Soil Compaction Monitoring Technique Using Deep Learning

Abstract

It is commonly known that the dynamic behavior of a vibratory drum of a soil compaction machine changes with soil stiffness. Although real-time monitoring techniques of compaction quality by measuring the acceleration of the vibratory drum have already been put into practice use, their applicability depends on the soil type and condition. In this study, to extend the range of applicability and improve accuracy, we propose a deep learning-based technique that allows the regression estimation of soil stiffness from the acceleration responses of a vibration drum. To collect a large amount of noise-free training data, the acceleration responses of a vibratory drum were simulated by numerically solving the equations of the motion mass-spring-damper system. We also conducted a field experiment to verify the proposed technique. The experimental results show that the estimated values of soil stiffness correlate with the measured values, with the correlation coefficient of approximately 0.79. Thus, the proposed method has potential as a new real-time monitoring technique for soil compaction quality.