Experimental Investigation on Effects of Water Flow to Freezing Sand around Vertically Buried Freezing Pipe

Artificial ground freezing (AGF) involves thermal, hydrological, mechanical, and even chemical processes that may cause some considerable changes in the geological material. Understanding the entirety of changes which occur during and after the freezing process is necessary to ensure the success and effectiveness of ground freezing projects such as frozen ground wall isolation, underground liquified natural gas (LNG) storage, tunnel construction, and mining operations, among others. Some previous experimental studies of the freezing of geological material focused on the mechanical properties of rock or soil as the porous medium, with the strain characteristic being one of the most frequently discussed topics. Rock or soil may deform by frost heave and rock segregation (Matsuoka 1990; Murton et al. 2006), which are closely related to pore-water distribution during freezing (Chen et al. 2004). A number of researchers have reported that freezing and thawing cycles caused deterioration of the mechanical properties of various types of rock and porous materials (e.g., Hori and Morihiro 1998; Yamabe and Neupane 2001; Altindag et al. 2004; Chen et al. 2004; Yavuz et al. 2006; Tan et al. 2011). The deterioration of mechanical properties indicates that the development of the frozen body inside the porous medium modified the pore structure, which also changed the hydraulic properties in the unfrozen state.

In addition to the mechanical properties of rock or soil as the porous medium, other important factors that determine the success of AGF projects are the groundwater condition and heat transfer mechanism. The characteristics of groundwater associated with AGF are determined by several aspects, including water content, seepage movement, and the solute material content, which can affect the freezing point of water (Bing and Ma 2011). Some numerical and analytical studies that evaluated the heat transfer mechanism in geological material considered the partial or full coupling between thermal, hydrological, and mechanical mechanisms (e.g., Harlan 1973; Neupane et al. 1999; Neupane and Yamabe 2001; Lackner et al. 2005; Nishimura et al. 2009; Liu and Yu 2011). Specifically, Goldstein and Reid (1978) provided an analytical solution for the effect of seepage flow on the freezing and thawing of saturated porous media around a freeze pipe. Those studies showed that seepage movement determines the characteristic of thermal energy exchange, which is influenced by its fundamental heat transfer modes. A thermal convection process is dominant when a significant amount of water is contained inside a porous medium. Groundwater seepage may accelerate heat transfer, but becomes a major problem when velocity is too high because it may prevent the frozen-body development.

Considering seepage flow effects in the heat transfer process, especially AGF, is important because high velocity of groundwater flow exists in the natural condition and underground engineering project sites. However, the number of experimental investigations of this topic is limited. Only a few groups of researchers have conducted freezing experiments that consider the effect of water flow. Sres et al. (2005) conducted an experiment to investigate the effect of low seepage velocity on the thermal distribution of soil surrounding a freezing pipe. A more significant effect of higher flow rate was observed in a larger-scale experiment conducted by Pimentel et al. (2010) and Huang et al. (2013). They focused on the temperature change during freezing that was measured by using multiple thermocouple sensors, which only provide temperature data at several designated points and do not give comprehensive thermal distribution. The use of infrared thermography of thermal distribution measurement provides better visualization of thermal distribution (Sudisman et al. 2016). However, that study was mainly conducted under cooling (positive temperature) conditions. In general, the previous experimental results showed that temperature change and distribution around the pipe are affected by seepage flow. However, none of them provided information on flow changes due to decreasing temperature.

Japan has been struggling to control the spread of radioactive contaminated groundwater at the Fukushima nuclear power plant after the 2011 Great Tohoku Earthquake. Frozen underground wall construction was chosen as one of the countermeasure efforts to isolate the contaminated groundwater so that it would not flow into the ocean. In this project, a high velocity of underground water flow is the main challenge to be controlled during the freezing of the ground, and it has motivated the authors to investigate the effects of fluid flow on freezing around the freezing pipes. Heat transfer, hydraulic conductivity, and deformation are the three most important factors investigated in ground freezing study. As the first step of continuous research, the authors investigate the heat transfer and hydraulic conductivity behaviors in this paper; the deformation behavior is not included here and will be investigated subsequently.

Based on the general objective and the limitations of the previous studies, the authors conducted a fundamental experiment that focused on the flow–thermal relationship and the visual observation of heat distribution of saturated sand with seepage flow around a freezing pipe. This basic study is important to investigate the thermal-flow characteristic and to evaluate the applicability of the method itself. Change of flow caused by thermal variation is an important parameter that gives information about the hydraulic property changes of the porous medium during freezing or unfreezing, and indicates progress of the frozen-body development in the porous medium. Furthermore, physical property variations of rock or soil as a consequence of different geological layers are inevitable in the field. This condition may lead to differences in the thermal distribution characteristics during freezing that affect its effectiveness. Hydraulic conductivity variation caused by material discontinuity of porous media is one of the most important factors that affect the freezing mechanism of geological material. To explain that hydraulic conductivity effect, the authors also conducted experiments with different grain sizes of sand layers.

For a single-layer sand specimen, Toyoura sand filled a height of 8.5 cm in the acrylic box, and the remainder was filled with oil clay (Fig. 1). A freezing pipe with a 1.0-cm diameter was inserted through the sand down to the bottom of the container. For this specimen, the tests were performed at an uncontrolled room temperature. Slight fluctuation of the experimental temperature was expected to occur because actual natural conditions also fluctuated. Initial temperatures of each test were varied with the room temperature and water flow temperature. The only controlled temperature was that of the brine solution temperature inside the refrigerator, $-20.0°\mathrm{C}$. However, it was affected by room temperature and warmed to between $-11.0°\mathrm{C}$ and $-7.0°\mathrm{C}$ when circulated inside the copper pipe. A Testo 876 thermal imaging camera (Testo, Lenzkirch, Germany) with $-20°\mathrm{C}$ to 100°C measurement range captured the surface temperature of this specimen.

To investigate the effects of flow on freezing, the experiments were conducted under four variations of flow condition as follows:

The temperature and flow conditions of the tests are summarized in Table 1.

Fig. 2 is a schematic of the layered sands inside the acrylic box container. Layer arrangements from the bottom to the top were 3.0 cm Toyoura sand, 2.0 cm silica sand, 3.5 cm Toyoura sand, and 2.0 cm impermeable oil clay. Two types of silica sand were used in this experiment: Silica #3 and Silica #5. A freezing pipe with a 1.0-cm diameter was inserted through the sand layers down to the bottom of the container. For the thermal measurement system, an Avio InfRec R500EX thermograph (Nippon Avionics, Tokyo) captured the surface temperature of this specimen. In addition to the thermograph, a tip-end thermocouple rod was inserted through the sand layers down to the middle of the silica sand layer. It was located 0.5 cm upstream side (left side) of the pipe. This combination gave comprehensive accurate thermal measurements along with visualization of the thermal distribution. To improve the effectiveness of freezing in this experiment, a better pipe insulation was used that was able to maintain a lower freezing pipe temperature, $-11°\mathrm{C}$.

Grain size, hydraulic conductivity, and thermal conductivity values of each sand material are listed in Table 2. Fig. 3 shows particle-size distribution curves for each type of sand. Hydraulic conductivity values were obtained by performing flow tests, whereas the thermal conductivity properties were obtained from thermal conductivity tests using a quick thermal conductivity meter (QTM-500, Kyoto Electronics Manufacturing Co., Ltd., Kyoto, Japan). Silica #3 had a higher hydraulic conductivity compared with other two sand types. However, the average thermal conductivity values of all types were relatively similar. Wet sand had almost 10 times higher thermal conductivity than dry sand. These measurement results were within the standard range value (Tarnawski et al. 2009, 2011; Kiyohashi et al. 2003).

In this study, thermal distribution and water flow behaviors were investigated under four different flow scenarios as follows:

Those four scenarios were used to compare two main conditions: comparison of the different specimens (Silica #5 and Silica #3) with the same flow head difference, and comparison of the same specimen (Silica #3) with different flow rates.

Test A1 was conducted by freezing the specimen with no initial water flow. Fig. 4 shows the measured temperature of specimen throughout Test A1 and continuing into Test A2. The result of a single thermograph measurement was a set of digital temperature data ($168\times 130$) that is represented by colors of the respective pixels of the thermal image. The maximum and minimum temperatures were the highest and the lowest values from 21,840 temperature data of a single thermal image at one time. The average temperature value was obtained by averaging all 21,840 data of a single thermal image. The maximum, minimum, and average temperatures were plotted with respect to the $y$-axis. They decreased rapidly immediately after the coolant was circulated. The minimum temperature was 0.0°C after 7 min cooling, and reached a quasi-steady state temperature of about $-5.0°\mathrm{C}$ within the first hour of freezing. At this stage, it is expected that the frozen body was formed in some part around the pipe and continued to develop as a frozen wall. The temperature was maintained by continuous coolant circulation for the next 2.5 days and slightly fluctuated due to day–night cycles, indicated by yellow and blue, respectively.

The presence of a formed frozen body is also supported by Fig. 5, which shows the surface thermal distribution of the specimen within 2 h of freezing. After 20 min of freezing, negative temperatures were equally distributed to both sides of the pipe because no flow affected the heat convection, which also indicates that a frozen wall may have formed inside the box. The lower part of the pipe had cooler temperatures than the upper part because the temperature of the pipe was not uniform. The freezing pipe was designed so that the coolest part was located at the bottom end.

The A2 test was intended to ensure that the frozen body had been formed during A1 test and to investigate how the water flow affected it. In addition to the measured temperature discussed previously, Fig. 4 also shows the measured discharged flow, which is plotted with respect to the $r$-axis. The water flow was introduced into the specimen 63.53 h (3,812 min) after the test started. At that time, the coolant was still in a circulating condition so that the frozen body was maintained. Therefore, no discharge flow was recorded even about 10 h after the water flow system was installed. It is clear evidence that a frozen wall formed and was able to prevent water flow. During these 10 h, the specimen’s temperature increased, which may have been caused mostly by the daytime room temperature as on the previous day, rather than the effect of the water flow installation.

To investigate the flow characteristic in response to the major escalated temperature, coolant circulation was stopped at 4,400 min. Fig. 6 is a more detailed graph of Fig. 4 that shows the last phase of the test once the freezing system was turned off. Both the average and minimum temperatures started to increase at 4,400 min in conjunction with the postfreezing condition, and the first discharged flow was recorded 11 min after that. The average temperature increased more slowly than the minimum temperature. The minimum temperature shows a stacked graph that may indicate the required latent heat during the melting process of ice from 4,410–4,430 min. During this test, a frozen body was formed inside and outside the box. Therefore, there are two steps in the minimum temperature graph. The first step indicated the remaining ice outside the box, which later slipped down due to gravity. The second step indicated the melting process of ice inside the box close to 0°C, followed by a more significant temperature increment. The flow flux suddenly increased after 11 min of postfreezing condition, which coincided with the melting process. Subsequently, the flow rate continued to increase by smaller increments until the test ended.

The melting of the frozen body is better visualized by the series of temperature profiles in Fig. 7. At 4,409 min (10 min after freezing was stopped), a small part of the lower-temperature ice (darker blue) existed at the bottom middle of the box, which was actually outside the box [Fig. 7(a)]. Figs. 7(b–d) show the melting process of the frozen body inside the box. The ice melted due to a warmer room temperature and the incoming water flow from the upstream (top left corner), which removed the upper ice wall gradually, and the flow increased with the melting of the frozen body. In addition, as described for the previous result, the lower part of the specimen had a lower temperature due to the freezing pipe design. Therefore, it is obvious that the lower part could have a thicker frozen body and require a longer period to melt than the upper part.

Tests B1 and B2 were conducted under initial flow rate variations $4.68\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ ($4.04\mathrm{m}/\mathrm{day}$) and $1.47\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ ($1.27\mathrm{m}/\mathrm{day}$), respectively. Fig. 8 shows the measured temperature and flow rate of Test B1 over time. Similar to the freezing test with the no-flow condition (Test A1), the measured temperature in this test also decreased significantly immediately after the coolant was circulated. The minimum temperature was below 0.0°C after 7 min cooling. The flow rate also decreased in conjunction with the declining temperature. This result indicated that major formation of a frozen body was ongoing this early in the test. After the major decrease, the temperature reached a quasi-steady-state condition within the first hour of freezing, whereas the flow rate continued to decrease with a smaller decline. However, water flow could not be stopped after more than 24 h of freezing, and it was predicted to have flowed constantly despite the freezing process continuing for the next few days. Considering the minimum pipe temperature, $-7°\mathrm{C}$ of the freezing pipe may not be sufficient to completely freeze the flow path. It was only able to reduce the flow rate to 38.72% of the initial flow.

The presence of the partially formed frozen body is also supported by Fig. 9, which shows the surface thermal distribution of the B1 test within two hours of freezing. Similar to the Test A1 result, negative temperatures occurred starting from 20 min after freezing. However, the thermal distribution was not equal for the left and right sides of the pipe because the thermal distribution was affected by the water flow. Furthermore, regardless of the lower initial temperature at the beginning of freezing test, Fig. 9 shows a lighter blue than Fig. 5, which indicates that the test with water flow had a less-formed frozen body than the test of without water flow. This is because a continuous influx heat of flowing water prevents ice formation in the sand pores, especially at a relatively high freezing temperature close to the freezing point of water.

In the next phase, the coolant circulation was turned off after running for 27 h (1,620 min), which caused both temperature and flow rate to increase. The result of this postfreezing phase is shown clearly in Fig. 10. The average and minimum temperatures abruptly increased, whereas flow rate had a milder increase, which was most likely unrecoverable in a short period. A similar characteristic was obtained for Test B2 despite the lower initial flow rate than for Test B1.

Fig. 11 compares the two tests, and mainly shows the difference in the magnitude of flow rate. The declining value of flow rate or hydraulic conductivity at the time of freezing and its recovery during thawing is a common result in frozen porous material (Horiguchi and Miller 1983). However, it does not guarantee that the water flow can be stopped completely.

In addition to the uncontrolled room temperature and the unmaintained temperature of brine when it was circulated into the freezing pipe, all the previously discussed results indicated that water flow had a very important role during the freezing of sand in this experiment, which affected the failure of frozen wall formation. A previous study reported that a frozen soil column would not merge for a critical groundwater velocity (Sanger and Sayles 1979), which is expressed aswhere $K_1$ = thermal conductivity of frozen soil ($\mathrm{W}/\mathrm{m}°\mathrm{C}$); $S$ = expected frozen wall diameter (i.e., spacing) (m); $d_0$ = diameter of freezing pipe (m); $v_s$ = difference between the freezing pipe surface temperature and freezing point of water (°C); and $v_0$ = difference between original ground temperature and freezing point of water (°C). This equation could not be used for this experiment due to the small value of $S$, i.e., $\mathrm{S}<2d_0$. In general, the critical water flow velocity varies within a range of $1–3\mathrm{m}/\mathrm{day}$ for $S>2d_0$ (Pimentel et al. 2007; Sanger and Sayles 1979). Therefore, logically it can be assumed that the critical water flow velocity of this experiment should not be less than $1–3\mathrm{m}/\mathrm{day}$ or even higher if the freezing pipe temperature is sufficient to freeze the ground, which means a better possibility to stop the water flow. However, this assumption is inappropriate for our result and has to be proven by further investigation.

Pimentel et al. (2010) reported that water flow could be stopped with a longer freezing duration despite a high water flow velocity expected to be larger than the critical value. In the present tests, Test B1 was conducted in less than 30 h and Test B2 in about 35 h, which were relatively short compared with Pimental et al.’s experiments. Even so, as mentioned previously, the water flow in these experiments was unlikely to stop even though the freezing process continued for several days.

The flux of both Tests B1 and B2 decreased similarly, to about 38% of their initial values at the time of freezing (Table 1). They increased to 43% 1 h after the end of the freezing phase. Based on these facts, it is expected that sectioned areas of the formed frozen body of both tests were the same, so that the unfrozen areas were also same.

The unfrozen area was most likely located around the acrylic wall surface. The authors expect that the flow near the acrylic interface had a higher velocity than that in other areas. Moreover, when the pores in the middle section were covered by ice, flow would be concentrated to pass through the sides of the pipe, so the flow velocity on that side was expected to be higher than before. Furthermore, a transition layer was expected to be formed between the sand–acrylic interfaces, which allowed water molecules to flow (Hoekstra and Miller 1967; Miller et al. 1975; Horiguchi and Miller 1980).

Figs. 12 and 13 show the relation between temperature and flow of freezing without initial water flow and freezing with initial water flow, represented by Tests A1 (followed by A2) and B1, respectively. These figures show the flow condition in four steps during the tests and clarify the differences of flow behavior, especially during postfreezing processes. For Test A1, there was no flow during the freezing because it was applied after the frozen wall formed (Fig. 12). Subsequently, the flow rate increased significantly with increasing temperature from the frozen condition to 0°C. After the thawing process, the flow rate increased gently with increasing temperature up to the initial temperature condition. For Test B1, the flow slightly decreased with decreasing temperature to 0°C (Fig. 13). Then it decreased more significantly in the negative temperature condition, corresponding to the formation of a frozen body. For the increasing temperature condition, the flow rate only slightly increased and could not recover to its initial value.

Steps 1 and 2 of these tests are not comparable due to the different flow conditions. For Steps 3 and 4, the main difference between these tests was the thawing process, whereas the heating process had a relatively similar trend. However, the reason for this difference is not clear and could not be explained by these experiments. One possible reason is that these two tests experienced different deformations during freezing, which generated distinct flow behavior after the ice melted. Specimen A1 experienced full freezing around the pipe, so the sand arrangement could have been deformed (expanded) due to the ice formation. On the other hand, Specimen B1 only experienced partial freezing around the pipe, so the expansion may not have been as significant. Moreover, continuous flow at the unfrozen part acted as a pressure force that confined the grains. More investigation is required to clarify the deformation behavior, because it was not part of the present study.

Fig. 14 shows the first 20-min results of the thermal profile of Tests C1 and C2 obtained by a thermograph camera. Each $10.5\times 8.5\text{-}\mathrm{cm}$ rectangle represents a single measurement of the box specimen’s surface at 4-min intervals. The boundary between layers is marked with black dashed lines with the silica sand layer located in between the lines. Due to the flow mechanism of the coolant fluid inside the freezing pipe, coolant fluid was concentrated at the bottom part of the freezing pipe so that it had a lower temperature than the upper part. However, regardless of the difference in the concentration of lower temperature, the Toyoura–Silica #5 specimen (Test C1) had a relatively straight low temperature distribution from top to bottom along the pipeline. This indicates that there was no significant difference in thermal conduction or advection among the layers inside the specimen. Water flow affected the thermal distribution uniformly in the horizontal direction so that the downstream tended to have a slightly lower temperature than the upstream. This effect did not last long due to the decreasing flow rate once the frozen body gradually developed and shut off the water flow.

In contrast, a slightly curved shape of low temperature distribution, especially in the silica sand layer, occurred from the first 12 min of Toyoura–Silica #3 freezing test (Test C2) results. The deviated shape indicates that there was some variation of thermal conduction or advection between layers. There was no significant difference in the thermal conductivity measurements (Table 2), so the variation in temperature distribution caused by the thermal conduction factor was minor. On the other hand, Silica #3 has a very high hydraulic conductivity compared to the other two materials. Therefore, it is almost certain that the curved shape was caused by the hydraulic conductivity variation of the layers. This effect disappeared after 16 min freezing. Hydraulic conductivity variation decreased once the water fused into ice and blocked the pores.

Test C1 had a lower temperature and froze earlier than did Test C2 (Fig. 14). The high flow velocity in Test C2 had more impact to prevent the accumulation of cold temperature in pores. Therefore, a longer time was required in Test C2 to fuse water into ice. This time variation is evident from the thermal-flow history (Figs. 15 and 16).

Figs. 15 and 16 show the relation between the measured temperatures and discharge flow of Tests C1 and C2, respectively. Thermocouple measurement results are denoted $T_c$, whereas thermograph measurement results are divided into minimum ($T_{g\text{\_}\min }$), average ($T_{g\text{\_}\mathrm{average}}$), and maximum ($T_{g\text{\_}\max }$) values. Surface thermal measurement was affected by the room temperature change, whereas the temperature inside the specimen was more independent. The difference between the inner and surface temperatures occurred as expected, and was confirmed by numerical simulation reported in previous research (Sudisman et al. 2016). In both tests, the inside temperature ($T_c$) and minimum surface temperature ($T_{g\text{\_}\min }$) of the specimens decreased rapidly immediately after the coolant was circulated. Although both tests were conducted under the same water head condition, the generated flow rates were different due to their hydraulic conductivity variation. Average initial discharge flow rates before freezing started were $0.43\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ and $1.83\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ for C1 and C2, respectively.

The freezing process can be observed from the flow-temperature profiles; $T_c$ continuously decreased to reach the quasi-steady-state condition, in which we can expect that a frozen body formed, and flow completely stopped. At 10 min of freezing in C1 (Fig. 15), the $T_c$ and $T_{g\text{\_}\min }$ values increased suddenly before continuing to decrease. The minimum freezing values of recorded $T_c$ were $-4.36°\mathrm{C}$ for C1 and $-3.37°\mathrm{C}$ for C2.

Freezing is an exothermic process that releases 79.72 calories of latent heat for each gram of ice fusion. However, there should not be any temperature change in the freezing body itself when the energy is released to the surroundings. In our experiment, water at the downstream side of freezing pipe was expected to freeze first, before the other parts. When this happened, no temperature change occurred in that part and heat was released to the surroundings, so a sudden increment was measured by the thermocouple and thermograph, which were located near the freezing part. This effect occurred in a limited area, so it did not increase the $T_{g\text{\_}\mathrm{average}}$ value. The values were stable or gently decreased. Another fact that supports this hypothesis is the cessation of water not long after that heat release, which confirmed that the frozen body had formed, and water could not pass through. The same mechanism was also found in Test C2 and other tests as well.

Figs. 17 and 18 prove the release of latent heat that slightly increased the temperature around the pipe. In Test C1, the temperature at 10 min was higher than at 9 min. In Test C2, the temperature at 12 min was higher than at 11 min.

Coolant circulation was stopped at 60 min so that temperature increased subsequently, followed by flow recovery. Thawing of ice is an endothermic reaction. Therefore, a larger volume of ice means that more energy is required in the melting process. Required latent heat is indicated by the relatively constant $T_c$ and $T_{g\text{\_}\min }$ values for some period after coolant circulation is stopped; In Fig. 15, $T_c$ was constant near 0°C (melting point) for approximately 13 min, from 64 to 76 min, whereas in Fig. 16, $T_c$ was constant for 12 min, from 63 to 74 min.

Flow in C1 started to recover at 74 min, which was 5 min later than the recovery in C2. The reason for the late flow recovery in Test C1 test was that it had a lower temperature and most likely a larger volume of ice at the end of the freezing period, which is confirmed by Fig. 19. For both C1 and C2, the recovered discharge flows were lower than their initial values, 73% for C1 and 43% for C2. This result may indicate that Specimen C2 had more deformation induced by freezing and thawing. Larger pores between Silica #3 grains were compacted and filled by Toyoura sand grains during flow and freeze-thaw processes.

The average flow rates of Q1, Q2, and Q3 were $1.83\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ ($1.58\mathrm{m}/\mathrm{day}$), $2.30\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ ($1.99\mathrm{m}/\mathrm{day}$), and $17.14\times {10}^{-3}{\mathrm{cm}}^3/{\mathrm{cm}}^2·\mathrm{s}$ ($14.82\mathrm{m}/\mathrm{day}$), respectively. Q1 and Q2 were defined based on the critical seepage velocity of freezing ground at $1–3\mathrm{m}/\mathrm{day}$ (Pimentel et al. 2007; Sanger and Sayles 1979). The critical velocity value does not apply to this experimental method because of the low pipe diameter–spacing ratio, which does not represent the real case in the field. Nevertheless, both had similar flow velocity magnitudes. For a different approach, Q3 was defined as extreme seepage with 10 times of normal flow velocity.

Fig. 20 shows the first 20-min results of the temperature profile of Tests Q1, Q2, and Q3. The crooked shape of thermal distribution in the 4th-min images of each test confirm that Q2 had slightly higher initial flow rate than Q1 but much lower than Q3. By considering only this condition, it is natural that the water flow effects in Test Q2 lasted slightly longer than in Test Q1. However, it disappeared earlier than expected. The influence of water flow is not visible in the 8th-min (Q2-08) profile, whereas it still exists in Q1-08. A possible reason for this condition is the initial temperature variation. Test Q2 had a lower initial temperature (18.29°C) than Test Q1 (21.55°C), so Test Q2 required less time to fuse water molecules into ice. The effect of flow is decreased once water is fused into ice, blocking the pores. From this stage, cold temperature distributes equally to the surrounding of the freezing pipe, and the effect of hydraulic conductivity variation no longer exists.

Unlike Q1 and Q2, Q3 had a very significant influence of water flow for a longer period. An intense flow rate prevents cold temperature accumulation and frozen-body formation, especially at the intermediate porous sand layer. Based on the Q3 test data, water was flowing for more than 15 h before it finally stopped. If this test were conducted on a large box specimen with a considerable pipe diameter–spacing ratio, it is reasonable that the water would remain unfrozen and flow could not be stopped.

From this result, it is obvious that hydraulic conductivity and flow rate variations have a very substantial influence on the freezing of sand layers. The intermediate layer that has higher porosity becomes the most critical part, and may remain unfrozen when the flow is too fast so that ice formation is hampered. Porous sand beds, continuous cracks, rock joints, and faults may become significant contributing factors that influence the artificial ground freezing practice in the field.