Characterization and prediction of compressive strength in ultralow-temperature frozen soil using nuclear magnetic resonance and WOA-ENN Model

Abstract

Artificial ground freezing（AGF）has been utilized in a range of engineerings, including tunneling, shaft sinking, deep excavations, and the construction of underground structures. Currently, the strength and efficacy of artificial ground freezing technology have become a prominent subject of research. To reveal the trend of changes in the uniaxial compressive strength of ultralow-temperature frozen soil, uniaxial compression tests were carried out on soil samples with varying water contents (17 %, 20 %, and 23 %) and temperatures (-10 °C to −180 °C). To analyze the relationship between the unfrozen water content and compressive strength, the unfrozen water content in the frozen soil was quantitatively measured by nuclear magnetic resonance (NMR). Additionally, a prediction model for the compressive strength of ultralow-temperature frozen soil was built with an Elman neural network based on the whale optimization algorithm (WOA-ENN). The results indicate that the frozen soil exhibits elastic–plastic failure at temperatures above −80 °C and brittle failure at temperatures below −80 °C. In the temperature range from −10 °C to −80 °C, the compressive strength of the frozen soil increases linearly with decreasing temperature, and stabilizes at temperatures ranging from −80 °C to −180 °C. The turning point in compressive strength at −80 °C arises because the content of unfrozen water remains relatively constant below this temperature. Furthermore, the relationship between the unfrozen water content and compressive strength follows a power function. To predict the compressive strength of frozen soil at ultralow-temperatures, a WOA-ENN prediction model is proposed as it has a significantly lower average absolute percentage error (4.00 %) compared to the ENN prediction (24.49 %). Overall, the proposed prediction model effectively addresses the complex nonlinear relationship between the compressive strength of ultralow-temperature frozen soil and its influencing factors, thus providing valuable insights into the mechanical response of frozen soil under ultralow-temperature conditions.