A PROPOSAL OF IN-SITU SOIL COMPACTION QUALITY MEASUREMENTS USING DEEP LEARNING

Abstract

This paper proposes a method for evaluating in-situ soil compaction quality using a convolutional neural network. Dynamic behaviors of a vibrating wheel during soil compaction were simulated by a numerical calculation of a vibrating wheel-soil interaction model, and large datasets of the simulated acceleration responses of vibrating wheel and the soil spring stiffnesses were used to train a quality estimation model. A field experiment of soil compaction was also demonstrated to verify the trained estimation model. The estimated soil spring stiffnesses showed a positive correlation with measured values.