudinal acceleration signal of the roller’s vibrating steel wheel to judge the soil’s compaction quality. The low-frequency component of the signal is used to identify the surface stiffness of the soil and, thus, indirectly estimate the degree of compaction. High-frequency components are considered noise. However, the high-frequency component can reflect the intensity of the collisions between particles. Therefore, the high-frequency component may be more effective than the low-frequency component in evaluating the compaction quality of deep soil. This study establishes a nonlinear model using Morse wavelet transform and deep neural network to evaluate the compaction quality. The influence of high and low-frequency components on evaluation results is analyzed by controlling the frequency band range of input. The results show that, compared with the low-frequency component, the high-frequency component can more accurately evaluate the degree of soil compaction. In order to eliminate the influence of the “double jump phenomenon” of the roller, high-frequency and low-frequency components should be considered simultaneously. This method can not only accurately distinguish between under-compaction and over-compaction but also has the potential to take the actual degree of soil compaction as output.