Direct numerical simulation of the thermal dehydration

reaction in a TGA experiment

Shuiquan Lana,b,∗, Mohammadreza Gaeinia , Herbert Zondaga,c , Anton van
Steenhovena , Camilo Rindta
a
Department of Mechanical Engineering, Eindhoven University of Technology, Den Dolech
2, 5612AZ Eindhoven, The Netherlands
b
Kuang-Chi Institute of Advanced Technology, 518057 Shenzhen, China
c
Energy research Center of the Netherlands - ECN, P.O. Box 1, 1755ZG Petten, The
Netherlands

Abstract
This work presents a detailed mathematical model of the coupled mass and
heat transfer processes in salt hydrate grains in a TGA experiment. The
purpose of developing this numerical model is to get a more fundamental
understanding of the influence of parameters like particle size, nucleation
rate and vapor pressure on the dehydration/hydration reaction in a TGA
experiment. Such a model needs a detailed description of the fluid flow and
water vapor distribution between the particles. The dehydration reaction
of grains of TCMs is described by the nucleation and nuclei growth model
presented in our earlier work. The flow around grains is solved by means of
the finite volume method using OpenFOAM including heat and mass transfer.
Direct numerical simulations of TGA-experiments under various conditions
are performed. Such simulations provide direct insight into the physics of
mass and heat transport processes coupled with detailed reaction kinetics at
grain scale. The numerical results are compared to the experimental results.
The developed CFD model can be a promising tool to calculate the overall
kinetics for dehydration reactions under realistic heat storage conditions. To
that end, the effect of buoyancy should also be included in the model to get
a more accurate description of convection within the sample.
Keywords: Reaction kinetics, direction numerical simulation, nucleation and

∗
Corresponding author
Email address: [email protected]; Fax number:2433445 (Camilo Rindt)

Preprint submitted to Applied Thermal Engineering December 6, 2016

growth, TGA measurement, thermochemical heat storage, salt hydrate

1. Introduction
Thermochemical heat storage materials (TCMs) are given more and more
attention for the application in seasonal heat storage [1, 2]. A number of salt
hydrates have been identified as promising candidates such as MgSO4 · 7 H2 O
[3], Mgcl2 · 6 H2 O [4] and SrBr2 · 6 H2 O [5]. The main principle of thermo-
chemical heat storage using salt hydrates is based on reversible reactions
like:
A(s) + heat ↔ B(s) + nH2 O(g) (1)
where n is the number of water molecules per unit of species A. In this
reaction, a thermochemical solid material (A) absorbs energy in the form of
heat and is converted chemically into two components (solid B and water
vapor), which can be stored separately. The reverse reaction occurs when
material B and water vapor are combined together and A is formed. Heat is
released during this reaction and the storage system is ready for recharging.
Salt hydrates are most frequently characterized by thermoanalytical tech-
niques like TGA-DSC [2]. Materials showing promising features like high
energy storage density, fast reaction kinetics and good structural stability are
used in reactor and system tests [6]. For better understanding and optimizing
thermochemical heat storage systems using salt hydrates, several numerical
models are developed to simulate the heat and mass transfer processes to-
gether with the gas-solid reactions at reactor- and system-scale [7–14]. In all
these models, the reactive medium is treated as a continuum at macroscopic
scale and the reaction kinetics is described by simplified kinetic models de-
rived from thermal analytical experiments like TGA-DSC. It is clear that the
reaction kinetics, as the root of such models, has a significant influence on the
numerical predictions. It is necessary to implement more fundamental kinetic
models to gain more detailed insight. Helbert et al. [15] presented a stochastic
and deterministic model for nucleation and growth in powder transformation,
which is applicable in non-isothermal and non-isobaric situations and is valid
for all kinds of grain shapes. Favergeon et al. [16] used this model to study
the dehydration reaction of Li2 SO4 · H2 O single crystals under isothermal and
isobaric conditions. With the assumption of constant values of nucleation and
growth, the rates of nucleation and growth were estimated by model-fitting of
the corresponding TGA results. Our earlier work [17] characterized the rates

2
of nucleation and growth under non-isothermal conditions and calculated the
reaction kinetics in terms of the nucleation and nuclei growth processes.
Besides the importance of the reaction kinetics the thermal decomposi-
tion reactions derived from TGA-DSC also depend largely on the sample
environment, particularly the atmosphere in the vicinity of the sample [18].
Unfortunately, such detailed information on the local atmosphere is not
accessible in a TGA-DSC measurement. Moreover, thermal analysis most
frequently focuses on the overall kinetic rates and the derivation of kinetic
models, but less attention is paid to the local atmosphere around the sample
particles. To take the influence of the sample environment into account, heat
and mass transport processes in the void space between packed grains have to
be included. A detailed numerical model of the TGA samples based on the
actual packed bed geometry can be helpful to find out the physical phenomena
taking place inside the TGA furnace.
In the present work, a traditional CFD method with body-fitted meshes is
used to simulate the dehydration reaction of a powdery sample of Li2 SO4 · H2 O
grains in a TGA experiment. The nucleation and growth model developed
earlier[17] is coupled to the mass and heat transfer processes between the
Li2 SO4 · H2 O grains. Direct numerical simulations using the CFD package
OpenFOAM are performed to study the mass and heat transport processes
during the dehydration reaction of multiple grains. The overall rate of the
reaction is computed and compared to TGA results under various conditions.
The influence of various parameters like particle size, nucleation rate and
vapor pressure is then discussed.

2. Numerical model
The simulated configuration is schematically shown in Fig. 1. This is a
typical TGA configuration in our lab (type: Netzsch STA 449 F3 Jupiter). It
should be noted here that only a part of the furnace in the height direction is
simulated. A humidified air flow from top to bottom is used to control the
humidity of the experimental environment. Temperature of all solid walls
including the side walls of the furnace and the crucible is set to a programmed
temperature profile. All dehydration tests are performed under controlled
temperature and humidity conditions. In the numerical model, a real-size 3D
geometry of the TGA apparatus is implemented. The computational domain
is divided into two regions. The first region is the powdery sample, which
is composed of N spherical grains that are randomly packed. The second

3
region is the flow region, which is a moist air flow at a low Reynolds number
in the furnace chamber and in the void space between particles. These two
regions are bounded by the solid walls of the crucible and the furnace. The
reactions at grain level are described by the nucleation and nuclei growth
models as presented in our earlier work [17]. Coupling of heat and mass
transfer with chemical reaction is realized at the gas-solid boundary, where
heat is extracted from the flow to the solid particles for the dehydration
reaction and meanwhile water vapor generated during the reaction is released
to the flow.
u=(0 5 0) mm/s c=0.39 mol/m3 T=Ts

u=(0 0 0）

∂ c/ ∂ n=0 d=6 mm
T=Ts d=10
h=3 mm H=15 mm

T=Ts
l=5 mm
∂ u/ ∂ n=(0 0 0） ∂ T/ ∂ n=0 ∂ c/ ∂ n=0

D=30 mm

Figure 1: Calculation domain including dimensions and boundary conditions for the
numerical model.

Assumptions for derivation of the model are as follows:

1 Each grain is considered to be spherical throughout the reaction. The

shrinkage of the crystal grains is less than 10% and not taken into
account in the present model.

2 A uniform temperature distribution in the grain is assumed, as the Biot

number is in the order of 0.1.

3 Because of computational time, the effect of buoyancy around the grain

is not considered. Also the influence of temperature on density and

4
 viscosity is neglected. This led to acceptable computational time of
several hours for the simulation of the dehydration experiment.

4 Water vapor is diluted in air. Water production by the reaction is

assumed not to disturb the total pressure distribution of the flow around
the particles. Also, the physical properties of the moist air are considered
to be constant and are the same as those of dry air.

5 A quasi-steady flow is considered. In this coupled problem, the time

scales of the fluid dynamics problem (d/u ≈ = O(second)) and the
solid reaction (O(minutes)) are very different. Hence the temperature
and vapor concentration in the flow domain are calculated at discrete
moments in time based on an update of the particle boundary conditions.

Flow region. Based on these assumptions the governing equations for the
quasi-steady flow problem can be written as follows:

∇·u=0 (2)
1
u · ∇u = ν∇2 u − ∇p (3)
ρ
u · ∇T = DT ∇2 T (4)
u · ∇c = Dm ∇2 c (5)
where u is the velocity vector, ν the kinematic viscosity, ρ the density, p the
pressure, DT the thermal diffusivity of the moist air (DT = κ/ρcp ), c the
molar concentration and Dm the mass diffusivity of the moist air.

Particle region. The dehydration reaction in terms of nucleation and nuclei

growth takes place in all particles. As presented in our earlier work [17],
the reaction kinetics of the dehydration reaction can be described by a
nucleation and nuclei growth model successfully. The conversion of the
particles is attributed to these two elementary processes. By using optical
microscopy experiments, the kinetics of nucleation and growth processes
were characterized as function of temperature and water vapor pressure
[17, 19]. Mathematically, the overall reaction kinetics is determined by
the systematic changes in area of active interface formed by appearing and
growing nuclei. With the rate correlations of nucleation and nuclei growth, the
fractional conversion α can be calculated from the appearance of nucleation

5
to completion of reaction. The temperature and water vapor pressure in the
surrounding atmosphere of each particle are unknown yet and will be solved
by the model. For the sake of simplicity, mean values of these variables over
the particle surface are used in the reaction rate calculation. The rate of
reaction ∂α/∂t can generally be written as:

∂α
= φ(T̄p , t)S(γ, φ, r0 , t)/V0 (6)
∂t
where α is the fractional conversion, φ(T̄p , t) is the growth rate (as function
of time and particle temperature) and S(γ, φ, r0 , t) is the active interface
between reactant and product (as function of nucleation rate γ, of growth
rate φ, of the grain dimension r0 and time t).
The rates of nucleation and growth are written as follows:
−a4
γ = a1 (−a2 cRT̄p + a3 )exp( ) [1/(mm2 min)] (7)
RT̄p
−a6
φ = a5 exp( ) [µm/min] (8)
RT̄p
with constants a1 -a6 given in Table 1. These parameters are derived from
microscopy experiments presented in previous work [17, 19].

Coupling via gas-solid interface. In order to couple the solid reaction and the
moist air flow, the conservation laws for mass and energy are considered for
the mass and heat transfer between the flow and the solid particles.

• The mass balance at the fluid-solid interface expresses that the water
vapor diffusing away via the flow equals the water vapor released by the
reaction:
∂c ρs V dα
ADm = (9)
∂n Ms dt
∂c
where A is the interface area, n is the normal vector, ∂n is the mean
value of the concentration gradient in the flow domain at the interface,
ρs is the density of the particle, V is the volume of the particle and Ms
is the molar mass of the solid reactant.

• The heat balance at the fluid-solid interface expresses that the heat
taken away from the flow is used for the endothermal reaction and for

6
 heating up the particle:

∂T ρs V dα ∂ T̄p
−Aκ = ∆H + ρs csp V (10)
∂n Ms dt ∂t

where κ is the thermal conductivity of the moist air, ∂T

∂n
is the mean
value of the temperature gradient in the flow domain at the interface,
T̄p is the particle temperature, ∆H is the reaction enthalpy and csp is
the specific heat capacity of the particle.

After reorganization of Eqs. 9 and 10, the conditions at the interface

between the flow and the solid particles are as follows:

u=0 (11)

∂ T̄p ∆H ∂α κA ∂T
=− s
− (12)
∂t Ms cp ∂t ρs csp V ∂n
∂c ρs V ∂α
= (13)
∂n Dm Ms A ∂t
Numerical scheme. The above stated PDEs, Eqs. 2–5, are numerically solved
through a Finite Volume method, using the CFD package OpenFOAM. The
standard flow solver in OpenFOAM, simpleFoam, is modified to solve the
PDEs in the flow region. SimpleFoam is a steady-state solver for incompress-
ible, turbulent flow. Since only laminar flow is considered in the present
work, no turbulent model is used. The SIMPLE algorithm is applied to the
Navier-Stokes equations in the steady-state case. Eqs. 2 and 3 are only solved
once because the flow is steady (due to the assumptions presented earlier).
The algorithm is of second order accuracy in space. The reaction of the solid
region (Eq. 6) determines the boundary conditions to be applied (Eqs. 9 and
10 ) and is solved by an external solver programmed in C++. Integrating from
time level j∆t to time level (j + 1)∆t, the following procedure is executed:
j j
(1) With the values of ∂α
∂t
and ∂T
∂n
, calculate T̄pj+1 by using Eq. 12. For the
time integration a first order Euler explicit scheme is used. Together
with the other boundary conditions, the new temperature field T j+1 of
the flow is calculated, using Eq. 4.

7
 j j+1
(2) With the value of ∂α∂t
∂c
, calculate ∂n at the fluid-solid interface by
using Eq. 13. Again a first order explicit Euler scheme is used for the
time integration. Together with the other boundary conditions, the new
concentration field cj+1 is calculated, using Eq. 5.
j+1
(3) Calculate average values of ∂T ∂n
and c̄j+1 in the flow domain at the
interface. Export c̄j+1 with the particle temperature T̄pj+1 to the external
j+1
solver. Calculate the new reaction rate ∂α ∂t
by using the kinetic model.
All parameter values used in the calculations are listed in Table 1[17, 19].

Table 1: Parameters used for calculations

Symbol Description Value
Ms Molar mass of Li2 SO4 .H2 O (kg/mol) 0.128
ρ Density of moist air (kg/m3 ) 1.18
ρs Density of Li2 SO4 .H2 O (kg/m3 ) 2060
csp Specific heat of Li2 SO4 .H2 O (J/(kgK) 1180.8
cp Specific heat of moist air (J/kgK) 1006.3
Dm Mass diffusivity of moist air (m2 /s) 2.82 × 10−5
ν Kinematic viscosity of moist air (m2 /s) 1.59 × 10−5
κ Thermal conductivity of moist air (W/mK) 0.026
κs Thermal conductivity of salt hydrate (W/mK) 0.5
∆H Enthalpy of dehydration (J/mol) 5.7 × 104
R Universal gas constant (J/(K mol) 8.314
g Acceleration due to gravity (m2 /s) 9.8
a1 Constant in Eq. 7 (1/(mm2 min)) 8.287 × 1033
a2 Constant in Eq. 7 (m3 /J) 3 × 10−4
a3 Constant in Eq. 7 1.4
a4 Constant in Eq. 7 (J/mol) 2.4 × 105
a5 Constant in Eq. 8 (µm/min) 3.931 × 1012
a6 Constant in Eq. 8 (J/mol) 8.84 × 104

3. Parameter choices for TGA-simulation

3.1. TGA measurements
TGA measurements of powdery Li2 SO4 · H2 O samples were performed for
comparison. The powdery samples were obtained from Sigma-Aldrich, with a

8
purity of ≥ 99.0%. The hydrated crystals in the samples have a particle size in
the range of 0.1–0.4 mm. Figure 1 shows the geometry of the TGA apparatus
with dimensions. In each measurement, an aluminum crucible was filled with
crystals without pressing. This yields a macro-porosity of around 0.54 in the
packed bed, measured by fully filling the crucible. If not stated otherwise,
the following experimental conditions were used in the present study: sample
mass around 60 mg, a temperature program (Ts ) characterized by constant
heating with a heating rate (kh ) of 1 ◦ C/min from ambient temperature till
150 ◦ C and then constant temperature for half an hour, constant moist air
flow with a flow rate of 230 ml/min (5 mm/s) and relative humidity of 41 %
at 30 ◦ C (13 mbar of partial water vapor pressure or 0.39 mol/m3 of water
vapor concentration).

3.2. Geometry and meshing

To create a discrete model of a 3D packed bed in CFD, an accurate
representation of the geometry of the bed is required. However, due to
computational limits, it is assumed that the powdery sample in the TGA
measurements can be approximated with a limited number (60) of packed
spherical particles with an identical diameter (d=1 mm), as shown in Fig.
1. The discrete element method (DEM) is used to generate the packing of
spherical particles under the influence of gravity. A packed bed is practically
constituted by dumping particles into an annular tube. In this study, a DEM
solver in OpenFOAM, icoUncoupledKInematicParcelFoam, is employed for
constituting the packing. Default settings in the solver are used. Once the
spherical particles are set in the tube by the DEM simulation, the coordinates
of the particle centers are extracted. Based on the coordinates and the
diameter, the geometry of the computational domain is generated and meshed
in Salome (an open-source mesh generator).
The gap model for dealing with contact points is adopted to reduce the
complexity of the 3D geometry and the computation time of each calculation.
To avoid the generation of mesh cells of poor quality at the points of contact
between particles mutually and between particles and crucible wall, all parti-
cles were shrunk in size but the particle position remained unchanged during
the geometry buildup. Shrinking particles in the CFD model is a common
approach adopted in published works [20–28] using a body-fitted mesh for the
packed bed. Typically, 1% shrinkage was used in those publications. When us-
ing this value of 1%, a huge amount of mesh elements are needed to discretize
the computational domain, since a very fine mesh around the contacting

9
points, especially between particles and the crucible wall, is required. This
makes each calculation very time-consuming. A value of 2% shrinkage reduced
the typical computing time of one calculation from several days to several
hours. Changing from 1% to 2% shrinkage increases the packed bed porosity
from 0.59 to 0.60. The original value of the porosity without shrinking is
about 0.58. It is expected that by using the shrinking particle method, the
mass and heat transfer processes will be influenced to some extend. This will
be discussed later on. As shown in Fig. 2, a tetrahedral mesh is used, where a
surface mesh size is specified and then the volume mesh based on the surface
mesh is generated. A size of 9% of the particle diameter for the triangle
edge was used on the surface, which falls in the range 4 − 10% suggested
by Eppinger et al. [29]. Before computation, the mesh-independence of the
numerical solution was checked by increasing the mesh resolution.

(a) (b)

Figure 2: Cross-section of the 3D mesh of (a) the computational domain and (b) the gap
areas between particles. Note that a cross-section of a regular tetrahedron mesh around
equally sized particles leads to an irregular triangular mesh and non-equally sized particles.

3.3. Boundary and initial conditions

Figure 1 shows the boundary conditions of the computational domain.
Walls of the furnace and the crucible are set to the same values. The inlet
temperature, which is the same as the temperature of the walls, is set by the
temperature program Ts as mentioned above. For all particles, the boundary
conditions are specified as in Eqs. 11–13. At the start of the calculation, the
initial velocity, pressure, temperature and concentration are set to 5 mm/s, 0

10
Pa, 370 K and 0.39 mol/m3 , respectively. The temperature program of the
furnace and the crucible is linear for the first part with a constant heating rate
(1 ◦ C/min) starting from 97 ◦ C (It is noted here from the experiments that
below 370 K no dehydration takes place. Therefore, to reduce computational
time the initial temperature is set to 370 K.) to 150 ◦ C and then remains
constant for the second part.

3.4. Time-step
In the present model, a key parameter is the time-step of the simultaneous
calculation of reaction and flow. The dehydration reaction of the particles
is computed by using the loosely coupled solver. Here the influence of the
time-step (10 s and 1 s) is studied on the physical processes. Figure 3(a)
shows the programmed setting temperature profile by a dashed line and the
calculated surface temperature of all particles in solid lines for a time-step of
10 s. During the reaction proceeding at increasing temperatures, it is clear
that there are temperature lags between the particles and the crucible.

160 160

150 150

140 140
Tmean (°C)

Tmean (°C)

130 130

120 120

110 Setting temperature 110

Setting temperature
Particle temperature Sample temperature
100 100

90 90
0 20 40 60 80 0 20 40 60 80
Time(min) Time(min)

(a) (b)

Figure 3: (a) Calculated temperature profile of each particle and the set crucible tempera-
ture; (b) Temperature profile of the sample and the set crucible temperature. (p(H2 O)=13
mbar, N=60, kh = 1 ◦ C/min, ∆t = 10 s)

To investigate the influence of the time-step further, an average tempera-

ture of all particles is calculated and denoted as the sample temperature (see
Fig. 3(b)). The temperature difference between this sample temperature and
the crucible temperature is then plotted in Fig. 4(a) as the solid line. The
sample temperature is much lower than the crucible temperature, especially

11
at 30 minutes, which is corresponding to the period of a fast reaction. The
dashed line in Fig. 4(a) shows the temperature difference for a calculation
without the dehydration reaction (in the order of 1 to 1.5 ◦ C,). This difference
with the baseline is attributed to numerical errors when a too large time-step
is used. This error can be reduced using a smaller time-step like ∆t = 1
s, as shown in Fig. 4(b). The remaining peak is attributed to the nature
of an endothermal reaction, which is inevitable and usually recognized as
the self-cooling effect [30–32]. So in order to reduce the numerical error,
a time-step in the order of ∆t = 1 s is required, but the computing effort
is increased then up to 100 h compared to 10 h with a time-step of 10 s.
Hence, a time-step ∆t = 10 is used in the following calculations realizing that
the corresponding numerical error results in a shift of the reaction kinetics
towards slightly higher temperatures, which will be taken into account in the
discussion in Section 5.

10
10
With reaction
Without reaction 8
8

6
∆ T(°C)

6
∆ T(°C)

4
4

2 2

0 0
0 20 40 60 80 0 20 40 60 80
Time(min) Time(min)

(a) (b)

Figure 4: Temperature difference between the sample temperature and the setting value of
the crucible with a time-step (a) 10 s and (b) 1 s. (p(H2 O)=13 mbar, N=60, kh = 1 ◦ C/min)

4. Velocity, temperature and concentration fields

A typical example of the temperature distribution of the flow at t = 30
min (setpoint temperature 127 ◦ C) is shown in Fig. 5(a). The temperature
around the crucible is high, as indicated in red, but inside the crucible the
temperature is much lower, as indicated in blue. The maximum difference
is up to 9 ◦ C, as also shown before in Fig. 3(a). As discussed above, the

12
particle temperature lags behind the heater temperature. One reason is the
endothermal reaction. The other reason is probably related to the shrinking
particle method used for the mesh generation. All contact points are removed,
so no direct heat conduction between the particles and the crucible wall is
possible. Thus, the heat transfer from the crucible to the sample is reduced.
The environment around the crucible is controlled by the moist air flow.
When the moist air with specified humidity flows through the furnace from
the top inlet to the bottom outlet, it is cooled by the particles due to the
endothermal reaction. As shown in Fig. 5(b), the flow is accelerated from a
velocity at the inlet of 5 mm/s to around 7.9 mm/s halfway the reactor due to
the blockage by the crucible. A stagnant blocked region is formed around the
crucible, indicated as the blue region shown in Fig. 5(b). It is clear that the
convection inside the crucible is negligible. Thus, the water vapor released by
the dehydration reaction from the solid particles is mainly transported by gas
diffusion towards the flow beyond the crucible. As shown in Fig. 5(c), water
vapor transfer by diffusion is not effective and a large amount of water vapor
is accumulated at the bottom of the crucible. The maximum concentration is
up to 10 times higher than the value of 0.39 mol/m3 of the surrounding flow.
The average vapor concentration around each particle is shown in Fig.
6(a). As the dehydration reaction takes place, the vapor concentration
reaches a peak value and then decreases till the ambient value. During
reaction, the self-generated atmosphere between different particles is very
different. For example, the maximum value of the average vapor concentration
of the particles ranges between 4 mol/m3 and 2.5 mol/m3 . Together with
Fig. 5(c), it is not difficult to find out that the average vapor concentration
around particles at the top layer of the packed bed is always lower than
for particles deeper in the bed. Due to these different ambient conditions
including temperature and water vapor pressure, the fractional conversion of
each particle is spread over a range as shown in Fig. 6(b).
In the present model free convection induced by the temperature difference
between the crucible and the particles, as well as forced convection by the flow
induced by the generation of water vapor, are not considered. It is however
possible that convective transfer of water vapor in the voids of the packed
bed should be taken into account.

13
 (a) (b)

(c)

Figure 5: (a) Temperature distribution (K) of the flow at t = 30 min; (b) Velocity
distribution (m/s) of the flow at t = 30 min; (c) Concentration distribution (mol/m3 )
of water vapor at t = 30 min. (Ts = 400 K, p(H2 O)=13 mbar, N=60, kh = 1 ◦ C/min,
∆t = 10 s)

5. Comparison to experiments and discussion

Influence of time-step. Fig. 7 shows the overall fractional conversion of the
powdery sample, which will be used in the following studies for comparison.
The numerical result for different time-steps ∆t is compared to the experi-
mental result from a TGA measurement. It is important to mention that the
temperature on the horizontal axis of all figures is the setting temperature
which can be different from the sample temperature (see Fig. 4). From Fig.
7 one can conclude that the numerical result for the fractional conversion is
delayed compared to the experimental result, which is attributed to several
factors including the numerical error of using a large time-step as discussed
above and under-estimation of the nucleation rate as discussed in our earlier

14
 4.5 1
4
3.5 0.8
c(H2O)mean(mol/m3)

3
0.6
2.5

β
2
0.4
1.5
1 0.2
0.5
0 0
0 20 40 60 80 100 110 120 130 140 150
Time(min) Temperature(°C)

(a) (b)

Figure 6: (a) Water vapor concentration profile of each particle; (b) Fractional conversion
of each particle. (p(H2 O)=13 mbar, N=60, kh = 1 ◦ C/min, ∆t = 10 s)

work [17]. Using a smaller time-step ∆t = 1 s moves the fractional conversion

towards the experimental result. However, the difference remains large.

1
Experimental result
Numerical result(∆t=10 s)
Numerical result(∆t=1 s)
0.8

0.6
α

0.4

0.2

0
100 110 120 130 140
Temperature(°C)

Figure 7: Overall fractional conversion as function of temperature: the solid line is the
experimental result and the dashed lines are the numerical results with different time-steps.
(p(H2 O)=13 mbar, N=60, kh = 1 ◦ C/min)

Influence of nucleation rate. There is a considerable difference between the

experimental and numerical results in Fig. 7. This is partly due to under-
estimation of the nucleation and growth rate in TGA as determined from our

15
microscopy experiments [17]. For example, Kirdyashkina et al.[31, 33] found a
much larger value of the growth rate. By using the nucleation and nuclei grain
model developed in our earlier work [17], a factor 4 difference of the nucleation
rate and 1.5 difference of the growth rate were derived from the best-fit of the
fractional conversion of a single salt hydrate crystal. To see the influence of the
nucleation and nuclei growth processes on the dehydration reaction of powdery
samples, different rate values are used in the simulations. Figure 8 shows
the fractional conversion as function of temperature for various nucleation
rates and a comparison to the experiment. Using a larger nucleation rate in
the simulation moves the kinetic curve towards lower temperatures, which
reduces the difference between the numerical result and the experimental
result as shown in Fig. 8(a). When the growth rate is also corrected by
a factor of 1.5, the fractional conversion from experiment is satisfactorily
predicted by the numerical result as shown in Fig. 8(b). The main difference
is at the first part of the fractional conversion, where the nucleation process
is involved. It seems that the correction factor 4 derived from a single crystal
study is not large enough for the powdery sample. It can be imagined that
there are more defects on a unit grain surface than on a unit large crystal
surface. Thus, the reaction of grains should have a larger nucleation rate than
the reaction of large crystals. In the following calculations, the nucleation
rate and growth rate measured for a crystal without defects are still used
without correction. The focus will be on the influence of various parameters
on the kinetics. In that sense, the absolute difference between numerical and
experimental results is not that important.

Influence of sample mass. The dehydration kinetics of powdery samples with

different masses was calculated and compared with experimental results ob-
tained from the TGA measurements. Figure 9 shows the numerical and
experimental fractional conversion of different sample masses. In the simula-
tions, the packed bed with larger mass constitutes of 60 spherical particles,
while the smaller sample constitutes of 20 particles with the same diameter of
1 mm. In both the numerical and experimental results, the first part of the
kinetic curves for different sample masses is the same. Then, the fractional
conversion rate for a larger sample decreases compared to the fractional con-
version rate for a smaller sample. Reaction of the larger samples takes longer
to finish, as observed in both experimental and numerical results. Except
the large delay between the numerical and experimental results, the influence
of sample mass on the reaction kinetics is predicted well by the numerical

16
 1 1
γ
0.8 4× γ 0.8

0.6 0.6
α

α
0.4 0.4

0.2 0.2 Numerical result

Experimental result

0 0
100 110 120 130 140 100 110 120 130 140 150
Temperature (°C) T (°C)

(a) (b)

Figure 8: (a) Overall fractional conversion as function of temperature for different nucleation
rates; (b) Comparison of the experimental and the numerical fractional conversion calculated
from a 4 times larger nucleation rate and 1.5 times larger growth rate. (p(H2 O)=13 mbar,
N=60, kh = 1 ◦ C/min, ∆t = 10 s)

results. Only the influence of mass on the conversion rate is larger in the
experiments than predicted by the numerical model.

0.8
61.41 mg
0.6
18.11 mg
α

64.72 mg
0.4

21.57 mg
0.2 Numerical result
Experimental result
0
100 110 120 130 140
Temperature(°C)

Figure 9: Overall fractional conversion as function of temperature for different sample

masses. (p(H2 O)=13 mbar, kh = 1 ◦ C/min, ∆t = 10 s)

The influence of the sample mass can be explained as follows. Figure 10

shows the detailed numerical results of the small size sample with 20 particles.
The average temperature difference between the particles and the crucible is
smaller than the value for the large sample as shown in Fig. 10(a). This also

17
applies to the water vapor accumulation at the bottom of the crucible. Since
less water vapor is produced from the reaction, the water vapor pressure at the
bottom of the crucible is lower. Effective heating enhances the reaction and
the lower water concentration also enhances the reaction. Thus, the reaction
kinetics is faster for the small sample than for the large sample. Meanwhile,
the more uniform surroundings of each particle in the small sample mass case
leads to a lower spread in the fractional conversion of each particle, which is
shown in Fig. 10(b). The overall kinetics of the small sample is consequently
faster compared to the overall kinetics of the large sample.

10 1
N=20
N=60
8 0.8
Max. ∆ T (°C)

6 0.6
α

4 0.4

2 0.2

0 0
0 10 20 30 40 50 100 110 120 130 140
Time(min) Temperature(°C)

(a) (b)

Figure 10: (a) Temperature difference as function of time between the sample temperature
and the setting value of the crucible for different sample mass; (b) Fractional conversion
of each particle as function of temperature. (p(H2 O)=13 mbar, N=20, kh = 1 ◦ C/min,
∆t = 10 s)

Influence of water vapor pressure. The influence of the water vapor pressure
on the reaction kinetics was examined under dynamic conditions. Figure
11(a) shows the numerical and experimental results at two different water
vapor pressures. In both numerical and experimental results, the kinetics is
hindered by the increase of water vapor pressure. It can be noticed from the
TGA results that the shape of the kinetic curve is similar, but shifted towards
higher temperatures at a higher water vapor pressure of 25 mbar. This feature
is also predicted by the developed model, although a smaller difference in
the numerical results between 13 mbar and 25 mbar is observed than the
difference in the experimental results. The reduction of the influence of water
vapor pressure on the kinetics is probably attributed to the large permeability

18
of the packed bed used in the simulations. In the experimental sample, water
vapor transfer is much more difficult because of the low permeability. To
investigate the vapor transport in the sample, simulations with lower diffusion
coefficient of water vapor were carried out. Results are shown in Fig. 11(b),
where the kinetic curve moves slightly towards higher temperatures for a
decrease in mass diffusivity by a factor 10.

1 1
Numerical result Dm
Experimental result
0.8 0.8 0.1× Dm
25 mbar
0.6 0.6
α

α
13 mbar
0.4 25 mbar 0.4

13 mbar
0.2 0.2

0 0
100 110 120 130 140 100 110 120 130 140
Temperature(°C) Temperature (°C)

(a) (b)

Figure 11: Overall fractional conversion as function of temperature (a) at different water
vapor pressures; (b) with different diffusion coefficients. (p(H2 O)=13 mbar, N=60, kh =
1 ◦ C/min, ∆t = 10 s)

Influence of heating rate. Figure 12(a) shows the comparison of numerical

and experimental results at different heating rates. It is clearly seen that
the heating rate has an important influence on the overall kinetics. The
shapes and the slopes of the kinetics curves are quite different at different
heating rates. This can be explained by the large temperature difference
between the sample and the setting value, as shown in Fig. 12(b). There
are two direct consequences caused by the large temperature difference. One
is that a higher setting temperature is required for the sample to reach
the reaction temperature. Therefore, the kinetic curve is shifted towards
higher temperatures. The other consequence is that the temperature variation
between grains is large. It is possible that a grain close to the crucible is
completely reacted, while a grain in the middle of the sample did not reach
the reaction temperature yet. This leads to an extended reaction period for
large heating rates as shown in the figure, although this effect is found to

19
be more pronounced in the experimental results compared to the numerical
results.

1 25
Numerical result
Experimental result
0.8 20

0.6 1 °C/min 15

∆ T (°C)
α

0.4 10

5 °C/min
0.2 5

0 0
100 120 140 160 0 5 10 15
Temperature(°C) Time(min)

(a) (b)

Figure 12: (a) Overall fractional conversion as function of temperature at different heating
rates; (b) Temperature difference between the sample temperature and the setting value of
the crucible for a heating rate of 5 ◦ C/min. (p(H2 O)=13 mbar, N=60, ∆t = 3 s)

6. Conclusions
This work presents a detailed mathematical model of the coupled mass
and heat transfer processes in salt hydrate grains in a TGA experiment to
investigate the influence of various conditions on the overall reaction kinetics
of powdery samples. The study focuses on the physics of mass and heat
transfer in the voids of a packed bed. It is demonstrated by simulating the
TGA experiment that the model is able to be used in such a complicated
system. Numerical results at various conditions are presented and compared
to experimental results. The influence of sample mass, water vapor pressure
and heating rate on the overall kinetics is discussed in terms of mass and heat
transfer within the sample. For the sample mass of TGA experiments, it is
recommended to keep it as small as possible to avoid too large variation of
temperature and vapor pressure over the sample. For the partial water vapor
pressure, one should note that besides the water vapor from the air flow the
self-generated water vapor during the dehydration reaction will also affect
the overall kinetics. For the heating rate, it is clear that small heating rates
are preferable in order to attain a uniform temperature distribution in the

20
sample. Isothermal methods instead of linear heating/nonisothermal methods
may be preferred to determine the overall kinetics. Also, it is shown that the
self-cooling effect cannot be ignored and can affect the results significantly,
complicating the identification of reliable and consistent kinetic parameters.
From the above simulations of various TGA measurements, it can be
concluded that several aspects of the model have to be improved in order to
calculate the mass and heat transfer more accurately. Firstly, the quality of
the geometry of the packed bed is relatively low. The micro-structure of the
packed bed plays a crucial role in the mass and heat transfer processes. For
example, the permeability of the packed bed is in general a function of porosity
and particle size and directly associated with the convective transport of water
vapor. In the present study, the porosity of the sample used in the simulations
(0.60) is larger than the value in the experiments (0.54). More important, the
structure of the void space is expected to be very different. The experimental
sample is composed of irregular particles with a size range of 0.1 mm – 0.4
mm, which is smaller than the sphere diameter (1 mm) used in simulations.
So, the particles in the TGA experiments are more closely packed, which
generates micro-pores between particles with a different structure/geometry.
The permeability of water vapor through the sample is believed to be much
lower in the experiments than in the simulations. Secondly, the quality of
the meshing of the packed bed is low. To avoid contacting points and reduce
computing effort, particles are shrunk by 2% in diameter. Thus, direct heat
transfer by conduction via contact between particles and between particles
and walls is not possible anymore. Thirdly, the flow solver is rather simple.
The physical properties in the present model are assumed to be constant. The
validity of this assumption is doubtful, particularly in the TGA system where
a large amount of water vapor is accumulated inside the sample during the
reaction. Besides, temperature dependent material properties of the moist
air will result in a changing velocity field for increasing furnace temperature.
Finally, the effect of buoyancy should also be included when a more accurate
description of convection within the sample is required. But despite these
simplifications the present model gives a good qualitative description of the
influence of various parameters influencing the deduced fractional conversion
calculated from a TGA-experiment.

21
7. Acknowledgements
The project is sponsored by the Advanced Dutch Energy Material (ADEM)
program in cooperation with the Energy research Center of the Netherlands
(ECN).

References
[1] P. Tatsidjodoung, N. Le Pierrès, L. Luo, A review of potential materials
for thermal energy storage in building applications, Renewable and
Sustainable Energy Reviews 18 (2013) 327–349.

[2] A. Solé, I. Martorell, L. F. Cabeza, State of the art on gas–solid thermo-

chemical energy storage systems and reactors for building applications,
Renewable and Sustainable Energy Reviews 47 (2015) 386–398.

[3] V. Van Essen, H. Zondag, J. C. Gores, L. Bleijendaal, M. Bakker,

R. Schuitema, W. Van Helden, Z. He, C. Rindt, Characterization of
MgSO4 hydrate for thermochemical seasonal heat storage, Journal of
Solar Energy Engineering 131 (4) (2009) 041014.

[4] H. Zondag, B. Kikkert, S. Smeding, M. Bakker, Thermochemical seasonal

solar heat storage with MgCl2 · 6 H2 O: first upscaling of the reactor, in:
IC-SES Conference, 2011.

[5] S. Mauran, H. Lahmidi, V. Goetz, Solar heating and cooling by a

thermochemical process. First experiments of a prototype storing 60kWh
by a solid/gas reaction, Solar Energy 82 (7) (2008) 623–636.

[6] H. Zondag, B. Kikkert, S. Smeding, R. de Boer, M. Bakker, Prototype

thermochemical heat storage with open reactor system, Applied Energy
109 (2013) 360–365.

[7] M. Lebrun, B. Spinner, Models of heat and mass transfers in solidgas

reactors used as chemical heat pumps, Chemical Engineering Science
45 (7) (1990) 1743–1753.

[8] V. Goetz, A. Marty, A model for reversible solid-gas reactions submitted

to temperature and pressure constraints: simulation of the rate of reaction
in solid-gas reactor used as chemical heat pump, Chemical Engineering
Science 47 (17) (1992) 4445–4454.

22
 [9] P. Neveu, J. Castaing-Lasvignottes, Development of a numerical sizing
tool for a solid-gas thermochemical transformer I. Impact of the micro-
scopic process on the dynamic behaviour of a solid-gas reactor, Applied
Thermal Engineering 17 (6) (1997) 501–518.

[10] M. Mbaye, Z. Aidoun, V. Valkov, A. Legault, Analysis of chemical heat

pumps (CHPS): basic concepts and numerical model description, Applied
Thermal Engineering 18 (3) (1998) 131–146.

[11] H.-J. Huang, G.-B. Wu, J. Yang, Y.-C. Dai, W.-K. Yuan, H.-B. Lu,
Modeling of gas–solid chemisorption in chemical heat pumps, Separation
and Purification Technology 34 (1) (2004) 191–200.

[12] G. Balasubramanian, M. Ghommem, M. R. Hajj, W. P. Wong, J. A.

Tomlin, I. K. Puri, Modeling of thermochemical energy storage by salt
hydrates, International Journal of Heat and Mass Transfer 53 (25) (2010)
5700–5706.

[13] M. Ghommem, G. Balasubramanian, M. R. Hajj, W. P. Wong, J. A.

Tomlin, I. K. Puri, Release of stored thermochemical energy from dehy-
drated salts, International Journal of Heat and Mass Transfer 54 (23)
(2011) 4856–4863.

[14] A. F. Lele, F. Kuznik, H. U. Rammelberg, T. Schmidt, W. K. Ruck,

Thermal decomposition kinetic of salt hydrates for heat storage systems,
Applied Energy 154 (2015) 447–458.

[15] C. Helbert, E. Touboul, S. Perrin, L. Carraro, M. Pijolat, Stochastic and

deterministic models for nucleation and growth in non-isothermal and/or
non-isobaric powder transformations, Chemical Engineering Science 59 (7)
(2004) 1393–1401.

[16] L. Favergeon, M. Pijolat, F. Valdivieso, C. Helbert, Experimental study

and Monte-Carlo simulation of the nucleation and growth processes dur-
ing the dehydration of Li2 SO4 · H2 O single crystals, Physical Chemistry
Chemical Physics 7 (21) (2005) 3723–3727.

[17] S. Lan, H. Zondag, A. van Steenhoven, C. Rindt, Kinetic study of

the dehydration reaction of lithium sulfate monohydrate crystals using
microscopy and modeling, Thermochimica Acta 621 (2015) 44–55.

23
[18] H. Tanaka, N. Koga, J. Šesták, Thermoanalytical kinetics for solid state
reactions as exemplified by the thermal dehydration of Li2 SO4 · H2 O,
Thermochimica Acta 203 (1992) 203–220.

[19] S. Lan, Grain-scale analysis of thermochemical heat storage materials,

PhD thesis, Eindhoven University of Technology, 2016.

[20] S. Logtenberg, M. Nijemeisland, A. Dixon, Computational fluid dynamics

simulations of fluid flow and heat transfer at the wall–particle contact
points in a fixed-bed reactor, Chemical Engineering Science 54 (13) (1999)
2433–2439.

[21] H. Calis, J. Nijenhuis, B. Paikert, F. Dautzenberg, C. Van Den Bleek,

CFD modelling and experimental validation of pressure drop and flow pro-
file in a novel structured catalytic reactor packing, Chemical Engineering
Science 56 (4) (2001) 1713–1720.

[22] A. G. Dixon, M. Nijemeisland, CFD as a design tool for fixed-bed reactors,

Industrial & Engineering Chemistry Research 40 (23) (2001) 5246–5254.

[23] M. Nijemeisland, A. G. Dixon, Comparison of CFD simulations to

experiment for convective heat transfer in a gas–solid fixed bed, Chemical
Engineering Journal 82 (1) (2001) 231–246.

[24] S. Romkes, F. Dautzenberg, C. Van den Bleek, H. Calis, CFD modelling

and experimental validation of particle-to-fluid mass and heat transfer
in a packed bed at very low channel to particle diameter ratio, Chemical
Engineering Journal 96 (1) (2003) 3–13.

[25] C. F. Petre, F. Larachi, I. Iliuta, B. Grandjean, Pressure drop through

structured packings: Breakdown into the contributing mechanisms by
CFD modeling, Chemical Engineering Science 58 (1) (2003) 163–177.

[26] M. Nijemeisland, A. G. Dixon, CFD study of fluid flow and wall heat
transfer in a fixed bed of spheres, AIChE Journal 50 (5) (2004) 906–921.

[27] T. Atmakidis, E. Y. Kenig, CFD-based analysis of the wall effect on

the pressure drop in packed beds with moderate tube/particle diameter
ratios in the laminar flow regime, Chemical Engineering Journal 155 (1)
(2009) 404–410.

24
[28] H. Bai, J. Theuerkauf, P. A. Gillis, P. M. Witt, A coupled DEM and
CFD simulation of flow field and pressure drop in fixed bed reactor with
randomly packed catalyst particles, Industrial & Engineering Chemistry
Research 48 (8) (2009) 4060–4074.

[29] T. Eppinger, K. Seidler, M. Kraume, DEM-CFD simulations of fixed bed

reactors with small tube to particle diameter ratios, Chemical Engineering
Journal 166 (1) (2011) 324–331.

[30] H. Tanaka, N. Koga, Self-cooling effect on the kinetics of nonisothermal

dehydration of lithium sulfate monohydrate, Journal of Thermal Analysis
and Calorimetry 36 (7) (1990) 2601–2610.

[31] B. V. L’vov, Mechanism of thermal dehydration of Li2 SO4 · H2 O, Ther-

mochimica Acta 315 (2) (1998) 145–157.

[32] B. L’vov, V. Ugolkov, The self-cooling effect in the process of dehydration

of Li2 SO4 · H2 O, CaSO4 · 2 H2 O and CuSO4 · 5 H2 O in vacuum, Journal
of Thermal Analysis and Calorimetry 74 (3) (2003) 697–708.

[33] N. Kirdyashkina, V. Okhotnikov, Kinetic studies of isothermal dehy-

dration of compressed Li2 SO4 · H2 O powders, Reaction Kinetics and
Catalysis Letters 36 (2) (1988) 417–422.

25