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

net/publication/23897012

Overview of numerical codes developed for predicted electrothermal deicing

of aircraft blades

Article · February 1988

DOI: 10.2514/6.1988-288 · Source: NTRS

CITATIONS READS

3 63

4 authors, including:

Karl-Josef Witt William B. Wright

Bauhaus-Universität Weimar NASA
172 PUBLICATIONS   1,591 CITATIONS    45 PUBLICATIONS   741 CITATIONS   

SEE PROFILE SEE PROFILE

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

Packing algorithm for widely graded random dense sphere packings View project

Hydraulic Heave in Cohesive Soil View project

All content following this page was uploaded by William B. Wright on 22 October 2014.

The user has requested enhancement of the downloaded file.

 AIAA-8 8-0 2 8 8
Overview of Numerical Codes
Developed for Predicted
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

Electrothermal De-Icing of Aircraft

Blades
T.G. Keith, K.J. De With w.B=Wright and
K.C. Masiulaniec,Univ. of Toledo,
Toledo, OH

AlAA 26th Aerospace Sciences Meeting

4 January 11-14, 1988/Reno, Nevada
For permission to copy of republish, contact the American Institute ot Aeronautics and Astronautics
370 L’Enfant Promenade, S.W., Washington, D.C. 20024
 OVERVIEW OF NUMERICAL CODES DEVELOPED FOR
PREDICTED ELECTROTHERMAL DEICING
OF AIRCRAFT BLADES

The0 G. Keith
AlAA Associate Fellow
Professor
Department of Mechanical Engineering
University of Toledo
Toledo, OH 43606

Kenneth J. De Witt
Professor
Department of Chemical Engineering
University of Toledo
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

Toledo, OH 43606

William B. Wright
Graduate Assistant
Department of Chemical Engineering
University of Toledo
Toledo, OH 43606

K. Cyril Masiulaniec
Assistant Professor
Department of Mechanical Engineering
University of Toledo
Toledo, OH 43606

ABSTRACT
An overview of the deicing computer codes that
have been developed at The University of Toledo
under sponsorship of the NASA-Lewis Research
Center is presented. These codes simulate the tran-
sient heat conduction and phase change occurring in
an electrothermal deicer pad that has an arbitrary
accreted ice shape on its surface. The codes are one-
dimensional rectangular, two-dimensional rectangular,
and two-dimensional with a coordinate traneformation
to model the true blade geometry. All modifications
relating to the thermal physics of the deicing problem
that have been incorporated into the codes will be dis-
cussed. Recent results of reformulating the codes using
different numerical methods to increase program effi-
ciency are described. In particular. this reformulation
has enabled a more comprehensive two-dimensional
code to run in much less CPU time than the original
version. The code predictions are compared with
experimental data obtained in the NASA-Lewis Icing
Research Tunnel with a UHlH blade fitted with a 8. F.
Goodrich electrothermal deicer pad. Both continuous
and cyclic heater firing cases are considered. The
major objective in this comparison is to illustrate which
codes give acceptable results in different regions of the
airfoil for different heater firing sequences.
INTRODUCTION
The formation of ice on aircraft components during
flight can severely affect aircraft performance. The ice
presence increases drag and decreases lift which not
only drastically increases fuel consumption but also
jeopardizes the ability of the aircraft to f\y. Aircraft with-
out any means of removing ice or preventing it from
forming cannot fly in atmospheric conditions which
promote ice accretion. As a result, there is a need to
develop effective syst6ms which can either keep ice
from forming on these surfaces (anti-icing) or can
remove the ice once it has formed (de-icing).
There are two commonly used systems for anti-
icing aircraft components: chemical and thermal.
Chemical systems lower the freezing point of water so
that it will not freeze, much like anti-freeze in an auto-
mobile. However, these systems are of limited use
during flight. Installation costs for supplying enough of
the chemical during flight are prohibitive as are costs
for maintenance of the system. Consequently, its use is
mainly for pre-flight deicing and for windshield anti-
icing.
Thermal systems prevent ice from forming by main-
taining the surface temperature of the aircraft compon-
ent above the melting point (32 OF). However, the
energy requirement for this is quite high and is imprac-;
tical for most aircraft. Jet airplanes use this method'
because of the availability of hot compressed air from
the engines.
There are three principal methods for de-icing air-
Released lo A l A A lo publkh in ell forms.
P.CYr 1%drdm.4 a work of Ihc u.S. Corrrnrntil snd 11
no# sua)^! I D roprriphl pwlwllon In Ihc Unlld 5I.W.
 The objective of the current work is not only to put
craft components: thermal, mechanical and electro-
the computer codes developed at The University of
impulse. Mechanical systems often employ a pneu-
Toledo into proper perspective, but also to indicate the
matic boot which is laminated to the surface to be de-
conditions when each is appropriate for use.
iced. The boot is a flexible. rubber-like material which,
when inflated. breaks the ice off the surface. Boots are W
relatively simple and efficient. but require frequent
maintenance to ensure reliability.
NUMERICAL MODELS
Electromechanical impulse is a new method still in
development. A series of electromagnets are pulsed in
Figure 1 is an illustration of a poltion of an ice
cycles, flexing a metal abrasion shield and thereby
covered airfoil that is equipped with an electrothermal
mechanically cracking the ice. This method appears to
deicer pad. As can be seen, the heating elements are
be energy efficient, but it has had limited application to
arranged around the leading edge of the airfoil and the
the deicing of aircraft components.
pad consists of five layers, exclusive of the ice layer.
Electrothermal systems use electrical heater ele- These layers are: a metal substrate, a layer of inner
ments which are laminated to the surface to be deiced. insulation, the heater, a layer of outer insulation, and
These heater elements consist of metal ribbons which
an abrasion shield. The insulation is generally made of
emit heat when an electrical current is passed through a glass cloth impregnated with resin. Moreover, in
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

them. This ribbon is surrounded by insulation and is

order to direct more heat towards the ice, the inner
covered with a top metal layer for protection. These
insulation is at least twice the thickness of the outer
heaters melt a small layer of ice at the surface which
insulation. The pad may be more complicated than
destroys the ice adhesion so that the aerodynamic
illustrated and may have additional layers. Also, ice
forces can remove the ice from the surface. Energy
may or may not be present at any particular point on
requirements for de-icing are much less than for anti-
the airfoil surface. in general, all layers are assumed to
icing, making this method practical for use on aircraft
be in perfect thermal contact; however, air gaps may
components.
exist. Marano (3) accounted for these by incorporating
Both electrothermal and pneumatic systems are
additional thermal resistances between the layers.
used to deice aircraft components. However, pneu-
matic boots are not used on smaller aircraft because of
the increase in drag caused when the boot inflates.
This is especially true on helicopter blades. At this time,
the electrothermal deicer system is considered to be
the most effective means to deice helicopter blades.
Recently, much work has been done in the area of
electrothermal deicer pad design. This is especially
true in the development of numerical models to simu-
late deicer pad performance. The first numerical study
of a deicer pad was performed by Stallabrass (l), who
developed both one and two-dimensional computer
models. Baliga (2), Marano (3). Gent (4), and Roelke
(5) all developed one-dimensional models using Figure 1 iced A i r f o i l Equipped w i t h
various numerical techniques. Chao (6) developed a an E l e c t r o t h e r m a l D e i c e r P a d
two-dimensional model which was later revised by
Leffel (7). This work was recently extended by Wright Geometric Considerations
(a), who investigated numerous numerical procedures
in an attempt to improve the numerical efficiency of Thus far. the numerical models developed at The
two-dimensional computer codes. Masiulaniec (9) used University of Toledo have taken three forms:
a coordinate transformation in order to account for
various geometric effects including blade curvature and one-dimensional rectangular (2). (3). and (5)
irregularity of the accreted ice shape. two-dimensional rectangular (6),(7),and (8)
Because there has been a steady evolution of elec- two-dimensional mapped (9)
trothermal deicer pad computer models over the past
few years, it was decided that a review of this work was A one-dimensional simulation of an electrothermal
warranted. This review will include brief descriptions of: deicer pad in a rectangular coordinate system is illus-
trated in Figure 2. It can be seen that this approach
(1) geometric considerations, neglects spanwise and transverse heat transfer effects,
considering only effects normal to the blade surface.
(2) methods used to simulate phase change,
Obviously, segmented heaters can not be treated by
(3) various numerical procedures used and
(4) capabilities of the computer codes developed. this approach. Nonetheless, because a one-dimen-
sional approach is uncomplicated numerically, requir-
Also, computed results in the form of a parametric study ing the solution of a single column of nodal tempera-
will be presented and predictions will be compared to tures, and because it yields acceptable results in a
existing experimental data. minimum amount of computer time, this approach has v
become popular for design studies.
The two-dimensional rectangular simulation, shown

-\"-
-\
03
Figure 2 A One-Dimensional, Rectangular
Deicer Pad Model
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288
in Figure 3, allows the gap between heaters to be taken
into account as well as other two-dimensional phenom-
ena such as multiple heaters, heater cycling times, and
variable ice thickness. However, because of the coor-
dinate system upon which the model is based, this ap-
proach can not treat blade curvature in any convenient
manner. Irregularity of the ice shape has been modeled
using a "staircase" grid approach, as seen in Figure 4.
Y I-i- I 1
I I
I I
Figure 3 A Two-Dimensional, Rectangular
Delcer Pad Model
Figure 4 A "Staircase" Grid Model of a Deicer
Pad with an Irregular ice Shape
A two-dimensional mapped simulation is diagram-
med in Figure 5.This approach, developed by Masiul-
aniec (9). uses a body-fitted coordinate transformation
to incorporate the full blade geometry. including all of
the heaters. The general strategy of this approach is to
map an irregularly shaped layered body into a set of
rectangular, stacked slabs. Ail computations are per-
formed in these transformed layers using the Crank-
Nicolson method with Gauss-Seidel iteration. This
code was developed so that the complete two-dimen-
sional problem could be modeled.
Figure 5 A Two-Dimensional, Mapped
Deicer Pad Model
Phase Change
Stallabrass (1) modeled the phase change by
simply holding a node at the melting point until suffi-
cient energy had accumulated to completely melt the
node. Baliga (2) used a formulation proposed by Bona-
cina et al. (8).which associates the latent heat effect
with a finite temperature interval about the phase
change isotherm. The most recent models have ail
used a technique described by Voller and Cross
(11,12) known as the enthalpy method. The enthalpy
method is also termed a weak solution method
because it is based on an integral formulation. In this
procedure, temperature is calculated from the enthalpy
and is therefore not calculated directly.
The enthalpy method uses the conservation of
energy equation formulated in terms of two dependant
variables, temperature and enthalpy. Predicting the
location of the solid-liquid interface is not required
because this is determined by nodal enthalpy alone.
The temperature at any point is then calculated using
the known enthalpy-temperature relationship. The
equivalence of this method and the classical formuia-
tion of the ablation problem was proven by Atthey (13).
 The previous investigations which used the
enthalpy method employed a non-linear relationship
between enthalpy and temperature during melting by
keeping the melt temperature constant at 32 OF. This
non-linearity requires that an iterative solution proced-
ure be used in order to find the appropriate tempera-
ture at each point. There is a modified enthapy method
which makes use of a small temperature interval at the
melting point, similar to that used by Baliga ( 2 ) . This
modification is part of a recent technique by Schneider
and Raw (14) to predict temperatures during phase
change in a more expedient fashion. Their technique
assumes the phase at each point in the ice so that the
enthalpy can be eliminated in favor of the temperature
prior to calculation. This in turn permits the use of non-
iterative solution methods which allow the new temp-
erature at each point to be determined more efficiently
than by the use of iterative methods. The assumed
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288
phase of each node is then updated as the solution
proceeds. This technique, termed the Method of
Assumed States, has been used by Roelke (5) in a
one-dimensional deicer pad analysis and by Wright (8)
in a two-dimensional rectangular analysis.
Two rules are used to establish the state of a node
within a control volume. The state of the node may be
solid, phase transition, or liquid. The two rules are:
Rule 1 : In a single iteration, any given control volume
is allowed to change by only one state.
For example, in heating, a node is only allowed to
change from solid to phase transition or from phase
transition to liquid in a single iteration. A nodal control
volume may not directly change from solid to liquid.
Rule2: In a single iteration, the state of a control
volume will not change, if in the previous
iteration, its state and that of its neighbors was
the same.
For example, if the control volume under consideration
and all of its neighbors were solid, but the calculations
predict its state should be liquid, the state of the control
volume will be maintained as solid.
These two rules were developed to reflect the physical
situation of a phase change in a melt process.
Numerical Procedures
Determination of the transient temperature distribu-
tions in a layered body containing heat sources is
impractical by any means other than by numerical
calculation. All of the models developed at The Univer-
sity of Toledo have used numerical finite difference
procedures to obtain results. The finite difference meth-
od discretizes the continuous time and space domains
into a grid of nodes. A system of algebraic equations
based on this grid replaces the governing differential
equations and boundary conditions. This system of
equations can be solved to determine the value of the
dependant variable at a particular node at any point in
time and space. Accuracy of these methods depends
on the particular method used and on the grid spacing.
For example, the explicit method is first order accurate
in time and second order accurate in space, whereas
the Crank-Nicolson scheme is second order accurate
in both time and space.
Many finite difference schemes can be found in the L
',
literature for the type of problem under study here.
However, few of these have been applied to deicer
modeling. Stallabrass (1) and Gent (4) both used an
explicit method with forward-time, central-space differ-
encing. Roelke (5) used a purely implicit method with
forward-time, central-space differencing. Furthermore,
this approach was combined with a cubic spline tech-
nique and the resulting set of equations solved using a
tridiagonal matrix technique. Baliga (2), Marano (3),
Chao (6), and Leffel(7) used the Crank-Nicolson
scheme which is an implicit scheme that has central-
time and central-space differencing. Implicit methods
have been used to eliminate stability requirements of
the explicit methods.
After finite differencing the equations, a variety of
numerical techniques can be used to solve the corres-
ponding matrix system of equations. The technique
used may be determined by the finite differencing
method, however. The simple explicit scheme needs
no solution technique as all new temperatures are calc-
ulated from previously known values. For the one-
dimensional heat conduction equation, this method
requires that the time step (At) be less than (Ax)*/Za to
ensure stability, where a is the thermal diffusivity and At
and Ax are the time and space increments, respect-
ively.
The Crank-Nicolson scheme is implicit and there-
fore is often solved using an iterative solution tech-
nique. Baliga ( 2 ) used Gaussian elimination to solve for
temperatures, which is very time consuming for a two-
dimensional problem. Marano (3), Chao (6) and Leffel
(7) used Gauss-Seidel iteration instead. This was used
because of the non-linearity in the phase change pro-
cedure employed. Anderson, et al. (15) indicated that
the Crank-Nicolson scheme is preferred over other
methods for the type of differential equation that gov-
erns the problem under consideration here.
Recently, Wright (8) investigated a variety of numer-
ical schemes in order to determine which is the most
efficient for modeling deicer pad designs. These meth-
ods were selected from those found in Anderson, et at.
(15) to be the most promising for this problem. Three of
these, the Simple Explicit, the Crank-Nicolson, and the
Simple Implicit methods, have been used by previous
investigators. The other six are the Hopscotch, ADE
(Alternating Direction Explicit), AD1 (Alternating Direc-
tion Implicit), time-splitting (the Method of Fractional
Steps), SIP (Strongly Implicit Procedure), and MSIP
(Modified Strongly Implicit Procedure). Two cases
involving a standard deicer pad were used to decide
which procedure was the most efficient. It was found
that the ADi (Alternating Direction Implicit ) method is
clearly the most efficient procedure for this problem. It is
approximately twice as fast as its nearest competitor,
MSIP, and is up to 25 times faster than the original
formulation. AD1 is faster than the other methods for two
main reasons. First, AD1 is a direct solution method and
 therefore does not have to iterate to find a solution. As
stated previously, direct methods can be used for the
phase change problem presented because of the
Method of Assumed States. Second, the time step
necessary to obtain an accurate solution with this
4
method is larger than that needed for most of the other
methods considered.
Comparison of Computer Codes
Baliga (2) and Marano (3) used a Crank-Nicolson
finite difference method in their one-dimensional
codes, while Roelke (5) used a simple implicit method.
Both of these methods are implicit schemes which are
unconditionally stable. Baliga (2) and Roelke (5) used
direct solution techniques to solve for temperature,
which could be used because of the method used to
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288
than a two-dimensional program, this analysis will be
useful in decreasing the computational time needed for
the problem. A comparison will also be made with
experimental data obtained by Leffel (7). This case
illustrates the usefulness of all three of the models
discussed previously (one-dimensional rectangular,
two-dimensional rectangular, and two-dimensional
mapped).
Parameter Variation
In this section, a parametric study of some of the key
variables of the problem will be performed in order to
determine their effect on the solution. A multitude of
variables could be studied; however, this discusion will
center on those variable's which were thought to be
responsible for making the problem two-dimensional.

simulate phase change. Marano (3) used Gauss- Table 1

Seidel iteration. All three of these codes calculate
temperatures quickly, although the most recent of these Summary of Computer Codes
codes (Roelke's) is the most expedient.
Chao's two-dimensional code (6) used the Crank-
Nicolson finite difference method and solved for
temperature using Gauss-Seidel iteration, following the
M?ibM&&
same path as Marano's (3) work with the one-dimen- One-Dimensional
sional model. Although this scheme yielded accurate
results, it was found to be inefficient for a two-dimen- Baliga High Heat Capacity, 1-D programs
sional problem. Leffel (7) modified Chao's code to Crank-Nicolson, have variable
include variable ice thickness. These codes do not take
Gauss Elimination Ay in each
full advantage of two-dimensionality in that only the
layer, variable
heater is two-dimensional, i.e.. a heater gap is inclu- Marano Enthalpy Method, heater cycling
ded. However, many other two-dimensional effects are (NASA2) Crank-Nicolson, times and ice
._. not incuded. Gauss-Seidel Iteration growth.
Recently, Wright (8) modeled many other two-dimen-
sional effects, including: variable properties in the
Roelke Method of Assumed States,
heater layer; variable ice thickness with respect to both
(MELT) Cubic Splines Technique,
space and time; a variable number of heaters with van- Tridiagonal Matrix Algorithm
able firing sequences: a nonuniform outer heat transfer
coefficient; noninsulated side boundary conditions; and
a variable grid. A significant decrease in computation
time was also achieved by the application of more Chao Enthalpy Melhod, One heater,
efficient numerical techniques such as the AD1 method. (SHED) Crank-Nicolson, variable grid
The body-fitted coordinate transformation technique Gauss-Seidel Iteration in y-direction
used by Masiulaniec (9) incorporated the Crank- only, heater
Nicolson method with Gauss-Seidel iteration. This Enthalpy Method, can be cycled,
Leffel
program allows simulation of the true blade shape. The (VARICE) Crank-Nicolson. heater con-
only drawback of this code is that a great deal of cpm- Gauss-Seidel Iteration ductivities
puter time is necessary at present to perform the coor- used in gap,
dinate transformation and to solve for the temperatures. Leffel code
Table 1 presents each of these codes along with a only has vari-
brief description of their capabilities. able ice shape
DISCUSSION OF RESULTS Method of Assumed States, >1 heater, var.
Wright
(DEICE) Several Numerical Methods grid in both
In the previous sections, numerous numerical (AD1 best), AD1 uses Tridiag- directions,
computer codes have been presented, all of which are onal Matrix Algorithm heaters cy-
capable of modeling the transient heat transfer in an cled. var. ice
electrothermal deicer pad. These codes will now be Two-Dimensional Eodv-Fitted shape and
evaluated to determine the conditions when one- growth, var

- dimensional codes are sufficient for deicing analysis, or

when two-dimensional rectangular or two-dimensional
mapped procedures are necessary. Since a one-
dimensional program uses less computational time
Masiulaniec Enthalpy Method
Crank-Nicolson
Gauss-Seidel
outer heat
x-fer coef-
ficient
Iteration

5
 ~

These variables are chosen so that a determination
can be made as to when the two-dimensional nature of
the problem disallows use of a one-dimensional
program. If a one-dimensional model is adequate for a
particular problem, the use of codes such as those
developed by Marano (3) or Roelke (5) will decrease
the amount of computer time needed to solve the
problem.
There are three variables which determine the two-
dimensional nature of the problem. First and foremost
is the size of the heater gap and the physical properties
of the gap. Second, the grid spacing may affect the
solution if an inadequate number of nodes is present. A
determination of the number of nodes necessary for
solution is also necessary for the purpose of determ-
ining an accurate solution. Third, a variable ice shape
and a variable outer ambient heat transfer coefficient
can significantly alter the temperature profiles, espec-
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288
A standard deicer, which is shown in Figure 6 , is
simulated using Wright's code (8) with the AD1 method.
Physical propetties for the layers and typical data for a
standard deicer are shown in Table 2. Figure 7 pre-
sents temperature vs time at the centerline of the heater
for the case where the physical properties of the gap
are the same as the surrounding insulation. This is a
more realistic case than using the properties of the
heater in this section which Chao's code considers.
The same cases are run using the properties of the
heater in the gap, and are shown in Figure 8. As can be
Seen in these plots, significant deviation from the one
dimensional result (zero gap thickness) is evidenced
even at small gap thicknesses. This deviation is, of
course, larger for the first cases which use a lower con-
ductivity in the gap than the cases which use the same
conductivity. These two cases are compared directly in
Figure 9, which plots heater temperature after 20 sec-

ially if there are regions on the blade which have no onds of simluated time versus gap thickness for both
ice. sets of cases considered.

1. Gap Size and Material

The size of the heater gap is the primary determin-

ant of the extent of the two-dimensional effect in a
deicer pad. An analysis of heater gap thickness can
therefore determine when a two-dimensional model is
needed or when the faster one-dimensional models
developed by Marano (3) or Roelke (5) are adequate.
The material used in this heater gap is also critical ice
since this can drastically alter the temperature of the
heater. For simplicity, Chao (6) used the same material
in the gap as was present in the heater. This analysis
will show when this is not a good approximation.

Table 2

Physical Properties of a Standard Deicer

Thermal Thermal
Conductivity Diffusivity
Layer Thickness(in) (BTU/hr-R2-"F) (ft2/hr)
__
Aluminum
D-Spar

Epoxy/Glass
Insulation
Heater Element
0.087

0.050

0.004
66.50

0.22

7.60
1.6500

0.0087

0.1380
Figure 6
- substrate
insulation
Pa b r a s i o n shield
heater e l e m e n t
A Standard Deicer Pad Model

Epoxy/Glass 0.010 0.22 0.0087 A similar set of plots are made which plot gap cen-
Insulation terline temperature instead of heater centerline temper-
ature for the same two sets of cases described previ-
Stainless Steel 0.012 8.70 0.1500 ously. These plots (Figures 10-12) show an even
Abrasion Shield greater dependence on the gap size, as well as a
greater dependence on the properties used in the gap.
Ice 0.250 1.29 0.0446 The reason that gap temperatures are significant is that
it may be necessary for the entire bottom layer of the
Heater Power Density =30. W/in2
ice to melt for the adhesion of the ice to the metal abra-
Outer Convection Coefficient = 150. BTUlhr-ft2-OF sion shield to be broken. Once this adhesion is broken,
Ambient Temperature = 10. OF dynamic forces will remove the ice from the surface. If W
Heater Length = 0.25 in. the gap temperatures are severely affected (lowered]
Heater Gap Length = 0.25 in.

6
 by the size of the gap, it will take longer to remove the
ice from the surface because the temperatures at the
0 Gap=0.25
ice-abrasion shield interface will be likewise affected. 0

160 7 0 Gap.0.30
D

-
.m.
W
5
150

140

130
0
0
Gap.0.25
Gap.3.05
.. Q
0

Gap.0.30

.,,I,
Q U
E, 4 Gap.O.50

ij,
120 8
t-
g 110
:
-
.-

, , , I I , , , , , , , ,

g 80
0 2 4 6 8 10
time (sec)
12 14 16 18 20

I
Figure 10 Variation in Gap Temperature
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

70
0 2 4 6 8 10 12 14 16 18 20
with Gap Size (k gap = k heater)
time (sec)

160 A

-5W
150

140
0 Gap-0.50
A

c
A
$a 130 D

0
Gap=O.25
Gap-0.05
0
I-
5 80 A 0
B
+5 120
n 1-D Solution m m
0 Gap.O.30 -
._
C 0

-
._ 110
8 Gap.O.50
0
0 ..
&
60

0
0

:L

100
s
0
g a
40
n
n
0
0
0

P
v
(3
L 90
20
0
iY R !! II
W
c
g 80
0 . , . , . , . I . , . ( . , . I
I 0 2 4 6 8 10 12 14 16 18 20
I
70 I . I . I , I , 1 ~ I . I , I , I I time (sec)

%
0
140- %In 140
N
.*.
0
.!! 138-' N

G
136: ..
.!! 120
m
-?? 134-
0
- 100

3 132- D SamekinGup p!
F! D 0 Oiff k in Gap
130- 0 W
E
I-" 128- 0 0
o 0
t-
126-
-
._ D
4 * . . g
40
5C 124-
0
D
D
m
L

+-
c 20
G
L
122- O D
0
W
0
W
5 120-t. ........,......... I...". ...................... 1
$ 0

v
 A two-dimensional analysis is needed to accurately
model the problem when the ratio of the heater length
to the gap length is less than 2.5. For heater-to-gap
ratios larger than this, it is felt that the decrease in
heater and gap temperatures is not large enough to
seriously affect the rate of ice removal. For ratios
smaller than this, the combination of increasing gap
size and decreased temperatures at the surface make
the use of a two-dimensional model necessary.
Furthermore, the gap properties should always be
modeled using the physical properties of the sur-
rounding insulation, as the use of different properties in
this gap has been shown to be a significant parameter
in determining heater temperatures. The impact of a
two-dimensional mapped code such as that devel-
oped by Masiulaniec (9) will not have a significant
effect on this ana!ysis unless the heater is in a region of
large curvature of the blade.
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288
Figure 13 shows a plot of temperature vs. time at the
ice-abrasion shield interface for four different ice thick-
nesses and for four different outer heat transfer coeffi-
cients, for a total of 16 cases. The four ice thicknesses
used were 0.03125, 0.0625, 0.125, and 0.25 inches.
The four heat transfer coefficients used were 40, 80, v
120, and 160 BTUlhr-ft2-OF.As can be seen, there is
little effect due to ice thickness or heat transfer coeffi-
cient. Similarly, the difference in heater temperature for
these cases is negligible. This shows that the solution
is not highly dependent on either ice thickness or heat
transfer coefficient, especially for cases with reason-
ably large ice thicknesses ( >0.25in.).
110, -
"
I
i 0 d-.0625.h-40
O

2. Grid Spacing

The number of nodes in both the heater and the

gap were varied to study what effect, if any, this had on
the results. The same cases that were used for deter-
mining the affect of gap size were run using from three
to twenty nodes in the heater, and from one to twenty
nodes in the gap, again using Wright's code (8) with
the AD1 scheme. For all of the cases studied, six nodes
were found to be necessary in the heater in the x- 30! . .,. . . . . . . .
I
0
I I I I I I I 1
0 2 4 6 8 10 12 14 16 18 20
direction, regardless of its size. Four nodes were nec-
essary in the gap for large gap sizes (heater length/gap
time (sec)
length < I), but as few as two nodes could be used in
the gap for large ratios (heater length/gap length > 5). Figure 13 Effect of Ice Thickness and u
In addition, it was found that the grid spacing in the Heat Transfer Coefficient
heater in the y-direction should be at least an order of
magnitude smaller than it is in the x-direction. This is Variable ice thickness cases and variable heat
especially true for small heater-to-gap ratios. This is transfer coefficient cases were also run with the stand-
understandable since the majority of the heat flow and ard deicer described previously. Figure 14 shows the
the larger temperature gradients are in the y-direction. variation in the ice-abrasion shield interface tempera-
A finer grid in regions which have high temperature ture as a function of ice thickness for heat transfer coef-
gradients is necessary to ensure an accurate solution. ficients of 40, 80, 120, and 160 BTU/hr -ft2-OF where
This result also holds for Masiulaniec's (9) two-dimen- these heaf transfer coefficients are constant for that
sional mapped system. particular run, but ice thickness is a function of x-

3. Ice Shape and Heat Transfer Coefficient

The size and shape of the ice layer can have a

significant effect on the temperature profiles in the
deicer pad, especially at the interface between the pad
and the ice. The same standard deicer case as de-
scribed previously was run using various ice thick-
nesses and various outer heat transfer coefficients in
order to determine at what point, if any, these para-
meters cease to have an impact on the results. This
becomes especially important in modeling cases which
have variable ice shapes and variable heat transfer
coefficients. It is important to determine how precise
each of these parameters needs to be measured.
These cases, however, do not contain a gap in the $ 4 0 1 . , . , . I . , . , . $ . I
heater so that the affect of these parameters on the - 0 2 4 6 8 10 12 14 16 18 20
solution could be distinguished from the affect of the x-node (20 max.)
gap size on the solution. Figure 14 Effect of Variable Ice Thickness U
on Ice-Abrasion Shield Temperature

8
 position. Figure 15 shows a similar plot, but plots the
ice-abrasion shield interface temperature as a function
of heat transfer coefficient for constant ice thicknesses
of 1/16, 1/8, and 1/4 inches. In this plot, the outer heat
transfer coefficient is a function of the x-position. These B
two plots show that the ice-abrasion shield interface n
temperature is more dependent on ice thickness than a
on heat transfer coefficient. Very little variation in this

m
temperature is noticed with a variable heat transfer
coefficient, especially for ice thicknesses greater than
0.25 inches. Also, if all the poin!s on a variable ice
shape are greater than about 0.125 inches, little vari- D h-120
0 h-160
ation in temperature is seen with a variable ice shape.
Finally, melt times for the cases presented in Figs. 14
and 15 are compared. B
5 ....,....
$ o.oo o.05 0.10 0.15 0.20 0.25
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

567
ice thickness (in)
c
I1 54 - Figure 16 Effect of Ice Thickness on Melt Times
iii 1’
52- D d=.25
10
*
-
- ’
a,
3 50-
t ’ Q C I C I . S n

.
;, ;, ; B
Q 0 Q
5 48- 1

,,I,
0
m
j
I-
“8

-
-
E

-
0,

g
46/.

44
40
8 ,

60
E,;, 0 ,

80 loo
E,;, 0 ,

120 140
, 0 ,

160
i,;
I
, , , , , , ,

Y h (BTUlhr’R * F)
Figure 15 Effect of Variable Outer Ambient
Heat Transfer Coefficient
Figure 16 shows a plot of melt time versus ice thick-
ness for the four heat transfer coefficients used in these
cases and Fig. 17 shows melt time versus heat transfer
coefficient for the four ice thicknesses run. The same
points are being plotted in these two figures. but are
plotted in two different ways for comparative purposes.
Melt time is defined as the time needed for the temper-
ature at the ice-abrasion shield interface to reach 32 OF.
Again, not much variation in temperature can be seen
due to different heat transfer coefficients. Figure 16,
however, shows an interesting phenomena at very low
ice thicknesses. A slight increase in melt time is noticed
at a heat transfer coefficient of 160 BTU /hr-ft*-OF. This
point shows that for very thin ice, heat is being removed
from the surface by convection so fast that melting is
taking longer to occur than it would for larger ice thick-
nesses. When the ice thickness becomes larger than a
certain value, however, the convective surface is so
distant that its affect on the abrasion shield tempera-
tures is less than the affect of the increasing ice thick-
ness, and the melt times start to increase again. This
suggests that there is a value of ice thickness at each
heat transfer coefficient for which the meit time is a
minimum.
U’
In conclusion, for the effect of ice thickness and heat
transfer coefficient on the solution, it is felt that the heat
transfer coefficient does not need to be variable except
CI ds.25
dn.125
6 II dm.0625
0 d1.03125
5
40 60 80 100 120 140 160
h (BTU/hr’fi * F)
Figure 17 Effect of Outer Ambient Heat
Transfer Coefficient on Melt Times
for cases which have sections of very thin ice, or no ice
at all, and that ice thickness can be modeled as a con-
stant if the ice thickness at every point is at least 0.125
inches. It should be noted that the critical ice thickesses
mentioned are dependent upon the heater power den-
sity. Smaller wattages will lower these values and high-
er wattages will raise them. Experience with using this
code reveals that if the power density is one-half this
value (15 W/in2), then the critical ice thicknesses are
likewise cut in half. This ratio seems to hold for other
power densities as well. This can be seen in the com-
parison with experimental data, which is discussed
next.
Comparison with Experimental Data
The computer program was run for comparison with
previously obtained experimental data ( 7 ) , once again
using Wright‘s (8) two-dimensional rectangular code for
the simulation. This code is used in these runs because
it can adequately run a two-dimensional case very

9
 quickly whereas the two-dimensional codes produced Where ice is present on the blade, it is modeled as
by Chao (6). Leffel (7). and Masiulaniec (9) would take having a constant value of 0.0625 inches due to the
a significantly longer time. The particular data set was previous discussion on the effect of ice shape on the
selected because it was a case of cyclic heating. The solution. Also, since the heater gap is very small in
ice shape and thickness were known from accretion
data taken at the time of the experiment. The data for
relation to the heater, a finer mesh is placed there in
order to avoid possible accuracy problems. A constant
"
the deicer pad is given in Table 2 and the various test heat transfer coefficient of 70 BTU/ft2-hr-"F is assumed
conditions for this case are given in Table 3. Fourteen to be adequate for this simulation. A constant value can
layers are modeled, which includes the ice layer. The be assumed due to the previous discussion of the affect
actual blade used had different physical properties in
the Substrate at the stagnation point ( P o s . ~ )but
, this
can not be modeled using this code, and the properties
at this position were assumed to be the same as for the Table 3
rest of the composite blade. Six heaters are modeled.
all of which are fired separately and have slightly dif-
ferent power densities. A case in which the heaters are Test Conditions for Run 305
not all fired at the same time was used to show two-
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

dimensional effects which could not be simulated

before since Chao's code contained only one heater. Heater Power Density Pos. 3 = 0.0 W/in2
Heater Power Density Pos. 4 = 15.7 W/in2
TABLE 2 Heater Power Density Pos. 5 = 15.8 W/in2
Heater Power Density Pos. 6 = 15.9 Wiin2
Physical Properties of Experimental Data Case
Heater Power Density Pos. 7 = 16.0 W/in2
Heater Power Density Pos. 8 = 16.3Wiin'
Thermal Thermal
Conductivity Diffusivity
Layer Thickness(in) (BTU/hr-R-OF) (ft2/hf) Pos. 4-6 are off the first 10 sec. of the run, on the next
__ 10 sec., and then off again for the remainder of the run.
Stainless Steel Pos. 7 is on for the first 10 sec. and off for the remainder
Abrasion Shield 0.0300 8.70 0.1500 of the run. L
Epoxy Adhesive 0.0168 0.10 0.0058
Pos. 8 is off for the first 20 sec., on for 10 sec., and off
Epoxy/Glass i 0.0138 0.22 0.0087
Insulation for the remainder of the run.

Epoxy Adhesive 0.0082 0.10 0.0058

Variable ice shape is modeled using 0.0625 in. ice
Copper 0.0065 60.00 1.1500 over Pos. 4,5, and 7 with no ice over Pos. 3,6. or 8
Heater Element
Each heater element is 1 in. wide with a 0.061 in. gap
Epoxy Adhesive 0.0082 0.10 0.0058
between heaters.
EpoxyiGlass 0.0138 0.22 0.0087
Insulation
Outer Convection Coefficient = 70. BTU/hr-f12-oF
Epoxy Adhesive 0.0082 0.10 0.0058 Inner Convection Coefficient = IO. BTU/hr-ft2-oF

Stainless Steel Ambient TemDerature = 20. OF

Blade Skin 0.0200 8.70 0.1500

of heat transfer coefficient on the solution. Figures
Epoxy Adhesive 0.0100 0.10 0.0058 18-23 plot temperature vs time at the ice-abrasion
shield interface, the bottom ofthe heater and the sub^
Aluminum 0.0500 102.00 2.8300 strate for both the experimental and numerical results.
Doubler The experimental results are shown as a solid line due
to the fact that this is the manner in which the graphics
Epoxy Adhesive 0.0100 0.10 0.0058 routine plots the data. The different positions given in
these plots are also shown in Figure 24. The three
Aluminum 0.1750 102.00 2.8300 points plotted at each position reflect the position of the
D-Spar thermocouples at each position. Beyond the six

10
 160, 160

Position 5
140 Position 3 140
d (On Second)

120
-..
Y
0
, 100
01
=-
L
:
L
80
01
.c
E
01
+ 60
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

40

20
165 225 285
0 I
150 250 350 150 250 350
TIM (Sec.) Time (Sec.)
Figure 20 Comparison of Numerical
Figure 18 Comparison of Numerical
Model to Experimental Data:
Model to Experimental Data:
Heater Position 5
Heater Position 3

a
160 , - 160 1
-
ion 4
cond)

t t-
5 225 285
1 I

150 250 350 150 250 350

Time (Sec.)
Time (Sec.)
Figure 21 Comparison of Numerical
Figure 19 Comparison of Numerical Model lo Experimental Data:
4 Model to Experimental Data: Heater Position 6
Heater Position 4

11
 Position 7
1401
(On First) v
6

Figure 24 Positions of Heaters

Around the Blade

positions given here, the experimental results agree

with Marano's (3) one-dimensional code and therefore
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

were not modeled with Wright's code (8). At Position 6 ,

and possibly Positions 5 and 7 as well, Masiulaniec's
(9) two-dimensional code could be used instead of
Wright's (8) two-dimensional rectangular code, but a

0
155
I

150
I
215
I

250
275
8 1

350
I comparison could not be obtained in time.
As can be seen from these figures, excellent agree-
ment with the experimental data is obtained at posi-
tions 5, 7, and 8. At positions where ice is present,
Time (Sec.) melting can be observed in both the numerical and
Figure 22 Comparison of Numerical experimental results by a slight change in slope which
Model to Experimental Data: can be seen in the ice-abrasion shield interface temp-
Heater Position 7 eratures.This can be seen more clearly by comparing
the response at Position 6, which has no ice, with the
response at Position 5 which does have ice. There is a
change of slope at 32 O F at Position 5 which indicates
the ice is melting. At Position 6 , the abrasion shield b
160, temperature exceeds 40 O F without any change in
slope, which indicates that ice is not present at this
Position 8 position.
The response given at Position 3 shows a large
140 (On Third)
amount of transverse heating since a significant temp-
erature response is noted even though this position
120 does not contain a heater. The numerical results also
show a response at this position, although the simu-

- lation under-predicts the experimental values. As can

I

100
be seen from the experimental results, however, the
0 temperature of the abrasion shield at Position 3 is
=
L
+J higher than the corresponding value at Position 4,
E
. 80 which does have a heater. Since the heat being re-
L
0
0. ceived at Position 3 comes from Position 4, it does not
!, 60 seem to be possible for the response to be higher at
Position 3 than at Position 4.
The maximum heater temperatures which are pre-
40 dicted at Positions 4 and 6 are also lower than the
experimental results despite the fact that the numerical
simulation predicts the substrate and abrasion shield
temperatures at these positions very well. However, the
175 235 2 program predicts the maximum heater tempeiatures at
I I the other positions adequately even though these posi-
150 250 350 tions have roughly the same power density as Posi-
Time (Sec.) tions 4 and 6. It is believed that air gaps near the heat-
Figure 23 Comparison ofNumerical
ers at Positions 4 and 6 are responsible for the differ-
ence in heater temperature. An air gap would cause
Model t o Experimental Data: the heater to obtain a much higher temperature, but u
Heater Position 8 would not significantly change the response at other
locations. This is, in fact, what is seen at these posi-

12
 tions. Position 6, however, is at the stagnation point of
ACKNOWLEDGEMENTS
the blade, and it is unknown at this point if the code
developed by Masiulaniec (9) could more accurately Numerical work on the electrothermal deicer pad has
predict the temperatures there.
been supported by the NASA Lewis Research Center,
This case is clearly two-dimensional in this region
Cleveland, Ohio. The authors gratefully acknowledge
owing to the different cycles on the heaters and to the
,. absence of ice at some points on the blade. Any case
this continuing support and the encouragement of their
work by the grant monitor, Dr. R. J. Shaw.
which has heaters firing in different cycles should be
modeled with a two-dimensional code since a one-
REFERENCES
dimensional code can not account for the transverse
heating which results from sequential firing of the
1. Stallabrass. J. R., "Thermal Aspects of Deicer
heaters. Transverse heating is also very evident at
Design", presented at the International Helicopter Icing
Position 3, which does not have a heater. Similarly, a
Conference, Ottawa, Canada, 1972.
one-dimensional code can not handle having an ice
shape such as the one present in this experimental
2. Baliga, G., "Numerical Simulation of One-
case. However, the heater gap in the experimental
Dimensional Heat Transfer in Composite Bodies with
case is not large enough to warrant a two-dimensional
Phase Change", M. Sc. Thesis, University of Toledo,
analysis, as this does not create much of a two-dimen-
Toledo, Ohio, 1980.
Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

sional effect. This can be seen from the fact that Posi-
tions 9-11 are adequately modeled using an one-
3. Marano, J. J., "Numerical Simulation of an Electro-
dimensional analysis even though these positions
thermal Deicer Pad", M S c . Thesis, University of
contain a heater gap, but display no other two-dimen-
Toledo, Toledo, Ohio, May 1982.
sional behavior.
4. Gent, R. W. and J. T. Cansdale. "One-Dimensional
CONCLUSIONS AND RECOMMENDATIONS
Treatment of Thermal Transients in Electrically Deiced
Helicopter Rotor Blades", RAE Technical Report 80159.
The computer codes developed here for the trans-
1980.
ient heat transfer in an electrothermal deicer pad have
been shown to predict accurate results for one-dimen-
5. Roelke. R. J., "A Rapid Computational Procedure for
sional rectangular, two-dimensional rectangular, and the Numerical Solution of a Heat Flow Problem with
two-dimensional mapped systems. Parametric studies Phase Change", M. Sc. Thesis, University of Toledo,
were performed to investigate the effect of gap width, Toledo, Ohio, August 1986.
nodal spacing, ice thickness, and outer heat transfer
coefficient. An analysis was performed to determine 6. Chao. D. F., "Numerical Simulation of Two Dimen-
4
when a two-dimensional rectangular model is preferred sional Heat Transfer in Composite Bodies with Appli-
over a one-dimensional rectangular model and when a cation to Deicing of Aircraft Components", PhD Thesis,
two-dimensional mapped model is preferred over a University of Toledo, Toledo, Ohio, Nov. 1983.
two-dimensional rectangular model. Then, a compar-
ison was made with existing experimental data. The
7. Leffel, K. L., "A Numerical and Experimental Investi-
results of the numerical simulation were found to gation of Electrothermal Aircraft Deicing", M. Sc. The-
compare very well with previous numerical calcu-
sis, University of Toledo, Toledo, Ohio, Jan. 1986.
lations, as well as with the experimental data.
An additional comparison was made with experi-
mental data supplied by the French Research Organ-
ization (ONERA). The results were found to agree very
well with this data and with those obtained via codes
8. Wright, W. B. "A Comparison of Numerical Methods
developed by ONERA and the British Royal Aircraft
for the Prediction of Two-Dimensional Heat Transfer in
Establishment (RAE).
an Electrothermal Deicer P a d , M. SC. Thesis, Univer-
The analysis performed shows that a two-dimen-
sity of Toledo, Toledo, Ohio, Dec. 1987.
sional rectangular model is preferred over a one-
dimensional rectangular model when the parameters
9. Masiulaniec, K. C. "A Numerical Simulation of the
discussed in the results section exceed certain values
Full Two-Dimensional Electrothermal Deicer Pad", PhD
or when the computational grid is small enough so that
the extra computer time used by the two-dimensional Thesis, University of Toledo, Toledo, Ohio, June 1907.
model is minimal. A two-dimensional mapped analysis
I O . Bonacina, C.. G. Comini. A. Fasano, and N. Primi-
is preferred when the body curvature of the blade is too
great for a rectangular model to provide adequate cerio, "Numerical Solution of Phase Change Prob-
lems", Int. J. Heat& Mass Trans.,vol. 10, p.1825, 1973.
results. provided the computational time expended on
the two-dimensional mapped simulation is not too
11. Voller, V. and M. Cross, "Accurate Solutions of
much.
Moving Boundary Value Problems Using the Enthalpy
Method", Int. J. Heat & Mass Trans., vol. 24, p.545,
J 1981.
 12. Voller, V.,M. Cross and P. G. Walton, "Assessment
of Weak Solution Numerical Techniques for Solving
Stefan Problems', 172, Dept. of Mathematics &
Computer Studies, Sunderland Polytechnic, U. K.,
1979.

13. Atthey. D. R., "A Finite Difference Scheme for Melt-

ing Problems", J. Inst. Math. Appl., vol. 13, p. 353, 1974.

14. Schneider, G. E. and M. J. Raw, "An Implicit Solu-

tion Procedure for Finite Difference Modeling of the
Stefan Problem", AlAA JournalVol. 22, Nov. 1984.

14. Carslaw, H. S. and J. C. Jaeger, Conduction of

Heat in Solids, Clarendon Press, Oxford, 1959.

15. Anderson, D.A,, J. C. Tannehill and R. H. Pletcher,

Downloaded by NASA GLENN RESEARCH CENTER on October 22, 2014 | http://arc.aiaa.org | DOI: 10.2514/6.1988-288

Computational Fluid Dynamics and Heat Transfer,

McGraw-Hill. New York, 1984.

14

View publication stats