Accepted Manuscript

Study on the influence of water flow on temperature around freeze pipes and its
distribution optimization during artificial ground freezing

Shibing Huang, Yunlin Guo, Yanzhang Liu, Lihua Ke, Guofeng Liu, Cheng
chen

PII: S1359-4311(17)36158-6
DOI: https://doi.org/10.1016/j.applthermaleng.2018.02.090
Reference: ATE 11872

To appear in: Applied Thermal Engineering

Received Date: 23 September 2017

Revised Date: 10 February 2018
Accepted Date: 25 February 2018

Please cite this article as: S. Huang, Y. Guo, Y. Liu, L. Ke, G. Liu, C. chen, Study on the influence of water flow
on temperature around freeze pipes and its distribution optimization during artificial ground freezing, Applied
Thermal Engineering (2018), doi: https://doi.org/10.1016/j.applthermaleng.2018.02.090

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Study on the influence of water flow on temperature around freeze pipes and its

distribution optimization during artificial ground freezing

Shibing Huanga,b, , Yunlin Guoa,Yanzhang Liua,b, Lihua Kea, Guofeng Liuc, Cheng

chena

a
School of Resources and Environmental Engineering, Wuhan University of Science and

Technology, Wuhan, Hubei 430081, China

b
Hubei Key Laboratory for Efficient Utilization and Agglomeration of Metallurgic Mineral

Resources, Wuhan, 430081, China.

c
School of Highway, Chang’ an University, Xi’ an, Shaanxi, 710064, China.

Abstract：Artificial ground freezing is usually used to improve ground and provide

temporary support. However, the freezing process is dramatically influenced by water

flow. Evidently, water flow with high velocity brings large heat energy that prevents

the freezing of porous media around freeze pipes. In this paper, for safety and

energy-saving, the influence of water flow on freezing process is simulated by a

developed coupled hydro-thermal model considering water/ice phase transition and

the positions of freeze pipes around circular tunnel are optimized through combining

this model with Nelder–Mead simplex method based on COMSOL multiphysics

platform. According to the evolution law of freezing band under well-distribution of

freeze pipes, three kinds of potential frequently-used probability distribution functions

are adopted to reduce control parameters and improve the efficiency of optimization
program including normal distribution, Poisson distribution and chi-square

distribution. The results show that the proposed combining simulation method is

suitable for optimization of freeze pipes arrangement. The freezing time is

significantly reduced no matter which one of the above distribution functions is

employed through careful design.

Keywords: Artificial ground freezing; Coupled hydro-thermal model considering

water/ice phase transition; Seepage flow; Nelder–Mead simplex method; Distribution

function.


Corresponding author. Tel.: +86 18502755916. E-mail address: huang1989.9@163.com

1. Introduction

Freezing of porous media has been widely studied, either natural or artificial

ground freezing [1]. Artificial ground freezing method, providing temporary support

and waterproof layer in geotechnical engineering construction, are extensively used in

soft rock and soils [2, 3]. The fundamental principle in ground freezing is to inject

cold energy and convert pore water into ice [4]. When the underwater tunnels pass

through the rivers or straits, the process of construction in soft rock or soils usually is

very difficult because of inrush disaster. Moreover, water may flow around the

underwater tunnels before excavation because of the difference of water head.

Artificial ground freezing technique is very suitable to improve the soil strength and

prevent water flow into the excavated area in this case. To meet the demand of frozen

soil strength, the thickness of frozen arch should be greater than 1.5 m [5]. However,
water flow brings much thermal energy against the formation of freezing arch and

changes the freezing path. The required freezing time of forming a specified frozen

arch is also considerably influenced by high seepage flow. It has been investigated the

area between two freeze pipes may not freeze if the flow velocity of water exceeds

1-2 m/d in high permeability soils or fractured media [6]. Moreover, when the

temperature drops below freezing point, partial water in porous media gradually

freezes but some liquid water is also existed [7-8]. Therefore, the freezing process of

porous media is very complicated but important to conduct artificial freezing

construction, which is associated with the coupled hydro-thermal action under low

temperature.

Many freezing models of coupled heat and water flow, freezing experiments and

related numerical studies on porous media have been reported in the past decades

[9-12]. However, the influence of high seepage flow on freezing process is

rarely considered [13]. Besides, at present, there are a lot of researches on freezing

construction technology and case analysis about artificial ground freezing, in which

freeze pipes are arranged evenly or empirically [14-16]. Nevertheless, the freezing

rates between two adjacent freeze pipes are unequal in the presence of seepage flow

and freeze pipes may satisfy a certain distribution law which needs to be further

studied [17]. To our knowledge only Marwan et al. [4] has tried to optimize the

positions of freezing pipes by Ant Colony Optimization and it results in a significant

decrease of freezing time. Due to the special freezing process of porous media under

high seepage velocity, it is necessary to study the freezing process and propose
corresponding optimization method during artificial ground freezing for safety,

energy saving and high efficient construction.

In this paper, the Nelder–Mead simplex method is adopted to optimize the

positions of freeze pipes through combining it with a developed coupled

hydro-thermal model in order to derive a required frozen arch as fast as possible. The

Nelder-Mead simplex method is an efficient and derivative free optimization

algorithm to find the minimum or maximum value of an objective function, which has

been widely applied in many fields, including engineering materials, geotechnical and

hydrologic engineering [18,19]. It is relatively simple and suitable for not too many

control parameters, in which only a numerical evaluation of the objective function is

required [20]. The basic idea for the simplex algorithm from geometry is shown in

Fig. 1. In the three dimensional space, a simplex is a special tetrahedron determined

by four points (F1, F2, F3 and F4) and their connected lines. The objective function is

estimated at every point. The highest point, where the objective function is largest

(e.g. Point F3), will be perpendicularly mirrored against the opposite plain segment to

a lower point, which is also accompanied with an expansion or contraction to modify

step size in order to reach the optimization valley floor. A termination criteria should

be given to stop this optimization procedure, generally including the maximum

number of reflections or a tolerance for critical variables. Similarly, it can be extended

to the N dimension in which the simplex is a special polytope of N+1 vertices.

 Fig. 1. Nelder-Mead simple for three optimization parameters [21]

Here a coupled hydro-thermal model considering the influence of water/ice

phase transition, which is firstly studied by Tan et al. [22], is proposed and validated

by a previous large-scale experiment of artificial ground freezing considering the

effect of water flow on the freezing process in Section 2. Then an optimization

method of freeze pipes arrangement, combining the coupled hydro-thermal model

with an efficient optimization algorithm (Nelder-Mead simplex method), is proposed

and applied in a circular tunnel constructed by artificial ground freezing method in

Section 3. The freezing temperature around freeze pipes and the minimum freezing

time under different seepage flow conditions is investigated before and after

optimization in Section 4. In Section 5, some issues about the topic are discussed.

Finally, some significant conclusions are drawn in the last section.

2. Coupled hydro-thermal model considering water/ice phase transition

Thermal transfer and water flow during freezing around freeze pipes should be

considered when investigating the influence of water flow on freezing temperature

around freeze pipes during artificial ground freezing. Therefore, two crucial functions

should be presented, heat conduction equation with phase change deduced from
energy conservation and continuity equation considering water flow deduced from

water mass conservation.

2.1 Governing equations

2.1.1 Basic assumptions

Considering the actual physical process, some basic assumptions are introduced

as below: (1) The rocks or soils are saturated, homogeneous and isotropic porous

media; (2) The evaporation process of water is ignored, and the Darcy’s law is

suitable to describe the groundwater flow in porous media; (3) The heat conduction in

freezing porous media satisfies Fourier’s law.

2.1.2 Heat conduction equation

A fully coupled hydro-thermal model has been firstly proposed by Tan et al. [22].

The heat conduction equation deduced from energy conservation in freezing porous

media is expressed as
T
Cv  l cl vl  (T )    (eT )  0 (1)
t
Where  and c are density and specific heat capacity, respectively. Subscript s, l

and i represent solid matrix, water and ice, respectively. T is temperature. v l is the

seepage velocity vector of water.

Water density is the function of pressure and temperature:

l  l 0 [1  T (T  T0 )   p ( pl  p0 )] (2)

Where l 0 is the density corresponding to the initial pressure p0 and initial

temperature T0 ;  T is the thermal expansion coefficient of water related with

temperature;  p is water compressibility; pl is water pressure. When p0  101 KPa and

T0  20 ℃, there are T  (9T  80) 106 /℃and  p =5 106 /Pa. The change of

water density with temperature when p0  101 KPa is shown in Fig. 2.

1000

999

l (kg/m3) 998

997

996

995

994

993
-25 -20 -15 -10 -5 0 5 10 15 20 25 30
T (℃ )

Fig. 2 The change curve of water density with temperature [23]

e is the effective thermal conductivity of porous media, which depends on the

thermal conductivity of its components (solid, unfrozen water and ice), their

volumetric fractions and the spatial distribution of its components. The exponential

weighted mean model is usually used to describe its change rule [24, 25]:

e  s1-nlnw in(1w )
u u
(3)

Where n is porosity and  is thermal conductivity. wu is the unfrozen water

content, which could be expressed as an exponential function of temperature [26]:

wu  eM (T Tm ) (4)

Where M is a parameter related with the distribution of pore radius. Tm = 0 ℃ is

the freezing point of bulk water.

Cv is the equivalent volumetric thermal capacity, which can be expressed as

follows:
wu
Cv  (1  n)  s cs  n(1  wu ) i ci  nwu l cl  nl  (5)
T
Where is the latent heat of water/ice phase transition; c is specific heat.
2.1.3 Continuity equation

The continuity equation of water/ice has been derived by Tan et al. [22] as

below:
( l wu n) [ i (1  wu )n]
    ( l v l )  0 (6)
t t
Assume water seepage in porous media satisfies Darcy’s law. Considering the

effect of segregation potential, water migration in freezing porous media is

kr
vl   k (pl  l g)  SP0T (7)
l

 ks 0
Where k = 
ks 
is the intrinsic permeability matrix, in which k s is the
0

0
permeability of saturated porous media before freezing; g =   is the gravity
g 

acceleration vector. SP0 is segregation potential, which is a function of average

suction in the frozen fringe, the applied pressure on the porous media and the rate of

cooling [27-29]. For example, the average value of segregation potential of Calgary

silt is approximately 2.3×10-9 m2/℃/s in segregation interval -0.3 ~ -0.1℃ [30].

l is the viscosity of liquid water which could be expressed as a function of

freezing temperature [23]:

1808.5
l  2.1106 exp( ) (8)
273.15  T
k r is the relative permeability which could be written as a function of water

saturation [31]:

kr  Sl [1  (1  Sl1/ ) ]2 (9)

Where  is material constant and Sl is water saturation.

The water saturation is equal to unfrozen water content during freezing, then we
can derive the final expression of relative permeability as below:

kr  eM (T Tm ) [1  (1  eM (T Tm )/ ) ]2 (10)

2.2 Model validation

Pimentel et al. [32] had conducted a large-scale laboratory test on artificial

ground freezing under different initial seepage flow velocities from 0 m/d to 2.1 m/d.

Medium sand was filled in a watertight box with inner dimensions 1.3 m×1.0 m×1.2

m to simulate porous material in this experiment. The box has a stable water inlet at

left and water outlet at right. All the surfaces have thermal insulation with at least 18

cm of foam to create adiabatic boundary conditions. Three freeze pipes with diameter

of 0.041 m perpendicular to the seepage-flow direction are located in the middle line.

Several thermistors of type Pt 100 are also placed along two middle lines (ML1 and

ML2) inside the porous media to monitor the freezing temperature field, which are

indicated by black solid points (the detailed coordinate positions can be found in the

original literature). This large-scale physical simulation model can be simplified to a

plane problem as is shown in Fig. 3.

The unfrozen water content of this porous material under different freezing

temperature was measured using Time-Domain Reflectometry (TDR) in the

experiment. Then, we can derive the unknown parameters M=3.93 from Eq. (4) by

fitting the experimental results as is shown in Fig. 4. Obviously, unfrozen water

content can be well expressed by the exponential function of temeprature. It should be

noted that the calculated freezing point of 0.1℃ is slightly greater than theoretical

value 0℃, which may be caused by measurement error. The other physical and
thermodynamics parameters for porous material and water/ice media are listed in

Table. 1 and Table. 2, respectively. The segregation potential of this medium sand is

very small, which may be negligible [33]. Two different seepage boundary conditions

(Initial seepage velocities: v0 = 0 m/d and v0 = 2 m/d) were investigated in this

experiment. Thermal boundary conditions of freeze pipes and water inlet gauged by

the temperature sensors are given in Fig. 5. Numerical analysis is carried out using the

developed coupled hydro-thermal model under the same conditions. The calculated

results of temperature at all measured positions are in good agreement with the

experimental values after different freezing time as is shown in Fig. 6. Obviously, the

freezing rates in porous media without water seepage are much faster. After freezing

of 45 hours, all the measured points on Line ML2 are frozen with the absence of water

flow while many measured points on Line ML2 but far from freeze pipes are unfrozen

under water flow of 2 m/d due to the inflow of heat. Therefore, the seepage flow will

hinder freezing around freeze pipes and result in the increase of freezing time when a

frozen band connecting freeze pipes is created. Actually, the calculation accuracy of

the proposed coupled hydro-thermal model under low temperature are greatly

influenced by the critical coupling parameters, mainly including the unfrozen water

content and the effective thermal conductivity of freezing porous media. Through the

comparisons between the calculated and measured temperature as is shown in Fig. 6,

those two important parameters can be well expressed by the exponential weighted

mean model and the exponential function, respectively.

 Table 1 The values of thermo-mechanical parameters of experiment material [32]

ρs (kg λs (W/ cs (J/ ks (m SP0 (m2/ M 

n
/m3) m/°C) kg/°C) /s) °C/s) (/°C)

0.41 1571 4.3 816 0.0001 0 3.93 0.5

Table 2 The values of calculation parameters of water/ice

 ρi cl ci λi (W/ λl (W/

(J/kg) (kg/m3) (J/kg/°C) (J/kg/°C) m/°C) m/°C)

334.88×103 917 4200 2100 2.2 0.6

Fig. 3 Plan view at measuring plane with location of thermistor

 Experiment values
1.0
Fitting curve

0.8

Unfrozen water content

0.6

0.4 wu  e3.93(T 0.1)

0.2

0.0

-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5

Temperature (°C)

Fig. 4 Measured and calculated unfrozen water content against temperature

25 25
20 20
Th (Freeze pipe)
15 15
10 10 Th (Freeze pipe)
Tw0 (Water inlet)
5 5
T (°C)

T (°C)

0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
t (h) t (h)

(a) v0 = 0 m/d (b) v0 = 2 m/d

Fig. 5 Thermal boundary conditions of freeze pipes and water inlet

30 20
Experiment results T10 T11 T13 Experiment results T28 T29 T30
25 T22 T36 T37 15 T31 T33 T34 T35
20 Developed model T10 T11 T13 Developed model T28 T29 T30
10 T31 T33 T34 T35
T22 T36 T37
15
5
10
0
T (°C)

5
T (°C)

0 -5

-5 -10
-10 -15
-15
-20
-20
-25
-25 0 10 20 30 40 50 60 70 80 90 100
0 10 20 30 40 50 60 70 80 90
t (h) t (h)

(a) v0 =0 m/d (ML1) (b) v0 =0 m/d (ML2)

 30 25
Experiment results T10 T11 T13
T22 T36 T37 Experiment results T28 T29 T30
25
Developed model T10 T11 T13 20 T31 T33 T34 T35
T22 T36 T37 Developed model T28 T29 T30
20
15 T31 T33 T34 T35
15

T (°C)
10
T (°C)

10
5
5
0
0

-5 -5

-10 -10
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
t (h) t (h)

(c) v0 = 2 m/d (ML1) (d) v0 = 2 m/d (ML2)

Fig. 6 Calculated and measured temperature against freezing time

3. Optimization of artificial ground freezing considering seepage flow

3.1 Model description

Consider artificial ground freezing for circular tunnel excavation under seepage

flow similar to the case studied by Marwan et al. [4], which is originally performed by

Ziegler et al. [17], as is shown in Fig. 7. Eighteen brine freeze pipes with steady

cooling temperature of -30℃ and diameter of 0.1 m are arranged to freeze soils

around designed section of this tunnel in order to form a closed frozen arch with

expected thickness of 1.5 m. The frozen arch is cooled to -3℃ for ensuring the

strength and stability [33] [35]. A stable water flows from the left to right insides

saturated porous media with initial temperature of 10℃. According to the scale and

characteristic of freezing project, half of the study area is taken to build a numerical

model (width: 30 m, depth: 10 m). The related parameters of porous media are chosen

as the same as that in the above experiment as is shown in Table 1.

 Fig. 7 Calculated model of artificial ground freezing

3.2 The influence of water seepage on freezing temperature

The development of frozen arch ( T  3C ) under well-distributed pipes before

optimization could be simulated by solving the developed coupled hydro-thermal

model under related boundary conditions above using finite element method as is

shown in Fig. 8. The initial flow velocity of water v0 is 1.5 m/d in this case. It shows

that a long frozen band is formed to connect the freeze pipes arranged along the flow

direction on the top of designed arch after freezing of 10 days. However, there is no

freezing intersecting band produced on the upstream or downstream side. This frozen

band grows more quickly on the downstream side along with freezing time because of

the resistance of heat inflow by freeze pipes on the upstream side as is shown in Fig. 8

(b). Water flows much faster through the unfrozen area of adjacent freeze pipes on the

upstream and downstream sides due to the freezing of the top side before a complete

frozen arch is formed (Larger arrows represent higher velocity in Fig. 8). After

freezing of 30 days, a complete frozen arch arises. Then, the seepage flow of water
insides excavation zone stops because of this waterproof arch. Water keeps flowing to

the downstream sides through the unfrozen area above the frozen arch. However, the

frozen arch does not occour in the designed ring and shifts to the left. The width of

frozen band on the upstream side is much smaller than that on the downstream owing

to the influence of water flow. Therefore, it is necessary to optimize the positions of

freeze pipes in order to generate a designed frozen arch using

significantly less freezing time.

(a) 10 d (b) 20 d (c) 30 d

Fig. 8 Frozen arch by well-distributed pipes (v0 =1.5 m/d, including the velocity

vector of water flow)

3.3 Optimization of freezing pipes arrangement

The positions of N freeze pipes could be determined by the polar coordinates

 ri ,i  . 2 × N unknown parameters need to be optimized if the distribution law of

freeze pipes is not considered. Therefore, some potential distribution functions may be

adopted to reduce the numbers of optimized parameters and improve the efficiency of

optimization. The radial coordinates of the freeze points (the centers of those freeze

pipes) could be written as (Fig. 9)

ri  r0   r (i ) (11)

Where r0 is the radius of center circle of designed arch;  r (i ) is the radially shift
distance of freeze points after optimization.

Fig. 9 Positions of freeze pipes after optimization in polar coordinate system

 r (i ) may be assumed to be linear variation with  i on the downstream side

and constant on the upstream side, which could be expressed as below [4]:

d 2 i   / 2

 r (i )=  2 (12)
 (d1  d 2 ) i +2d 2  d1 i   / 2
 

Where d1 and d 2 are the maximum radially shift distance of freeze points on the

downstream side and on the upstream side, respectively. Considering the designed

thickness of frozen arch (1.5 m) and the radial development of frozen band around

freeze pipes as is shown in Fig. 8, they can be set as 0.5  d1  0 and 0  d2  0.5 ,

respectively. It means that all the freeze pipes should be move from downstream

toward upstream side as is shown in Fig. 9.

The other position parameter need to be characterized is the angular coordinate

 i .There is

0 i 1
i =  (13)
i 1 +i 1 i 1

Where i 1 is the difference of angular coordinate between the i th and (i-1) th

freeze points. The positions of freeze pipes after optimization in polar coordinate
system are shown in Fig. 9 (red solid points).

As we know, a complete frozen band between two adjacent freeze pipes must

occur on the top of designed arch firstly and extend to both sides under

well-distribution freeze pipes as is shown in Fig. 8. Besides, the freezing of porous

media on the upstream side is much harder due to water flow. Therefore, the following

three condition should be satisfied when rearranging the freeze pipes: (1) Freeze pipes

should be more compactly rearranged on the upstream side to reduce freezing time of

total designed arch; (2) i 1 between two adjacent freeze pipes should increase with

the increasing of  from left to right until reaches the maximum value; (3) The

maximum i 1 should be located on the position after the fourth freeze pipe where i ≥

5. By the above analysis, three potential and common distribution functions are

adopted to rearrange the freeze pipes, including normal distribution, Poisson

distribution and chi-square distribution, because it’s hard to determine which one is

much better before optimization.

(1) Normal distribution

The normal distribution is a very common continuous probability distribution.

The probability density function is

1 ( x   )2
f ( x,  ,  )  exp( ) (14)
2 2 2

Where  and  are distributed parameters representing mathematical expectation

and standard deviation, respectively.

As i 1 should gradually increase from upstream side to the maximum value on

the downstream side. It may be expressed as

 f (i  1,  ,  )
i 1   10

 f (i  1,  ,  )
(15)
i 2

Obviously, when i  1    0 , i 1 gets the maximum values. Therefore, the

maximum i 1 occurs between the  th and ( 1+  )th freeze pipe. Where

is the rounding symbol. It has been validated that the maximum i 1 should occur

after the fourth freeze pipe where 5  i  10 , because the frozen band starts from the

fourth freeze pipe as is shown in Fig. 8. Therefore, we have 4    9 and

4   9.

(2) Poisson distribution

Poisson distribution is an important discrete probability distribution, which can

express the probability of a given number of events occurring in a fixed interval space

if these events happen with a known average rate. The probability density function

can be expressed as

 x e 
g ( x,  )  (16)
x!

Where x  0 ;  represents the mathematical expectation and standard deviation.

According to the properties of probability density function of Poisson

distribution, when x= , g ( x,  ) gets the maximum value. As we know the

maximum i 1 should be located between the 4+k th and 5+k th freeze pipes

where k  0 . Then we can define i 1 as below

g (  (5  k )  i,  )
i 1 = 10

 g (  (5  k )  i,  )
i 2
(17)

Obviously, when   5  k  i = , i 1 reaches the maximum value. Then we

have i  5  k . k is a unknown control parameter need to be optimized. Then we can

derive 0  k  5 , because there is 5  i  10 when i 1 takes the peak value.

Another unknown parameter  is from 10 ~ 70 in order to ensure that the freeze

pipes can be sufficiently intensive layout on the upstream. For convenience, λ is

replaced by 10 λ to reduce the optimization interval. Then, we have

g (10  (5  k )  i,10 )
i 1 = 10

 g (10  (5  k )  i,10 )
(18)
i 2

Where 1    7 and 0  k  5 .

(3) Chi-square distribution

The chi-squared distribution with q freedom degrees is the distribution of a sum

of the squares of q independent standard normal random variables, which is widely

used in inferential statistics. The probability density function is

q x
1 1 
p ( x, q )  q
x2 e 2
q (19)
2 ( )
2
2
q
Where x  0 ; ( ) denotes the Gamma function.
2
Similarly, when x  q  2 , p( x, q) takes the maximum value because x is equal

to the mode. As the same, the maximum i 1 is between the 4+s th and 5+s th

freeze pipes where s  0 . Then we can define i 1 as below:

p(q  (7  s)  i, q)
i 1 = 10

 p(q  (7  s)  i, q)
(20)
i 2

From Eq. (20), i 1 reaches the maximum values when q  7  s  i  q  2 .

Then we can have i  5  s . s and q are unknown control parameters need to be

optimized. Similarly, there is 0  s  5 for 5  i  10 at the peak value of i 1 .

According to characteristics of probability density function, set 10  q  24 to ensure

that i 1 could be smaller enough on the upstream sides. Therefore, the ranges of

optimized control parameters are 0  s  5 and 10  q  24 .

From the above analysis, no matter which distribution function is chosen to

determine the optimal positions of freeze pipes, there are four control parameters need

to be optimized for rearranging the freeze pipes by Nelder-Mead simple method. The

basic idea of determining the optimal control parameters by the Nelder-Mead simple

method including the following steps: (1) Several initial values of control parameters

are inputted and the corresponding values of objective function can be calculated; (2)

The decline direction of objective function is determined by evaluating the values of

objective function; (3) New control parameters are chosen on this decline direction

and the corresponding new values of the objective function are derived; (4) By

comparing the new with the original values of the objective function, a new decline

direction can be determined. Therefore, during any optimization process, new values

of the control parameters are chosen and optimized by Nelder-Mead simple method.

New parameter values will cause rebuilt of geometry model and regeneration of FEM

model until the optimal solution is obtained or the maximum number of objective

evaluations is reached. This whole optimization procedure is shown in Fig. 10.

In the optimization process, the objective function is the freezing time calculated

by FEM model when a closed frozen arch (T≤ -3℃) with expected thickness of 1.5 m

is created, which can be expressed as

 min t (x) subject to Tfa ≤ -3℃ (21)
X R

Where t is the freezing time. x is the vector of the control parameters. R is

the bound set of the control parameters which have been derived above. Tfa is the

maximum temperature in the expected frozen arch.

Therefore, the less the completed freezing time, the better the freezing quality.

Fig. 10 Flowchart of the optimization.

4. Results

4.1 Temperature around freeze pipes

In this section, the optimization simulation of ground freezing under different

initial flow velocities from 0 m/d to 2.5 m/d has been performed. The freezing process

of designed arch under seepage flow of 1.5 m/d after optimization are shown in Fig.
11 ~ Fig. 13. After freezing of 5 days, the maximum frozen zone arises around the

sixth or the seventh freeze pipe on the top because of the prevention of heat inflow by

the fourth and fifth freeze pipes (the number of freeze pipes increases along

counterclockwise). It is encouraging that the frozen bands occur not only on the top

but also on the corners of designed arch because of the space reduction of freeze pipes

on the corners and extend to the middle until a long and complete frozen arch is

formed. Therefore, it is different from the growth process of frozen band with

well-distributed pipes. Moreover, the optimal freezing process is very close to each

other no matter which distribution function is adopted. As a results, the freezing

process is faster and a thicker complete frozen arch is created after about 25 days after

optimization.

(a) 5 d (b) 15 d (c) 25 d

Fig. 11 Frozen arch after optimization adopting normal distribution (v0 =1.5 m/d)

(a) 5 d (b) 15 d (c) 24 d

Fig. 12 Frozen arch after optimization adopting Poisson distribution (v0 =1.5 m/d)
 (a) 5 d (b) 15 d (c) 25 d

Fig. 13 Frozen arch after optimization adopting chi-square distribution (v0 =1.5 m/d)

4.2 The minimum freezing time

What’s more, the minimum freezing time of designed arch exponentially

increases with the increase of water flow velocity before and after optimization. A

desirable prediction function for estimating the minimum freezing time has been

derived as is shown in Fig. 14. The completed freezing time of designed frozen arch

increases faster under well-distributed pipes, which is about 108 days in the presence

of seepage flow of 2.0 m/d. However, only 38 days is needed to create a designed

frozen arch under the same seepage flow condition after optimizing the positions of

freeze pipes by the Nelder-Mead simplex method combining with any one of the three

distribution functions mentioned above. It means that the Nelder-Mead simplex

method is very effective for optimization of freeze pipes arrangement and the chosen

distribution functions (normal distribution, Poisson distribution and chi-square

distribution) can well characterize the optimal distribution law of freeze pipes under

water flow condition by careful design.

 120
Normal distribution
Poisson distribution
100 Chi-square distribution
Uniform distribution

Completed freezing time tf (d)

Fitting curve before optimization
80 Fitting curve after optimization

60 t f  16  2.53e1.79v0

40

20
t f  16  0.42e2v0

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Initial seepage velocity v0 (m/d)

Fig. 14 Comparison of completed freezing times versus initial seepage velocity

4.3. The optimum positions of freeze pipes

Actually, the minimum freezing time corresponds to several groups of values of

control parameters after optimization no matter which distribution function is adopted.

Fig. 15 has shown that many combinations of values of control parameters may result

in the minimum freezing time of designed arch. All of those values of the control

parameters belong to the optimal parameter set, which are represented by the solid

points in Fig. 15. The minimum freezing time cannot be obtained by the other

parameter set, so those values are attributed to error parameter set, which are

represented by the hollow points in Fig. 15. Evidently, the control parameters will

converge to be stable values corresponding to the minimum freezing time after dozens

of optimization steps. However, in order to reduce the excavation and disturbance area

of frozen zone, the excess frozen zone beyond designed frozen arch should be located

outside the excavation section as far as possible. Actually, the excess frozen zone on

the upstream side is small and there is a little different between the maximum and the

minimum d2 of the optimal parameter set as shown in Fig. 11~13 and Fig. 15,
respectively. Therefore, the minimum d1 of the optimal parameter set may be adopted

to reduce the frozen zone in the excavation area on the downstream side. Based on the

above principle, the most optimum group of parameters under different flow

velocities are derived and listed in Table. 3. It is clear that the values of common

parameters (d1 and d2) from different distribution functions are very close under the

same flow condition. Because the radial movement of freeze pipes is decided by Eq.

(12) no matter which distribution function is adopted. Compared with the

optimization method raised by Marwan et al. [4], in which the Ant Colony

optimization algorithm is applied to search the optimal control parameters, the

proposed combining optimization method in this paper have much better stability and

reliability, including the control parameters and the minimum freezing time. Moreover,

the most optimum d1 gradually increases with the increasing of water flow velocity.

The influence of water flow velocity on the most optimum positions of freeze pipes

can be seen in Fig. 16. Obviously, the freeze pipes should be increasingly closely

arranged on the upstream side and the total freeze pipes gradually move to left as

water flows faster. It should be noted that our model is limited to the initial flow

velocity of water less than 3 m/d, which is the nature flow velocity of groundwater

inside porous media in most cases. When the water flows much fast, there may be

turbulence flow, which is beyond the scope of this research. The freezing of porous

media under turbulence flow of water need to be researched in the future, including

the freezing process of fractured media under much high seepage velocity.
 0.6
Optimal solution set d1 d2
9 Optimal solution set 
Error solution set d1 d2
8 Error solution set 
0.4 7
6


0.2 5
4
d1(d2) (m)

3
0.0 0 10 20 30 40 50 60 70 80 90
9 n
Optimal solution set 
-0.2 8
Error solution set 
7
-0.41 6


-0.4
-0.50 5
4
-0.6
3
0 10 20 30 40 50 60 70 80 90
0 10 20 30 40 50 60 70 80 90
n
n

(a) Normal distribution

6
0.6 Optimal solution set k
Optimal solution set d1 d2 5 Error solution set k
Error solution set d1 d2 4
0.4
3
k

2
0.2
1
d1(d2) (m)

0
0.0 0 10 20 30 40 50 60 70 80
7 n
Optimal solution set 
6
-0.2
5 Error solution set 
-0.38 4


-0.4 3
-0.50 2
1
-0.6 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
n n

(b) Poisson distribution

0.6 6
Optimal solution set d1 d2 5 Optimal solution set s
Error solution set d1 d2 Error solution set s
4
0.4
3
s

2
0.2 1
0
d1(d2) (m)

0.0 0 10 20 30 40 50 60 70 80 90
24
n
Optimal solution set q
-0.2 22 Error solution set q

-0.40 20
q

-0.4
-0.50 18

-0.6 16
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
n n

(c) Chi-square distribution

Fig. 15 Determination of the optimum parameters (v0 =1.5 m/d)

 Fig. 16 The most optimum positions of freeze pipes using normal distribution

Table. 3 The most optimum group of parameters for arranging freeze pipes

v0 (m/d) d1 (m) d2 (m)  

0.5 -0.15 0.19 6.20 7.59

Normal 1 -0.19 0.22 6.67 6.48

distribution 1.5 -0.41 0.24 6.82 5.75

2 -0.49 0.19 8.58 6.51

2.5 -0.49 0.18 7.10 5.93

v0 (m/d) d1 (m) d2 (m) k λ

0.5 -0.23 0.24 2.67 3.74

Poisson 1 -0.21 0.25 3.10 3.84

distribution 1.5 -0.38 0.25 4.00 4.00

2 -0.43 0.21 4.36 4.00

2.5 -0.49 0.17 4.53 4.13

Chi-square v0 (m/d) d1 (m) d2 (m) s q

distribution 0.5 -0.13 0.20 1.08 21.82

 1 -0.16 0.23 1.05 23.52

1.5 -0.40 0.27 1.67 20.00

2 -0.45 0.28 2.00 20.90

2.5 -0.48 0.17 0.94 17.06

5. Discussion

Because of the injection of heat by water flow, the freezing process of designed

arch during artificial ground freezing is much harder, especially the freezing of porous

media on the upstream sides. In order to generate a complete frozen arch within

limited time, it necessary to improve freezing intensity on the upstream sides or

adding straight-row freezing pipes to cut off water flow [36]. Optimization of freeze

pipes arrangement is an energy-saving and effortless technique. Therefore, in this

paper an optimization method to rearrange freeze pipes by combining finite element

analysis of coupled hydro-thermal model with the Nelder–Mead simplex method is

proposed. As we known, the release of latent heat caused by water/ice phase transition

will prevent the freezing process [37]. Therefore, the influence of latent heat of

water/ice phase transition on freezing is also considered. For the sake of reducing the

control parameters and improving calculation efficiency, the optimum position of

freeze pipes is assumed to satisfy a particular distribution function including normal

distribution, Poisson distribution and chi-square distribution. Then, only four control

parameters are chosen to be optimized and related ranges of values are determined by

analysis. It should be noted that there may be some other reasonable distribution
functions can be used to promote optimization process. We cannot try one by one

owing to the limited time and so many probability distribution functions. Therefore,

three commonly-used distribution functions are employed in this paper by careful

design. As a result, the completed freezing time of designed arch is greatly reduced by

adopting this combining simulation method. However, in this paper there are still two

issues need to be addressed in the future research:

(1) Optimization of freeze pipes arrangement around non-circular cross section

It can be seen that our combining simulation method is suitable for circular

section excavation, around which the freeze pipes are also arranged in a circle.

However, there are many shapes of cavern including U-shaped, arch, square section

and so on [3] 34]. The optimum distribution of freeze pipes are more complicate in

those cases. Therefore, it’s necessary to establish the possible distribution function of

optimal positions of freeze pipes around non-circular cross section.

(2) Combining optimization of artificial ground freezing with other

pre-supported technology

It has been derived the freezing time exceeds 38 days after optimization of freeze

pipes arrangement when the initial seepage flow is more than 2.0 m/d in the case

study as is shown in Fig. 14. Therefore, the proposed optimization method combining

with other pre-supported method may be more economical and time-saving, such as

the combining method of pipe-roofing and artificial ground freezing [14]. The

application of tubular roof between the freeze pipes could well prevent water flow,

enhance heat transfer and decrease the deformation of surrounding environments [15].
Therefore, a more effective and energy-saving optimization method of artificial

ground freezing combined with other pre-supported method may need to be studied

further.

6. Summary

A new combining optimization method of freeze pipes arrangement considering

water flow has been proposed and applied to the artificial ground freezing process

during tunnel excavation. From this study, the following conclusions can be drawn:

(1) The developed coupled hydro-thermal model considering water/ice phase

transition is suitable to simulate the freezing process of porous media during

artificial ground freezing under water flow condition.

(2) Through the comparisons between the calculated and measured temperature, the

unfrozen water content and the effective thermal conductivity of freezing porous

media could be well expressed by the exponential weighted mean model and the

exponential function, respectively.

(3) The development of freezing band around freeze pipes is significantly influenced

by water flow and the completed freezing time of designed arch tremendously

increases with the increase in the initial flow velocity of water, which can be well

predicted by an exponential equation.

(4) The proposed combining optimization method is very appropriate for searching

out the most optimum positions of freeze pipes under seepage flow no matter

which one of the common distribution functions is adopted to arrange freeze pipes

including normal distribution, Poisson distribution and chi-square distribution.

 (5) Our proposed method is limited to optimize freeze pipes around circular section at

present. More complicated distribution functions need to be established if this

optimization method is applied to the other section shapes when using artificial

freezing method.

Acknowledgements

This work was supported by the National Natural Science Foundation of China

(Grant No. 41702291), the Natural Science Foundation of Hubei Province (No.

2015CFA142), the Hubei Key Laboratory for Efficient Utilization and Agglomeration

of Metallurgic Mineral Resources (2017zy005).

References
[1] O.B. Andersland, B. Ladanyi, Frozen ground engineering, John Wiley & Sons,

Inc. 2004.

[2] G. Russo, A. Corbo, F. Cavuoto, et al., Artificial Ground Freezing to excavate a

tunnel in sandy soil, Measurements and back analysis, Tunn. Undergr. Space

Technol. 50 (2015) 226-238.

[3] E. Pimentel, S. Papakonstantinou, G. Anagnostou, Numerical interpretation of

temperature distributions from three ground freezing applications in urban

tunnelling, Tunn. Undergr. Space Technol. 28 (2012) 57-69.

[4] A. Marwan, M. M. Zhou, M. Z. Abdelrehim, et al., Optimization of artificial

ground freezing in tunneling in the presence of seepage flow, Comput. Geotech.

75 (2016) 112-125.

[5] X. D. Hu, S. J. Deng, Ground Freezing Application of Intake Installing

 Construction of an Underwater Tunnel. Procedia Engineering, 165 (2016)

633-640.

[6] M. Vitel, A. Rouabhi, M. Tijani, et al., Thermo-hydraulic modeling of artificial

ground freezing: Application to an underground mine in fractured sandstone,

Comput. Geotech. 75 (2016) 80-92.

[7] E. J. A. Spaans, J. M. Baker, The soil freezing characteristic: Its measurement and

similarity to the soil moisture characteristic, Soil Sci. Soc. Am. J. 60 (1996)

13-19.

[8] Y. M. Lai, W. Pei, M. Zhang, et al., Study on theory model of

hydro-thermal–mechanical interaction process in saturated freezing silty soil, Int.

J. Heat. Mass. Tran. 78 (2014) 805-819.

[9] K. Hansson, J. Šimůnek, M. Mizoguchi, et al., Water flow and heat transport in

frozen soil, Vadose Zone J. 3 (2014) 693-704.

[10] J. M. McKenzie, C. I. Voss, D. I. Siegel, Groundwater flow with energy transport

and water–ice phase change: numerical simulations, benchmarks, and application

to freezing in peat bogs, Adv. Water Resour. 30 (2007) 966-983.

[11] W. B. Yu, W. B. Liu, Y. M. Lai, et al., Nonlinear analysis of coupled

temperature-seepage problem of warm oil pipe in permafrost regions of Northeast

China, Appl. Therm. Eng. 70 (2014) 988-995.

[12] W. Song, Y. Zhang, B. Li, et al., A lattice Boltzmann model for heat and mass

transfer phenomena with phase transformations in unsaturated soil during

freezing process, Int. J. Heat Mass Tran. 94 (2016) 29-38.

[13] M. Vitel, A. Rouabhi, M. Tijani, et al., Modeling heat and mass transfer during

ground freezing subjected to high seepage velocities, Comput. Geotech. 73 (2016)

1-15.
[14] X. D. Hu, S. J. Deng, H. Ren, In Situ Test Study on Freezing Scheme of

Freeze-Sealing Pipe Roof Applied to the Gongbei Tunnel in the Hong

Kong-Zhuhai-Macau Bridge, Appl. Sci. 7 (2017): 1-13.

[15] Q. X. Yan, Y. J. Xu, W. B. Yang, et al., Nonlinear transient analysis of

temperature fields in an AGF project used for a cross-passage tunnel in the

Suzhou Metro, KSCE Journal of Civil Engineering, (2017), 1-11. DOI

10.1007/s12205-017-1118-4

[16] Y. S. Kang, Q. S. Liu,Y. Cheng, et al., Combined freeze-sealing and New

Tubular Roof construction methods for seaside urban tunnel in soft ground, Tunn.

Undergr. Space Technol. 58 (2016) 1-10.

[17] M. Ziegler, C. Baier, B. Aulbach, Numerical simulation of artificial ground

freezing applications in tunnelling, Computational Methods in Tunneling (EURO:

TUN 2009), Aedificatio: Ruhr University Bochum , 2009, pp. 343-350.

[18] M. A. Luersen, R. Le Riche, Globalized Nelder–Mead method for engineering

optimization, Comput. Struct. 82 (2004) 2251-2260.

[19] R. Barati, Parameter estimation of nonlinear Muskingum models using

Nelder-Mead simplex algorithm, J. Hydrol. Eng. 16 (2011) 946-954.

[20] J. A. Nelder, R. Mead, A simplex method for function minimization, Comput. J.

7 (1965) 308-313.

[21] A. Ouria, M. M. Toufigh, Application of Nelder-Mead simplex method for

unconfined seepage problems, Appl. Math. Model. 33 (2009) 3589-3598.

[22] X. J. Tan, W. Z. Chen, H. M. Tian, et al., Water flow and heat transport including

ice/water phase change in porous media: numerical simulation and application,

Cold Reg. Sci. Technol. 68 (2011) 74-84.

[23] D. R. Lide, handbook of chemistry and physics, 84th edition, CRC press Inc,
 (2003-2004), pp. 980－981.

[24] X. J. Tan, W. Z. Chen, D. S. Yang, et al., Study on the influence of airflow on the

temperature of the surrounding rock in a cold region tunnel and its application to

insulation layer design, Appl. Therm. Eng. 67 (2014) 320-334.

[25] S. B. Huang, Q. S. Liu, A. P. Cheng, et al., A fully coupled

thermo-hydro-mechanical model including the determination of coupling

parameters for freezing rock, Int. J. Rock Mech. Min. 103 (2018) 205-2144.

[26] S. B. Huang, Q. S. Liu, Y. Z. Liu, et al, Freezing strain model for estimating the

unfrozen water content of saturated rock under low temperature, Int. J. Geomech.

18 (2018) 04017137. Doi:10.1061/(ASCE)GM.1943-5622.0001057

[27] J. M. Konrad, N. R. Morgenstern, The segregation potential of a freezing soil.

Can. Geotech. J. 18 (1981) 482-491.

[28] J. M. Konrad, Influence of cooling rate on the temperature of ice lens formation

in clayey silts. Cold Reg. Sci. Technol. 16 (1989), 25-36.

[29] J. M. Konrad, N. R. Morgenstern, Effects of applied pressure on freezing soils.

Can. Geotech. J. 19 (1982) 494-505.

[30] J. M. Konrad, M. Shen, 2-D frost action modeling using the segregation potential

of soils, Cold Reg. Sci. Technol. 24 (1996) 263-278.

[31] S. Nishimura, A. Gens, S. Olivella, et al., THM-coupled finite element analysis of

frozen soil: formulation and application, Geotechnique. 59 (2009) 159-171.

[32] E. Pimentel, A. Sres, G. Anagnostou, Large-scale laboratory tests on artificial

ground freezing under seepage-flow conditions, Geotechnique. 62 (2012):

227-241.

[33] J. M. Konrad, Frost susceptibility related to soil index properties. Can. Geotech.

J. 36 (1999) 403-417.
[34] S. Y. Li, Y. M. Lai, M. Y. Zhang, et al., Minimum ground pre-freezing time

before excavation of Guangzhou subway tunnel, Cold Reg. Sci. Technol. 46

(2006) 181-191.

[35] Y. Yamamoto, S. M. Springman, Axial compression stress path tests on artificial

frozen soil samples in a triaxial device at temperatures just below 0℃, Can.

Geotech. J. 51 (2014) 1178-1195.

[36] X. D. Hu, J. Z. Yu, H. Ren, et al., Analytical solution to steady-state temperature

field for straight-row-piped freezing based on superposition of thermal potential,

Appl. Therm. Eng. 111 (2017) 223-231.

[37] R. Lackner, A. Amon, H. Lagger, Artificial ground freezing of fully saturated soil:

thermal problem, J. of Eng. Mech. 131 (2005) 211-220.

Highlights

1. A developed coupled hydro-thermal model was applied for artificial

ground freezing.

2. The influence of water flow on freezing temperature around freeze

pipes was investigated.

3. A new combining optimization method of freeze pipes arrangement

was proposed.

4. Three common distribution functions were suitable to determine the

optimum positions of freeze pipes.