Intelligent prediction of unfrozen water content of artificially frozen clay along one-dimensional column based on LF-NMR

Abstract

The unfrozen water content (UWC) plays a crucial role in frozen soil engineering, however, traditional unfrozen water measurement or calculation methods are time-consuming and costly, and it is difficult to express the non-linearity among the influencing factors. Currently, LF-NMR is widely recognized as an effective tool for measuring unfrozen water. In this study, a calibration curve that can describe the relationship between the NMR signal and water content was derived. Moreover, the effects of temperature, initial water content, and soil height on UWC are explored along one-dimensional large-diameter columns based on LF-NMR data. Combined with the ML algorithm and the UWC test data under different factors based on LF-NMR, an intelligent prediction method that can consider the nonlinear characteristics of multiple factors is proposed. The results showed that the NMR accuracy is 176.64 semaphore per 1% water content, the error between the water content calculated from the calibration curve and that obtained from the drying method is small (average error is 0.97%), and the linearity degree of the calibration curve is good (R² = 0.999). Moreover, due to the strong correlation between unfrozen water and temperature, initial water content, and other factors, the results indicated that the GPR model can better describe this correlation and has the best prediction effect, which was evaluated with the quantitative indicators: R² = 0.96, RMSE = 1.07, MAE = 1.00, and again verifies the superiority of this model in combination with other literature data. Overall, this paper offers a technical basis for controlling and preventing freezing and thawing disasters in cold regions and artificial freezing engineering projects.