Ultrasonic assessment of cement-stabilized soils: Deep learning experimental results

Soil-cement materials are widely used in civil engineering. Previous studies described problems related to errors in ultrasonic measurements of the unconfined compressive strength (UCS) of low-strength soil-cements. In the presented experiments, a standard ultrasonic pulse velocity (UPV) measuring device was equipped with an oscilloscope to extend its functionality. An analysis of the ultrasound signal revealed the phenomenon of masking key elements of the longitudinal wave, which caused difficulties in the detection of noise and the correctness of measurements using classical detection methods (without prediction). The important goals of this study were to build an error removal method and UPV universal detection mechanism without human interference. To improve the measurement efficiency, a deep learning method was implemented. High-quality pulse and strength recognition results were obtained, improving the accuracy of the standard ultrasonic pulse measuring device. Based on the results, the multivariate correlations of UCS with the material UPV, maturation time and cement content are described.