approximate solution methods for one-dimensional
TRANSCRIPT
Available online at www.sciencedirect.com
ARTICLE IN PRESS
Applied Mathematics and Computation xxx (2008) xxx–xxx
www.elsevier.com/locate/amc
Approximate solution methods forone-dimensional solidification from an incoming fluid
S.L. Mitchell a,*, T.G. Myers b,1
a MACSI, Department of Mathematics and Statistics, University of Limerick, Limerick, Irelandb Department of Mathematics and Applied Mathematics, University of Cape Town, Rondebosch 7701, South Africa
Abstract
This paper concerns a one-dimensional model for solidification due to incoming supercooled liquid impacting on a sub-strate that is maintained at a fixed temperature. Using a boundary immobilisation method, and assuming that both thesolid and liquid layers remain thin throughout the process, a second-order accurate perturbation expansion is determined.An alternative approximate solution, found using the heat balance integral method, is also described to analyse the prob-lem, and the liquid height and temperatures in the solid and liquid are subsequently found for both approximate solutions.These are then compared with a numerical scheme which solves the full Stefan problem. The perturbation solution isshown to be more accurate, but the HBI method is simpler to implement and avoids complications that arise in the order-ing of terms in the perturbation expansion as the difference between the substrate and melting temperature changes.� 2008 Elsevier Inc. All rights reserved.
Keywords: Solidification; Heat balance integral; Boundary immobilisation; Thin film; Stefan problem
1. Introduction
The solidification of a molten material sprayed onto a substrate that is maintained at a temperature belowthe melting temperature has numerous natural and industrial applications. Perhaps the most common examplebeing when atmospheric water freezes on a structure. This has been studied in the context of icing on powertransmission and generating equipment, aircraft and seacraft, for example, see [6,12,13,16,17,20]. In an indus-trial setting solidification from a flowing liquid or a droplet spray is of interest in the casting of metals and sprayforming, lava flows and hydrate build-up in oil pipelines [4,5,9]. In this paper, we focus on a particular situation,relevant to atmospheric icing, where the liquid spray is supercooled. Consequently, when it impacts on a surfacethat is below the solidification temperature the liquid immediately solidifies. If there is sufficient energy in thesystem then, once an insulating layer has built up, a liquid layer can form on the top of the solid layer. For iceaccretion the initial layer is known as rime ice, the ice formed from a liquid layer is known as glaze ice.
0096-3003/$ - see front matter � 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2008.02.031
* Corresponding author.E-mail address: [email protected] (S.L. Mitchell).
1 Present address: Department of Mathematical Sciences, KAIST 373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, South Korea.
Please cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-put. (2008), doi:10.1016/j.amc.2008.02.031
2 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
One of the most challenging aspects of modelling ice accretion on structures is when there is a significantwater layer [6,21]. This occurs when temperatures are relatively high and can lead to ice forming away fromthe main impact region. In a series of papers Myers and co-workers have developed a hierarchy of modelsstarting with one-dimensional ice growth from an incoming supercooled water spray [14,18] to three-dimen-sional accretion and flow on a flat surface, which has then been extended to an arbitrary shaped surface [15–17]. In the current work we return to the basic problem, namely one-dimensional accretion, since an accuratemodel at this stage can be implemented at all subsequent stages of the hierarchy.
In the case of ice accretion on aircraft, the ice layer is never permitted to become large. Based on theassumption that both the ice and water layers remain thin throughout the process the analysis in [14–18] usesa leading order perturbation to determine the temperatures in the two layers. In the following analysis, weemploy a boundary immobilisation method (BIM) which has been applied numerically to problems with amoving boundary by Crank [3] and more recently by Kutluay et al. [11]. This technique has also been usedto obtain a perturbation solution to Stefan problems, see [2,10] and references therein. By changing the timevariable to the ice thickness we can determine the perturbation solution up to second-order, and so improve onthe accuracy of previous work. We also analyse the same problem using a heat balance integral (HBI) method,see [7,8]. Finally a numerical scheme is described, adapted from that of Brakel et al. [1], which is used to com-pare the accuracy of the perturbation and HBI solutions.
Although we are interested in general solidification problems we will confine our discussion to ice and waterunder aircraft icing conditions. Data for this situation are readily available, see [1,18]. We will also restrict ourresults to relatively high temperatures (typically down to around �6 �C).
2. Governing equations
Consider a situation where a supercooled spray impacts on a surface. If the surface is below the solidifica-tion temperature then the initial droplets will freeze almost instantaneously. In mild temperatures, or when thesolidified layer is sufficiently thick, a fluid layer may subsequently appear. In the following work we will focuson an ice and water system, although the model is equally applicable to other materials. The problem config-uration is depicted in Fig. 1.
The temperatures in the ice and water are denoted by bT ðz; tÞ and hðz; tÞ, respectively, and the thickness ofeach layer is bðtÞ and hðtÞ. The problem is described by two heat equations:
Pleaput.
obTot¼ ki
qici
o2bToz2
; 0 6 z 6 b; ð1Þ
oh
ot¼ kw
qwcw
o2hoz2
; b 6 z 6 bþ h; ð2Þ
Fig. 1. Schematic of the 1D model problem.
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 3
ARTICLE IN PRESS
where k, q and c are the conductivity, density and specific heat capacity, respectively; a Stefan condition
Pleaput.
qiLfdbdt¼ ki
obToz� kw
ohoz; at z ¼ bðtÞ; ð3Þ
where Lf is the latent heat of freezing; and a mass balance,
qi
dbdtþ qw
dhdt¼ _m; ð4Þ
where _m is the rate at which mass enters from the spray.We impose a fixed temperature boundary condition at the substrate z ¼ 0
bT ð0; tÞ ¼ bT s; ð5Þwhere bT s is the substrate temperature. This is a reasonable approximation to the true state when the substrateis a good conductor with a relatively high thermal mass. A more general cooling condition is dealt with in[1,18], but we concentrate on this case here since it reduces the number of ice growth regimes and thereforeallows us to focus on the solution methods. The extension to a cooling condition is straightforward. Theremaining boundary conditions for (1)–(4) depend on the type of ice growth, either rime or glaze. To simplifythe algebra we will restrict attention to the case, where bT s is the same as the air temperature, bT a. This corre-sponds to the substrate being a good conductor with a region exposed to the air (as occurs with aircraft icing).
We now consider the rime and glaze ice growth separately.
� Rime ice growth: In this case there is no water layer present and so h � 0. We therefore only need to solve(1) over the domain ½0; b�, where b ¼ _mt=qi is determined from the integrated form of (4) with bð0Þ ¼ 0. Anenergy balance gives the boundary condition at the top of the ice layer z ¼ b (at the ice–air interface). Weconsider two different conditions, either a fixed or variable energy condition:
ðiÞ ki
obToz¼ Qi or ðiiÞ ki
obToz¼ Qi þ qlðbT s � bT Þ; at z ¼ b; ð6Þ
where the terms in Qi may represent droplet kinetic energy, aerodynamic heating and latent heat, respectively,and ql is composed of the droplet thermal energy, evaporation and convective heat transfer. The energy termsare discussed in detail in [1].The rime phase ends at time t ¼ tw, when the ice surface reaches the melting tem-perature bT ðbw; twÞ ¼ bT f where bw ¼ bðtwÞ.� Glaze ice growth: In this phase h > 0 and we must solve (1)–(5) with the following boundary conditions. At
the ice/water interface z ¼ b, the temperatures are constant and equal to the melting temperature of the icebT f , i.e.
bT ðb; tÞ ¼ bT f ¼ hðb; tÞ: ð7Þ At the top of the water layer z ¼ bþ h, similar conditions hold to those in (6), namelyðiÞ kwohoz¼ Qw or ðiiÞ kw
ohoz¼ Qw þ qmðbT s � hÞ: ð8Þ
Provided _m is constant, Eq. (4) integrates to
qiðb� bwÞ þ qwh ¼ _mðt � twÞ; ð9Þ
using the initial condition hðtwÞ ¼ 0, and this allows us to eliminate h in subsequent calculations.3. Non-dimensional analysis
The system is now non-dimensionalised in order to identify the dominant terms. We set
z ¼ zH; t ¼ t
s; b ¼ b
H; h ¼ h
H; T ¼
bT � bT f
DT; h ¼ h� bT f
DT; ð10Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
4 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
where DT ¼ bT f � bT s. Applying this scaling to the mass balance relation (4) gives
Pleaput.
dbdtþ q
dhdt¼ 1; q ¼ qw
qi
; ð11Þ
provided we choose the timescale s ¼ qiH= _m. The Stefan condition (3) is now
dbdt¼ oT
oz� k
ohoz
� �z¼b
; k ¼ kw
ki
; ð12Þ
where the height-scale is
H ¼ kiDTLf _m
: ð13Þ
The choice of height and timescales is motivated by the fact that the mass balance and Stefan conditions obvi-ously define the growth rates. The heat equations (1) and (2) therefore become
�i
oTot¼ o2T
oz2; �i ¼
ciH _mki
; 0 6 z 6 b; ð14Þ
�wohot¼ o
2hoz2
; �w ¼ qcwH _m
kw
; b 6 z 6 bþ h: ð15Þ
The non-dimensional boundary condition (5) at the substrate–ice interface is now
T ð0; tÞ ¼ �1: ð16Þ
The rime boundary conditions (6) at the ice–air interface, z = b, are
ðiÞ oToz¼ P i; or ðiiÞ oT
oz¼ P i � pið1þ T Þ: ð17Þ
The glaze boundary conditions (7) and (8) become
T ðb; tÞ ¼ 0 ¼ hðb; tÞ; ð18Þ
and at z = b + h
ðiÞ ohoz¼ P w or ðiiÞ oh
oz¼ P w � pwð1þ hÞ; ð19Þ
where
P i ¼QiH
kiDT; pi ¼
qlH
ki
; P w ¼QwH
kwDT; pw ¼
qmH
kw
: ð20Þ
The non-dimensional problem is now set up, so we move on to describe two approximate solutions, onefound using a perturbation method and one found using a heat balance integral (HBI) method, and comparethem to a numerical solution adapted from that given in [1].
4. Rime problem 0 < t < tw
In the rime period there is no water layer and so b(t) = t. Then from the previous section we have to solve
o2Toz2¼ �i
oTot; 0 < z < bðtÞ; ð21Þ
T ð0; tÞ ¼ �1; and ðiÞ oToz¼ P i
����z¼b
or ðiiÞ oToz¼ P i � pið1þ T Þ
����z¼b
: ð22Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 5
ARTICLE IN PRESS
4.1. The perturbation solution
The glaze problem, which is dealt with in the following section, is simplified by using b as the time variable(since b(t) is monotonic it is valid to set t = t(b)). For the rime problem, b = t and so this transformation isirrelevant. However, for consistency with the following section we will work with t(b). We also use a co-ordi-nate system moving with the freezing boundary y = z � b(t), otherwise known as the BIM [2,3,10,11], anddenote T(z, t) = S(y, b).
Under this co-ordinate change Eqs. (21) and (22) become
Pleaput.
o2Soy2¼ �i
oSob� oS
oy
� �dbdt¼ �i
oSob� oS
oy
� �; �b < y < 0 ð23Þ
Sð�b; bÞ ¼ �1; and ðiÞ oSoy¼ P i
����y¼0
or ðiiÞ oSoy¼ P i � pið1þ SÞ
����y¼0
: ð24Þ
Provided �i is small, S may be expanded in powers of �i as
S ¼ S0 þ �iS1 þ �2i S2 þ � � � ð25Þ
and substituted into (23) and (24) to give
o2S0
oy2¼ 0;
o2Sk
oy2¼ oSk�1
ob� oSk�1
oy; ð26Þ
for k = 1,2, . . . , with S0(�b, b) = �1, Sk(�b, b) = 0 and
ðiÞ oS0
oy¼ P i
����y¼0
;oSk
oy¼ 0
����y¼0
;
ðiiÞ oS0
oy¼ P i � pið1þ S0Þ
����y¼0
;oSk
oy¼ �piSk
����y¼0
:
Let us first consider the fixed energy condition (i) at the ice–air interface. The solution is given by
S0 ¼ P iðy þ bÞ � 1; S1 ¼ S2 ¼ � � � ¼ 0; ð27Þ
and so S(y, b) = Pi(y + b) � 1. The rime period ends when T(bw, tw) = 0 or, equivalently, S(0, bw) = 0. Hencethe ice thickness and time at which the rime period ends occurs at bw = tw = 1/Pi. In terms of the physicalparameters this is tw ¼ Lf _m=Qi. It should be noted that back in the original co-ordinate system, the perturba-tion solution is T(z, t) = Piz � 1 which is the exact solution in this case.For the cooling condition (ii) at the ice–air interface, the leading order solution S0 is
S0ðy; bÞ ¼ A0ðbÞðy þ bÞ � 1; A0ðbÞ ¼P i
1þ pib: ð28Þ
In this case Sk 6¼ 0 for k > 0. To determine S1 we must solve
o2S1
oy2¼ oS0
ob� oS0
oy¼ A00ðbÞðy þ bÞ; S1ð�b; bÞ ¼ 0;
oS1
oy¼ �piS1
����y¼0
;
which leads to
S1ðy; bÞ ¼ A1ðbÞðy þ bÞ þ 1
6A00ðbÞðy þ bÞ½y2 þ 2bðy � bÞ�; A1ðbÞ ¼
piA00ðbÞb3
3ð1þ pibÞ: ð29Þ
Also, the second-order solution S2 is
S2ðy; bÞ ¼ A2ðbÞðy þ bÞ þ 1
6ðA01ðbÞ � A00ðbÞbÞðy þ bÞ½y2 þ 2bðy � bÞ� þ 1
120A000ðbÞðy þ bÞ½y2 þ 2bðy � 2bÞ�2;
ð30Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
6 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
where
Pleaput.
A2ðbÞ ¼ �2piA
000ðbÞb5
15ð1þ pibÞþ piðA01ðbÞ � A00ðbÞbÞb3
3ð1þ pibÞ: ð31Þ
The time tw when the rime period ends is then the solution of
S0ð0; twÞ þ �iS1ð0; twÞ þ �2i S2ð0; twÞ ¼ 0;
which becomes
tw ¼ A0ðtwÞ þ �i A1ðtwÞ �1
3A00ðtwÞt2
w
� �þ �2
i A2ðtwÞ �1
3ðA01ðtwÞ � A00ðtwÞtwÞt2
w þ2
15A000ðtwÞt4
w
� �� ��1
¼ P itw
1þ pitw
þ �ipiP it3w
3ð1þ pitwÞ3� �
2i piP it4
wð5þ pitw þ p2i t2
wÞ15ð1þ pitwÞ5
( )�1
: ð32Þ
We can approximate tw to the same level of accuracy as the temperature by taking
tw ¼ tw0þ �itw1
þ �2i tw2þ � � � ¼ tw0
��ipit
3w0
3ð1þ pitw0Þ þ
�2i pit
4w0ð15þ 18pitw0
þ 8p2i t2
w0Þ
45ð1þ pitw0Þ3
; ð33Þ
where
tw0¼ 1
P i � pi
: ð34Þ
The leading order term is singular when Pi = pi. This reflects the possibility that a water layer may never ap-pear and occurs when the incoming energy balances the heat loss. Using the expressions for Pi and pi in (20)leads to the condition DT < Qi/ql (this may be seen as the requirement to have a positive temperature gradientin (6)). Since the fixed energy condition (24i) has no cooling, pi = 0, and so provided Pi > 0 water will alwaysappear in that case.
4.2. The quadratic HBI solution
We now consider the quadratic HBI solution of the systems (21) and (22). In [19] a cubic HBI solution, withno quadratic term, is used to approximate the temperature in a finite ice block. That choice was motivated bycomparison with an exact solution, since the small z expansion gave this cubic form. Furthermore, a pertur-bation solution in the thin water layer also indicated the cubic profile was appropriate. In the current situationour perturbation solution includes a quadratic term at first-order, and so we stick with the standard quadraticHBI approximation. The solution method is now described for the boundary condition (ii) at the ice–air inter-face and then the solution for condition (i) follows directly by setting pi = 0. In general, the quadratic heatbalance integral method [7,8] involves assuming a solution of the form,
T ðz; tÞ ¼ a0ðtÞ þ a1ðtÞ 1� zb
þ a2ðtÞ 1� z
b
2
; ð35Þ
where the coefficients ai(t) are unknown. We can immediately eliminate two of these by applying the boundaryconditions in (22). Thus (35) reduces to
T ðz; tÞ ¼ �1þ P ib� ð2þ pibÞa2
1þ pib
� �zbþ a2
z2
b2: ð36Þ
The remaining unknown coefficient a2(t) must now be determined by integrating the heat equations (2) fromz = 0 to z = b,
oToz
����z¼b
� oToz
����z¼0
¼ �i
d
dt
Z b
0
T ðz; tÞdz� dbdt
T ðbðtÞ; tÞ� �
: ð37Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 7
ARTICLE IN PRESS
Then substituting T from (36) gives an ODE to solve for a2(t). At this stage Goodman and Shea [8] introducethe function wðtÞ ¼
R b0
T ðz; tÞdz, which has initial condition w(0) = 0, as there is not always an initial conditionfor the remaining unknown coefficient. In the present case we can determine that a2(0) = 0 and so could solvefor a2(t) directly. However, for the glaze problem it is more convenient to work with w, and the correspondingintegral of h, and so for consistency we also use this form here. Therefore (37) becomes
Pleaput.
dwdt¼ 2
�ib� 1
1þ pib
� �a2 þ
P ib1þ pib
� 1: ð38Þ
Before solving (38) we must eliminate a2(t) by re-writing it in terms of w as
a2ðtÞ ¼�6ð1þ pibÞðbþ wðtÞÞ þ 3P ib
2
bð4þ pibÞ: ð39Þ
Since b = t Eq. (38) is now a first-order differential equation for the single unknown w(t). Once w is known wemay determine a2(t) and hence T(z, t) via Eq. (36).
Finally, the rime period ends at t = b = tw, where T(tw, tw) = 0. Using (36) this leads to
1 ¼ P itw � ð2þ pitwÞa2ðtwÞ1þ pitw
� �þ a2ðtwÞ: ð40Þ
The solution for boundary condition (i) is simply found by setting pi = 0 in the above equations. In fact, forthe fixed boundary condition, it can be shown that a2(t) = 0 for all t. The simplest way to deduce this it tosolve the ODE for a2(t) found from (37). The general solution, when b = t and pi = 0, is given bya2ðtÞ ¼ C
ffiffitp
e3=ð�itÞ and we must set C = 0 to ensure a finite solution as t ? 0. Then the expression for T in(36) reduces to T = �1 + Piz which is identical to the perturbation solution (and the exact solution).
4.3. The numerical solution
In this section we describe a semi-implicit finite difference scheme which is used to solve the systems (21) and(22), as developed for a related problem in [1].
The problem involves solving the heat equation in (21) with a moving boundary z = b(t) and so the tem-perature profile T is calculated on a moving grid with a constant number of equally spaced mesh points. Sincethe height of the ice layer varies with time, the size of the mesh cell must be recalculated at each time step. Thiseffect is dealt with by incorporating a convection term into the heat equation. Then the total derivativereplaces the time derivative in the following way:
DTDt¼ oT
otþ oT
ozozot¼ 1
�i
o2Toz2þ oT
ozozot; ð41Þ
where ozot is the speed of the moving grid and oT
ot has been eliminated using (14). The discretised form of (41) is
T nþ1j � T n
j
Dt¼ 1
�i
T nþ1jþ1 � 2T nþ1
j þ T nþ1j�1
Dznþ1i
2
!þ 1
2
T nþ1jþ1 � T nþ1
j
Dznþ1i
ozi
ot
����jþ1
2
þT nþ1
j � T nþ1j�1
Dznþ1i
ozi
ot
����j�1
2
" #; ð42Þ
which holds for j = 1, . . . , Ji � 1 and n = 0, . . . ,N � 1. The subscript ‘‘i”, denoting ice, is introduced here forconvenience as it is necessary to distinguish between the meshes for ice and water in the numerical solution ofthe glaze problem, described in Section 5.3.
We write the discretisation in (42) as
ri þmi
2
ozi
ot
����jþ1
2
" #T nþ1
jþ1 � 1þ 2ri þmi
2
ozi
ot
� ����jþ1
2
� ozi
ot
����j�1
2
!" #T nþ1
j þ ri �mi
2
ozi
ot
����j�1
2
" #T nþ1
j�1 ¼ �T nj ; ð43Þ
where
ri ¼Dt
�iDznþ1i
2; mi ¼
DtDznþ1
i
: ð44Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
8 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
Since dbdt ¼ 1 in the rime period we have bn+1 = bn + Dt and so the parameters ozi
ot and Dznþ1i are evaluated as
Pleaput.
ozi
ot
����j
¼ jJ i
dbdt¼ j
J i
; Dznþ1i ¼ bnþ1
Dt: ð45Þ
We must now approximate the boundary conditions in (22). At the substrate–ice interface the conditionT(0, t) = �1 is simply T nþ1
0 ¼ �1. To discretise condition (ii) at the ice–air interface we set j = Ji in (42)and eliminate the first occurrence of T nþ1
J iþ1 using the discretised form,
T nþ1J iþ1 � T nþ1
J i�1
2Dznþ1i
¼ P i � pið1þ T nþ1J iÞ:
The second occurrence of T nþ1J iþ1 is approximated slightly differently using
T nþ1J iþ1 � T nþ1
J i
Dznþ1i
ozi
ot
����J iþ1
2
¼ ½P i � pið1þ T nþ1J iÞ�ozi
ot
����J i
¼ P i � pið1þ T nþ1J iÞ;
since ozi
ot
��J i¼ 1. Then the boundary condition (ii) becomes
� 1þ 2ri þ 2piDznþ1i ri þ piDznþ1
i
mi
2� mi
2
ozi
ot
����J i�1
2
" #T nþ1
J iþ 2ri �
mi
2
ozi
ot
����J i�1
2
" #T nþ1
J i�1
¼ �T nJ i� 2ðP i � piÞDznþ1
i ri � ðP i � piÞDznþ1i
mi
2ð46Þ
with (i) recovered by setting pi = 0 in (46). This solution is valid whilst T nþ1J i
< 0, and the time at which thiscondition is violated gives the numerical value of tw.
5. Glaze ice growth t > tw
During this phase the growth is governed by Eqs. (11), (12), (14), (15) subject to conditions (16), (18), (19).The initial condition t = tw, b = bw, h = 0 is determined from the solution in the rime stage.
5.1. The perturbation solution
With the change of variables y = z � b(t) and t = t(b), we write T(z, t) = S(y, b), h(z, t) = /(y, b). The ther-mal problem for the variable energy boundary condition (ii) may then be written as
o2S
oy2¼ �i
oSob� oS
oy
� �dbdt; Sð0; bÞ ¼ 0; Sð�b; bÞ ¼ 0; ð47Þ
o2/oy2¼ �w
o/ob� o/
oy
� �dbdt; /ð0; bÞ ¼ 0;
o/oy¼ P w � pwð1þ /Þ
����y¼h
ð48Þ
with the problem for the simpler fixed energy boundary condition (i) the same but with pw = 0. The Stefancondition (12) is now
dbdt¼ oS
oy� k
o/oy
� �y¼0
ð49Þ
and the mass balance (11) still holds which relates h to b. Note that we obviously require dhdt > 0 and the solu-
tion below satisfies this at leading order for all DT P 0.Before finding a perturbation solution we need to think about the size of the small parameters, �i and �w,
which are plotted against DT in Fig. 2. Clearly different terms dominate over different temperature regions,indicating that a variety of expansions are needed. For DT 2 [0, 1] both �i and �w are very small, with �w � 10�i.In this range, a first-order expansion in �w should provide accurate results (with errors of around 0.4%). ForDT 2 [1, 5.5] we note �2
w � �i and an expansion to Oð�2wÞ leads to errors typically less than 0.7%. In this case we
may set �i ¼ b�2w. In the range DT 2 [5.5, 8.5] we may set �i ¼ b1�
3w and so on.
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
0 2 4 6 8 10 12 14 16 180
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
temperature difference
εw
εw2
εi
εiε
w
εi2
Fig. 2. Variation of �i, �w and second-order terms, plotted against Dt.
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 9
ARTICLE IN PRESS
As discussed in Section 1 our primary interest is in relatively high temperatures, where a significant waterlayer appears. For this reason we will analyse DT 2 [1, 5.5]. The solution for DT < 1 can be taken from theleading order terms of this solution. Expanding the temperatures as
Pleaput.
S ¼ S0 þ �2wS2 þ � � � ; / ¼ /0 þ �w/1 þ �2
w/2 þ � � �
and substituting into (47) and (48) leads too2S0
oy2¼ 0; S0ð0; bÞ ¼ 0; S0ð�b; bÞ ¼ �1
o2S2
oy2¼ b
oS0
ob� oS0
oy
� �oS0
oy� k
o/0
oy
� �y¼0
; S2ð0; bÞ ¼ 0; S2ð�b; bÞ ¼ 0
and
o2/0
oy2¼ 0; /0ð0; bÞ ¼ 0;
o/0
oy¼ P w � pwð1þ /0Þ
����y¼h
;
o2/1
oy2¼ o/0
ob� o/0
oy
� �oS0
oy� k
o/0
oy
� �y¼0
; /1ð0; bÞ ¼ 0;o/1
oy¼ �pw/1
����y¼h
;
o2/2
oy2¼ o/0
ob� o/0
oy
� ��k
o/1
oy
� �y¼0
þ o/1
ob� o/1
oy
� �oS0
oy� k
o/0
oy
� �y¼0
; /2ð0; bÞ ¼ 0;o/2
oy¼ �pw/2
����y¼h
:
For brevity, in this section we only give the results for boundary condition (i). The expansions for boundarycondition (ii) are quoted in Appendix.
The leading order solutions with pw = 0 are therefore
S0 ¼ A0ðbÞy; A0ðbÞ ¼1
b; and /0 ¼ B0y; B0 ¼ P w; ð50Þ
and the Oð�wÞ solution is given by
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
10 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
Pleaput.
/1 ¼ B1ðhÞy �k2
B0y2; B1ðb; hÞ ¼ B0kh; ð51Þ
where
k � kðbÞ ¼ oS0
oy� k
o/0
oy
� �y¼0
¼ A0ðbÞ � kB0 ¼1
b� kP w: ð52Þ
Then the Oð�2wÞ solutions are
S2 ¼ A2ðbÞy þkb6½A00ðbÞy3 � 3A0ðbÞy2�; ð53Þ
/2 ¼ B2ðb; hÞy þ1
2B1ðb; hÞðkB0 � kÞy2 þ k
6ðB1b þ B0kÞy3 � k
24B0kby4 ð54Þ
with
A2ðbÞ ¼kb6½A00ðbÞb2 þ 3A0ðbÞb�;
B2ðb; hÞ ¼ �B1ðb; hÞðkB0 � kÞh� k2ðB1b þ B0kÞh2 þ k
6B0kbh3;
where we have used B1b to denote oB1
ob . Substitution of expansions S and / into the Stefan condition (49) gives
dbdt¼ oS0
oyþ �2
w
oS2
oy� k
o/0
oyþ �w
o/1
oyþ �2
w
o/2
oy
� �� �y¼0
¼ A0ðbÞ þ �2wA2ðbÞ � kfB0 þ �wB1ðhÞ þ �2
wB2ðhÞg; ð55Þ
subject to the initial condition b = bw at t = tw = bw. The substitution h = (t � b)/q reduces Eq. (55) to a first-order ordinary differential equation for b. Note that the expression for /2 contains the term hb = (tb � 1)/q,where tb may be replaced using tb = 1/k. This is a leading order approximation which is sufficient here sincehb only appears at second-order. However, for the boundary condition (ii) this quantity appears at first-orderand so we must expand tb as tb,0 + �wtb,1. This is discussed in more detail in Appendix.
Once Eq. (55) is solved, we can calculate h(t) and consequently the temperature profiles in the two layers.
5.2. The quadratic HBI solution
The method for determining the quadratic HBI solution for glaze growth is similar to that discussed in Sec-tion 4.2. We briefly describe the solution method for the boundary condition (ii) at the water–air interface andthen the solution for condition (i) follows directly by setting pw = 0. The solution for T is again of the form(35) and similarly for h we assume
hðz; tÞ ¼ c0ðtÞ þ c1ðtÞbþ h� z
h
� �þ c2ðtÞ
bþ h� zh
� �2
; ð56Þ
where the coefficients ai(t) and ci(t) are unknown. Note that we have divided by h in (56) as this is the width ofthe water region, and it simplifies the algebra when dealing with the boundary condition at z = b + h.
We can immediately eliminate four of the coefficients by applying the boundary conditions in (16), (18),(19ii). Thus T in (35) is simply
T ðz; tÞ ¼ �1þ ð1� a2Þzbþ a2
z2
b2; ð57Þ
and h in (56) becomes
hðz; tÞ ¼ ðP w � pwÞh� c2
1þ pwh� ðP w � pwÞhþ c2pwh
1þ pwhbþ h� z
h
� �þ c2
bþ h� zh
� �2
: ð58Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 11
ARTICLE IN PRESS
The remaining unknown coefficients a2(t) and c2(t) must now be determined by integrating the heat equation(14) from z = 0 to z = b and (15) from z = b to z = b + h. Hence
Pleaput.
dwdt¼ 2a2
�ib;
dvdt¼ 2c2
�whþ db
dtþ dh
dt
� �ðP w � pwÞh� c2
1þ pwh
� �; ð59Þ
where
wðtÞ ¼Z b
0
T dz; vðtÞ ¼Z bþh
bhðz; tÞdz: ð60Þ
As discussed for the HBI solution in the rime region in Section 4.2, we find it convenient to use these integralsas it is not clear how to choose initial conditions for a2 and c2 and the resulting ODEs are more complicated.Using these expressions we can eliminate a2 and c2 from (59) by re-writing them in terms of w and v, i.e.
a2 ¼ �3� 6wb; c2 ¼
3ðP w � pwÞh4þ pwh
� 6ð1þ pwhÞvhð4þ pwhÞ : ð61Þ
The pair of first-order differential equations in (59) must be coupled with (11) and (12) to give a system of fourequations for the four unknowns w, v, b and h. However, h can immediately be eliminated by writing it interms of b as h = (t � b)/q. The Stefan condition (12) becomes
dbdt¼ 1þ a2
b� k
ðP w � pwÞh� ð2þ pwhÞc2
hð1þ pwhÞ
� �ð62Þ
and the system (59) and (62) can be solved subject to initial conditions w(tw) = ww (found from the rime solu-tion), v(tw) = 0 and b(tw) = tw.
Again note that condition (i) at the water–air interface is recovered by setting pw = 0 in the above analysis.
5.3. The numerical solution
We now explain how to extend the semi-implicit finite difference scheme described in Section 4.3 to includethe water layer which develops in the glaze period. The rime model is run until T exceeds zero. The final valueprofile in the rime region which satisfies T nþ1
J i< 0 is the initial condition for the glaze solution. At the top of the
ice layer the boundary condition (17) is replaced by (18) and so hnþ10 ¼ 0.
The ice and water growth thicknesses b and h are determined by
bnþ1 ¼ bn þ Dtbnt ; hnþ1 ¼ hn þ Dthn
t ; ð63Þ
for n = 0, . . . ,N � 1, where bnt and hnt are found by discretising (11) and (12) as follows: the former is simply
hnt ¼ ð1� bn
t Þ=q and then
bnt ¼
3T nJ i� 4T n
J i�1 þ T nJ i�2
2Dzni
� k�3hn
0 þ 4hn1 � hn
2
2Dznw
� �: ð64Þ
Note that to determine b1 in (63) we need to use the final profile in the rime region which satisfies T nþ1J i
< 0.This gives T 0
J i, T 0
J i�1 and T 0J i�2 in b0
t , where the second term is zero since h = 0 at z = b, and so b1 is known sinceb0 is simply bw = t.
The numerical scheme (43) remains valid but now
ozi
ot
����j
¼ jJ i
bnt ; ð65Þ
where bnt is given in (64). Inside the water layer the numerical scheme is derived identically to that for the ice
layer (43), namely
rw þmw
2
ozw
ot
����jþ1
2
" #hnþ1
jþ1 � 1þ 2rw þmw
2
ozw
ot
� ����jþ1
2
� ozw
ot
����j�1
2
!" #hnþ1
j þ rw �mw
2
ozw
ot
����j�1
2
" #hnþ1
j�1 ¼ �hnj ; ð66Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
12 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
which holds for j = 1, . . . ,Jw � 1 and where
TablePhysic
ci
ki
qi
Lf
Qi
qlbT f
TableNume
Mesh
m = 1m = 2m = 4m = 8m = 16
The nu
Pleaput.
rw ¼Dt
�wDznþ1w
2; mw ¼
DtDznþ1
w
; Dznþ1w ¼ hnþ1
J w
;ozw
ot
����j
¼ jJ w
hnt : ð67Þ
Finally, the boundary condition (ii) at the water/air interface is similar to (46), and so
� 1þ 2rw þ 2pwDznþ1w rw þ pwDznþ1
w hnt
mw
2� mw
2
ozw
ot
����Jw�1
2
" #hnþ1
Jwþ 2rw �
mw
2
ozw
ot
����Jw�1
2
" #hnþ1
Jw�1
¼ �hnJw� 2ðP w � pwÞDznþ1
w rw � ðP w � pwÞDznþ1w hn
t
mw
2; ð68Þ
with (i) again found from setting pw = 0.
6. Results
6.1. Rime results
Table 1 shows standard physical parameter values for ice and water, applicable to aircraft icing conditions,taken from [1]. We begin with the case of the fixed energy boundary condition (22i). As discussed above, boththe perturbation and HBI solutions in fact equal the exact solution in this case, and we therefore use this caseto test the validity of the numerical scheme. The end of the rime phase is tw = 1/Pi = 0.893533 which is inde-pendent of DT. Table 2 gives the numerical predictions of tw as we reduce the mesh size (i.e. we double m
where Dt = 1 10�4/m and Ji = 40m) for two values of DT. Observe that both columns converge to theexact tw. As discussed in Section 6.1 we only consider DT 2 [1, 5.5] since this is the range of validity for theperturbation solution in the glaze phase. However, in the rime phase the restriction is DT < Qi/ql � 18.89,which is needed to ensure that there is a positive gradient in the variable energy condition (ii), as discussedat the end of Section 5.1. Although not shown in Table 2, choosing larger values of DT in the numerical solu-tion also shows the same predictions. When m = 16 we find agreement with the exact solution to 4 d.p. and sowe use m = 16 in numerical calculations from now on.
The results for the variable energy boundary condition (22ii) are of more interest since tw now changes asDT varies. In Table 3 we present values of �i, H , Pi and pi for DT = 5,2. We refer to the perturbation solutionsT = T0, T = T0 + �iT1 and T ¼ T 0 þ �iT 1 þ �2
i T 2 as ‘‘Pert0”, ‘‘Pert1” and ‘‘Pert2”, respectively (after convert-
1al parameter values for ice and water for aircraft icing conditions [1]
2050 J/kg K cw 4218 J/kg K2.18 W/m K kw 0.57 W/m K917 kg/m3 qw 1000 kg/m3
3.34 105 J/kg _m 0.05 kg/m2 s1.869 104 J/m2 s Qw 1990 J/m2 s989.3 W/m2 K qm 956.9 W/m2 K273 K
2rical prediction of tw for DT = 5 and DT = 2
size DT = 5 DT = 2
tw = 0.8932000 tw = 0.8934000tw = 0.8933500 tw = 0.8934500tw = 0.8934500 tw = 0.8935000tw = 0.8934875 tw = 0.8935125tw = 0.8935125 tw = 0.8935250
merical time-step is Dt = 1 10�4/m and mesh spacing is Ji = 40 m.
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
Table 3Scaling parameters in the rime period for various DT
�i H Pi pi
DT = 5 0.03069 0.0006527 1.1191528 0.29620DT = 2 0.01228 0.0002611 1.1191528 0.11848
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 13
ARTICLE IN PRESS
ing back from S(y, b) to T(z, t)). For DT = 5 the numerical solution gives tw = 1.211219 whilst the predictionsfor the approximate solutions (and the differences from the numerical solution) are: tw = 1.215137 (0.32%),tw = 1.211171 (0.0040%), tw = 1.211258 (0.0032%) and tw = 1.21221 (0.082%) for the Pert0, Pert1, Pert2and HBI solutions, respectively. We can also look at the errors between the temperatures T for the variousapproximate solutions. If we define E� = jTnum � T�j and calculate the L2 norm L2;� ¼ kE�k2 ¼
PE2�
� 1=2
then L2,pert0 = 1.566 10�2, L2,pert1 = 4.78 10�4, L2,pert2 = 1.838 10�4 and L2,hbim = 2.192 10�3. Usingsmaller values of DT gives slightly smaller times to melting tw, i.e. tw � 0.9989 for DT = 2. However, the result-ing errors all show the same trend as for DT = 5, and in particular that Pert2 is the most accurate solution.
6.2. Glaze results
Typical solutions for the temperatures and ice heights b are shown in Figs. 3 and 4, respectively, for bothcases pi, pw = 0 and pi, pw 6¼ 0, with DT = 5. Table 4 shows the L2 errors between each of the approximatesolutions, Pert0, Pert1, Pert2 and HBI, and the numerical solution. In both cases, the L2 errors for tempera-ture in the ice, T, show that Pert2 is significantly more accurate than the other approximations. The HBI solu-tion is better than both Pert0 and Pert1 which is also true for the errors in the ice height b. However, for theerrors in the water layer h, the HBI solution is very inaccurate for boundary condition (i) but much better forboundary condition (ii). The errors for T for Pert0 and Pert1 are identical which is due to the fact that both ofthese expansions use T = T0 (i.e. S = S0) because the next order corrections are Oð�2
wÞ. The high errors in b aredue to the fact that the glaze calculation starts at a different tw for each solution.
The errors in Table 4 and the profiles in Figs. 3 and 4 seem to indicate that it is acceptable to use the HBIsolution in the ice layer but the perturbation solution (up to second-order accuracy) must be used in the waterlayer. This will avoid some of the lengthy calculations required for the perturbation expansion. In [19] a sim-ilar approximate solution was determined for the one-dimensional melting of a finite thickness layer. A per-turbation expansion was used in the thin liquid layer and an HBI solution used in the solid thicker layer.
Eq. (55) only involves first-order derivatives of b. If we use t as the time variable, the second-order term inthe corresponding version of (55) involves btt and b2
t so we end up with a second-order equation, with a small
1.8 2 2.2 2.4 2.6 2.8
–0.1
–0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
ice
heig
ht b
(i)
ice
water
pi, p
w = 0
position z 2 2.2 2.4 2.6 2.8
–0.22
–0.2
–0.18
–0.16
–0.14
–0.12
–0.1
–0.08
–0.06
–0.04
–0.02
ice
water
pi, p
w≠ 0
(ii)
position z
Fig. 3. Plots of the temperatures T and h for glaze growth for the two boundary conditions (19) with DT = 5. Shown are the numericalsolution (solid line), Pert1 (dotted line), Pert2 (dashed line) and HBI (dot-dashed line).
Please cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-put. (2008), doi:10.1016/j.amc.2008.02.031
2 2.2 2.4 2.6 2.8 31.6
1.65
1.7
1.75
1.8
1.85
1.9
1.95
2
2.05
2.1
t
ice
heig
ht b
(i)
pi, p
w = 0
2 2.2 2.4 2.6 2.8 31.8
1.9
2
2.1
2.2
2.3
2.4
2.5
t
pi, p
w≠ 0
(ii)
Fig. 4. Plots of ice height b for glaze growth for the two boundary conditions (19) with DT = 5. Shown are the numerical solution (solidline), Pert1 (dotted line), Pert2 (dashed line) and HBI (dot-dashed line).
Table 4Table showing the L2 errors for T, h and b (between each approximate solution and the numerical solution) from Figs. 3 and 4 in the glazeperiod
Error pi = 0, pw = 0 pi 6¼ 0, pw 6¼ 0
Pert0 Pert1 Pert2 HBI Pert0 Pert1 Pert2 HBI
T 1.90e�2 1.90e�2 9.02e�5 2.42e�3 3.17e�2 3.17e�2 1.70e�4 3.91e�3
h 4.32e�2 4.45e�2 3.69e�2 2.33e�1 5.06e�2 4.30e�2 2.05e�2 3.59e�2
b 0.350 0.283 0.097 0.128 0.578 0.592 0.292 0.422
14 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
parameter in front of the leading order term. We also only have a single initial condition on b. Carrying on tohigher order terms then introduces higher order derivatives. Using t = t(b) results in a first-order equation, andthis is the reason for making this transformation.
Lower temperature ranges generally involve one more calculation. For example if we look at DT 2 [5.5, 8.5]and set �i ¼ b�3
i then
Pleaput.
S ¼ S0 þ �3wS3 þ � � � ;
/ ¼ /0 þ �w/1 þ �2w/2 þ �3
w/3 þ � � � :
We find S3 = S2 from the solution above, the / terms remain the same and we are left to calculate /3.
7. Conclusion
In this paper we have investigated two approximate solution methods to describe one-dimensional solidi-fication due to incoming supercooled liquid impacting on a fixed temperature substrate, which is kept belowthe melting temperature. Firstly, we applied a perturbation expansion to the non-dimensional system. Usingthe boundary immobilisation method allowed us to carry out the expansion to second-order, making it moreaccurate than in previous investigations. The use of the solid thickness as the time variable ensures that thegoverning ODEs are first-order and linear. Secondly, we used an alternative approximate solution, knownas the heat balance integral method which is a popular technique for solving one-dimensional heat conductionproblems involving a change of phase. We have also described a numerical scheme which solves the full Stefansystem, and this was used to test the accuracy of the approximate perturbation and HBI solutions.
We showed that the second-order perturbation solution gave the most accurate predictions for both thetemperature and height of the solid (and consequently liquid) layer. However, the perturbation solution is ana-
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx 15
ARTICLE IN PRESS
lytically more involved since the second-order correction must be included to ensure the best accuracy, andthis requires the use of the boundary immobilisation method and a change in time variable. In addition, extracomplications arise with �i and �w changing orders depending on the size of DT. The HBI method is mucheasier to implement, though from examining the results in Section 7 it is clear that it is not accurate in thewater layer. Therefore, it seems that the best compromise between simplicity and accuracy is to use a pertur-bation solution in the liquid layer and the HBI solution in the solid layer.
Acknowledgements
T.M. acknowledges the support of the National Research Foundation of South Africa, under Grant Num-ber 2053289 as well as the Department of Mathematical Sciences at the Korean Advanced Institute of Scienceand Technology (KAIST) where a large part of this work was carried out. SM acknowledges the University ofCape Town Post-Doctoral Fellowship and KAIST.
Appendix. Glaze solution for the variable energy boundary condition
In Section 6.1 we determined the perturbation solutions for both the temperatures in the water and ice lay-ers, given by T and h, respectively, (or S and / after the transformation y = z � b(t) and t = t(b)). We wrotedown results for the simpler fixed energy boundary condition (i) and here we give the results for condition (ii).These are far more complicated, especially the second-order correction /2, but this term is shown to be nec-essary to give a more accurate approximate solution than the HBI solution, as discussed in the results sectionin Section 7.
The leading order solutions corresponding to (50) are
Pleaput.
S0 ¼ A0ðbÞy; A0ðbÞ ¼1
b; and /0 ¼ B0ðhÞy; B0ðhÞ ¼
P w � pw
1þ pwh; ð69Þ
with the Oð�wÞ solution given by
/1 ¼ B1ðb; hÞy �k2
B0ðhÞy2 þ k6
B00ðhÞhby3; ð70Þ
where
B1ðb; hÞ ¼k2
B0ðhÞh2þ pwh1þ pwh
� k6
B00ðhÞhbh2 3þ pwh1þ pwh
; ð71Þ
and
k � kðb; hÞ ¼ oS0
oy� k
o/0
oy
� �y¼0
¼ A0ðbÞ � kB0ðhÞ ¼1
b� kðP w � pwÞ
1þ pwh: ð72Þ
Note that the expression in (70) involves hb and so we will have hbb terms in /2. The second-order solution S2 isidentical to that in (53) but with k replaced by (72). However, /2 is now
/2 ¼ B2ðb; hÞy þ1
2B1ðb; hÞðkB0ðhÞ � kÞy2 þ 1
6½kðB1b þ B0kÞ � kB1ðb; hÞB00ðhÞhb�y3 � k
24½kbB0ðhÞ
þ 2kB00ðhÞhb�y4 þ k120½kbB00ðhÞhb þ kðB000ðhÞh2
b þ B00ðhÞhbbÞ�y5; ð73Þ
where
B2ðhÞ ¼ �1
2B1ðb; hÞðkB0ðhÞ � kÞh 2þ pwh
1þ pwh� 1
6½kðB1b þ B0kÞ � kB1ðb; hÞB00ðhÞhb�h2 3þ pwh
1þ pwhþ k
24
½kbB0ðhÞ þ 2kB00ðhÞhb�h3 4þ pwh1þ pwh
� k120½kbB00ðhÞhb þ kðB000ðhÞh2
b þ B00ðhÞhbbÞ�h4 5þ pwh1þ pwh
: ð74Þ
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031
16 S.L. Mitchell, T.G. Myers / Applied Mathematics and Computation xxx (2008) xxx–xxx
ARTICLE IN PRESS
The Stefan condition is again given by (55) and so b can be determined using the initial condition b(tw) = tw.Note that the expressions for /1 and /2 both involve hb and the expression for /2 also involves hbb. Hence wemust determine hb to first-order and hbb to leading order. As discussed in Section 7, hb = (tb � 1)/q. Sincetb ¼ db
dt
� �1we use (55) to give
Pleaput.
tb ¼ fA0ðbÞ � kB0ðhÞ � �wkB1ðb; hÞg�1: ð75Þ
However, B1(b, h) also involves hb and so we write tb = tb,0 + �wtb,1. After substitution of expressions for A0(b),B0(h) and B1(b, h) into (75) we find that
tb;0 ¼1
k; tb;1 ¼
1
kk2
B0ðhÞh2þ pwh1þ pwh
� k6q
B00ðhÞðtb;0 � 1Þh2 3þ pwh1þ pwh
� �:
Also, to leading order in �w it can easily be shown that
tbb ¼ �1
ðA0ðbÞ � KB0ðhÞÞ2A00ðbÞ �
kq
B00ðhÞ1
ðA0ðbÞ � KB0ðhÞÞ2� 1
!" #:
References
[1] T.W. Brakel, J.P.F. Charpin, T.G. Myers, One dimensional ice growth due to incoming supercooled droplets impacting on a thinconducting substrate, Int. J. Heat Mass Transfer 50 (2007) 1705–2694.
[2] J. Caldwell, Y.Y. Kwan, Perturbation methods for the Stefan problem with time-dependent boundary conditions, Int. J. Heat MassTransfer 46 (2003) 1497–1501.
[3] J. Crank, Two methods for the numerical solution of moving boundary problems in diffusion and heat flow, Quart. J. Mech. Appl.Math. 10 (2) (1957) 220–231.
[4] A.C. Fowler, Mathematical Models in the Applied Sciences, Cambridge University Press, 1997.[5] I.A. Frigaard, Solidification of spray formed billets, J. Eng. Math. 31 (1997) 411–437.[6] R.W. Gent, N.P. Dart, J.T. Cansdale, Aircraft icing, Phil. Trans. R. Soc. Lond. A 358 (2000) 2873–2911.[7] T.R. Goodman, The heat-balance integral and its application to problems involving a change of phase, Trans. ASME 80 (1958) 335–
342.[8] T.R. Goodman, J.J. Shea, The melting of finite slabs, J. Appl. Mech. 27 (1960) 16–27.[9] E. Gutierrez-Miravete, E.J. Lavernia, G.M. Trapaga, J. Szekely, N.J. Grant, Mathematical model of the spray deposition process,
Metall. Trans. A 20 (A) (1989) 71–85.[10] C.L. Huang, Y.P. Shih, Perturbation solutions of planar diffusion-controlled moving-boundary problems, Int. J. Heat Mass Transfer
18 (1975) 689–695.[11] S. Kutluay, A.R. Bahadir, A. Ozdes, The numerical solution of one-phase classical Stefan problem, J. Comp. Appl. Math. 81 (1997)
135–144.[12] L. Makkonen, Models for the growth of rime, glaze icicles and wet snow on structures, Phil. Trans. R. Soc. Lond. A 358 (2000) 2913–
2939.[13] L. Makkonen, M. Auttti, Wind Energy-technology and Implementation, The Effects of Icing on Wind Turbines, Elsevier Science
Publishing, 1991.[14] T.G. Myers, An extension to the Messinger model for aircraft icing, AIAA J. 39 (2) (2001).[15] T.G. Myers, J.P.F. Charpin, A mathematical model for atmospheric ice accretion and water flow on a cold surface, Int. J. Heat Mass
Transfer 47 (2007) 5483–5500.[16] T.G. Myers, J.P.F. Charpin, S.J. Chapman, The flow and solidification of a thin fluid film on an arbitrary three-dimensional surface,
Phys. Fluids 14 (8) (2002) 2788–2803.[17] T.G. Myers, J.P.F. Charpin, C.P. Thompson, Slowly accreting ice due to supercooled water impacting on a cold surface, Phys. Fluids
14 (1) (2002) 240–256.[18] T.G. Myers, D.W. Hammond, Ice and water film growth from incoming supercooled droplets, Int. J. Heat Mass Transfer 42 (1999)
2233–2242.[19] T.G. Myers, S.L. Mitchell, G. Muchatibaya, M.Y. Myers, A cubic heat balance integral method for one-dimensional melting of a
finite thickness layer, Int. J. Heat Mass Transfer 50 (2007) 5305–5317.[20] G.I. Poots, Ice and Snow Accretion on Structures, Research Studies Press, 1996.[21] S.K. Thomas, R.P. Cassoni, C.D. MacArthur, Aircraft anti-icing and de-icing techniques and modelling, J. Aircraft 33 (5) (1996) 841–
854.
se cite this article in press as: S.L. Mitchell, T.G. Myers, Approximate solution methods for ..., Appl. Math. Com-(2008), doi:10.1016/j.amc.2008.02.031