Abstract: How to determine the setting state of fresh concrete is crucial to the quality control of dam concrete construction. The traditional measurement of concrete setting time uses the penetration resistance method, which is detrimental and difficult to achieve real-time continuous monitoring of the setting. Most of the emerging ultrasonic nondestructive monitoring is a method that adopts a single transmission or reflection, but its measurement is affected greatly by the incident angle, and its calculation of wave velocity requires an accurate determination of the arrival time of the first wave through manual calculations. Aimed at these problems, this paper develops a method of ultrasonic transmission-reflection joint measurement to accurately calculate the setting time of dam concrete. We conduct an experimental study of monitoring concrete setting state, construct a neural network model based on the Levenberg-Marquaredt (L-M) algorithm, and train it using the time point data set determined by energy index. This new method has been applied to the setting test of dam concrete C25 and compared with the reflection method and the transmission method both based on energy analysis, and the transmission method based on wave velocity analysis. The results show the maximum relative error of its setting time calculations is reduced by 2.5% - 18.7%. The correlation coefficients of its calculated initial setting time and final setting time with the real values are 0.89 and 0.96 respectively, verifying its accuracy and theoretical and practical significance in the quality control of dam concrete construction.

Key words: dam concrete, setting time, ultrasonic, transmission-reflection method, L-M neural network, energy