Remolding Water Content Effect on the Behavior of Frozen Clay Soils Subjected to Monotonic Triaxial Loading

Abstract

Understanding the mechanical behavior of frozen clay subgrade soils was essential for ensuring the safe and stable operation of transportation lines. However, the influence of remolding water content w on this behavior remained unclear. To address this gap, this study examined the effect of w through monotonic triaxial testing. Three typical remolding water contents (w = 19%, 27.5% and 35%) and three confining pressures (σ₃ = 200 kPa, 700 kPa and 1200 kPa) were considered. Results showed that the mechanical behavior of frozen clay soils displayed a clear dependence on w, which was controlled by microstructural evolution. As w increased, the shear strength q_max, resilient modulus E₀ and cohesion c increased, which resulted from the progressive development of ice bonding within the shear plane. A threshold w value was found at w_opt = 27.5%, marking a structural transition and separating the variations of q_max, E₀ and c into two regimes. When w ≤ 27.5%, the soil fabric was controlled by clay aggregates. As w increased, the growth in ice cementation was confined within these aggregates, leading to limited increase in q_max, E₀ and c. However, as w exceeded 27.5%, the soil fabric transitioned into a homogeneous matrix of dispersed clay particles. In this case, increasing w greatly promoted the development of an interconnected ice cementation network, thus significantly facilitating the increase in q_max, E₀ and c. The friction angle φ decreased with w increasing, primarily due to the lubrication effect caused by the growing ice. In addition, the enhanced lubrication effect in the clay particle-dominated fabric (w > 27.5%) resulted in a larger reduction rate of φ. Regarding Poisson’s ratio v and dilation angle ψ, the w increase led to growth in both parameters. This phenomenon could be explained by the increased involvement of solid ice into the soil structure.

1. Introduction

In China, the seasonal frozen soil region occupies about 53.5% of the country’s total land area. These regions were predominantly located in northern and northeastern China, with extensions into parts of the central and southwestern areas. Rapid economic and social development has caused an increase in transportation needs, resulting in expanded railway/highway construction in seasonal frozen soil area. As an essential part of the railway/highway structure, the subgrade played an important role in load bearing, and its mechanical behavior was crucial for ensuring the safe and stable operation of transportation lines. Till now, the relevant aspects have drawn many attentions and were widely investigated [1,2,3], and these studies were the basis for numerical modeling [4]. During the subgrade construction, the clay soils were usually used as fillings materials. According to the JTG D30-2015 [5], the compaction of clay soils should be carried out at optimum water content w_opt. However, owing to complex in situ construction conditions, the actual remolding water content w during compaction could not be strictly controlled and differed across different sites, which caused variations in soil fabric [6,7,8]. Such variations in soil structure have been shown to significantly influence the mechanical behavior of thawed soils. Moreover, after being frozen, the mechanical behavior of frozen clay soils might exhibit variability. From a practical point of view, this aspect should be studied.

Till now, many investigations have been carried out on frozen fine-grained soils. Chang et al. [9] investigated the strength of the frozen clay soils at the water content corresponding to the in situ one, and found that under high confining pressure, the stress–strain behavior exhibited characteristic of strain hardening. In the study of Sun et al. [10] with silty clay at one given water content, it was found that the increase in confining pressure enhanced the plastic deformation capacity of frozen soils. Lai et al. [11] and Li et al. [12] reported that for frozen clay soils at saturated state, a strain softening pattern was identified for stress–strain relationship. Regarding the mechanical behavior of frozen fine-grained soils at various water contents, Ma et al. [13] carried out uniaxial compression creep test on frozen clay soils, and found that the peak strength occurred at the water content w = 40%, followed by w = 120%, while the minimum strength was observed at w = 80%. Zhang et al. [14] investigated the strength of silty clay using uniaxial compression test under various water contents. They reported that this type of frozen soil achieved its maximum strength when approaching the saturated state. Nevertheless, these studies were based on uniaxial testing condition, which failed to reflect the actual triaxial stress state encountered by the soils in the field. With regard to triaxial loading test, Sayles and Carbee [15] observed that, in silty clay, the water content increase resulted in a shift in stress–strain relationship from ductile to brittle behavior. In the study of Liu et al. [16] with a silty clay, a monotonic increase in q_max with w was observed. However, Zhang et al. [17] reported that the silty clay from Tibet Plateau exhibited a non-monotonic change in shear strength as water content increased, with an initial decrease followed by a subsequent increase. In summary, although numerous studies have investigated the mechanical behavior of frozen fine-grained soils, the majority have focused on silty clay soils with relatively low plasticity. The low plasticity of these soils rendered their soil fabric less sensitive to the variation in remolding water content w compared to clay soils with high plasticity. To the authors’ knowledge, the mechanical behavior of frozen clay soils under varying w values has not been systematically studied.

This study aimed to address this gap by providing a comprehensive experimental investigation into the influence of water content on the triaxial mechanical response of frozen clay soils. Three confining pressures (σ₃ = 200 kPa, 700 kPa and 1200 kPa) and three typical w values (optimum water content w_opt = 27.5% as well as its dry side w_d = 19% and wet side w_w = 35%) were considered. The mechanical parameters including shear strength q_max, Young’s modulus E₀, cohesion c, friction angle φ, Poisson’s ratio v and dilatancy angle ψ were analyzed. In addition, scanning electron microscopy (SEM) and the mercury intrusion porosimetry (MIP) were jointly performed to identify the soil fabric. The obtained results allowed the effect of remolding water content on the mechanical behavior of frozen clay soils to be clarified.

2. Materials and Methods

2.1. Testing Materials

The clay soils used in this study were typical materials used in subgrade construction. The hydrometer method (ASTM D422) [18] was used to determine its grain size distribution, and the testing result was shown in Figure 1. Given that this clay soil exhibited a liquid limit w_L of 63% and a plasticity index I_p of 26% (determined accroding to ASTM D4318-17 [19]), it was classified as CH under the Unified Soil Classification System (ASTM D2487-11 [20]). The specific gravity G_s of the clay soils were determiend as 2.6 according to ASTM D854-23 [21]. Standard proctor compaction tests (ASTM D698-12 [22]) revealed that the optimum water content w_opt was 27.5% and the maximum dry density ρ_dmax was 1.50 Mg/m³, as shown in Figure 2.

Table 1. Soil properties.

Soil Type	Index Property	Value
Clay soils	Specific gravity, Gs	2.6
	Clay content (<2 μm)	85%
	Liquid limit, wL	63%
	Plastic limit, wP	26%
	USUC a classification	CH
	Optimum water content, wopt	27.5%
	Maximum dry density, ρdmax	1.50

^a Unified Soil Classification System.

summarizes the basic parameters of the tested clay soils.

2.2. Experimental Methods

2.2.1. Sample Preparation

Cylindrical samples were fabricated for triaxial testing, which had a diameter D of 39.1 mm and a height H of 80 mm. This dimension was selected aiming to adapt to the testing cell of triaxial apparatus. In addition, the sample diameter D significantly exceeded the maximum grain size d_max = 0.01 mm of the clay soils, which was consistent with the ASTM D2850 [23] specification that the D/d_max ratio should be larger than 6 to eliminate the scale effect.

The sample preparation began by directly adding the amount of water corresponding to the target water content (optimum water content w_opt = 27.5% as well as its dry side w_d = 19% and wet side w_w = 35%) to the oven-dried clay soils. Then, the moistened soils were sealed in an airtight container for 24 h to achieve moisture homogenization. Subsequently, the homogenized clay soils were compacted in five layers to prepare a sample with D = 39.1 mm and H = 80 mm within a cylindrical mold (39.1 mm in diameter and 120 mm in height). The shape and size were selected to ensure the compatibility with the triaxial testing apparatus utilized. A static compaction method was utilized for the compaction procedure. Note that the soil mass used for each layer was carefully controlled to be the same, and to guarantee proper connection between adjacent layers, the surface of each layer was slightly scratched prior to placing the subsequent layer. These standardized protocols ensured that all samples exhibited consistent shape and size. Following compaction, the sample was carefully wrapped in a plastic membrane to prevent moisture loss and stored in a sealed container for 24 h for stabilization. Note that this stabilization period was determined by assessing the water content across the sample diameter. The water contents at the quarter diameter, half diameter and the diameter points were considered, and the results showed that 8 h equilibration duration was sufficient to achieve uniform water distribution. Also note that the compaction degree D_c of all the samples was controlled at D_c = 0.80. This level was selected based on the in situ measurement of the D_c of the subgrade by Liu et al. [24], which reported that D_c ranged between 0.79 and 0.96. A D_c value near the lower boundary of this range was chosen, aiming to represent the most unfavorable condition encountered in the field. The prepared samples are presented in Figure 3.

To ensure accurate measurement and control of water content w during the experiment, several standardized procedures were implemented. Before wetting the soils to the target water content, the clay soils were oven-dried at 105 °C for 24 h to remove any residual water. The target water content was then achieved by evenly spraying water onto the dry soils using a high-precision digital scale. The mixtures were sealed and left for 24 h to ensure full moisture homogenization. Prior to compaction, representative soils were taken and oven-dried to verify the actual water content. Any minor deviation from the target value was corrected by adjusting the amount of added water before final compaction. During the compaction stage, the moistened soils were placed into a rigid metal mold, and the compaction was achieved through a steel ram. Due to the enclosed nature of the mold and the relatively short duration of the compaction process, water loss via evaporation was considered negligible. After compaction, each sample was immediately wrapped in plastic membrane to prevent moisture loss and stored in a sealed container for 24 h to allow stabilization. Following this stabilization period, the water content of the sample was measured, and the results showed that the actual water content deviated from the target value by less than ±0.1%. After sample preparation, a pre-freezing process was conducted prior to the frozen soil triaxial test. This pre-freezing step was carried out while the sample was wrapped in plastic membrane. Once the water within the sample was fully frozen into ice, moisture evaporation became extremely limited. In addition, the triaxial tests were performed within rubber membrane, which also helped minimize any potential water loss. As a result, the water content of the sample can be considered unchanged throughout the entire testing process.

2.2.2. Triaxial Tests

In this study, the triaxial tests were carried out utilizing a triaxial testing apparatus with computerized control (see Figure 4). This apparatus comprised three primary components: a servo-controlled loading system, a multi-channel data acquisition unit equipped with precise sensors and a pressure chamber assembly.

The servo-controlled loading system incorporated a rigid structural frame integrated with a high-precision axial actuator, enabling the applications of both stress-controlled and strain-controlled loadings. The sensors equipped into the data acquisition unit encompassed a confining pressure sensor for measuring the radial pressure applied around the sample, an axial force transducer for measuring vertical force applied to the sample, a pore pressure transducer for monitoring the water pressure within the sample, a linear variable differential transformer (LVDT) for measuring axial displacement, and a volumetric change monitoring sensor for tracking the sample volume change.

The triaxial pressure chamber possessed a thermal insulation capability, which enabled the liquid (ethylene glycol) inside the chamber to be kept at a consistent negative temperature throughout the testing period. This insulation was achieved by enveloping the standard chamber wall with an external thermally insulating organic shell, creating an annular gap between the two shells that was also filled with ethylene glycol. Temperature regulation was managed through a closed-loop cryogenic system. In this process, the ethylene glycol in the annular area cooled down and its temperature dropped to the target value. This cooling effect was transferred to the ethylene glycol inside the chamber, causing its temperature to decrease to the desired level as well. Note that the temperature for all the triaxial tests carried out during this study was set at −10 °C.

Before being introduced to the triaxial chamber, the sample underwent a 48 h freezing period at −10 °C. Following installation of the sample, the coolant in the chamber were cooled to −10 °C as well. Then the sample stood for over 12 h for thermal equilibrium. Each sample was subjected to triaxial shearing at a constant axial strain rate of 1 mm/min under three confining pressure levels (σ₃ = 200 kPa, 700 kPa and 1200 kPa). The three different confining pressures used in this study were selected based on both field conditions and experimental requirements. Specifically, the confining pressure of σ₃ = 200 kPa corresponded to a high subgrade with a height of around 20 m. The mechanical behavior of high subgrade was critically important in road and railway engineering, as the failure of such structure often led to catastrophic engineering accidents.

The other two higher σ₃ levels (σ₃ = 700 kPa and 1200 kPa), although not directly representing typical field stress levels, were chosen to ensure reliable determination of the shear strength parameters of cohesion c and internal friction angle φ. Due to the significantly high cohesion of frozen soils, narrow spacing between σ₃ levels may diminish the detectable influence of internal friction [25], making it difficult to accurately determine φ from the failure envelope. At the same time, the upper limit of the confining pressure was carefully selected to prevent excessive compression that could induce pressure melting, triggering additional mechanical responses of frozen soils [26], which fell outside the focus of this study. The testing procedure used in this study was made by referring to the previous studies (Xu et al., Yang et al.) [27,28]. The testing process was stopped when the axial strain ε₁ reached 20%, which indicated that the failure had been achieved.

2.2.3. Microscopic Tests

The microscopic structure of the clay soils was examined through the scanning electron microscopy (SEM) and the mercury intrusion porosimetry (MIP). The SEM test was concentrated on the geometrical appearance of the soil, while the MIP test was focused on the pore size distribution. Before carrying out each test, a freeze-drying technique was utilized to eliminate the water from the sample. During the freeze-drying process, each sample (around 2 g) was firstly placed in a freezer set at −80 °C for an initial freezing period of 24 h. Subsequently, the frozen samples were transferred to a freeze dryer, where the chamber temperature was also maintained at −80 °C, and a vacuum condition was applied. The drying phase then commenced, during which ice sublimated directly from the solid to the gaseous state without passing through the liquid phase. This sublimation process lasted approximately 48 h. Freezing the sample at −80 °C minimized the formation of large ice crystals that could damage the soil fabric, thereby preserving the original arrangement of soil particles and aggregates, which has been verified by many scholars (Gillott JE, 1970; Gillott JE, 1976; Shi et al., 1999) [29,30,31]. And the 48 h freeze-drying period ensured complete dehydration of the sample.

The MIP test was performed with consideration of optimum water content w_opt = 27.5% as well as its dry side w_d = 19% and wet side w_w = 35% using Autopore IV 9510 porosimeter (Micrometritics Corporate, Georgia, America). The mercury intrusion process was carried out by progressively increasing the mercury pressure, during which mercury infiltrated gradually into the large pores at low pressure and into the small pores at high pressure. At a given pressure, the total volume of the infiltrated pores was measured and the corresponding pore diameter was calculated following the method used by Delage et al. [32].

Regarding SEM test, three remolding water contents (w = 19%, 27.5% and 35%) were considered. During the test, the apparatus of SU8020 (Hitachi Corporate, Tokyo, Japan) was run at a voltage of 5 kV with a magnification of 5000×. Prior to SEM observation, aurum was sprayed onto the soil surface to form a thin layer, with which the electric conductivity of the soil was enhanced, and thus the imaging clarity was improved.

3. Results

3.1. Shear Behavior

Figure 5 depicted the relationship between deviator stress q and the axial strain ε₁ for samples with optimum water content w_opt = 27.5%, dry side w_d = 19%, and wet side w_w = 35%. Figure 5a–c showed that for each sample, as ε₁ increased, generally q exhibited a trend of initial growth followed by a reduction, defining a typical strain-softening pattern. This characteristic was also observed by Zhang et al. [10]. This mechanical response resulted from a dual-phase deformation mechanism: at the initial loading phase, the soil sample was densified through rearrangement of soil grains, ultimately reaching peak deviator stress q_max. Subsequently, the initiation and development of microcracks within the soil matrix triggered a progressive reduction in soil skeleton, resulting in macroscopic stress deterioration. In addition, this strain-softening characteristic was less pronounced at higher confining pressure σ₃, manifested by a smaller attenuation of post-peak stress. This σ₃-dependent characteristic can be explained by the more enhanced soil fabric densification under higher σ₃ level, which more effectively restrained the fabric degradation. Additionally, Figure 5a–c demonstrated that at each σ₃ value, the curve of the sample with higher water content lay above that of the sample with lower water content. Furthermore, the curve showed a more significant rise when w increased from 27.5% to 35%, compared with the increase in w from 19% to 27.5%.

Regarding the volumetric strain ε_v, Figure 6 presents a visual view of samples after shearing under σ₃ = 700 kPa. It could be observed that the increase in w favored dilation behavior. In this study, negative and positive values represent dilatancy and contraction, respectively. Figure 7 presents the ε_v–ε₁ relationship at different σ₃ levels. As can be observed, at a confining pressure σ₃ = 30 kPa, each sample initially contracted and then exhibited dilation behavior. In addition, the increase in w led to a decrease in contraction and an increase in dilation. When σ₃ increased, the sample displayed increased contraction and diminished dilation. This σ₃ effect was also observed by Wang et al. [23]. At σ₃ = 120 kPa, the sample with w = 19% even only presented pure contractive behavior without any dilation.

3.2. Shear Strength q_max

To further investigate the effect of remolding water content w, the q_max value of each curve was extracted and plotted against remolding water content w in Figure 8. Note that the shear strength q_max was determined as the peak shear stress observed in the stress–strain curve (see Figure 5). The peak shear stress represented the maximum resistance of the soil to shearing deformation under a given confining pressure, and was therefore commonly regarded as the shear strength in laboratory testing. This method was well-established and has been extensively used in many studies (Wang et al., 2018; Su et al., 2020) [33,34]. As could be observed, for each σ₃ value, the variation of q_max with w exhibited a bi-linear pattern characterized by two distinct slopes, the transition between which occurred at w = 27.5%. In addition, the slope of w = 27.5–35% was obviously larger than that of w = 19–27.5%. Specifically, at σ₃ = 200 kPa, the slope for w = 27.5–35% was k₁ = 1.07 and for w = 19–27.5% it was k₂ = 0.54. At σ₃ = 700 kPa, the slope for w = 27.5–35% was k₁ = 0.89 and for w = 19–27.5% it was k₂ = 0.46. At σ₃ = 1200 kPa, the slope for w = 27.5–35% was k₁ = 0.80 and for w = 19–27.5% it was k₂ = 0.38. Note that k₁ and k₂ refer to the slopes of the fitting lines corresponding to two different remolding water content ranges (w ≤ 27.5% and w > 27.5%). These values were derived from the angle between the trend line and the horizontal axis. The purpose of calculating these slopes was to qualitatively compare the variation trends of a given mechanical parameter with respect to w within the same coordinate system, under two distinct moisture regimes. Given that the comparison was limited to one single parameter and conducted within the same coordinate system, it was believed that this method was appropriate. The calculated results are summarized in

Table 2. Slopes of the variation of mechanical parameter.

	Slope k (w ≤ 27.5%)	Slope k (w > 27.5%)
qmax at σ3 = 200 kPa	0.54	1.07
qmax at σ3 = 700 kPa	0.46	0.89
qmax at σ3 = 1200 kPa	0.38	0.80
c	0.42	1.38
φ	0.65	0.98
E0	0.33	0.93

.

In addition, it could be observed that the increasing rate of q_max, which reflected the sensitivity of mechanical behavior to w, decreased as σ₃ increased. Specially, at w ≤ 27.5%, the slope decreased from 0.54 at σ₃ = 200 kPa to 0.46 at σ₃ = 700 kPa and further to 0.38 at σ₃ = 1200 kPa. Similarly, for w > 27.5%, the slope declined from 1.07 to 0.89 and then to 0.80 with increasing σ₃. This trend could be attributed to the relative contributions of cohesion and internal friction to the overall shear strength q_max. Under low σ₃ level, q_max tended to be more sensitive to w variation because cohesion (primarily attributed to ice cementation between particles) contributed more significantly to the overall q_max. In contrast, at higher σ₃ level, the contribution of internal friction increased, thereby reducing the relative influence of water-induced cohesion on q_max.

3.3. Cohesion c and Friction Angle φ

To better understand the effect of remolding water content w on shear strength q_max, the cohesion c and friction angle φ, which determined the magnitude of q_max, were calculated using the critical state method [35]:

$$\sin \mathit{\phi }={\displaystyle \frac{3M}{6+M}}$$

(1)

$$c={\displaystyle \frac{S\left(3-\sin \mathit{\phi }\right)}{6\cos \mathit{\phi }}}$$

(2)

where M and S, respectively, correspond to the slope and intercept of the critical state line established in the p-q plane. p represented the bulk stress.

The results of calculated c and φ are plotted in Figure 9a,b as a function of w, respectively. As can be observed, when remolding water content w rose, cohesion c increased, while friction angle φ exhibited a decreasing trend. In addition, similar to the case of q_max, the increasing trend of c and decreasing trend of φ could be divided into two stages with w = 27.5% as the boundary. Moreover, for both c and φ, the slope within the range of w = 27.5–35% was larger than the slope within the range of w = 19–27.5%. Specifically, regarding c, the slopes in the intervals of w = 27.5–35% and w = 19–27.5% were 1.38 and 0.42, respectively. For φ, the corresponding slopes were, respectively, 0.65 and 0.98.

3.4. Young’s Modulus E₀

The Young’s modulus E₀ was defined as the ratio between deviator stress q and axial strain ε₁ when ε₁ ranged from 0 to 1% [33]:

$$E_0={\displaystyle \frac{q}{{\mathit{\varepsilon }}_1}}$$

(3)

This strain range of 0–1% corresponded to the early stage of loading, during which complex particle rearrangement and initial stress adjustment occurred. Despite this inherent complexity, the variation of all testing E₀ values with w fell within a bounded range (Figure 10) and exhibited a non-linear trend that could be regarded as bi-linear. Notably, the transition point in this behavior coincided with the threshold water content w_opt = 27.5%, which was also identified in other mechanical parameters.

3.5. Poisson’s Ratio v and Dilation Angle ψ

The Poisson’s ratio v and dilation angle ψ were calculated based on the volumetric strain ε_v–axial strain ε₁ relationship according to the equations below:

$$v={\displaystyle \frac{1-k_{\mathrm{C}}}{2}}$$

(4)

$$\sin \mathit{\psi }={\displaystyle \frac{k_{\mathrm{D}}}{-2+k_{\mathrm{D}}}}$$

(5)

where k_C and k_D, respectively, represent the slopes of the contraction phase and dilation phase of the volumetric strain–axial strain curve.

The variation in Poisson’s ratio v with w under different σ₃ levels are illustrated in Figure 11a. As can be observed, at each w value, the σ₃ increase led to a decrease in v. This could be explained by the fact that the stiffness of the sample at higher σ₃ value was larger, which led to a smaller lateral strain. In addition, the increase in w led to an increase in v: when w increased from 19% to 35%, v grew from 0.25–0.28 to 0.31–0.39. The dependency of dilation angle ψ on w is shown in Figure 11b. It can be seen that no dilation behavior was observed under σ₃ = 1200 kPa. In addition, the dilation angle ψ was larger at higher w values.

4. Discussion

4.1. Microscopic Mechanism of the Mechanical Behavior

The above phenomena could be explained by the variations of soil fabrics at various remolding water contents. As can be observed from SEM testing results (Figure 12), clay particles at w_opt and its dry side exhibited a tendency to cluster, forming clay aggregate. For both w_opt and its dry side, the pore structure could be categorized into two distinct types: macro-pores located between clay aggregates and micro-pores situated within clay aggregate. In contrast, on the wet side, clay particles remained dispersed without aggregation, and micro-pores were relatively uniformly distributed among particles.

Although the SEM testing offered qualitative observation of the soil morphology, a precise quantification of the pore size distribution was achieved through MIP testing, as shown in Figure 13. During MIP testing, mercury intrusion was performed by incrementally increasing the applied pressure. As the intrusion pressure increased, mercury progressively filled the pores from smaller ones to larger ones within the sample. The cumulative volume of mercury intruded at each pressure step was carefully recorded. The corresponding pore diameter d was calculated using the following equation:

d = 4σcosγ/p

(6)

where σ = mercury–solid interfacial tension (0.484 N/m); γ = mercury–solid contact angle (141.3°); and p = intrusion pressure. Based on these measurements, the cumulative pore size distribution curve was firstly constructed by plotting the cumulative intruded pore volume against pore diameter on a logarithmic scale. Subsequently, the differential pore size distribution curve was obtained by differentiating the cumulative data with respect to the logarithm of pore diameter (Figure 13). This differential curve illustrates the percentage of pores at a specific pore diameter d.

Figure 13 indicated that the clay soils at optimum water content (w_opt) and its dry side exhibited a bimodal pore structure, consisting of micro-pores (d ≤ 0.4 μm) and macro-pores (d > 0.4 μm). Compared to the dry side, the clay soils at w_opt contained fewer macro-pores and a higher proportion of micro-pores. However, on the wet side of w_opt, the pore structure became unimodal, predominantly composed of micro-pores.

The pore structure variation was consistent with the soil fabric evolution, as illustrated in Figure 14. When on the dry side of w_opt, large clay aggregates were formed due to high matric suction [36]. Consequently, the soil fabric was defined by an aggregate-aggregate structure with numerous large pores between aggregates (Figure 14a). When reaching w_opt, the aggregation tendency was reduced, and the quantity of clay particles remaining in dispersed state increased, partially occupying the large pores between aggregates. This resulted in an aggregate-dispersed structure (Figure 14b). After w_opt, the formation of clay aggregate was inhibited due to the insufficient matric suction. In this case, the soil fabric was dominated by dispersed clay particles (Figure 14c). Based on the testing results of SEM and MIP, the soil fabric evolution with increasing w could be inferred [32,36], as illustrated in Figure 14. On the dry side of w_opt, the soil fabric was controlled by an aggregate–aggregate structure (Figure 14a), with water primarily confined within the micro-pores inside the clay aggregate. At w_opt, part of the clay aggregate disintegrated into dispersed clay particles, leading to a transition in soil fabric towards an aggregate-dispersed structure (Figure 14b). In this case, most of the water was still retained within the aggregate, while a portion was distributed among the dispersed clay particles. On the wet side of w_opt, the soil fabric was dominated by dispersed structure (Figure 14c), with water occupying the micro-pores between dispersed clay particles.

The shear strength q_max, resilient modulus E₀ and cohesion c of frozen clay soils were observed to increase with w increasing, which was predominantly governed by the variation of microstructural property of the shear plane. The remolding water content w exerted its influence through varying the formation of ice cementation within this plane. As remolding water content w increased, the overall quantity of ice cementation on the shear plane grew, reinforcing the shear plane and leading to an increase in cohesion c. As a result, the q_max and E₀ increased correspondingly. When w ≤ 27.5%, the soil fabric was dominated by clay aggregates. Since water mainly resided inside the aggregate, ice cementation merely reinforced the connections between clay particles inside aggregate. Since numerous large pores (non-cemented zones) existed between aggregates, the effective cemented area induced by the moisture increase remained limited. As a result, cohesion c, q_max and E₀ increased at a relatively low increasing rate. When w > 27.5%, the soil structure transitioned to a homogeneous system dominated by dispersed clay particles with uniformly distributed small pores. In this case, increased moisture content w induced full development of ice cementation networks within inter-clay space. As a result, cohesion c as well as q_max and E₀ increased in a high rate.

Regarding friction angle φ, it was observed to diminish with increasing w, possibly due to the fact that growing ice exerts a lubricating effect within the soil skeleton, thus reducing the friction angle φ. And since this lubricating effect was smaller in aggregate-dominated skeletons compared with particle-dominated skeletons, the decreasing rate of φ at w ≤ 27.5% was smaller. Note that in frozen soils, the shear strength q_max and E₀ were predominantly controlled by cohesion c; this decrease in φ had little effect on q_max and E₀, which still exhibited increasing trends.

When it came to the parameters related to the volumetric variation (Poisson’s ratio v and dilation angle ψ), both showed a tendency to increase as w increased. This was due to the involvement of more ice at higher w values, which introduced more solid materials into the soil structure, favoring horizontal deformation and volumetric expansion.

4.2. Comparison with the Previous Studies

A comparison of the current testing results with the previous studies is summarized in Figure 15. Previous studies have primarily focused on silty clay with plasticity index I_p ranging from 11.7 to 14.43, which was significantly lower than the I_p = 37 of the clay used in this study. These studies with silty clay soils showed that similar to clay soils, their mechanical behaviors were also highly dependent on w. In the case of silty clay, the peak strength qmax exhibited varying trends (either increasing then decreasing (Niu et al.) [37], continuously increasing (Liu et al. [16]), or continuously decreasing (Zhang et al. [17])) as the water content shifted from the dry side to the wet side of w_opt. These discrepancies might be attributed to the differences in sample dry density. In contrast, the clay used in this study was controlled at one dry density level. The effect of dry density on the relationship between q_max and w was not clear and should be studied in the future.

5. Conclusions

Monotonic triaxial tests were carried out on frozen clay soils under three confining pressures σ₃ (σ₃ = 200 kPa, 700 kPa and 1200 kPa). The effect of remolding water content w was investigated with three typical w values (optimum water content w_opt = 27.5% as well as its dry side w_d = 19% and wet side w_w = 35%). In addition, the microstructures of the tested samples were examined through MIP and SEM tests. The results allowed the following conclusions to be drawn.

The mechanical behavior of frozen clay soils exhibited distinct remolding water content w-dependent characteristic governed by microstructural evolution. The shear strength q_max, resilient modulus E₀ and cohesion c increased as w increased, which was driven by the progressive increase in ice cementation within the shear plane. Notably, a threshold w value was identified at w_opt = 27.5%, which marked a structural transition and divided the variation of q_max, E₀ and c into two zones. When w ≤ 27.5%, clay aggregate dominated the soil fabric and ice cementation was limited within the clay aggregate, resulting in subdued growth of q_max, E₀ and c. As w increased beyond 27.5%, the soil fabric transitioned into a homogeneous matrix of dispersed clay particles, where increased w facilitated the interconnected ice cementation network, thereby accelerating the enhancements of q_max, E₀ and c.

Regarding friction angle φ, it decreased with w increasing due to lubrication effect exerted by the growing ice. In addition, since this lubrication effect was enhanced at clay particle-dominated fabric (w > 27.5%), the decreasing rate of φ was larger at w > 27.5%. As for volumetric parameters (Poisson’s ratio v and dilation angle ψ), both exhibited an increasing trend as w increased. This can be attributed to the more involvement of solid ice into the soil structure, which enhanced the horizontal deformation and volumetric expansion.