International Journal of Heat and Mass Transfer 55 (2012) 7346–7351

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ijhmt

Technical Note

An analytical solution to the problem of frost growth and densification on

flat surfaces
Christian J.L. Hermes ⇑
Center for Applied Thermodynamics, Department of Mechanical Engineering, Federal University of Paraná, 81531990 Curitiba-PR, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: This paper advances an algebraic first-principles simulation model to predict the frost growth and den-
Received 28 May 2012 sification on flat surfaces. The model was put forward based on macroscopic heat and mass balances in
Received in revised form 21 June 2012 the frost layer, which were written according to a dimensionless formulation and solved analytically to
Accepted 22 June 2012
obtain an algebraic expression for the time evolution of the frost thickness as a function of the Nusselt
Available online 27 July 2012
number, the supersaturation degree and the air-to-surface temperature difference. The model predictions
for the frost thickness were compared with experimental data obtained elsewhere, when a very good
Keywords:
agreement between calculated and measured counterparts was observed. The sensitivity of the frost
Frost formation
Modeling
growth rate to key heat and mass transfer parameters is also assessed and reported.
Analytical solution Ó 2012 Elsevier Ltd. All rights reserved.
Sensitivity analysis

1. Introduction Na and Webb [10,15] investigated the fundamental phenomena

associated with the mass transfer in the frost layer, concluding that
Frost is observed in several engineering applications, spanning the water vapor at the frost surface is supersaturated and, there-
from aircrafts to refrigerating appliances [1]. In general, the pres- fore, the mass transfer tends to be overpredicted if the air is as-
ence of frost is undesired since it is associated with either the dete- sumed to be saturated. Na and Webb [10] also put forward a
rioration of the equipment performance (e.g., evaporator clogging transient, distributed model for frost growth and densification on
[2,3]) or critical safety aspects (e.g., in-flight wing icing [4]). Thus flat horizontal surfaces based on mass and energy balances in the
understanding and predicting frost growth and densification frost layer, which was modeled as a porous medium. An expression
through simulation models are up-to-date issues for modern engi- to account for the supersaturation of air on the frost surface was
neering design. also introduced. The model, which entails the numerical solution
Different frost growth and densification models have been pro- of partial differential equations, successfully predicted the experi-
posed in the past two decades [5–13]. Among the most influential mental data for frost thickness with an agreement of ±15%.
studies lays the one of Lee et al. [8], who presented a simplified Lee and Ro [11] carried out a critical survey of the frost forma-
algebraic model for frost growth and densification over a flat plate. tion models available in the open literature, concluding that the
The model accounted for both molecular diffusion and vapor information concerning frost conductivity and density is the crux
desublimation within the frost layer. The initial condition for frost for successful predicting frost growth and densification phenom-
density, the Lewis number and the heat transfer coefficient were ena. Such a conclusion was later endorsed by Kandula [13], who
all adjusted to improve the model predictions that, at the end, carried out a study in which advanced correlations for frost density
agreed with experimental data satisfactorily. The model of Lee and thermal conductivity were used with a numerical model based
et al. [8] was later improved by Cheng and Cheng [9], who incorpo- on that of Cheng and Cheng [9] to predict experimental data from
rated the empirical correlation for frost density devised by Hayashi different researchers. After a comprehensive model validation,
et al. [14], qf = 650  exp (0.277  Tf), so that the frost thickness be- Kandula [13] reported that, in general, the model predicted the
came the only dependent variable. Through such an approach, they experimental data satisfactorily.
not only solved the issue related to the initial condition for frost In a previous work [12], the author and coworkers put forward a
density but also enhanced the model predictions for frost densifi- theoretical and experimental investigation of the frost accretion on
cation, reporting improvements in comparison with the original flat surfaces focusing on the most important factors affecting the
model. frost growth and densification phenomena, such as the surround-
ing air temperature, humidity and velocity, and the surface tem-
⇑ Tel.: +55 41 3361 3239; fax: +55 41 3361 3123. perature. The processes of frost growth and densification were
E-mail address: chermes@ufpr.br investigated experimentally in order to provide a physical basis

0017-9310/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2012.06.070
 C.J.L. Hermes / International Journal of Heat and Mass Transfer 55 (2012) 7346–7351 7347

Nomenclature

a0, . . . , a2 coefficients of Eq. (13) t time, s

b0, . . . , b3 coefficients of Eq. (29) ua air velocity, m s1
Bi Biot number, Bi = hxs/kf X dimensionless frost thickness, X = xs/L
C coefficient of Eq. (30) xs frost thickness, m
cp specific heat of moist air, J kg1 K1
d0, d1 coefficients of Eqs. (33) and (34) Greeks
D diffusivity of water vapor in air, m2 s1 a thermal diffusivity, m2 s1
Df effective diffusivity of water vapor in air in the frost b slope of Eq. (14)
layer, m2 s1 d tortuosity
h heat transfer coefficient, W m2 K1 / relative humidity, / = pv/psat
hm mass transfer coefficient, kg m2 s1
pffiffiffiffiffiffiffiffiffiffi e porosity
Ha Hatta number, Ha ¼ xs k=Df k desublimation coefficient
isv latent heat of desublimation (=2.83  106), J kg1 v kinematic diffusivity, m2 s1
Ja Jakob number, Ja = cp(Ta  Tw)/(xa  xw)isv h dimensionless temperature, h = (Ts  Tw)/(Ta  Ts)
k thermal conductivity of moist air, W m1 K1 q density of moist air, kg m3
kf effective thermal conductivity of frost layer, W m1 K1 qf space-averaged frost density, kg m3
kfo intercept of Eq. (14), W m1 K1 x humidity ratio, x = 0.622/pv/(patm  /pv), kgv kg1
L plate length, m x
~ supersaturation degree, x~ ¼ xa  xw , kgv kg1
Le Lewis number, Le = a/D s dimensionless time, s = kft/qfcpL2
m mass flux, kg m2 s1
Nu Nusselt number, Nu = hL/k Subscripts
Pr Prandtl number, Pr = m/a a moist air
q sensible heat flux, W m2 calc calculated
Re Reynolds number, Re = uaL/m exp experimental
Sc Schmidt number, Sc = m/D f frost
Sh Sherwood number, Sh = hmL/D i ice
Ta air temperature, °C s frost surface
Ts frost surface temperature, °C sat saturation
Tw plate surface temperature, °C v water vapor
T~ air-to-surface dimensionless temperature difference, w plate surface
T~ ¼ a1 ðT a  T w Þ

Z xs
for the development of a theoretical model to predict the variation d
m¼ qðxÞdx ð1Þ
of the frost thickness with time. The governing equations for mass dt 0
and heat diffusion were integrated analytically, coming up with a
where m is the mass flux of water vapor going into the frosted med-
formulation which requires numerical integration of only one time
ium. Defining the space-averaged frost density, qf, as follows:
dependent ordinary differential equation. When compared with
Z xs
experimental data, the model predictions of the frost thickness as 1
qf ¼ qðxÞdx ð2Þ
a function of time agreed to within ±10% error bands. xs 0
Despite the abundant literature in the field, there is still a need
where xs is the frost thickness, thus Eq. (1) can be written in terms
for an explicit, algebraic relationship between the frost thickness
of two different mass fluxes (see Fig. 1), yielding
and the factors affecting the frost formation process. This paper,
therefore, puts forward a dimensionless model based on macro- dxs dq
m ¼ qf þ xs f ð3Þ
scopic heat and mass balances in the frost layer, which was solved dt dt
analytically to obtain an algebraic expression for the frost thick-
ness as a function of time. The model predictions for the frost
thickness were compared with experimental data obtained from
[16], when a very good agreement between calculated and mea-
sured counterparts was observed.

2. Formulation

2.1. Modeling of frost growth and densification

In general, the following key simplifying assumptions have

been considered for modeling the frost accretion on flat surfaces:
(i) the processes of mass and heat diffusion within the frost layer
were treated as quasi-steady and one-dimensional; (ii) the frost
thickness is uniform along the plate length; (iii) the air pressure
is uniform in the air stream and within the frost layer; and (iv)
the Lewis analogy is applicable. Thus, most frost growth and den-
sification models available in the open literature [5–13] have been
formulated based on the following mass balance in the frost layer Fig. 1. Physical model for frost growth and densification over a flat surface (adapted
represented in Fig. 1: from [12]).
7348 C.J.L. Hermes / International Journal of Heat and Mass Transfer 55 (2012) 7346–7351

where the first term of the right-hand side stands for the frost qf ¼ a0 expða1 T s þ a2 Þ ð13Þ
growth rate, whereas the second term is regarded with the frost
layer densification as water vapor migrates into the porous medium, kf ¼ kfo þ bqf ð14Þ
where it changes phase into ice crystals, thus increasing the frost
The formulation provides a mean to determine the time evolution
density and reducing the frost porosity, e, which in turn is defined as
of the frost thickness, which in addition of requiring numerical cal-
qf  qi
e¼ ð4Þ culations does not provide a straight indication of the parameters
qa  qi affecting the frost formation process. The present study revisits
Furthermore, the overall mass flux from moist air to the frost layer, the model of Hermes et al. [12], invoking additional assumptions
m, is calculated from and adding further simplifications so that an algebraic, explicit
expression for xs(t) is achieved.
m ¼ hm ðxa  xs Þ ð5Þ
where hm is the convective mass transfer coefficient, related to the 2.2. Proposed analytical solution
heat transfer coefficient h through the Lewis boundary layer anal-
ogy, Nu/Pr1/3 = Sh/Sc1/3 [17], yielding The model simplification starts by recognizing that the temper-
ature profiles in the frost layer (see Fig. 1) are fairly linear [12],
h
hm ¼ ð6Þ hence Eq. (10) can be re-written as follows:
cp Le2=3
Ts  Tw h
where Nu and Sh are the Nusselt and Sherwood numbers, respec- kf ¼ hðT a  T s Þ þ ðxa  xs Þisv ð15Þ
xs cp
tively, Pr and Sc are the Prandtl and Schmidt numbers, respectively,
and Le = Sc / Pr = a/D is the Lewis number, which has been calculated where Le  1, xs = xsat(Ts) and xw = xsat(Tw), so that the porous
through different ways in the literature. For example, Lee et al. [8] medium formed within the frost layer has not to be modeled. Defin-
proposed an empirical formula for the Lewis number considering ing the Biot number as Bi = hxs/kf, Eq. (15) can be written as
the air humidity ratio and velocity, while Na and Webb [15] modi-    
Ts  Tw xa  xs isv
fied the Lewis number to account for the influence of the properties h¼ ¼ Bi 1 þ ð16Þ
Ta  Ts T a  T s cp
of porous media, such as porosity and tortuosity. Most authors,
however, assumed the Lewis number to be the unity [5,9,13]. Additionally noting that a variation in the supercooling degree is
In addition to the overall mass balance of Eq. (3), local mass and proportional to a variation in the supersaturation degree [1], it fol-
energy balances are invoked to determine the water vapor concen- lows that
tration and also the temperature distribution in the frost layer (see xa  xs xa  xw
Fig. 1),  ð17Þ
Ta  Ts Ta  Tw
2
d x
Df ¼ kx ð7Þ Now defining the dimensionless frost thickness, X = xs/L, and noting
2
dx that Bi = NuX(k/kf), where Nu = hL/k is the Nusselt number associ-
ated with the boundary layer over the frost surface, it follows that
2
d T    
kf 2
¼ qa isv kx ð8Þ 1 k 1
dx h ¼ Bi 1 þ ¼ Nu 1þ X ð18Þ
Ja kf Ja
where Df is the effective diffusivity of water vapor in air in the frost
layer (=De/d), and kf is the effective thermal conductivity of the por- where Ja is the sensible-to-latent-heat ratio, which can be inter-
ous medium, being both dependant on the porosity e and the tortu- preted as a modified Jakob number,
osity d of the frosted medium, isv is the latent heat of desublimation, cp ðT a  T w Þ
and k is the desublimation coefficient to be determined from the Ja ¼ ð19Þ
isv ðxa  xw Þ
solution of Eq. (7) by assuming x(x = 0) = xw, ðdx=dxÞx¼xs ¼ 0 and
x(x = xs) = xs as boundary conditions, yielding [12]: Based on Eq. (18), it can be noted that the temperature differences
involving the frost surface temperature are functions of h(X), and
1
Ha ¼ cosh ðxs =xw Þ ð9Þ the temperatures Ta and Tw, as follows:
pffiffiffiffiffiffiffiffiffiffi
where Ha ¼ xs k=Df is the Hatta number, which represents the ra- Ts  Tw h
¼ ð20Þ
tio between the time scale of diffusion and of desublimation. Simi- Ta  Tw 1 þ h
larly, Eq. (8) can be solved considering a prescribed temperature
condition at the plate surface, T(x = 0) = Tw, and the following heat Ta  Ts 1
¼ ð21Þ
flux continuity at the frost surface, Ta  Tw 1 þ h

dT  Thus, once X, Ta and Tw are known, the frost surface temperature, Ts,
kf  ¼ q þ misv ð10Þ
dx x¼xs can be determined over time. Further advancements can be incor-
porated into the model by introducing the dimensionless time,
where q is the sensible heat flux from air to the frost layer, calcu-
s = qft/kfcpL2, which is a modified Fourier number based on the dif-
lated from
fusive length xs and frost properties, where the frost thermal con-
q ¼ hðT a  T s Þ ð11Þ ductivity is calculated from Eq. (14). Noting that kf/qf = kfo/qf + b
and also that kfo/qf  b [18], it follows that kf/qf  kfo/qf, yielding
The solution of Eq. (8) yields the following temperature profile in
the frost layer [12]: c p L2 cp L2
dt ¼ qf ds þ s dqf ð22Þ
q þ misv Df kfo kfo
Ts ¼ Tw þ þ qa xw ð1  coshHaÞisv ð12Þ
kf kf Thus, Eq. (3) can be re-written as follows:
In addition, empirical information for the frost density and thermal
conductivity of the frost layer is required, which usually are in the cp L2
Lqf dX þ LXdqf ¼ m ðqf ds þ sdqf Þ ð23Þ
following form [2,3,12,14]: kfo
 C.J.L. Hermes / International Journal of Heat and Mass Transfer 55 (2012) 7346–7351 7349

In addition, it can be shown that It can be noted that the positive root of Eq. (33) provides the follow-
ing explicit solution for Eq. (31),
cp L2 Nu x~ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m ¼ L ð24Þ 2
kfo ~ 1þh
k d1 þ 4d0 s  d1
X¼ ð34Þ
~ ¼ k =ka is a dimensionless thermal conductivity, and x 2
where k fo ~ is
the supersaturation degree [1] calculated from In addition, the C-value can be determined by noticing that s/X  X/
d0 after the early crystal growth period, yielding
x
~ ¼ xa  xw ð25Þ
~
ð1 þ TÞð1 þ 1=JaÞ
Thus, Eq. (23) can be re-written as follows: C ð35Þ
x ~
~ ð2 þ TÞ
 
Nu x~ dqf Nu x ~
dX þ X  s ¼ ds ð26Þ so that the coefficients of Eq. (33) can be re-written as follows:
~ 1þh
k qf ~ 1þh
k
~
ð2 þ TÞ ~
k ~
ð2 þ TÞ
In case when Eq. (13) is used for the sake of frost density calcula- d0 ¼ x
~ ; d1 ¼
~
ð1 þ TÞð1 þ 1=JaÞ ~
Nu ð1 þ TÞð1 þ 1=JaÞ
tion, it follows that
where one can see that X is a function of s, x ~ and Nu only, as
~ , T,
dqf h dX Ja ¼ Jaðx ~ Since the solution is a parabola passing through the
~ ; TÞ.
¼ T~ ð27Þ
qf ð1 þ hÞ X origin, X(s = 0) = 0, one should expect X to growth continually with
pffiffiffi
s, which is a typical behavior of diffusive dominant mass transfer
where T~ ¼ a1 ðT a  T w Þ is a dimensionless air-to-surface tempera- processes [17,18].
ture difference, as a1 has the units [°C1], thus Ja ¼ cp T=a~ 1 isv x
~.
Substituting Eq. (27) into Eq. (26), the following evolving equation
2.3. Closing information
for the frost thickness X over time s can be derived,
dX Nux~ The frost density correlation proposed in [12] was adopted in the
¼ ð28Þ present study, thus the coefficients of Eq. (13) are a0 = 207,
ds k ~ þ Nuð1 þ TÞð1
~ ~ T~ 1þh
þ 1=JaÞX  Nux h s
X
a1 = 0.266, a2 = –0.615  Tw, with –7.6 6 Ts 6 0°C, –15 6 Tw 6 –
One can note that Eq. (28) is an ordinary differential equation with 5 °C, and qf ranging from 82 to 318 kg m3. The effective frost con-
rational coefficients, whose solution for X(s = 0) = 0 is as follows ductivity was calculated based on the correlation presented in [8]
[19]: with kf expressed as a function of the frost density according to

2
ðb0 ðb1  2Þ þ b2 Þð1 þ b3 XÞb2 =b0 þ b1 b3 ðb0 þ b2 ÞX 2 þ b3 ð2b0  b2 ðb1 þ 1ÞÞX þ b0 ð2  b1 Þ  b2
2 2
s¼0 ð29Þ
b3 ð2b0 þ b2  3b0 b2 Þ

where
 
Nux~ ~ b2 ¼ Nux ~ b3 ¼ Nu 1 þ 1 second-order polynomial fit. However, in the present analysis the
b0 ¼ ; b1 ¼ 1 þ T; ~ T;
~
k ~
k Ja quadratic term was dropped out as it has no material influence
on the final result, thus Lee’s et al. correlation could be fitted
It can be noted, however, that Eq. (29) is implicit for X and, there- to the form of Eq. (14), with kfo = 0.132 W m1 K1 and
fore, require an iterative root finding procedure. To avoid such a b = 3  104 m4 s3 K1. Finally, the Nusselt number was calculated
cumbersome calculation, one should note, through a scale analysis,
pffiffiffi from Nu = 0.037Re0.8Pr0.43, valid for Re < 3  107 [20].
that h/(1 + h)1 and X  s [18], thus yielding

h s 3. Discussion
 CX ð30Þ
1þh X
Fig. 2 compares the model predictions with experimental data
where C is a constant to be determined. Substituting Eq. (30) into for the frost thickness obtained from Piucco [16] for Ta = 16°C,
Eq. (28), it can be shown that /a = 80%, and ua = 1.0 m s1. It can be clearly seen that the approx-
dX Nux~ imated analytical solution (Eq. (34)) followed closely the exact
¼ ð31Þ analytical solution (Eq. (29)), and that they both predicted well
ds k ~ þ Nuðð1 þ TÞð1
~ þ 1=JaÞ  C x ~
~ TÞX
the experimental trends for different surface temperatures, Tw,
which can be integrated straightforwardly, particularly for t > 30 min, where the rates of frost growth and den-
Z X Z X Z s sification are equally important. At the early stages, the model pre-
~
k ~
dX þ Nuðð1 þ TÞð1 þ 1=JaÞ  C x ~
~ TÞ XdX ¼ Nux
~ ds ð32Þ dictions showed poorer comparisons with experimental data
0 0 0 partly because the approximation kf/qf  kfo/qf breaks down in
thus yielding the following second-degree algebraic equation, the crystal growth regime. However, since the numerical model
of Hermes et al. [12] has also underpredicted the experimental
X 2 þ d1 X  d0 s ¼ 0 ð33Þ data for the early stages of frost growth, one can conclude that
the empirical information required for frost density and thermal
where
conductivity computations (Eqs. (13) and (14)) are not suitable
2x
~ ~
2k 1 for the early growth period.
d0 ¼ ; d1 ¼ Fig. 3 illustrates Eq. (30), where the bullets refer to the term sh/
~
ð1 þ TÞð1 ~ T~
þ 1=JaÞ  C x ~
Nu ð1 þ TÞð1 ~ T~
þ 1=JaÞ  C x
(1 + h)X calculated from the exact solution (Eq. (29)) for the condi-
7350 C.J.L. Hermes / International Journal of Heat and Mass Transfer 55 (2012) 7346–7351

Fig. 2. Comparison between model predictions and experimental data of [14] Fig. 4. Comparison between model predictions and experimental data of [16] for
(Ta = 16 °C, /a = 80%, ua = 1.0 m s1). other working conditions

tions described in Fig. 2, whereas the lines are linear best fits
considering a zero intercept. One can clearly see that sh/(1 + h)X
behaves linearly with X, thus corroborating the approximation of
Eq. (30).
Fig. 4 explores different humidity (50% and 65%), surface tem-
perature (9 and 5 °C) and air temperature (22 °C) conditions
with ua = 0.7 m s1. Again, a good matching between the predic-
tions of Eq. (34) and the experimental data was found, with differ-
ences of the same order of the experimental uncertainty bounds
(±0.01 mm), thus indicating that the model is suitable to be used
in a vast range of running conditions. However, it should be
emphasized that the model was not able to predict the experimen-
tal trends for 5 °C, 50% after 80 min, when the experimental curve
shows an inflexion and consequently an offset between the calcu-

Fig. 5. Comparison between model predictions and experimental data of [16] for all
~ 6 0:0090, 5:3 6 T~ 6 8:5, 29:3 6 Nu 6 40:6).
data (0:0057 6 x

lated and the experimental curves. Such an issue may be due either
to experimental errors or to physical phenomena not accounted for
by the model.
Fig. 5 compares the predictions of Eq. (34) with all experimental
data, where the 45° line indicates the perfect matching. Again, it is
clear that the model underpredicts the experimental data for
X < 0.01 (i.e., early stage), although an excellent matching is ob-
served for higher values of X. It should be noted that the indepen-
dent dimensionless parameters (x ~ Nu) used for the model
~ , T,
validation exercise span the following ranges, respectively:
0:0057 6 x ~ 6 0:0090, 5:3 6 T~ 6 8:5, and 29:3 6 Nu 6 40:6.
Fig. 6 explores the model sensitivity to x~ , T~ and Nu, using the
normalized derivatives, (f/X)(oX/of), with f ¼ fx ~ Nug within
~ ; T;
the range of the experimental data used for model validation
Fig. 3. Validity of Eq. (30) (Ta = 16 °C, /a = 80%, ua = 1.0 m s1). ~  0:0075, T~  7:5 and Nu  35). It is clear that both Nu
(i.e., x
 C.J.L. Hermes / International Journal of Heat and Mass Transfer 55 (2012) 7346–7351 7351

process, the proposed formulation (Eq. (34)) requires a very low

computational effort, being therefore suitable to be used for frost
prediction in complex geometries such as tube-fin heat exchangers
and aircraft wings, which require distributed simulation models
involving a large number of control volumes. Nonetheless, the
validity (and also the sensitivity) of the model in a wider range is
to be verified face of the uncertainties of the assumptions (e.g., lin-
ear temperature profile, unitary Lewis number) and the correla-
tions considered for the frost density, the frost conductivity, and
the heat transfer coeffcient.

References

[1] R.O. Piucco, C.J.L. Hermes, C. Melo, J.R. Barbosa, A study of frost nucleation on
flat surfaces, Exp. Therm. Fluid Sci. 32 (2008) 1710–1715.
[2] F.T. Knabben, C.J.L. Hermes, C. Melo, In-situ study of frosting and defrosting
processes in tube-fin evaporators of household refrigerating appliances, Int. J.
Refrig. 34 (2011) 2031–2041.
[3] D.L. Silva, C.J.L. Hermes, C. Melo, First-principles simulation of frost
accumulation on fan-supplied tube-fin evaporators, Appl. Therm. Eng. 31
(2011) 2616–2621.
[4] F.T. Lynch, A. Khodadoust, Effects of ice accretions on aircraft aerodynamics,
Prog. Aerosp. Sci. 37 (2001) 669–767.
[5] S.M. Sami, T. Duong, Mass and heat transfer during frost growth, ASHRAE
Transactions (1989) 158–165.
~ ¼ 0:0075, T~ ¼ 7:5 and Nu = 35.
Fig. 6. Sensitivity of Eq. (34) with regard to x
[6] Y.X. Tao, R.W. Besant, Y. Mao, Characteristics of frost growth on a flat plate
during the early growth period, ASHRAE Transactions: Symposia, CH-93-2-2
(1993) 746–753.
~ play major roles upon the frost growth processes, whereas T~
and x [7] S.A. Sherif, S.P. Raju, M.M. Padki, A.B. Chan, A semi-empirical transient method
has a minor influence. Looking at the coefficients of Eq. (34), one for modelling frost formation on a flat plate, Int. J. Refrig. 16 (1993) 321–329.
[8] K.S. Lee, W.S. Kim, T.H. Lee, A one-dimensional model for frost formation on a
~
can note the term ð2 þ TÞ=ð1 ~ in both do and d1 coefficients
þ TÞ cold flat surface, Int. J. Heat Mass Transf. 40 (1997) 4359–4365.
and also that, if T~  10 then ð2 þ TÞ=ð1
~ ~  1, which explains
þ TÞ [9] C.H. Cheng, Y.C. Cheng, Predictions of frost growth on a cold plate in
why T~ has a minor influence on X. atmospheric air, Int. Commun. Heat Mass Transfer 28 (2001) 953–962.
[10] B. Na, R. Webb, New model for frost growth rate, Int. J. Heat Mass Transf. 47
(2004) 925–936.
4. Conclusions [11] Y.B. Lee, S.T. Ro, Analysis of the frost growth on a flat plate by simple models of
saturation and supersaturation, Exp. Therm. Fluid Sci. 29 (2005) 685–696.
[12 C.J.L. Hermes, R.O. Piucco, C. Melo, J.R. Barbosa Jr., A study of frost growth and
A dimensionless algebraic expression was introduced to calcu- densification on flat surfaces, Exp. Thermal Fluid Sci. 33 (2009) 371–379.
late the thickness of a frost layer over time as a function of key [13] M. Kandula, Frost growth and densification in laminar flow over flat surfaces,
Int. J. Heat Mass Transf. 54 (2011) 3719–3731.
independent parameters such as the supersaturation degree, the
[14] Y. Hayashi, A. Aoki, S. Adashi, K. Hori, Study of frost properties correlating with
air-to-surface temperature difference, and the Nusselt number that frost formation types, ASME Journal of Heat Transfer 99 (1977) 239–245.
drive both the growth and densification of a frost layer. The model [15] B. Na, R. Webb, Mass transfer on and within a frost layer, Int. J. Heat Mass
predictions were compared with experimental data from [16] – Transf. 47 (2004) 899–911.
[16] R.O. Piucco, Análise teórico-experimental da formação de geada em
with the independent dimensionless parameters ranging as fol- refrigeradores domésticos, M.Sc. thesis, Federal University of Santa Catarina,
lows: 0:0057 6 x ~ 6 0:0090, 5:3 6 T~ 6 8:5, 29:3 6 Nu 6 40:6 – Florianópolis, Brazil, 2008.
when it was observed that the model followed the experimental [17] H.D. Baehr, K. Stephan, Heat and Mass Transfer, second ed., Springer, Berlin,
Germany, 2006.
trends closely. The formulation was also used to assess the influ- [18] B.D. Storey, A.M. Jacobi, The effect of streamwise vortices on the frost growth
ences of x ~ and Nu on the frost thickness, when it was found that
~ , T, rate in developing laminar channel flows, Int. J. Heat Mass Transfer 42 (1999)
the most important players are x ~ and Nu, whereas T~ has a minor 3787–3802.
[19] D. Zwillinger (Ed.), Standard Mathematical Tables and Formulae, 30th ed., CRC
influence. Press, Boca Raton, USA, 1996.
In addition of being easy-to-implement and providing a clear [20] J.H. Lienhard IV, V.J.H. Lienhard, A Heat Transfer Textbook, 4th ed., Dover,
indication of the key parameters that drive the frost formation Mineola, USA, 2011.