See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328515077

A generalized thermal conductivity model of geomaterials based on micro-

structures

Article  in  Acta Geotechnica · October 2019

DOI: 10.1007/s11440-018-0728-4

CITATIONS READS

7 528

4 authors:

Chong Wang Yuanming Lai

Lanzhou University Chinese Academy of Sciences
16 PUBLICATIONS   143 CITATIONS    248 PUBLICATIONS   4,191 CITATIONS   

SEE PROFILE SEE PROFILE

Mingyi Zhang Shuangyang Li

Chinese Academy of Sciences Chinese Academy of Sciences
108 PUBLICATIONS   1,450 CITATIONS    51 PUBLICATIONS   828 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Stability of embankment in permafrost regions View project

All content following this page was uploaded by Chong Wang on 07 November 2018.

The user has requested enhancement of the downloaded file.

Acta Geotechnica
https://doi.org/10.1007/s11440-018-0728-4 (0123456789().,-volV)(0123456789().,-volV)

RESEARCH PAPER

A generalized thermal conductivity model of geomaterials based

on micro-structures
Chong Wang1,2 • Yuanming Lai2,3 • Mingyi Zhang2,3 • Shuangyang Li2,3

Received: 3 November 2017 / Accepted: 9 October 2018

 Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract
Knowledge of particle shape and configuration-dependent thermal conductivity is necessary to investigate heat or water
transfer in geomaterials, especially under freezing states. Thermal conductivity of a porous medium is affected by its
matrix components and particle shapes. However, the particle shapes in geomaterials are various, and the effect of the ice
phase on the thermal conductivity may increase substantially as ice content increases from unfrozen to freezing. In this
study, a generalized thermal conductivity model for geomaterials is proposed based on phase transition theory and
geometry approximation with respect to unfrozen and freezing states. Volumetric contents of each component and shape
factors are required to predict thermal conductivities by the model. In order to evaluate the model, test results from both
ours and previous literatures are employed to evaluate the calculated ones, and they match very well. In addition, compared
with the other two models, i.e., a physical model and an empirical model, the proposed model is more reasonable and
effective.

Keywords Porous media  Shape factor  Thermal conductivity  Unfrozen water

1 Introduction rocks or soils, is difficult to be obtained, especially under

freezing conditions due to variational ice contents with
Thermal conductivity k (W m-1 K-1), as a basic physical temperature.
parameter to describe the transmit heat ability of a porous Various experimental studies have been performed to
medium, has been widely used in the field of geotechnical describe the thermal conductivity of porous media. These
engineering, especially in numerical modeling [21–23, 43]. researches mainly focused on the effects of components
Conduction, convection and radiation are three main and external factors on the thermal conductivity, such as
mechanisms of heat flow through porous media [11]. For the volumetric proportions of soil constituents, dry density,
solid materials, conduction is of paramount importance, water content, porosity, mineralogy, temperature, freeze–
and convection and radiation are usually negligible [3]. In thaw cycles and degree of saturation [1, 8, 9, 29, 42, 44].
practice, the thermal conductivity of porous media, such as Generally, the porous medium is described as a composite
material and includes solid particle, water, and air. The
experimental results have implied that the thermal con-
& Mingyi Zhang ductivity increases with the increase in water content
[email protected] [8, 9, 44]. Orakoglu et al. [29] investigated the effect of the
1 freeze–thaw cycle on the thermal conductivity of silty clay,
Key Laboratory of Mechanics on Disaster and Environment
in Western China, the Ministry of Education of China and which shows a decreasing trend with the increase in
School of Civil Engineering and Mechanics, Lanzhou freeze–thaw cycle. Zhang et al. [44] researched the effects
University, Lanzhou 730000, China of particle size and fines contents on the thermal conduc-
2
State Key Laboratory Frozen Soil Engineering, Northwest tivity of quartz sands. Their results indicated that the
Institute of Eco-Environment and Resources, Chinese thermal conductivity of fine sands is lower than that of
Academy of Sciences, Lanzhou 730000, China coarse sands. The research of Yuan et al. [42] demonstrated
3
University of Chinese Academy of Sciences, Beijing 100049, that the increase in porosity induces the reduction in the
China

123
 Acta Geotechnica

thermal conductivity. However, the general utility of the can also be taken as a reference for the study of the thermal
experimental study is lacking and test results are valid only properties in cold regions.
under specific situations. Furthermore, the experimental
research is expensive and time consuming, especially in the
field. Therefore, it is necessary to introduce some predic- 2 Experimental methodology
tion models to calculate the thermal conductivities of
porous media considering the physical properties. 2.1 Materials and sample preparation
Numerous empirical models have been proposed to
describe the thermal conductivity of porous media The porous media used in this study include silty clay and
[5–7, 14, 17, 19, 31, 32, 34–36, 45]. They were developed sandstone. The silty clay was sampled from the Beiluhe
by the physical properties (e.g., porosity, degree of satu- field site on the Qinghai-Tibet Plateau (34540 N and
ration, mineral content, grain-size distribution) of the por- 92560 E), and the sandstone was taken from Lanzhou, West
ous medium, and the predicted results were in agreement China (36030 N and 105500 E). The silty clays were dried,
with the measured results. For example, some empirical crushed, and then sieved separately by a sieve with 2-mm
models have been proposed to predict the thermal con- holes, and the soils with particle sizes below 2 mm were
ductivity of porous media as logarithmic functions of water used. The grain-size distribution was analyzed by the
content or power functions of dry density [19, 31, 32, 36]. Master Sizer-2000 Laser particle size analyzer (Malvern,
The predicted models mentioned above are proposed based UK), which is shown in Fig. 1. The test results showed that
on the underlying assumption that the solid particle of the silty clay sample contains approximately 30.14% clay
rocks or soils is spherical. However, the particle shapes of (\ 0.005 mm), 64.10% silt (0.005–0.075 mm), and 5.76%
geotechnical materials are various, and the influence of sand (0.075–2 mm). The sample had a volume-weighted
shape factor on the thermal properties is non-negligible. To mean diameter of 0.023 mm and a uniformity coefficient of
the authors’ knowledge, only a few researchers have 1.87 and had a low plasticity, with a liquid limit of 33.3%
investigated the thermal conductivity considering the par- and a plastic limit of 20.0%. The specific surface area was
ticle construction [5–7, 17, 34, 35]. Tarnawski and Wagner 1.480 m2 g-1. The initial water content of silty clay
extended the de Vires’s approach and proposed a new specimen was 10% (g g-1), dry density was 1.82 g cm-3,
model for the thermal conductivity of soils with different and the samples were remolded into cylinders with a
moisture contents and positive temperatures up to 95 C by diameter of 61.8 mm and a height of 50 mm. Silty clays
assuming particles to be ellipsoidal [7, 35]. But the cal- were loaded into a cylindrical mold and then compressed
culated model has not been extended to estimate the ther- into a cylindrical specimen at constant loading rate
mal conductivity of freezing porous media. Cote and [27, 46, 47]. For sandstones, the specimens were obtained
Konrad developed a generalized model integrating the by water drilling method and then remolded into cylinders
effects of porosity, degree of saturation, particle shape, with a diameter of 50 mm and a height of 100 mm [28].
mineral content, and grain-size distribution, which is used The constituent minerals of silty clay and sandstone were
to describe thermal conductivities of unfrozen or freezing analyzed by X-ray diffractometer (shown in Table 1). In
soils [5]. However, this model is a multi-factor model addition, the relative contents were calculated based on
without a detailed introduction about the particle con-
struction. Therefore, it is urgent and necessary to derive a
Cumulative volume percentage less than

superior model considering particle shapes in detail, and 100

for porous media under both unfrozen and freezing states.

The objective of this study is to develop a new model to 80
estimate the thermal conductivity of multiphase porous
a certain size [%]

media, such as soils and rocks. To simulate the variation of 60

thermal conductivity with shape factor, those are assumed
that the solid particle of porous media is ellipsoidal and the
40
configuration of porous media is SWIA (concentric ellip-
soid from the center to out layers are solid particle, water,
ice, and air, respectively). Furthermore, the proposed 20

model is evaluated using test results in both this study and

previous literatures. Additionally, the model is compared 0
1000 100 10 1 0.1
with two other general models, a physical model and an Grain-size [µm]
empirical model. The approach developed in this paper is
general and applicable to any type of porous media, which Fig. 1 The grain-size distribution curve of the silty clay

123
Acta Geotechnica

Table 1 Mineral constituents of sandstone and silty clay in this study unsteady hot-wire method includes a vertical and a cylin-
(Thermal conductivities of selected minerals are converted from drical system, which provides both heating and thermom-
Horai [15])
etry. Hot wire was heated by external power supply, and
Mineral Sandstone Silty clay Thermal conductivity then the temperature of samples increased due to heat
constituents Mass Mass (W m-1 K-1) releasing from hot wire. Further, the thermal conductivity
content content
(%) (%)
of porous media was calculated by using the measured
thermal diffusivity. Assuming a surface hot source trans-
Montmorillonite 0 0 3.19 fers heat to a boundless medium vertically, the theoretical
Illite 0 18.6 2.32 model was established [37].
Gypsum 12.2 0 1.26 The measured results can be obtained in a short time as
Kaolinite 2.8 5.3 8.14 the power of thermal systems changes rapidly; hence, the
Chlorite 0 0 5.15 process can be described as transient. For a general thermal
Quartz 42.3 42 7.69 equilibrium, the relationship between the sample temper-
Orthoclase 7.3 7 2.32 ature and time can be obtained by consulting with Fourier’s
Plagioclase 11.5 11.3 1.53–1.98 law [29]:
Calcite 12.7 13.4 3.59  
q r2
Dolomite 3.1 0 5.51 DT ¼ T  Tref ¼ Ei  ð1Þ
4pk 4at
Siderite 1.4 0 3.01
Hematite 5.4 2.4 11.28 where DT is the variation of temperature, T is the tem-
Pyrite 1.3 0 19.21 perature of the wire at time t, Tref is the initial temperature,
q is the heating power, k is the measured thermal con-
ductivity of a sample, r is the radius of hot wire, a is the
thermal diffusivity, t is the testing time, and Ei is the
mass, which can be transformed into volumetric content exponential integral function. The Taylor series of Ei can
according to mineral density. be obtained:
This study focuses on the effects of the constituents and  2 k
volume fractions on the thermal conductivities of the silty     r
 4at
r 2  r 2  X1
clay and sandstone. Four types of porous media were Ei  ¼ c þ ln  þ ð2Þ
4at 4at k¼1
kk!
chosen: unfrozen silty clay (soil, water, and air), freezing
silty clay (soil, water, ice, and air), unfrozen sandstone where c is the Euler–Mascheroni constant. If the radius of
(sand grain/mineral particle, water, and air), and freezing wire is small and the testing time is large, Ei approximate
sandstone (sand grain/mineral particle, water, ice, and air). to the first two terms of the Taylor series.
The disintegration occurred due to the cementation From Eqs. (1) and (2), the relationship between DT and
between sand grains eliminating, caused by the high water lnðtÞ is linear over the valid range between times t1 and t2 ,
contents and moisture migration. Hence, it is difficult to and the slope is q=ð4atÞ. The temperature T corresponding
prepare sandstone samples with higher saturability, espe- to the measured thermal conductivity k is given by:
cially for high-porosity samples. The initial volumetric 1
water contents of silty clay and sandstone were 18.20 and T ¼ Tref þ ½DT ðt1 Þ þ DT ðt2 Þ ð3Þ
2
13.72%, respectively. And three multiple samples were
prepared for each thermal conductivity measurement From Eqs. (1)–(3), the thermal conductivity can be
[27, 28]. The relative contents of four phases varied with expressed as:
 
negative temperatures because the unfrozen water content q 4at
decreases with decreasing temperature under freezing k¼ ln 2 c ð4Þ
2p½DT ðt1 Þ þ DT ðt2 Þ r e
states [41].

2.2 Test procedure

3 Prediction models
The thermal parameters were determined by a Quickline-
30 Thermal Parameter Analyzer (made in America, Man- 3.1 Model description
ufactured by Anter Corporation, shown in Fig. 2) for all
specimens with an accuracy of ± 0.01 W m-1 K-1. The Kleinberg and Griffin suggested that ice does not contact
transient hot-wire method was used in the analyzer with soil particle surfaces directly and forms in the interior
[24, 29]. As shown in Fig. 2, the heat transfer system of of pores [20]. Their results indicated that the representative

123
 Acta Geotechnica

Power Source
Processing system

Surface probe

A A

Heat flux

Thermistor

A A

Plastic Tube Wire

Hot wire Thermocouples Cylindrical sample

Fig. 2 The schematic diagram of the Quickline-30 thermal parameter analyzer

configuration of the freezing soil is the SWIA structure material in the x, y, and z directions, kj is the thermal
(soil, water, ice, and air, shown in Fig. 3). We can inves- conductivity of the jth material. Each material may have a
tigate the effect of particle shape on the thermal conduc- unique set of shape factors due to the different radii. The
tivity by assuming an ellipsoidal shape of matrix particles shape factors are closely related to the ratio of radii in x
with various shape factors, and the thermal conductivities axis and z axis of the ellipsoids. /j is the volume filling
of constituent components are kk , kk1 ,…, respectively ratio of the jth material, and Nx , Ny and Nz are the shape
(Fig. 3). The multiphase porous medium contains an inner factors along the main axes of the ellipsoid. The sum of the
P
ellipsoidal solid particle surrounded by air under air-satu- shape factors is always equal to 1 ( k¼x;y;z Nk ¼ 1).
rated condition, or water under water-saturated condition. The shape factors can be calculated by Campbell et al.
For unsaturated samples, from the inside to out layers are [2]:
solid particle, water and air under unfrozen condition, or
solid particle, water, ice, and air under freezing condition. Nyð jÞ ¼
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Therefore, the general form of the model for calculating the >
> m cos 1
ðmÞ  m 1  m2
>
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; 0\m\1
thermal conductivity of porous media keff can be given as: >
> 2
>
> ð 1  m 2 Þ3
>
>
keff ¼ k0 >
>
  
 > m¼1
Xn
/j   X kj ð jÞ < 1=3;
>
þ kj  k0 1= 1 þ   1 Nk 1
n k >
j¼1 k¼x;y;z >
> 2 ½ 1  mð2Þ 
>
>
ð5Þ >
> " pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi# m[1
>
> mð2Þ 1 þ 1  mð2Þ
>
>

where k is the effective relative thermal conductivity near >
>  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ln pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
>
: 3 1  1  mð2Þ
ð jÞ 4 ½1  mð2Þ 
boundaries, Nk (k ¼ x; y; z) is the shape factors of the jth
ð6Þ
1 h i
Multiphase ellipsoid Nxð jÞ ¼ 1  Nyð jÞ ð7Þ
1þm
y m h i
Nzð jÞ ¼ 1  Nyð jÞ ð8Þ
1þm
λk-3 where m ¼ c=a, and a, b, and c are the radius of the ref-
bk-2 λk-2
bk-1bk λk-1 erence ellipsoid along the main axes x, y, and z,
λk
respectively.
x
The relationships between shape factors N and m are
shown in Fig. 4. Equation (5) describes heterogeneous
mixtures with different thermal conductivities randomly
ak
ak-1
distributed and oriented in a host medium.
ak-2

Fig. 3 A multiphase porous system of ellipsoidal inclusions lying a

background of thermal conductivity k0

123
Acta Geotechnica

values and predicted values. It is a summary statistic of

statistical variability.

1X n
MAD ¼ jyi  f i j ð11Þ
n i¼1

4 Results and discussion

4.1 Sensitivity of the new model to shape

factors

The previous thermal conductivity models were proposed

Fig. 4 Shape factors N for various m values based on the assumption that the solid particle is spherical
and other media surround it uniformly, and some physical
parameters (such as bulk density, temperature, water con-
3.2 Model performance measures tent, porosity, and saturation) were included
[5, 10, 26, 33, 38]. However, these models neglected the
1. Coefficient of determination R2 fact that solid particle shape has a significant influence on
In statistics, the coefficient of determination, deno- the effective thermal conductivity of porous media [2, 7].
ted by R2 or r2, is a coefficient that indicates the pro- In order to study the effect of structure shape on the ther-
portion of the variance in the dependent variable, mal conductivity of porous media, a general model was
which is predictable from the independent variable proposed to predict the thermal conductivity by consider-
[12]. Based on the proportion of total variation of ing the shape of solid particle in this study.
outcomes, it indicates the deviation of the calculated To better understand behaviors of the new model, the solid
results from measured ones. The definition of the fraction of a porous medium could potentially be visualized
coefficient of determination is: as a mixture with a continuum of particle shapes (from
Pn spheres to disks). The geometry of water films (bound and
SSres ðyi  fi Þ2
R2 ¼ 1  ¼ 1  Pi¼1n
ð9Þ free) and ices would then be influenced by the variability and
SStot Þ2
i¼1 ðyi  y distribution of these shapes [13]. From the direct observa-
where a dataset has n values marked by y1,…, yn, each tions of particle shapes (i.e., microscopy), the assignment of
associated with a predicted value f1,…, fn, y is the different shape factors to different medium particles can be
P
mean of the observed data y ¼ 1n ni¼1 yi , SSres is the confirmed, and it is still unclear how the solid fraction would
residual sum of squares, and SStot is the total sum of potentially affect the shapes of the other phases (i.e., water
squares, respectively. The coefficient of determination and ice). Without additional information, assumption of
ranges from 0 to 1. simple geometries is likely adequate.
2. Root-mean-square deviation RMSD In order to discuss the effect of soil particle shape on the
The root-mean-square deviation (RMSD) is fre- effective permittivity, He et al. [13] assumed the aspects of
quently used to measure of the differences between soil particles (rotation ellipsoids) from less than 10-4 to
values predicted by a model and the values actually more than 50. Therefore, for different particle shapes,
observed. The RMSD represents the sample standard shape factors Nx;y;z ¼ (1, 0, 0), (0.50, 0.25, 0.25), (0.33,
deviation of the differences between predicted values 0.33, 0.33), (0.05, 0.47, 0.48), and (0.00, 0.50, 0.50) are
and observed values [16]. chosen to illustrate their effects, and relationships keff ð/w Þ
are simulated for unfrozen and freezing conditions of
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi porous media. For freezing states, it is assumed that the
Pn 2
i¼1 ðfi  yi Þ initial water content /ini is constant and no water migration
RMSD ¼ ð10Þ
n occurs. The increase in ice content /i is 1.09 times of the
decrease of water content /l , D/i ¼ 1:09D/l when water
3. Mean absolute deviation MAD is frozen.
The average absolute deviation of a dataset is the For the simulation, the total porosity of the soil–water
average of the absolute deviations between measured system is assumed to be 0.45 m3 m-3,

123
 Acta Geotechnica

kg = 0.024 W m-1 K-1 (gas thermal conductivity), demonstrates the effects of shape factors and initial water
kw = 0.55 W m-1 K-1 (water thermal conductivity), contents on the effective thermal conductivity. Shape fac-
ks = 1.80 W m-1 K-1 (solid matrix thermal conductivity) tors Nx;y;z = (1, 0, 0), (0.50, 0.25, 0.25), (0.33, 0.33, 0.33),
and ki = 2.22 W m-1 K-1 (ice thermal conductivity), and (0.05, 0.47, 0.48), and (0.00, 0.50, 0.50) are approximately
four initial water contents (/ini = 0.4, 0.3, 0.2, and represented by aspect \ 10-4, 0.5, 1, 10, and [ 50,
0.1 m3 m-3) are used. Without considering structure respectively, which reflects the particle shapes from disk to
effects, the series and parallel flow models are universally sphere and to needle. The arithmetic mean model
accepted as yielding the widest bounds. In order to evaluate [Eq. (12)] and geometric mean model [Eq. (13)] represent
the proposed model, the arithmetic mean model [Eq. (12)] two extremes of effective thermal conductivity.
and the geometric mean model [Eq. (13)] are introduced. As seen in Fig. 5, predicted values of the proposed
Xn model are located in the interval between arithmetical and
k¼ / j kj ð12Þ geometrical mean models. When the solid ice phase
j¼1 transforms into liquid water, it is found that the effective
Y
n
/
thermal conductivity decreases with the increase in water
k¼ kj j ð13Þ content at a special initial water content (i.e., the decrease
j¼1 in ice content), which can be explained by that the thermal
Different shape factors (particle shapes are represented conductivity of ice is larger than that of water. For exam-
by Nx;y;z and assumed to be the same for all components in ple, when the initial water content is 0.2 m3 m-3, the
the soil–water system) are set and tested. Figure 5 effective thermal conductivity is 1.267 W m-1 K-1 at

(a) (b)

(c) (d)

Fig. 5 Influence of shape factors and initial water contents on the new model

123
Acta Geotechnica

(a) (b)

(d)
(c)

Fig. 6 Influence of initial water contents on the normalized thermal conductivity

/w = 0 m3 m-3, which is larger than 0.872 W m-1 K-1 at of the effective thermal conductivity at different shape
/w = 0.2 m3 m-3 for Nx;y;z = (0.50, 0.25, 0.25). Addi- factors to that at Nx;y;z = (0.33, 0.33, 0.33) (i.e., sphere
tionally, it is found that the initial water content has an particle).
important effect on the thermal conductivity. For instance, j ¼ kNx;y;z =ksphere ð14Þ
as /ini increases from 0.10 m3m-3 to 0.40 m3m-3, the
effective thermal conductivity increases from 0.902 to where ksphere is the thermal conductivity of porous media at
1.179 W m-1 K-1 for Nx;y;z = (1, 0, 0). Furthermore, for Nx;y;z = (0.33, 0.33, 0.33).
spherical particles [i.e., Nx;y;z = (0.33, 0.33, 0.33)], the The sensitivity of the normalized thermal conductivity
calculated thermal conductivities are smaller than those of to the water content at different initial water contents is
other shape factors. It can be seen that aspect = 1 (i.e., shown in Fig. 6. As the water content increases (i.e., the ice
spherical particles) generally represents the smallest keff at content decreases), the normalized thermal conductivity
certain water content, and keff starts to increase when increases as well. For example, when the initial water
aspect decreases to \ 10-4 or increases to [ 50. Hence, content is 0.3 m3 m-3, the calculated normalized parame-
disk- and sphere-shaped particles represent the upper and ter is 1.005 at /w = 0 m3 m-3, and 1.039 at
lower bounds of the proposed model for keff ð/w Þ, /w = 0.3 m3 m-3 for Nx;y;z = (0.00, 0.50, 0.50). Moreover,
respectively. it is found that the normalized thermal conductivity is
Johansen presented a normalized concept for illustrating relatively sensitive to shape factor (1, 0, 0) (i.e., disk-
the soil thermal conductivity [18]. Similarly, a normalized shaped solid particle), in particular at high water contents.
parameter j is proposed in this study to describe the ratio The assumption that solid particle is sphere in porous

123
 Acta Geotechnica

media would underestimate the effective thermal conduc- [Eqs. (5)–(8)] with a measured thermal conductivity. In
tivity of the actual thermal system. In our examples, when this paper, the equivalent shape factors of silty clay and
the initial water content is 0.1 m3 m-3, the normalized sandstone are (0.27, 0.62, 0.11) and (0.16, 0.28, 0.56),
thermal conductivity reaches up to 1.210 at respectively. The related literature datasets and experi-
/w = 0.1 m3 m-3 for disk-shaped solid particles. mental results in this study are collected and analyzed. The
databases of Pei et al. [30] include 21 measured results of
4.2 Predicting the thermal conductivities porosity-dependent thermal conductivity in rock samples.
under unfrozen states The volumetric water contents range between 13.93 and
33.58%, and measured thermal conductivities range from
The multiphase porous medium (e.g., soil–water system or 0.794 to 1.975 W m-1 K-1 under unfrozen state (10 C),
sandstone specimen) under unfrozen state is a three-phase and the shape factors of samples from the E Lashan tunnel
medium and comprises solid particle, liquid water, and air. and Jiang Luling tunnel are (0.20, 0.45, 0.35) and (0.33,
The thermal conductivities of soil–water systems or sand- 0.26, 0.41), respectively. Orakoglu et al. measured the
stones under the unfrozen condition have been measured thermal conductivity of the temperature-dependent clayey
by previous researchers [29, 30, 39, 40] and in this study soil sample at three temperatures: - 15 C, -5 C, and
(see Table 2). These databases are reanalyzed by the pro- 0 C [29]. The dry density is 1.80 g cm-3, and measured
posed model [Eq. (5)]. In our experimental study, thermal conductivities range from 0.189 to
kg = 0.024 W m-1 K-1 and kw = 0.55 W m-1 K-1 are 1.142 W m-1 K-1, and the shape factor of clayey soils is
set. For the solid particle of rock or soil specimens, when (0.30, 0.58, 0.12).
the complete mineral compositions are known, the thermal To validate the new model, the calculated values of
conductivity of solid particles ks can be calculated by the thermal conductivity are compared with test results. The
generalized geometric mean method with thermal con- test data include the experimental results in this study and
ductivity data of forming minerals [5]. others referenced from previous researches [29, 30, 39, 40].
Yn Xn The shape factors of consolidated and unconsolidated rocks
ks ¼ kxmj j with xj ¼ 1 ð15Þ are (0.12, 0.64, 0.24) and (0.13, 0.59, 0.28), respectively
j¼1 j¼1 [39, 40]. Furthermore, it is necessary to compare the pro-
where kmj is the thermal conductivity of mineral j posed model with other previous models. In this study, two
(W m-1 K-1), and xj is the volumetric proportion of general models are employed in the following discussion.
mineral j. The physical model is proposed based on the thermal
To the authors’ knowledge, the shape factors have rarely performance of each component. The thermal conductivity
been considered to describe the effective thermal conduc- of media could be defined as [25]:
tivity of porous media. In the following, the proposed keff ¼ ð1  nÞks þ ð1  hu ÞnSr ki þ hu nSr kw þ ð1  Sr Þnkg
model [Eq. (5)] is evaluated by comparison with experi- ð16Þ
mental results, which are listed in Table 2. For the geo-
material, the content of each phase is constant under where n is the initial porosity of porous media, Sr is the
specified condition. Furthermore, the equivalent shape saturation, and hu is the unfrozen water content.
factor of the material can be obtained by curve fitting Based on the laboratory investigation, Chen developed
an empirical model which is expressed as a function of
porosity and degree of saturation [4]:
Table 2 Physical properties and the measured results of porous media
keff ¼ kðs1nÞ knw ½ð1  b1 ÞSr þ b1 b2 n ð17Þ
Porous Porosity Volumetric Test kmea
medium (%) water content temperature (W m-1 K-1) where b1 and b2 are empirical parameters.
(%) (C)
Overall, the proposed and two previous models (i.e.,
Sandstone 18.83 13.72 ? 20 1.537 physical and empirical) are adopted to predict the thermal
-5 1.443 conductivity within an error of 10% for various porous
- 10 1.416 media. Figure 7 shows the predicted results under unfrozen
- 20 1.433 states. The performances of the three models are listed in
Silty clay 29.07 18.20 ? 26 0.533 Table 3. Coefficients of determination R2 between calcu-
-1 0.585 lated values and experimental results for the proposed,
-2 0.603 physical, and empirical models are 0.952, 0.642, and 0.906,
-5 0.618 respectively. The higher R2 values (R2 [ 0.90) indicate
- 10 0.629 better match between the calculated results [Eqs. (5) and

123
Acta Geotechnica

8 Table 3 The performances of the three models

Woodside and Messmer (1961b)
7 Woodside and Messmer (1961a)
Model State R2 RMSD MAD
Pei et al. (2013)
6 Orakoglu et al. (2016) +10% New model Unfrozen 0.952 0.264 0.195
This study
-10%
Freezing 0.961 0.124 0.104
5 1:1 Physical model Unfrozen 0.642 1.175 0.754
4 Freezing 0.840 0.250 0.203
(a) Empirical model Unfrozen 0.906 0.603 0.439
3
Freezing 0.708 0.337 0.290
2

0 models are better than the physical model in predicting the

0 1 2 3 4 5 6 7 8 thermal conductivity of porous media under unfrozen
states.
8 In addition, to perfectly illustrate the validity of the
7 three models, the mean absolute deviations of calculated
and experimental results are calculated for each model by
6 +10% Eq. (11), respectively. The errors between the calculated
5 -10% and experimental results can be presented in the residual
plots. Totally, 16 different values are selected from Fig. 7,
4 and the residual plots for three models are described in
(b) Fig. 8. From Fig. 8a, it can be seen that the residuals are all
3 Woodside and Messmer (1961b)
Woodside and Messmer (1961a) in the range of [- 0.7, 0.7], and the residual case plot
2 Pei et al. (2013) reflects that the predictive effect is well. Nevertheless,
Orakoglu et al. (2016)
1 This study
there are larger errors between calculated and experimental
1:1 values for the 5th sample in Fig. 8b and the 12th sample in
0 Fig. 8c. As demonstrated in the boxplot of the deviations
0 1 2 3 4 5 6 7 8
for three models under the unfrozen state (Fig. 9), the
deviation range of calculated values by the proposed model
8
from the experimental values is the smallest
Woodside and Messmer (1961b)
7 Woodside and Messmer (1961a) +10% (- 0.697–0.594 W m-1 K-1), and the distribution of
Pei et al. (2013) deviation is the most concentrated. The maximum devia-
6 -10%
Orakoglu et al. (2016)
tion of the proposed model is 0.697 W m-1 K-1, which is
This study
5 1:1 the smallest among the three models.

4
(c) 4.3 Predicting the thermal conductivities
3 under freezing states
2
The freezing multiphase porous medium is a four-phase
1 medium and comprises of solid particle, liquid water, ice,
and air. Many researchers have measured the thermal
0
0 1 2 3 4 5 6 7 8 conductivity of the soil–water system or sandstones under
freezing states [29, 30]. Some experimental results in this
study are listed in Table 2. These databases are reanalyzed
Fig. 7 Comparison between measured thermal conductivities under by using the proposed model [Eq. (5)] and the other two
unfrozen states and predicted values by models a the proposed model,
b the physical model, and c the empirical model models. In our experimental study, ki = 2.22 W m-1 K-1
is chosen.
(17)] and the experimental results. Moreover, the root- For freezing states, it is assumed that all water freezes in
mean-square deviations for three models are 0.2644, the situ pore and no water migration occurs. The volume
1.1747, and 0.6035, respectively. The smaller RMSD val- expands to 109% when pore water phase changes into ice.
ues indicate that the calculated values are close to experi- The following relationship exists during the phase
mental results. Therefore, the proposed and empirical transition:

123
 Acta Geotechnica

(a)

(b)

(c)

123
 Acta Geotechnica

b Fig. 8 The residual plots of the three models a the proposed model, 5
b the physical model, and c the empirical model
Pei et al. (2013)
Orakoglu et al. (2016)
4 This study
1:1

+10%
3 (a)
-10%

1 2 3 4 5

5
Pei et al. (2013)
Orakoglu et al. (2016)
4 This study +10%
1:1

Fig. 9 The boxplot of deviations of the calculated values of three 3 (b) -10%
models from the tested values

/i ¼ 1:09ð/ini  /uw Þ ð18Þ 2

where /i is the volumetric ice content, /ini is the initial

water content, and /uw is the unfrozen water content in the 1
freezing porous media at different temperatures.
The empirical model has been employed by numerous 1 2 3 4 5
researchers to describe the relationship between unfrozen
water content and temperature. For example, the model 5
developed by Xu et al. [41] can be expressed as: Pei et al. (2013)
-10%
/uw ¼ c1 T ðc2 Þ ð19Þ Orakoglu et al. (2016)
4 This study
1:1 -10%
where the temperature T is taken as absolute value (C) and
c1 and c2 are the fitting constants.
3 (c)
The experimental results in this study and related liter-
ature datasets of freezing porous media are used to com-
pare the new model and the other two models. The 2
database of Pei et al. [30] includes 21 measured results of
porosity-dependent thermal conductivity in rock samples
under freezing condition. The initial volumetric water 1
content ranges from 13.93 to 33.58%, and the measured
thermal conductivity from 1.156 to 2.117 W m-1 K-1 1 2 3 4 5
(- 10 C). Orakoglu et al. [29] measured the temperature
dependence of thermal conductivity in the clayey soil
Fig. 10 Comparison of measured thermal conductivities under freez-
sample at different temperatures. The measured thermal ing states with predicted values by models a the proposed model,
conductivity ranged from 1.020 W m-1 K-1 at - 5 C to b the physical model, and c the empirical model
1.142 W m-1 K-1 at - 15 C.
To validate the proposed model under freezing condi- researches [29, 30]. To conclude, the three models predict
tions, the calculated values of thermal conductivity are the thermal conductivity under freezing condition within an
compared with test results. The test data include the error of 10%, as shown in Fig. 10. Coefficients of deter-
experimental results in this study and others from previous mination R2 between calculated values and experimental

123
 Acta Geotechnica

the predictive effect of the new model is better than that of

(a) others.
Furthermore, to perfectly illustrate the validity of the
three models, the deviations between simulated and mea-
sured values are calculated for each model by Eq. (11),
respectively. Sixteen values are selected from the dataset in
Fig. 10, and the residual plots for the three models are
described in Fig. 11. The errors between calculated values
and experimental values can be shown in the residual plot.
From Fig. 11a, it can be seen that the residuals of the
proposed model are all in the range of [- 0.35, 0.35], and
those of the physical and empirical models are in the range
of [- 1.00, 1.05] and [- 0.85, 0.70], respectively. More-
over, there are larger errors between the calculated and
measured values for the 15th sample in Fig. 11b and the
(b) 9th sample in Fig. 11c. As demonstrated in the boxplots of
the deviations for the three models under freezing states
(Fig. 9), the deviation range of calculated values by the
proposed model from the experimental values is the
smallest (- 0.223–0.243 W m-1 K-1), and its distribution
of deviation is the most concentrated. The maximum
deviation of the proposed model is 0.243 W m-1 K-1,
which is the smallest among the three models. In conclu-
sion, the proposed model can predict the thermal conduc-
tivity of porous media under freezing states well.

5 Conclusions
(c)
An experimental study of the effective thermal conduc-
tivities of silty clays and sandstones was performed under
both unfrozen and freezing states. Moreover, the effect of
particle structures on the effective thermal conductivity of
porous media is theoretically analyzed. The following
conclusions are drawn:
(1) The thermal conductivity of porous medium
increases rapidly as the temperature reduces from
positive to sub-zero. In addition, the thermal con-
ductivity increases slowly with the decrease in
temperature due to the decrease of unfrozen water
under freezing states.
Fig. 11 The residual plots of the three models a the proposed model, (2) Considering the configuration of porous media and
b the physical model, and c the empirical model the construction of solid particles, a new model for
predicting the thermal conductivity of porous media
under unfrozen and freezing states was proposed and
results for the proposed, physical, and empirical models are
verified by comparing with the experimental results.
0.951, 0.840, and 0.708, respectively. The higher R2 values
(3) A normalized parameter has been introduced to
(R2 [ 0.90) indicate a better agreement between the cal-
describe the relationship between the thermal con-
culated [Eq. (5)] and measured values. Moreover, the root-
ductivity and particle shape. The results indicated
mean-square deviations for three models are 0.1235,
that the effective thermal conductivity has been
0.2501, and 0.3373, respectively. The smaller RMSD val-
underestimated when the solid particle was assumed
ues indicate that the calculated values are closer to the
as sphere.
experimental results. Therefore, it can be concluded that

123
Acta Geotechnica

(4) The assumption that aspect = 1 (i.e., spherical 14. He HL, Zhao Y, Dyck MF, Si BC, Jin HJ, Lv JL, Wang JX (2017)
particles) generally represents the smallest keff at a A modified normalized model for predicting effective soil ther-
mal conductivity. Acta Geotech 12:1281–1300
certain water content, and when the aspect decreases 15. Horai K (1971) Thermal conductivity of rock-forming minerals.
to\ 10-4 or increases to[ 50, keff starts to increase. J Geophys Res 76(5):1278–1308
Therefore, the effective thermal conductivities of 16. Hyndman RJ, Koehler AB (2006) Another look at measures of
disk- and sphere-shaped particles stand for the upper forecast accuracy. Int J Forecast 22(4):679–688
17. Johansen O (1975) Thermal conductivity of soils. Ph. D. thesis,
and lower bounds of the proposed model, University of Trondheim, Trondheim, Norway. US Army Corps
respectively. of Engineers, Cold Regions Research and Engineering Labora-
(5) Compared with the physical and empirical models, tory, Hanover
the proposed model has minimum RMSE and MAD 18. Johansen O (1975) Varmeledningsevne av jordarter, Ph.D. thesis,
Norge tekniske hogskole, Trondheim
values and the maximum R2 value, indicating that 19. Kersten MS (1949) Laboratory research for the determination of
the proposed model is better than the two others. the thermal properties of soils. Research Laboratory Investiga-
tions, Engineering Experiment Station, University of Minnesota,
Minneapolits, Minn. Technical report 23
Acknowledgments This research was supported by Key Research 20. Kleinberg RL, Griffin DD (2005) NMR measurements of per-
Program of Frontier Sciences of Chinese Academy of Sciences (Grant mafrost: unfrozen water assay, pore-scale distribution of ice, and
No. QYZDY-SSW-DQC015), the National Natural Science Founda- hydraulic permeability of sediments. Cold Reg Sci Technol
tion of China (Grant Nos. 41230630, 41471063), Natural Science 42:63–77
Foundation of Gansu Province (1508RJZA100), the Program of the 21. Li SY, Lai YM, Pei WS, Zhang SJ, Zhong H (2014) Moisture-
State Key Laboratory of Frozen Soil Engineering (Grant No. temperature changes and freeze-thaw hazards on a canal in sea-
SKLFSE-ZT-23). sonally frozen regions. Nat Hazards 72:287–308
22. Li SY, Zhan HB, Lai YM, Sun ZZ, Pei WS (2014) The coupled
moisture-heat process of permafrost around a thermokarst pond in
References Qinghai-Tibet Plateau under global warming. J Geophys Res:
Earth Surf 119:836–853
23. Li SY, Niu FJ, Lai YM, Pei WS, Yu WB (2017) Optimal design
1. Bauer S, Urquhart A (2016) Thermal and physical properties of
of thermal insulation layer of a tunnel in permafrost regions based
reconsolidated crushed rock salt as a function of porosity and
on coupled heat-water simulation. Appl Therm Eng
temperature. Acta Geotech 11:913–924
110:1264–1273
2. Campbell GS, Jungbauer JDJ, Bidlake WR, Hungerford RD
24. Li DW, Liu J, Cai Q, Li Y, Zhou Y (2017) Experimental study on
(1994) Predicting the effect of temperature on soil thermal con-
thermal conductivity coefficient on underground engineering
ductivity. Soil Sci 158:307–313
artificial frozen clay. Chin J Undergr Space Eng 13(6):1452–1456
3. Carslaw HC, Jaeger JC (1986) Conduction of heat in solids.
25. Liu WM, He P, Zhang Z (2002) A calculation method of thermal
Oxford University Press, Oxford
conductivity of soils. J Glaciol Geocryol 24(6):770–774 (in
4. Chen S (2008) Thermal conductivity of sands. Heat Mass Transf
Chinese)
44(10):1241–1246
26. Lu Y, Lu S, Horton R, Ren T (2014) An empirical model for
5. Cote J, Konrad JM (2005) A generalized thermal conductivity
estimating soil thermal conductivity from texture, water content,
model for soils and construction materials. Can Geotech J
and bulk density. Soil Sci Soc Am J 78:1859–1868
42(2):443–458
27. National standards of P. R. C (1999) Standard for soil test
6. Cote J, Konrad JM (2009) Assessment of structure effects on the
method. GB/T 50123-1999
thermal conductivity of two-phase porous geomaterials. Int J
28. National standards of P. R. C (2013) Standard for test methods of
Heat Mass Transf 52:796–804
engineering rock mass. GB/T 50266-2013
7. de Vries DA (1963) Thermal properties of soils, in physics of
29. Orakoglu ME, Liu JK, Niu FJ (2016) Experimental and modeling
Plant Environment. In: Van WR (ed) Wijk. North-Holland,
investigation of the thermal conductivity of fiber-reinforced soil
Amsterdam
subjected to freeze-thaw cycles. Appl Therm Eng 108:824–832
8. Deng YS, He P, Zhou CL (2004) An experimental research on the
30. Pei WS, Yu WB, Li SY, Zhou JZ (2013) A new method to model
thermal conductivity coefficient of saline soil. J Glaciol Geocryol
the thermal conductivity of soil-rock media in cold regions: an
26(3):319–323 (in Chinese)
example from permafrost regions tunnel. Cold Reg Sci Technol
9. Deng A, Yu YT, Wang HB (2010) Thermal conductivity of WFS-
95:11–18
FA-EPS fills. Chin J Geotech Eng 32(8):1297–1302 (in Chinese)
31. Rao MG, Singh DN (1999) A generalized relationship to estimate
10. Dong Y, McCartney JS, Lu N (2015) Critical review of thermal
thermal conductivity of soils. Can Geotech J 36(4):767–773
conductivity models for unsaturated soils. Geotech Geol Eng
32. Salomone LA, Kovacs WD, Kusuda T (1984) Thermal perfor-
33:207–221
mance of soils. J Geotech Eng: ASCE 110(3):359–374
11. Ghanbarian B, Daigle H (2016) Thermal conductivity in porous
33. Sepaskhah AR, Boersma L (1979) Thermal conductivity of soils
media: percolation-based effective-medium approximation.
as a function of temperature and water content. Soil Sci Soc Am J
Water Resour Res 52:295–314
43:439–444
12. Gujarati DN, Porter DC (2009) Basic econometrics, 5th edn.
34. Sundberg J (1988) Thermal properties of soils and rock, Swedish
McGraw-Hill, New York
Geotechnical Institute, Report No. 35, Linkoping
13. He HL, Dyck M, Zhao Y, Si BC, Jin HJ, Zhang TJ, Lv JL, Wang
35. Tarnawski VR, Wagner B (1992) A new computerized approach
JX (2016) Evaluation of five composite dielectric mixing models
to estimating the thermal-properties of unfrozen soils. Can Geo-
for understanding relationship between effective permittivity and
tech J 29(4):714–720
unfrozen water content. Cold Reg Sci Technol 130:33–42

123
 Acta Geotechnica

36. Van RM, Winterkorn HF (1959) Structural and textural influ- embankments in permafrost regions. Appl Therm Eng
ences in thermal conductivity of soils. Highw Res Board Proc 36:252–259
39:576–621 44. Zhang N, Yu XB, Pradhan A, Puppala AJ (2015) Effects of
37. Wen Z, Sheng Y, Ma W, Deng YS, Wu JC (2005) Experimental particle size and fines content on thermal conductivity of quartz
Studies of Thermal Conductivity of Undisturbed Permafrost at sands. Transp Res Record 2510:36–43
Beiluhe Testing Site on the Tibetan Plateau. J Glaciol Geocryol 45. Zhang N, Yu XB, Pradhan A, Puppala AJ (2017) A new gener-
27(2):182–187 (in Chinese) alized soil thermal conductivity model for sand-kaolin clay
38. Woodside W, Messmer JH (1961) Thermal conductivity of por- mixtures using thermo-time domain reflectometry probe test.
ous media. J Appl Phys 32:1688–1706 Acta Geotech 12:739–752
39. Woodside W, Messmer JH (1961) Thermal conductivity of por- 46. Zhou ZZ, Ma W, Zhang SJ, Du HM, Mu YH, Li GY (2016)
ous media. I. Unconsolidated rocks. J Appl Phys 32:1688–1699 Multiaxial creep of frozen loess. Mech Mater 95:172–191
40. Woodside W, Messmer JH (1961) Thermal conductivity of por- 47. Zhou ZZ, Ma W, Zhang SJ, Mu YH, Li GY (2018) Effect of
ous media. II. Consolidated rocks. J Appl Phys 32:1699–1706 freeze-thaw cycles in mechanical behaviors of frozen loess. Cold
41. Xu XZ, Wang JC, Zhang L (2001) Frozen soil physics science. Reg Sci Technol 146:9–18
Science Press, Beijing
42. Yuan XZ, Li N, Zhao XY, Li J (2010) Study of thermal con-
ductivity model for unsaturated unfrozen and frozen soil. Rock Publisher’s Note Springer Nature remains neutral with regard to
Soil Mech 31(9):2689–2694 (in Chinese) jurisdictional claims in published maps and institutional affiliations.
43. Zhang MY, Lai YM, Li DQ, Tong GQ, Li JB (2012) Numerical
analysis for thermal characteristics of cinderblock interlayer

123

View publication stats