Phase Change UDF Curve Fitting
Phase Change UDF Curve Fitting
Phase Change UDF Curve Fitting
uk
Provided by Brunel University Research Archive
United Kingdom
Abstract:
This paper reports on the development, experimental validation and application of a semi-
empirical model for the simulation of the phase change process in phase change materials
(PCM). PCMs are now increasingly being used in various building materials such as
indoor thermal conditions. Unlike many commercial simulation packages that assume very
similar melting and freezing behaviour for the PCM and no hysteresis, the methodology
employed treats the melting and freezing processes separately and this allows the inclusion
of the effect of hysteresis in the modelling. As demonstrated by the results in this paper,
this approach provides a more accurate prediction of the temperature and heat flow in the
thermal conditions during thermal cycling. The difference in the prediction accuracy of the
two methods is a function of the properties of the PCM. The smaller the hysteresis of the
PCM, the lower will be the prediction error of the conventional approach, and solution time
will become the determining factor in selecting the simulation approach in practical
applications.
1
Keywords: Phase Change Materials (PCM), Heat source/sink, Temperature hysteresis
2
Nomenclature:
t Time (s)
ρ Density (kg/m3)
H Enthalpy (J/kg)
uj Velocity vector
T Temperature (K)
Direction vector
β Liquid Fraction
Subscripts:
sol Solidus
liq Liquidus
m Melting
f Freezing
t Current time
3
1.0 Introduction
Phase change materials (PCM), in the context of buildings, refer to materials with enhanced
heat storage capabilities in a specific temperature range through accessing the latent
capacity of the materials. Conventional building materials such as bricks and concrete
provide thermal mass by sensible processes, that is, through changes in temperature.
However the growing trend by architects to design aesthetically pleasing buildings, with
maximum exposure to the outdoors and maximum sunlight, produces thermally less
massive buildings and therefore reduces the influence of the thermal inertia of the building
Various studies have portrayed PCMs as a very effective way of enhancing the thermal
inertia properties of lightweight building materials [1,2]. Kuznik et al. [3] showed that a 5
thick concrete. The increased interest in PCMs has led to various companies developing
commercial products for new or retrofit applications. Such products include BASF Micronal(r)
PCM boards, DuPont Energain(r) the EBB Clay PCM building boards, as well as a growing
While the advantages of PCMs are clear, their effectiveness is heavily dependent on the
building fabric, weather conditions and their interaction with heating, ventilation and air-
performance of PCMs over a wide range of conditions and operating modes. Various
commercial simulation tools incorporate phase change modelling capabilities such as ESP-r
[10]; TRNSYS [11]; and FLUENT, amongst others. However, these simulation tools do not
provide a flexible enough way to introduce custom individual melting and freezing
4
processes. The validation of the heat source/sink method used in this study to overcome
these limitations will be performed with FLUENT, even though it can also be applied using
melting/solidification model but this does not provide the flexibility to vary the enthalpy-
temperature relationships and the possibility of introducing temperature hysteresis [4]. The
default enthalpy-porosity method used, developed from the Stefan Problem, does not
explicitly track the solid-liquid interface, but rather, a parameter known as the liquid
fraction (β), which indicates the fraction of liquid in a specific cell in the modelling domain
[5]. The liquid fraction allows the computation of the change in enthalpy from the energy in
Eq. (1)
Eqs. (2)
Where;
Eq. (3)
The solver constantly iterates between Eq. (1) and (3) to determine the temperature of each
cell. In conduction dominant materials where advective movements within the liquid phase
5
(4)
The limitations of the default enthalpy-porosity model presented above are that the
calculation of the liquid fraction is done via the lever rule [6] (that is, the
enthalpy/temperature relationships are assumed linear), and that the solver assumes that
observations show that these are not always the case, and can result in discrepancies
This paper presents an approach by which the process of melting and solidification are
solidification within the CFD model. This provides a better representation of the processes
involved and more accurate simulation results. This method is inspired by similar works
performed on evaporation/ condensation [8]. The concept relies on the addition of a heat
source term SE in equation (4) to mimic the melting and solidification process. The works in
ref. [8] focus on a single phase change temperature, incorporating source terms in the
energy and momentum equations to simulate phase change in pure water. This study, on
the other hand, focuses on phase change temperature ranges (mushy regions) and only
Contrary to the methods in commercial software, such as FLUENT which uses the enthalpy-
porosity method, the method proposed here calculates the energy stored/released directly
from the temperature in the form of a user defined function (UDF). The liquid fraction is
calculated as an extra parameter, also from a UDF, but is not used in the simulation process.
6
This eliminates the dependency of the enthalpy function to obey the lever rule and provides
The reason is that the quantification of the heat source/sink is dependent on the energy or
enthalpy change during the phase change process. This, therefore, is a semi-empirical
thermal techniques such as Differential Scanning Calorimetry (DSC) [4] or the T-history
method [13].
The approach presented in this study is valid for materials where advective movements
inside the PCM are negligible and the dominant heat transfer mode is conduction. This is the
case for PCMs impregnated in matrices such as plasterboard or tiles. In applications where
large quantities of PCM are used, for example PCM in large tank for thermal storage
applications, movement of the liquid phase will introduce heat transfer by convection. In
these cases, the source term in Eq. 1 must be modified to accommodate for the appropriate
convection effects.
7
2.0 Description of the model
A crucial aspect of the model is to differentiate between melting and freezing, so that the
solver uses the appropriate heat source function. Melting is an endothermic process, i.e.
absorbing heat as the temperature of the material increases, and freezing is an exothermic
process, releasing heat as the temperature decreases. As a result, melting will be mimicked
through a heat sink, while freezing as a heat source, with the corresponding change in
Fig. 1 - Differences between the step function used in commercial software to describe melting and
freezing, actual data from DSC results, and data produced by the modelling approach proposed in
this paper.
It can be seen from Fig. 1 that the default CFD curve is a step function, approximating the
DSC curve. The default CFD curve aims at equalising the area between the onset and the
end points, and is not an accurate representation of the actual DSC curve. On the other
hand, the UDF (SE) is a more flexible, accurate representation of the DSC curve. By
8
appropriately adjusting the regression coefficient of the UDF polynomial curve, an exact
For melting, the simulation cell temperature at the previous time-step should be lower than
at the ‘actual’ time-step, and vice-versa for freezing. This concept is therefore used to
provide the solver with the required information to determine whether the material in a
specific cell is melting or freezing, and consequently follow the appropriate enthalpy-
temperature curve. Additionally, the function of the liquid fraction can be defined based on
the curvature of the integral of the UDF DSC curve, and the onset and end temperatures of
For completeness, it is not sufficient to define the heat source term (SE) only inside the
phase change temperature range; it should also be defined for all the conditions in Table 1.
9
Heat sink conditions for Melting ( )
Table 1- Conditions required to fully defining the heat source/sink User Defined Source term. f(T)m and f(T)f are
the equations of the cumulative energy (J/kg) against temperature relationships from the UDF DSC curve in
Fig. 1, from the onset of melting and freezing, respectively.
70 mm × 80mm aluminium box as shown in Fig. 2. The box consisted of approximately 0.5 kg
of the composite PCM. The material investigated is a composite of low density polyethylene
and organic PCM, with phase change behaviour as shown in Fig. 4. The experiment was
10
Fig. 2) and a PICO data logger, calibrated for the temperature range encountered in the
experiment.
11
Fig. 4 - DSC result of PCM composite
The shaded areas underneath the DSC curves represent the latent heat energy during phase
change used in this study. Note that freezing is slightly underestimated in order to portray
the effect of inaccurately choosing the phase change parameters (onset and end
temperatures and latent heat capacity). More details are given in section 4.0.
The air temperature of the chamber was varied from below the solidus temperature and
above the liquidus temperature as shown in Fig. 3, and the temperature at different points
The physical and thermal properties of the composite PCM material shown in Table 2 were
12
Density 840 kg/m3
Following the DSC curves in Fig. 4, the cumulative energy for the PCM composite for melting
and freezing are shown in Fig. 5. It is important to note that DSC provides the heat flow
(mW) at each temperature for a specific sample mass, and that these values have to be
Fig. 5(a) - Experimental, UDF & Default CFD cumulative energy-temperature curves for the melting
process
The cumulative energy curve shown in Fig. 5(a) (equivalent to the enthalpy/energy-
temperature relation) provides the basis for the melting process. The onset of melting is
taken to be 290 K (16.85 oC) and the end temperature 302 K (28.85 oC), with a latent heat
capacity of 44360 J/kg, based on the original DSC curves in Fig 4. The equation (J/kg) of the
13
Fig. 5 (b) - Experimental, UDF & Default CFD cumulative energy-temperature curves for freezing
process
Similarly, the onset of freezing is taken to be 298.15 K (25.00 oC) and the end temperature at
289 K (15.85 oC), producing an average hysteresis of ≈ 2 K, based on the energy curves. The
latent heat capacity for freezing is 44117 J/kg. The equation of the UDF curve (J/kg) was
determined to be:
The liquid fraction equations for melting and freezing respectively are:
Validation of the model was done in two parts; melting and freezing. This is to provide
greater flexibility in the simulations, as well as to ease the investigation of the model in
regards to melting and freezing separately. The modelling results are obtained with a mesh-
14
converged model of 12,320 hexahedral elements, a converged time-step of 10s and all
simulations iteratively converged. The validation results are shown for a single point, in the
material domain, although the same trends and explanations apply to all other points.
For melting, the PCM box is initialised at 8oC and the air temperature gradually increases
with the same trend as in the experiment. A convective heat transfer coefficient of 5 W/m2K
between the box and the surrounding air is used for validation. The temperatures at the
four locations were monitored. The validation results for location A are shown in Fig. 6.
The results in Fig. 6 show the differences in using the different models in relation to the
experiment. The UDF model predicts the temperature trend at both points more accurately
than the default FLUENT model. From the energy-temperature curve in Fig. 5(a), the UDF
model produces a gradual energy change, as opposed to the default model where the
energy change is constant. Furthermore, the onset and the end temperatures of melting on
both models are accurate, confirming the reliability in the approximation of the DSC energy
15
curve. The difference towards the end of the simulation time (> 9000s) can be attributed to
an increase in specific heat capacity as the material changes from solid to liquid phase. This
The default FLUENT model predicts lower temperatures for both locations during melting.
Close attention to the models’ results in Fig. 6 shows that if the initial temperature trends
during melting are flawed, then the error is carried into the simulation affecting results at
For ease of comparison, the Root Mean Square (RMS) errors relative to the experimental
data were calculated for each location. For the case of melting, the UDF model produced an
error of 1.3%, while the Default CFD model produced a much higher error of 12.3%, and
For freezing, the PCM box is initialised at 59oC and the air temperature gradually decreases
with the same trend as in the experiment. A convective heat transfer coefficient of 8 W/m2K
between the box and the surrounding air, and a specific heat capacity of 2550 J/kgK are used
for validation. The validation results for locations C are shown in Fig. 7.
16
Fig. 7 - Experimental, UDF and Default CFD models’ temperatures of
point C during the freezing process
The results from Fig. 7 show that the experimental approximation of the energy-
temperature curve in Fig. 5 (b) is not accurate enough to exactly predict the freezing
temperature trends in the composite. This is due to the fact that a slightly inaccurate
freezing onset temperature and energy-temperature curve were deliberately chosen from
the DSC data (Fig. 4). Therefore, the discrepancy at the initial stages of freezing in the UDF
model in Fig. 7, is due to a sharper gradient in the UDF energy-temperature curve, relative
to the DSC results, at the onset of freezing (298 K) in Fig. 5(b). The temperature trend in Fig.
7 of the UDF model is however a good representation of the UDF energy-temperature curve
input.
Conversely, the default CFD model predicts a temperature trend with an average error of
approximately 2oC over the freezing range. Thus, even with a relatively inaccurate selection
of the UDF enthalpy-temperature relationship, the UDF model provides a more accurate
prediction than the default model. In this case, the RMS error for the UDF model is 0.5%,
while the default model produces a much higher error of 6.3%, over the phase change
range.
The main observation from sections 5.1 and 5.2 is that the errors with the UDF model are
lower than the default CFD model for both the melting and freezing cases.
5.3 Validation Contour plots of UDF model for Melting and Freezing
Figs. 8 and 9 show the contour plots of the PCM composite during the melting and freezing
processes using the UDF model. Melting completes after 12000s while freezing is a much
17
slower process (due to slower reduction in the surrounding experimental air temperature),
and completes in more than 20000s. The phase change effects start at the corners and
edges due to higher heat transfer rates and progresses towards the bulk of the composite.
The contour results, depicted in Figs. 8 and 9 for both the temperatures and the liquid
fraction for melting and freezing, respectively, are in accordance with the user defined
functions. Figs. 8 and 9 show five plane sections of the entire 3-D PCM model simulated, and
confirm the validity of the UDF model in regards to visualisation of heat transfer. In this case
where the advective movements are neglected, the liquid fraction does not affect the
momentum of the liquid phase of the PCM. It is simply aids visualisation of the mushy parts.
Fig. 8 - Temperature and Liquid Fraction contours during the melting process
Fig. 9 - Temperature and Liquid Fraction contours during the freezing process
18
6.0 Application of heat source/sink model
PCMs are a growing technology used in buildings to provide passive thermal control of the
indoor environment, by introducing additional thermal inertia in the internal space to lower
the air temperature swing, and maintain thermal comfort. Commercial products include
BASF Micronal(r) PCM boards, DuPont Energain(r) and the EBB Clay PCM building boards. To
test the performance of such systems, numerical simulations are often employed, but as
shown in Figs. 6 and 7, using the default enthalpy-porosity models of phase change often
leads to errors in the dynamic temperature predictions. The following sub-sections will
portray the use of the heat source/sink model, in comparison to the default CFD model for
The wall consists of 19mm render, 200mm concrete and 13mm plaster, and according to
CIBSE [14], has a decrement factor of 0.42 and decrement delay of 6.5 hrs. To this wall, a
20mm solid PCM layer was applied on the internal side, and the thermal response of the
wall examined with both the heat source/sink model and the default enthalpy-porosity
model. The walls were investigated in FLUENT by applying a sinusoidal air temperature
excitation: varying from peak temperatures of 273.15K and 318.15K, over a period of 24
hours, with heat transfer coefficient of 7.7 W/m2K on the internal surface [14], and the
19
Fig. 10 – Simulation Wall Description
The PCM used for the validation of the heat source/sink model (Fig. 4) is used for this non-
obtained from Figs. 4 and 5, and Table 2. Because of the inability to specify separate melting
and freezing temperatures and enthalpies in the default model, the solidus temperature
was taken to be 289.5 K, the liquidus temperature as 300.1 K and the latent heat capacity as
44240 J/kg, calculated as the average values for both melting and freezing.
Fig. 11 shows the temperature variation with time at the midpoints of each wall layer,
predicted by the default method in the CFD software and from the User Defined Model, for
20
Fig. 11(a) – PCM temperature development Fig. 11(b) – Plaster temperature development
Fig. 11(c) – Concrete temperature development Fig. 11(d) – Render temperature development
Fig. 11 – Simulated temperature trends at midpoints of simulation wall for validation PCM
Figs. 11 (a) – (d) show that the thermal dynamics of the entire wall is affected by the choice
of the phase change model. The phase change temperature trend in the PCM segment of
the wall (Fig. 11a) are in accordance with the individual model (default and UDF) enthalpy-
temperature curves, but the thermal dynamics of the entire wall is very different for each
As the default model assumes a higher enthalpy change at the beginning of melting (Fig. 5a),
the temperatures in the UDF model (Fig. 11a) are higher, due to the lower latent heat
storage capacity of the PCM. Subsequently, the onset of freezing is clearly shown in Fig.
11(a) at t ≈ 70000s by the change in temperature reduction rate for the UDF model, in
relation to the gradual change in the default model. It is noticed that the initial temperature
trends in the wall are very important in the subsequent development of temperature along
21
the materials. In this regard, the curvatures of the enthalpy-temperature relationships are
very important, as they dictate both the initial and subsequent development of temperature
22
7.0 Conclusion
The results in the paper show that the heat source/sink method developed is suitable for
the prediction of phase change phenomena and offers more accurate representation of the
method used by many commercial software packages. The importance of energy variation
with temperature during phase change and its direct effect to the solution dynamics in a
wall are portrayed. Furthermore, because the individual melting and freezing effects as well
as hysteresis are incorporated in the same code, this method allows accurate simulation of
the PCM behaviour under cyclic conditions imposed by the variation in external ambient
temperature.
As CFD simulations require detailed inputs, this semi-empirical model is also heavily
dependent on the thermal properties of the PCM, especially the temperatures at the start
and end of the melting and freezing processes, the energy-temperature relations as well as
the specific heat-temperature relations. Variation of the thermal conductivity of the PCM
with temperature may also have an influence if significant changes in its value take place
during phase change. Hence, the input to the simulation is crucial, and in this study, a
minimum regression coefficient (R2) of 0.992 is used in the development of correlations for
The higher the hysteresis of the PCM the more pronounced will be the advantages of the
8.0 Acknowledgements
23
This work was made possible through sponsorship from the Engineering and Physical
9.0 References
[1] L. Shilei, F. Guohui, Z. Neng, D. Li, Experimental study and evaluation of latent heat
storage in phase change materials wallboards, Energy and Buildings 39(2007): pp 1088 -
1091
[2] F. Kuznik, J. Virgone, Jean Noel, Optimization of a phase change material wallboard for
[3] F. Kuznik, J. Virgone, J.J Roux Energetic efficiency of a room wall containing PCM
wallboard: A full scale experimental investigation, Energy and Buildings, 40(2008): pp 148-
156
[4] G. Susman, Z. Dehouche, T. Cheechern, S. Craig, Tests of prototype PCM ‘sails’ for office
[5] V.R Voller, L Shadabi, Enthalpy method for tracking a phase change boundary in two
[6] ANSYS FLUENT Theory Guide (Release 13.0, November 2010), Solidification and melting,
[7] P.W Egolf, H Manz, Theory and modelling of phase change materials with and without
mush regions, International Journal of Heat and Mass Transfer 37(1994): pp 2917 – 2624
24
[8] S.C.K De Schepper, G.J Heynderickx, G.B Marin, Modeling the evaporation of a
material: Data for validation of numerical modelling, Energy and Buildings 41 (2009): pp
561–570
[10] D. Heim and J. Clarke, Numerical modelling and thermal simulation of phase change
[11] J. Bony, S Citherlet , Numerical model and experimental validation of heat storage with
[12] I. Ceróna, J. Neila, M. Khayet , Experimental tile with phase change materials (PCM) for
[14] Chartered Institute of Building Services Engineers (CIBSE), Guide A (2006): pp. 3-48
25