Soil thermal conductivity refers to the capacity of soil to transfer heat and is a critical consideration in various engineering applications, including the disposal of radioactive waste, the installation of high-voltage buried cables, and the configuration of ground heat pumps. Soil thermal conductivity is influenced by several factors, including mineral composition, compaction, moisture content, dry density, soil gradation, and temperature. Understanding these factors is essential for effective engineering practices and ensuring the safety and efficacy of projects involving soil interaction.
Previous research has concentrated on quantifying the factors that affect thermal conductivity in various soil types, utilizing methods such as the thermal needle probe and cast bronze sphere, while establishing correlations with these influencing factors. Findings indicate that moisture content, dry density, and particle size are significant determinants of thermal conductivity. Additionally, it is essential to consider the mineralogical composition and arrangement of soil grains in predictive analyses. Most investigations have been conducted under room temperature conditions, where thermal conductivity remains stable. However, for soils subjected to temperatures below 0 °C, the transition of pore water into ice modifies heat transfer mechanisms, underscoring the importance of temperature in the study of thermal conduction in frozen soils. Considerable research has been undertaken on frozen soils and those experiencing freeze-thaw cycles, resulting in the development of various predictive models that integrate temperature and phase transition phenomena.
A neural network approach has been employed to investigate heat transfer through soil and develop an accurate model for predicting thermal conductivity. This approach considers various input factors, including moisture content, dry density, mineralogical type, particle gradation, and other relevant environmental variables. The study encompasses five different soil types: silty, silty, clay, fine sand, and coarse sand. Soil samples were carefully prepared with varying moisture contents and dry densities to enhance the model’s predictive capabilities based on artificial intelligence techniques.
Neural networks were independently developed for each type of soil, designated as PM-C for clay, PM-S for silt, PM-SS for silty sand, PM-FS for fine sand, and PM-CS for coarse sand. In addition, a generalized model, referred to as PM-G, was created to accommodate five types of soils with varying characteristics. Individual models utilized two input parameters—moisture content (w) and dry density (γd)—while the
generalized model incorporated four parameters: moisture content (w), dry density (γd), clay content (c), and quartz content (qc). The dataset was divided into 80% for training purposes and 20% for testing.
The determination of the optimal number of neurons in the hidden layer of artificial neural networks (ANNs) was achieved through a systematic approach, beginning with a single neuron and progressively increasing the count until the network reached optimal training performance. The training and testing phases were conducted using Matlab 2017, with momentum factors set at 0.01 and 0.001, and a maximum of 1000 training cycles implemented. It is noteworthy that different soil types exhibit varying sensitivities to input parameters. For the models developed for clayey soils, the transfer functions utilized were tan-sigmoid for the hidden layers and log-sigmoid for the output layers, while other model configurations employed log-sigmoid for both layers.
The investigation, conducted on four types of soils—clay, silt, fine sand, and coarse sand—from the Nanjing area, aimed to explore the intricate relationships between thermal conductivity and various geotechnical properties. This report specifically emphasizes the properties of clay and fine sand, focusing on assessing how moisture content and dry density influence thermal conductivity.
The findings reveal that thermal conductivity exhibits a positive correlation with moisture content, particularly demonstrating a significant increase at lower moisture levels, specifically below 20% for clay and below 5% for fine sand. Beyond these thresholds, the thermal conductivity stabilizes after reaching a critical moisture content.
Furthermore, it has been observed that thermal conductivity increases with rising dry density, which can be attributed to the enhanced contact points among solid particles. Notably, substantial differences in thermal conductivity were evident irrespective of moisture levels, with a marked effect observed in clay. While other factors such as porosity and degree of saturation also play a role in influencing thermal conductivity, their impacts are primarily captured through the lenses of moisture content and dry density.
The solid phase of soil consists of a variety of minerals, including feldspar, calcite, mica, and quartz, which play a significant role in determining thermal conductivity. Among these minerals, quartz stands out with a notably high thermal conductivity of 7.69 W/K·m, while the thermal conductivities of other minerals typically range from 1.25 to 4.00 W/K·m. Soils that have a higher quartz content display enhanced thermal conductivity, suggesting that both the type and proportion of minerals, particularly quartz, are crucial factors in influencing the thermal conductivity of soil particles.
Additionally, fluctuations in temperature affect the thermal movement of molecules and the process of thermal conduction within the soil. Generally, an increase in temperature facilitates greater molecular movement, thereby improving heat transfer.
The factors influencing soil thermal conduction warrant a focused examination, particularly in relation to
particle size, distribution, shape, and temperature. Notable observations include the following: Smaller particle sizes increase the number of contact points, subsequently enhancing thermal resistance and reducing soil thermal conductivity. In general, sandy soils exhibit higher thermal conductivity than clayey soils due to inherent particle properties. In dry soils, heat transfer primarily depends on contact points, as air possesses significantly lower thermal conductivity compared to soil particles. The incorporation of binders can improve thermal conduction by establishing a more stable soil structure. Additionally, particle size and shape play a crucial role in determining soil configuration at the micro level, where finer particles can create large secondary aggregations that facilitate improved thermal conduction when aligned with the direction of heat flow. In summary, the characteristics of soil particles notably influence heat transfer within soils, particularly among fine-grained types.
The values of R² and VAF for the generalized model (PM-G) are either comparable to or slightly exceed those of the individual artificial neural network (ANN) models. This observation indicates that the predictive capability of the PM-G model is on parity with that of the individual models. Furthermore, the validation of the ANN models using testing data effectively compares the predicted results against laboratory measurements. Notably, for clayey soils, the PM-G model demonstrates R² values that are nearly equal to those of individual models (PMC and PM-S), reflecting strong predictive accuracy. In the case of sandy soils, the PM-G model outperforms the individual models (PM-SS, PM-FS, and PM-CS) with marginally higher R² values. In conclusion, both the individual and generalized models effectively predict the thermal conductivity of various soil types.
Reference
Zhang, T., Wang, C. J., Liu, S. Y., Zhang, N., & Zhang, T. W. (2020). Assessment of soil thermal conduction using artificial neural network models. Cold Regions Science and Technology, 169, 102907.