Energy 92 (2015) 87e99

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

A numerical study of air preheater leakage

Avinash Maharaj a, *, Walter Schmitz b, Reshendren Naidoo b
a
Eskom Holdings SOC Ltd., Enterprises Park, Building 1, Boiler Engineering, 1 Simba Road, Sunninghill, Johannesburg, 2191, South Africa
b
University of the Witwatersrand, School of Mechanical, Industrial and Aeronautical Engineering, North West Engineering Building, 1 Jan Smuts Avenue,
Braamfontein, Johannesburg, 2000, South Africa

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

Article history: The purpose of the research is to quantify direct leakage for a Ljungstro€ m air preheater. The leak flow
Received 12 December 2014 path was assumed to be similar to flow through an orifice and a theoretical approach was used to
Received in revised form calculate leakage. A 2D CFD model with a geometry similar to the one considered in the theoretical
18 June 2015
approach was also developed to calculate leakage. It was noted that for the different leak gaps investi-
Accepted 19 June 2015
gated, the theoretical calculated leakage was always lower than the CFD calculated leakage. This can be
Available online 16 July 2015
attributed to the fact that the leak gap geometry and actual orifice flow geometry are not identical. Using
the 2D CFD model, the leak flow was quantified for various seal settings across the operating air tem-
Keywords:
Ljungstro€m air preheater perature range.
Direct leakage An actual air preheater CFD model was also developed. This model was unable to accurately calculate
Orifice the hot end leak due to inaccurate temperature calculation in the hot end leak gap. After applying a
correction factor to the hot end leak, it was possible to determine the direct leakage.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction
Air preheaters are found in most steam generating plants to heat
combustion air and enhance the combustion process. In most ap-
plications, flue gas serves as the energy source and the air preheater
is considered to be a heat trap which collects and transfers heat
from the flue gas to incoming air. This process can increase the
overall boiler efficiency by 5%e10% [1]. Air preheaters are usually
found directly behind the boiler as shown in Fig. 1. In a typical coal
fired power plant, the air preheater will receive flue gas from the
economizer and combustion air from the FD (Forced Draught) fans.
The hot air produced by the air preheater enhances combustion and
is required for drying and transportation of fuel in pulverized coal
fired boilers [2].
Air preheaters have a general operating fault known as leakage [3].
Leakage occurs when part of one gas stream flowing through a sealing
system leaks into the other gas stream. The quantity of leakage is
* Corresponding author. Tel.: þ27 11 800 8285, þ27 83 999 4291.
E-mail addresses: avinash.maharaj@eskom.co.za (A. Maharaj), walter.schmitz@
wits.ac.za (W. Schmitz), reshendren.naidoo@wits.ac.za (R. Naidoo).
dependent on seal clearance area and static pressure difference be-
tween the air and flue gas streams flowing through the air preheater.
€m air preheater
1.1. Ljungstro
A regenerative heat exchanger transfers heat indirectly by
convection as a heat storage medium is periodically exposed to hot
and cold flow streams. Ljungstro €m is the most common type of
regenerative air preheaters', it is commonly referred to as a rotating
matrix air preheater. Fig. 2 shows a typical arrangement drawing of
€m air preheater.
a Ljungstro
The air preheater consists of a cylindrical shell and rotor which
is packed with bundles of heating surface elements (typically
corrugated and undulated plates) which is rotated between counter
flowing air and flue gas streams. Bearings in the upper and lower
support beams guide the rotor at the central shaft. A typical rotor
speed of one to three revolutions per minute is achieved by a motor
driven pinion engaging a rotor encircling pin rack. The shaft can
either be aligned horizontally or vertically, however the vertical
arrangement is more common [2].
In the vertical shaft arrangement as shown in Fig. 2, hot flue
gas enters from the top and cold air enters from the bottom. In

http://dx.doi.org/10.1016/j.energy.2015.06.069
0360-5442/© 2015 Elsevier Ltd. All rights reserved.
88 A. Maharaj et al. / Energy 92 (2015) 87e99

List of symbols Re Reynolds number

T temperature,  C
b orifice diameter ratio t time, s
m dynamic viscosity, Pa$s !
V velocity field
r density, kg/m3 v velocity, m/s
A cross sectional area, m2 X correction factor
Cd orifice coefficient Z number of radial seals under the sector plate
Dh hydraulic diameter, m hO2 im_ oxygen content weighted by mass flow, %
f Fanning friction factor
h specific enthalpy, kJ/kg Subscripts
kexp expansibility factor APH air preheater
L characteristic length, m fg flue gas
m_ mass flow rate, kg/s in inlet
p pressure, Pa out outlet
q heat transfer rate, W

such a flow arrangement, the hot end is at the top, and the cold
end is at the bottom. During operation, the rotor experiences a
temperature differential between the hot and cold ends causing
the rotor to expand and distort. This distortion creates gaps
between moving and stationary parts allowing air to leak into
the flue gas stream. The leak gap exists even without rotor
distortion however under operating conditions the gap in-
creases. Air to flue gas leakage can be minimised by cold pre-
setting axial and radial seal plates to minimise gaps in the hot
operating condition [2].
1.2. Leakage
Air flow leaking from the air stream to the flue gas stream is
referred to as leakage. It can be reported in kilograms per seconds
(kg/s), but can also be expressed as a percentage of the flue gas inlet
flow. Leakage is undesirable because it represents fan power which
has been wasted in transporting air which bypasses the combustion
zone. Leakage also has the potential to reduce an air preheater's
thermal performance.
For a Ljungstro€ m air preheater, leakage can be categorized as
direct and entrained. Direct leakage occurs when higher pressure
air leaks into the lower pressure flue gas stream through gaps be-
tween the rotating and stationary parts.
This leakage path is usually formed when the radial seals pass
under the radial sealing plate or sector plate in a Ljungstro €m air
preheater. Fig. 3 shows the typical sealing system of a Ljungstro €m
air preheater.
Entrained leakage can be described as being a result of the
rotation of the matrix from one stream to the next [4]. Air is carried
into the flue gas stream as the heating surface components or
baskets are rotated from the air stream to the flue gas stream. This

Fig. 1. A typical coal fired boiler showing the position of the air preheater [2]. €m type air preheater [2].
Fig. 2. Vertical shaft Ljungstro
 A. Maharaj et al. / Energy 92 (2015) 87e99 89

 A lower flow rate of combustion air requiring additional auxil-

iary power consumption since the FD fan is required to supply
more air.
 Dilution and cooling of flue gas with cold incoming air.
 Overloading of the FD, PA (Primary Air) and ID (Induced
Draught) fans.

The previous author has suggested introduction of an automatic

€m air preheater sealing system [2].
Fig. 3. Ljungstro
leakage is directly proportional to the void volume of the rotor and
the motor speed.
The design leakage for regenerative air preheaters varies from
5% to 15% but increases with time because the sealing system de-
teriorates. Air leakage can be calculated directly as the difference in
flow between inlet and outlet ducts for either the flue gas or air
stream. However, it is difficult to obtain accurate velocity mea-
surements in large ducts as is the case in a power plant. Leakage can
be more accurately calculated based on oxygen content measured
in the flue gas stream at the inlet and outlet of the air preheater [2].
leakage control system and major redesign of the sealing system
with emphasis on the radial seals [5].
A model was developed to simulate the operation of a rotary air
preheater including the effect of various seal settings on flue gas
parameters [1]. A typical arrangement of seals for a rotary air pre-
heater is summarized and shown in Fig. 4.
Peripheral seals prevent the bypass flow of air or flue gas around
the matrix. This air does not contribute to heat exchange and must
also be reduced. The leakage flow rate through the peripheral seals
are smaller compared to other leak paths due to the fact that the
pressure gradients in the axial direction are much smaller than the
pressure difference between hot and cold streams. In the geometry
required for the model, only the radial seal settings were varied [1].
The mass flow through the radial seals was determined using
pressure differences between the air inlet and flue gas outlet ducts
(i.e. hot and cold streams).
The results of the CFD (computational fluid dynamics) model
produced distributions of the flue gas velocity, temperature and
carbon dioxide (CO2) concentration over the computational domain
which consisted of the flue gas ducting. When seal clearances were
increased, more air leaked into the flue gas stream resulting in a
decrease of temperature and CO2 content at the inner walls of the
flue gas duct.
It was possible to determine whether leakage increased at the
left or right hand side of the rotor, but it was not possible to
establish the leakage through the seals at the hot end or cold end.
The model confirmed the following assumptions; increased
leakage at the cold end has a significant impact on flue gas tem-
perature and increased leakage at the hot end has little impact on
the flue gas temperature.

1.3. Literature review

South African research into air preheaters has focused on the

effects of fouling and erosion effects on the thermal performance of
the heating elements. The reason for this is that most South African
air preheaters have been designed in Europe for power plants
which burn coals with a lower ash content but higher Sulphur
content than South African coals. Therefore the intent of the pre-
vious research has been to improve air preheater performance by
considering the effects of low grade coals as used in South African
power plants. A RAH (Regenerative Air Heater) simulation program
was developed [4]. The program was able to calculate the tem-
perature distribution within the rotating matrix, the fluid outlet
temperature and the pressure drop across the heating elements.
The program was unable to calculate air leakage but was able to
simulate the effects of leakage and blockage on the performance of
the air preheater. The simulation program requires the air leakage
rate as an input.
A previous author who performed an air preheater leakage test Fig. 4. Arrangement of seals. A e axial seals, P e peripheral seals, R e radial sealing
concluded that leakage causes the following problems [5]: plate or sector plate [1].
90 A. Maharaj et al. / Energy 92 (2015) 87e99

A comprehensive modelling tool for leakages and thermal per-
formance of rotary air preheaters was developed [6]. The models
use experimental correlations for Nusselt number, friction factor
and seal discharge coefficients. The results showed that even small
leakages of 5% can lead to significant reduction in thermal energy
transfer to the cold fluid.
The leakage mass flow rate depends on matrix rotation, seal
clearance and pressure drop across the seals. The leak flow through
the seals was modelled as orifices and discharge coefficients must
be known before modelling these regions [6]. Fig. 5 shows typically
what flow path the leak air would take at different regions. A 5%
leakage rate occurring through all leakage paths (radial, axial and
peripheral directions), caused a double reduction in the air stream
thermal energy when compared to a scenario with 0% leakage [6].
A methodology was presented to determine quantitatively the
influence of leakages on an air preheater heat transfer performance

€m air preheater with radial, axial and peripheral seals [6].

Fig. 5. A Ljungstro
 A. Maharaj et al. / Energy 92 (2015) 87e99 91

[7]. The results showed that leakages can cause a significant drop in
the effectiveness of an air preheater. A drop in effectiveness de-
pends almost linearly on the quantity of direct leakage. For the
combined case with all leakages (radial, axial and peripheral)
considered occurring simultaneously, for every 1% of leakage on the
cold end, the drop in air preheater effectiveness was approximately
7% [7].

1.4. Objective of the present study

Thermal design theory of air preheaters is well established;

however the influence of leakages is omitted or incorporated in a
simplified manner to make the problem manageable [6]. In most
models, the leak gap is modelled as an orifice; however authors do
not explicitly mention the procedure to calculate the discharge
coefficient which is highly dependent on specific geometry of the
air preheater being investigated. The objective of this research is to
€ m air preheater.
quantify air preheater direct leakage for a Ljungstro
2. Calculation and CFD simulation
The theoretical leak rate for the air preheater radial seal gap was
determined using an orifice flow analogy. A 2D CFD model was also
prepared to compare the CFD calculated leak rate with the theo-
retical leak rate. A CFD model was also prepared for the actual
computational domain of a Ljungstro €m air preheater to determine
the leak rate through the radial seal gap.
2.1. Theoretical air leak calculation
Direct leakage occurs when higher pressure air leaks into the
lower pressure flue gas through gaps between the rotating and
stationary parts. This leakage path is formed when the radial seals
pass under the sector plate. Fig. 6 shows the direct leakage path.
The leak gap dimensions have been exaggerated in the figure in
order to make the gap clear to visualize.
Equation (1) has been proposed to calculate leakage at the radial
seals if the flow between the sector plate and the radial seals is
considered to be similar to flow through an orifice [2,8e10].
Fig. 7 shows a side view of the air leaking between the radial seal
and sector plate. The higher pressure side is point 1 and the lower
pressure side is point 2. For the analysis of the air preheater, point 1
is equivalent to the air inlet stream and point 2 is equivalent to the
flue gas outlet stream.
Fig. 7. Air leak gap between radial seal and sector plate.
Equation (1) includes an orifice coefficient (Cd) which accounts
for the expansion effect of air as it passes through the leak gap. An
orifice coefficient of 0.65 has been recommended for most air
preheater applications [8]. A previous author performed labora-
tory tests and CFD simulations to calculate air leak mass flow rate
for the geometry given in Fig. 7 [8]. It was noted that significant
discrepancies exist when the flow is assumed to be incompress-
ible. The expansibility factor (kexp) for a non-viscous and
compressible ideal gas has been suggested to also account for the
expanding flow that will always happen as air passes through the
seal gaps.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m_ leak ¼ kexp Cd Agap 2ðp1  p2 Þr=Z (1)
where:
m_ leak ¼ Leak mass flow (kg/s)
kexp ¼ Expansibility factor
Cd ¼ Orifice coefficient
Agap ¼ Cross sectional area of leak gap (m2)
p1 ¼ Static pressure for air inlet stream (Pa)
p2 ¼ Static pressure for flue gas outlet stream (Pa)
r ¼ Density (kg/m3)
Z ¼ Number of radial seals under the sector plate
The geometry in Fig. 8 shows the leak gap and the throat region
(t) of the orifice which will be used in the analysis to calculate the
orifice coefficient. For the air preheater under investigation, typical
values for the air stream are given in Table 1.

Fig. 6. Side view of air preheater showing direct leakage path through gap between
radial seal and sector plate. Fig. 8. Leak gap geometry simplified to a typical orifice.
92 A. Maharaj et al. / Energy 92 (2015) 87e99
Table 1
Typical air preheater values for the air inlet stream.
Units
pair,in Pa
Tair,in K 306.89
mair,in Pa$s
vair,in m/s
rair,in kg/m3
The geometry in Fig. 8 is modelled as a concentric orifice with
corner pressure taps. Equation (2) can be used to determine the
orifice coefficient (Cd) [11].
Equation (2) predicts orifice coefficients within ±0.6 percent for
104 < ReDh;1 < 107 . Using Equation (2), a value of 0.596 is calculated
for the orifice coefficient.
91:71b2:5
Cd ¼ 0:5959 þ 0:0312b2:1  0:184b8 þ
Re0:75
Dh;1
where:
b ¼ Orifice diameter ratio
Re ¼ Reynolds number
Dh,1 ¼ Hydraulic diameter for cross section 1 in Fig. 8 (m)
2.2. CFD model of leak gap
To investigate leak flow further, a 2D CFD model was prepared
using STAR-CCMþ Version 8.02. A two dimensional view of the
computational domain is shown in Fig. 9. The geometry is similar to
the geometry used by a previous author to study air preheater
direct leakage numerically [8]. It was observed that recirculation
occurred immediately after the fluid passes the leak gap. To account
for the recirculation, the downstream length is elongated and ends
with a converging section to accelerate the flow out of the domain.
This is done in accordance with best practice guidelines to avoid
reversed flow at the outlet.
With reference to Fig. 9, the inlet was defined as a stagnation
inlet boundary (total pressure) and the outlet was defined as a
pressure outlet boundary (static pressure). In defining two pressure
boundaries, a pressure loss of 3.83 kPa can be achieved. This is the
typical pressure loss across the radial seal for the air preheater
being simulated. This pressure loss is defined as the difference in
static pressure between the air inlet and flue gas outlet streams.
Pressure loss will vary for different air preheaters depending on
rotating matrix construction and heating element type.
A typical value for the air temperature at the air preheater inlet
and outlet is 35  C (cold end) and 290  C (hot end) respectively.
Simulations were performed to calculate the air leak mass flow rate
Fig. 9. Computational domain of 2D CFD model.
for air temperatures of 30  C, 150  C and 300  C at seal settings of
15 mm, 20 mm, 25 mm and 30 mm. This temperature range en-
Value compasses the air temperature in the leak gap at both cold and hot
1978
end sides of the air preheater.
Figure A1 shows the volume mesh which was generated using
1.87E-05 polyhedral cells. Volumetric control was utilized to achieve a
5.42 smaller cell base size in the leak region around the radial seal and in
0.9703
the leak gap. The mesh size then grows rapidly to a larger base size
in all other regions. Figure A2 shows an enlarged image of the
volume mesh generated in the leak gap. Four prim layers with a
thickness of 2 mm were generated in the wall regions to capture the
effects of the viscous sublayer.
In order to ensure that the simulation results are independent of
the mesh, a mesh sensitivity study was performed. The study is
performed by increasing the number of cells such that a solution is
achieved such that no significant changes in the results occur [12].
Such a solution indicates the discretization error is reduced to an
acceptable error and mesh independence is reached.
(2)
A mesh sensitivity study was performed by reducing the base
size of the cells in the leak gap. The final cell size for the volumetric
control was 3 mm in the leak gap and 30 mm for all other cells. The
results of the mesh sensitivity analysis are shown in Table 2. The
change in leak mass flow rate between Run 4 and Run 1 is less than
5% and therefore the mesh setup in Run 4 was considered to be
acceptable.
The physics models used to define the flow are given in Table 3.
There is no heat transfer to the surrounding environment, i.e.
adiabatic process.
2.3. CFD model of actual air preheater
Fig. 10 shows the computational domain prepared for the air
preheater under investigation which is composed of the rotating
matrix, direct leak gap for both cold and hot ends, as well as the
ducting layout as constructed on-site. A leak gap of 20 mm as
measured during a unit outage was used for both cold and hot end
leak gaps. The computational domain was created with converging
regions for the air and flue gas outlet ducts. The reason for this is
that the outlet boundaries were defined as pressure outlet
boundaries which allow recirculation to occur. Therefore the
converging section is added to the outlet ducts to accelerate the
flow out of the computational domain in order to prevent conver-
gence problems caused by unsteady flow [13]. The inlet boundaries
are defined as mass flow inlets.
The physics models chosen are used to define the primary var-
iables of the simulation such as pressure, temperature and velocity
as well as the mathematical formulation used to generate the so-
lution [13]. In actual operation, the rotating matrix of the air pre-
heater is in a constant state of rotation. However in this simulation
the flow is considered as steady state. Steady state flow indicates
that properties at every point in a flow field do not change with
 A. Maharaj et al. / Energy 92 (2015) 87e99
Table 2
Mesh sensitivity analysis for 2D CFD model.
Run Total number of cells
1 21,701
2 36,074
3 67,157
4 468,979
Table 3
Physics models.
Time Steady
Material Single-component material
Flow Segregated flow
Equation of state Ideal gas
Viscous regime Turbulent
Reynolds-Averaged Turbulence K-Epsilon
Wall treatment model All yþ wall treatment model
time (t). Stated mathematically, the definition of steady flow with
!
respect to density (r) and the velocity field ð V Þ is as follows; vr/
! The air preheater being simulated utilises a 278 DU (Double
vt ¼ 0 and vvtV ¼ 0. In steady state flow, any property may vary from
point to point in the flow field, however all properties remain
constant with time at every point [11].
The viscous regime of the flow is defined as being turbulent. The
ideal gas model was chosen to calculate density as a function of
temperature and pressure. The segregated flow model was
selected. This model solves the flow equations (one for each
component of velocity, and one for pressure) in a segregated or
uncoupled approach. Although this model has its foundations in
constant density flows it can be used for mildly compressible flows.
The segregated flow model uses less memory than the coupled flow
model; however the number of iterations required increase with
mesh size [13].
The approach of modelling turbulence was based on selecting a
model that will provide closure of the RANS (Reynolds Averaged
Navier Stokes) equations. The challenge with regards to turbulence
modelling is to model the Reynolds stress tensor in terms of mean
flow quantities and in doing so provide closure of the governing
equations [13]. The K-Epsilon turbulence model was chosen. This
models' main advantage is that it offers a good compromise be-
tween robustness, computational cost and accuracy. The model is
well suited for industrial type applications that contain recircula-
tion with or without heat transfer.
A wall treatment model needs to be chosen in conjunction
with the turbulence model. yþ is defined as being a unitless
distance away from the duct wall. The main advantage of using a
Fig. 10. Computational domain with actual ducting layout.
93
wall model is the significant saving in near wall mesh resolution
which is required to resolve the viscous affected regions of the
m_ leak (kg/s) wall. The all yþ wall treatment model is chosen and is valid when
6.46
yþ is very low or when yþ is greater than 30. This model is a
6.56 hybrid treatment that attempts to imitate the high yþ treatment
6.56 for regions with coarse meshes, and the low yþ treatment for
6.22 regions with fine meshes [13].
In the air preheater simulation, the model has a variety of mesh
resolutions. The ducts have coarse meshes whereas the leak gaps
have fine meshes. Therefore the all yþ treatment model is chosen.
The Multi Component Material Model is chosen to define the
composition of flue gas on a volume basis. Table 4 gives the typical
flue gas composition from coal combustion [2].
The rotating matrix was modelled as a porous medium. This can
be done since the rotating matrix is only part of a more substantial
flow domain including flow through ducts and the leak gap. The
flow between the heating element plates is not of interest, rather
the main interest is the macroscopic effect of the porous medium
on the overall air and flue gas flows. A friction factor correlation was
used to derive the inertial and viscous resistances which are
required when defining the porous resistance in the CFD model [4].
Undulating) 0.8 mm thick heating element plate. For the heating
surface elements being modelled, the friction factor correlation is
given as f ¼ 1:1835ðReair;in Þ0:446 [4]. Where f is the Fanning fric-
tion factor and Reair,in is the Reynolds number at the air inlet side.
Equation (3) can be used to determine the pressure drop across the
rotating matrix for a range of air velocities.
 2
2f vair;in rair;in Lrotating matrix
Dpair ¼ (3)
Dh
where:
Dp ¼ Pressure loss (Pa)
vair,in ¼ Air velocity at inlet (m/s)
Lrotating matrix ¼ Rotating matrix depth (m)
Dh ¼ Hydraulic diameter of heating elements from manufac-
turer (m)
Fig. 11 shows a graph of Fanning friction factor across a large
Reynolds number range for the air inlet stream. The operating
Reynolds number is used to determine the Fanning friction factor.
The friction factor is used in Equation (3) to generate a curve of
pressure drop per unit meter length across a range of air velocities
shown in Fig. 12.
The curve is in the form of y ¼ ax2 þ bx, where a and b are the
inertial and viscous resistances respectively. The units for the in-
ertial and viscous resistances are kg/m4 and kg/m3$s respectively.
Since the main region of interest is the leakage mass flow rate,
the mesh size in the leak gap is varied from coarse to fine and the
leak flow rate is calculated. The mesh size in the gap is varied by
increasing the number of prismatic cells in the leak gap. The base
size of the polyhedral mesh in the ducts and rotating matrix was
Table 4
Typical volume composition of flue gas.
% v/v V (volume fraction)
N2 74.86 0.7486
O2 3.28 0.0328
CO2 13.97 0.1397
SO2 0.08 0.0008
H2O 7.81 0.0781
94
Fig. 11. Fanning friction factor versus Reynolds number for the air inlet stream.
Fig. 12. 2nd order polynomial curve to determine inertial and viscous resistances.
chosen to provide adequate resolution for the bulk flow of air and
flue gas.
The mesh size in the leak gap is required to be of adequate
resolution in order to accurately calculate the velocity profile in the
leak gap. Figure A3 shows the leak region with prismatic thin layers
used to generate the mesh in this region. The number of layers in
this gap is varied during the mesh sensitivity study.
The mesh sensitivity analysis was performed when the rotating
matrix is in such a position that there is only one radial seal under
the sector plate (i.e. Z ¼ 1). Fig. 13 shows that for Run 4 with 10
layers in the leak gap, the mass flow rate is no longer significantly
changing. For the mesh study, both air and flue gas streams were
considered to be a single-component material.
An enthalpy source option was used to introduce or remove a
specified amount of heat flow from the rotating matrix. The
enthalpy source option is only associated to the fluid stream to
which it is assigned and is a net source or sink within the model.
Fig. 13. Effect of increasing thin mesher layers in leak gap for different runs on the calculated leakage mass flow rate.
A. Maharaj et al. / Energy 92 (2015) 87e99
The rotating matrix was split into three distinct regions as shown in
Fig. 14; a region in contact with the flue gas stream (heat sink), a
region in contact with the air stream (heat source) and the region
under the sector plate (no enthalpy source option). Heat sink and
source values of 25 MW and 24 MW were used respectively for the
rotating matrix flue gas and air regions. These values were cali-
brated using sub-models [14].
The enthalpy source option is defined by the heat transfer rate
equation given in Equation (4).
_ in  hout Þ
q ¼ mðh (4)
where:
q ¼ Heat transfer rate (W)
m_ ¼ Mass flow rate (kg/s)
h ¼ Specific enthalpy (kJ/kg)
From Equation (4) it can be seen that the enthalpy source option
can only be applied to a region in the rotating matrix which is
subjected to flow. Since the region of the rotating matrix under the
sector plate is a dead zone, i.e. not exposed to a flow, the enthalpy
source option could not be applied. This resulted in an inaccurate
temperature calculation in the hot end leak gap, resulting in an
inaccurate hot end leak flow rate. Fig. 15 shows that the tempera-
ture in the hot end leak gap is similar to the cold end leak gap,
however it is expected that the temperature should be closer to the
air outlet temperature.
Fig. 16 shows the measuring planes created in the leak gaps to
measure the CFD calculated leakage. A measuring plane is created
normal to the air leak flow in the leak gap between the radial seal
and the sector plate.
3. Simulation results and discussion
3.1. Comparison of 2D leak gap CFD model and theoretical
calculation results
Fig. 17 compares the CFD and theoretical calculations for varying
seal settings across the air stream operating temperature range. The
theoretical calculations are shown as solid lines and the CFD cal-
culations are shown as dashed lines.
Fig. 17 shows that the CFD and theoretical calculation for leak
mass flow match closer for smaller leak gaps than for larger leak
gaps. For the 15 mm gap at an air temperature of 150  C, the
theoretical leak is 4.62 kg/s and the leak mass flow calculated from
CFD is 5.22 kg/s, resulting in a difference of 0.6 kg/s (13%). For the
30 mm gap at an air temperature of 150  C, the theoretical leak is
9.26 kg/s and the leak mass flow calculated from CFD is 10.73 kg/s,
resulting in a difference of 1.47 kg/s (16%).
 A. Maharaj et al. / Energy 92 (2015) 87e99 95

Fig. 14. Heat sink and source regions defined for the rotating matrix.

Fig. 15. Sectional view through rotating matrix.
It can be noted that for all leak gap settings investigated, the
theoretical air leak mass flow rate is always smaller than the CFD
calculation. When using Equation (1), an assumption is made that
the leak gap is similar to an orifice. Recall that an orifice plate is a
thin plate that is usually clamped between pipe flanges. Fig. 18
compares the radial seal leak gap geometry and actual orifice
plate geometry.
Fig. 18 shows that the flows are topologically not the same. The
flow through the orifice is symmetrical along the flow axis whereas
the flow through the radial seal gap is not. The non-identical ge-
ometries can contribute to an error in the determination of the
orifice coefficient and therefore have an effect on the theoretical
calculation of leak mass flow.
Fig. 19 is a cross sectional velocity plot for a 15 mm leak gap with
an air temperature of 30  C. The CFD simulation calculates an
average velocity of 65.28 m/s in the leak gap. As expected, the high
velocity in the leak gap causes a jetting effect immediately down-
stream of the radial seal causing the air velocity to reach approxi-
mately 90 m/s in the most extreme velocity regions. Velocity
vectors which are shown as black arrows show the recirculation
which is expected to occur immediately downstream of the radial
seal.
Fig. 20 shows that a higher air velocity is reached in the leak gap
at higher air temperatures. For a leak gap of 15 mm, an air tem-
perature of 30  C yields an air velocity of 65.28 m/s in the leak gap
whereas an air temperature of 300  C yields an air velocity of
89.12 m/s. This deviation results in a difference of 23.84 m/s (37%).

Fig. 16. Measurement planes in leak gaps.

96 A. Maharaj et al. / Energy 92 (2015) 87e99
Fig. 17. Characterisation of the leak mass flow for varying seal settings across the
operating air temperature range.
Fig. 21 shows the influence of differential pressure on air leak
mass flow rate. The study was performed on geometry with a leak
gap of 15 mm with an air temperature of 30  C.
The typical differential pressure experienced between the air
inlet and flue gas outlet streams for the air preheater under
investigation is 3.83 kPa which results in a leak mass flow rate of
6.22 kg/s. The effect of differential pressure on mass flow rate in the
leak gap is investigated across a range from 2.3 to 5.3 kPa in 500 Pa
intervals. For a differential pressure of 2.3 kPa, a mass flow rate of
4.81 kg/s is calculated, resulting in a 29% difference from typical
conditions. For a differential pressure of 5.3 kPa, a mass flow rate of
7.38 kg/s is calculated, resulting in a 19% difference from typical
conditions. Fig. 21 shows that the deviation of mass flow rate from
the typical value is more prevalent for lower differential pressures
than for larger differential pressures.
3.1.1. Discussion
Using a simple leak gap geometry, the air leak mass flow rate
could be calculated at different seal settings across the expected
operating air temperature range.
The theoretical calculated values for air leak flow matched the
CFD calculated values closer for smaller leak gaps than for larger
leak gaps. This indicates that the orifice coefficient is better pre-
dicted for smaller leak gaps than for larger gaps.
A velocity plot in the leak gap region and in the immediate
region surrounding the radial seal shows that the air velocity
reaches an average velocity of 65.28 m/s. A jetting effect can be
visualised immediately downstream of the radial seal as well as
recirculation.
Fig. 18. Comparison of leak and orifice plate geometry.
For the 15 mm gap using the CFD calculated values, Fig. 17 shows
that the air leak at the cold end (30  C) is 6.22 kg/s and at the hot
end (300  C) is 4.49 kg/s. Therefore the total direct leakage is
10.71 kg/s. This shows that the hot end leakage accounts for 42% of
the total leakage whereas the cold end leakage accounts for 58% of
the total leakage. This confirms that the cold end leakage accounts
for the larger portion of the total direct leakage.
For a 15 mm gap using the theoretical approach, Fig. 17 shows
that the leak at the cold end is 5.46 kg/s and the leak at the hot end
is 3.97 kg/s. Therefore the total direct leakage is 9.43 kg/s. The total
direct leakage calculated using the theoretical approach is 13.57%
lower than the value calculated using CFD.
It is important to note that these investigations were done for a
uniform leak gap. However, in operating conditions, the leak gap
changes due to thermal distortions of the rotor.
3.2. Actual air preheater CFD model results
Using the measurement planes created in the hot end and cold
end gaps, the CFD calculated leakage rates were measured and are
given in Table 5.
Table 5 shows that a hot end leak of 6.52 kg/s has been calcu-
lated. However this mass flow is inaccurate since the temperature
in the hot end leak gap is incorrect. A correction factor can be
applied using Equation (1). The CFD model calculates a pressure
difference between the air outlet stream and flue gas inlet stream of
2677.25 Pa (p1p2), and an average temperature in the hot end leak
gap of 55  C. Substituting these values into Equation (1), a value of
6.45 kg/s is calculated ðm_ 55 C Þ.
The CFD calculated air outlet temperature is 265  C, substituting
this temperature and a pressure drop of 2677.25 Pa into Equation
(1), a value of 5.04 kg/s is calculated ðm_ 265 C Þ. A correction factor (X)
can now be calculated as follows:
m_ 55 C
X¼ ¼ 1:28
m_ 265 C
Therefore the corrected CFD hot end leak can be calculated as
follows:
m_ leak;hot end 6:52
m_ leak;hot end;corrected ¼ ¼ ¼ 5:09 kg=s
X 1:28
The total leak flow calculated by the CFD model is therefore now
calculated as follows:
m_ leak;total;corrected ¼ m_ leak;cold end þ m_ leak;hot end;corrected
¼ 7:14 þ 5:09 ¼ 12:23 kg=s
 A. Maharaj et al. / Energy 92 (2015) 87e99
Fig. 19. Vector plot showing air velocity in leak gap and recirculation.
Fig. 20. Characterisation of the air velocity in the leak gap across the operating air
temperature range for varying seal settings.
Fig. 21. Influence of differential pressure on air leak mass flow rate for 15 mm gap.
3.2.1. Discussion
The total direct leakage calculated by the actual air preheater
model can be compared to the results from the 2D CFD model. From
Fig. 10, the CFD calculated cold and hot end leakages for a 20 mm
leak gap is 8.35 kg/s and 6.05 kg/s respectively. Therefore the 2D
CFD model, calculates a total direct leakage of 14.40 kg/s compared
to the actual air preheater model leakage of 12.23 kg/s. The actual
air preheater model calculates a leakage which is 15% lower than
the leakage calculated by the 2D CFD model. When comparing
calculated leakage for the simplified 2D model and actual air
97
preheater model, it is noted that ducting configuration has an effect
on the calculated leakage.
It must be noted that the actual air preheater model does not
account for thermal distortion and assumed a constant leak gap for
both the hot and cold ends.
4. On-site measurements to determine leakage
On-site measurements were performed for the air preheater
under investigation with the intention of measuring air pre-
heater leakage [14]. Flue gas inlet measurement points were
available. These points were located a distance of 1.4 m away
from the air preheater. However, air preheater flue gas outlet
measurement locations were unavailable. The next available
measurement location for flue gas downstream of the air pre-
heater was located at the ID (Induced Draught) fan inlet. This
measurement location is downstream of the FFP (Fabric Filter
Plant). Since the flue gas operates at negative gauge pressure, the
possibility exists for air ingress into the FFP and ambient air
leaking into the ducting between the air preheater outlet and ID
fan inlet. Fig. 22 shows a side view of the draught plant layout
and the measurement locations.
A portable flue gas analyser was used to extract flue gas samples
at the air preheater flue gas inlet and ID fan inlet. The leakage was
calculated based on the change in oxygen content in the flue gas
before and after the air preheater. The derivation of the leakage
formula can be found in Appendix B. Table 6 gives measurement
results for the LH (left hand) air preheater. The air preheater
leakage (% APH leak) was calculated using Equation B-5 (refer to
Appendix B). The air leakage mass flow rate ðm_ leak Þ was also
calculated using Equation B-5.
4.1. Discussion
Recall from the results of the actual air preheater model, the
calculated leakage was 12.23 kg/s. The measured air preheater
leakage was 15.14 kg/s. The measured leakage is 23.8% larger than
the corrected CFD calculated leakage.
Table 5
CFD simulation results.
Leakage Units Value
m_ leak;cold end kg/s 7.14
m_ leak;hot end kg/s 6.52
m_ leak;total kg/s 13.66
98 A. Maharaj et al. / Energy 92 (2015) 87e99

Fig. 22. Draught plant side view showing measurement locations. Drawing is not to scale.

Table 6 The velocity magnitude contour plot in Fig. 24 confirms the

Air preheater leakage calculations.
Units
hO2 im;APH % 4.79
_ inlet
hO2 im;ID
_ fan inlet % 6.56
_
mfg;measured;APH inlet kg/s
% APH leak % 11.11
m_ leak kg/s
It is important to note that the measured leakage given in
Table 6 includes air preheater direct and entrained leakage as well
as possible leakage into the FFP, whereas the CFD calculated leakage
only accounts for air preheater direct leakage.
For the air preheater under investigation, the conventional
method to determine leakage from measurements will not provide
a value for the air preheater direct leakage.
Using the actual air preheater CFD model, the measuring loca-
tion at the air preheater flue gas inlet can be investigated. A velocity
vector contour plot at the measuring location is shown in Fig. 23.
The velocity vector contour plot shows that the vectors are only
aligned in the z direction at the bottom right hand region of the
duct. For the remainder of the duct, the vectors have x, y and z
components indicating the presence of a recirculation zone. The
high degree of recirculation indicates that there is a likelihood of
errors when performing measurements at this location.
presence of a non-uniform flow field at the measuring location.
Value When performing site measurements, it will be necessary to do a
mass weighted average of O2 measurements in order to obtain a
representative value.
136.25 Figs. 23 and 24 confirm a non-uniform flow field at the
measuring location indicating a higher possibility of error when
15.14
obtaining measurements. The difficulty in obtaining accurate
measurements in large ducts becomes evident when analysing the
flow field at the existing flue gas inlet measuring location [2].
5. Conclusion
The theoretical leak rate for a Ljungstro €m air preheater was
determined by assuming that the air leak path is similar to an orifice
flow path. The orifice coefficient was determined based on the specific
geometry of the leak gap. A 2D CFD model was also prepared to
calculate leakage rate. The theoretical and CFD results were compared
across the operating air temperature range for varying leak gaps. It
was noted that the CFD and theoretical results matched closer for
smaller leak gaps than for larger leak gaps. For a 15 mm gap at an air
temperature of 150  C, the difference between the theoretical leak
calculation and CFD calculation was 13%. For a 30 mm gap at an air
temperature of 150  C, the difference between the theoretical leak
calculation and CFD calculation was 16%. For the different leak gaps
investigated, the theoretical leak calculation was always lower than

Fig. 23. Velocity vector contour plot at the air preheater flue gas inlet.
 A. Maharaj et al. / Energy 92 (2015) 87e99
Fig. 24. Velocity magnitude contour plot at the air preheater flue gas inlet.
the CFD calculation. A comparison of the leak gap geometry and actual
orifice geometry shows that the flows are topologically not the same.
It is possible that the non-identical geometries contribute to an error
in determining the orifice coefficient used in the theoretical leak
calculation. Using the 2D CFD model, the air velocity in the leak gap
was characterised for varying seal settings. An investigation was also
performed to determine the influence on differential pressure across
the radial seal on the air leak mass flow rate. A lower differential
pressure was found to have a larger impact on the deviation of leak
mass flow rate from typical values.
The CFD modelling was extended to include the actual geometry
of the air preheater as well as the flue gas and air ducting as con-
structed on-site. The model inaccurately calculated hot end leak flow
due to limitations in using the enthalpy source option to introduce or
remove a specified amount of heat flow from the rotating matrix.
Using the theoretical leak flow equation, a correction factor was
determined and applied to the calculated hot end leak. The actual air
preheater model calculated a leakage which was 15% lower than the
leakage calculated by the 2D CFD model. It was noted that ducting
configuration has an effect on the calculated leakage.
On-site testing was performed for the air preheater under
investigation. Oxygen content in flue gas measurements before and
after the air preheater were used to calculate leakage. The down-
stream measurement location was non-ideal as it was situated after
the FFP. As a result, the measured leakage includes air preheater
leakage as well as any air ingress into the FFP. As a result, the
measured leakage was 23.8% larger than the leakage calculated from
the actual air preheater model. The existing measurement locations
on-site do not allow for measurement of air preheater leakage only.
The actual air preheater model was used to investigate the
measuring location at the air preheater flue gas inlet. A velocity
vector contour plot showed a high degree of recirculation indi-
cating a likelihood of errors when performing measurements at this
location. The vector contour plot highlights the problem of
obtaining accurate measurements in large ducts as is the case in
power plants. The CFD model can be used to determine better
measurement locations by attempting to find a location where the
flow is more uniform.
For the air preheater under investigation, the existing mea-
surement locations cannot be used to measure only air preheater
99
leakage. The 2D CFD model can provide an initial estimation of air
preheater leakage which can be used in early stages of an inves-
tigation. However, the actual air preheater CFD model shows that
ducting configuration does have an effect on the leakage calcula-
tion. Furthermore, the actual model can be used to investigate the
effect that non-uniform flow fields have on measurement
accuracy.
A recommendation for future work is to improve the heat source
and sink implementation in the rotating matrix to further advance
the CFD simulation. A combination of CFD and Finite Element
Analysis can also be used to account for thermal distortion of the
rotor and its effect on the leak gap.
Acknowledgements
The authors acknowledge the support of the Eskom Power Plant
Engineering Institute (EPPEI), in particular for the financial support
of Avinash Maharaj as an EPPEI student at the Combustion Engi-
neering Specialisation Centre at the University of the Witwa-
tersrand. The Specialisation Centre provided support with analysis
and interpretation of data as well as review of reports.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.energy.2015.06.069.
References
[1] Drobnic B, Oman J, Tuma M. A numerical model for the analyses of heat
transfer and leakages in a rotary air preheater. Int J Heat Mass Transf
2006;25e26:5001e9.
[2] Kitto JB. Steam/its generation and use. 41st ed. Ohio: The Babcock & Wilcox
Company; 2005.
[3] Skiepko T. Experimental results concerning seal clearances in some rotary
heat exchangers. Heat Recovery Syst CHP 1988;6:577e81.
[4] De Klerk G. An improved simulation model for rotary regenerative air heaters.
MSc Dissertation (Eng). Johannesburg: University of the Witwatersrand; 2001.
http://www.wits.ac.za.
[5] Siddhartha Bhatt M. Effect of air ingress on the energy performance of coal
fired thermal power plants. Energy Convers Manag 2007;7:2150e60.
[6] Shah RK, Skiepko T. Influence of leakage distribution on the thermal
performance of a rotary regenerator. Appl Therm Eng 1999;7:685e705.
[7] Skiepko T, Shah RK. Modelling and effect of leakages on heat transfer
performance of fixed matrix regenerators. Int J Heat Mass Transf 2005;8:
1608e32.
[8] Cai M, Hui S, Wang X, Zhao S, He S. A study on the direct leakage of rotary air
preheater with multiple seals. Appl Therm Eng 2013;1e2:576e86.
[9] Harper DB. Analysis of the rotary regenerator for gas turbine applications and
investigation of regenerator seal leakage. Massachusetts Institute of Tech-
nology; 1954. Technical Report No. 5. Defense Technical Information Center.
http://www.dtic.mil/dtic/tr/fulltext/u2/042477.pdf.
[10] Maclaine-cross IL, Ambrose CW. Pressure drop in parallel plate rotary
regenerators. In: Fifth Australasian conference on hydraulics and fluid
mechanics. Christchurch. New Zealand: University of Canterbury; 1974.
[11] Fox RW, Pritchard PJ, McDonald AT. Introduction to fluid mechanics. 7th ed.
United States of America: John Wiley & Sons Inc; 2009.
[12] Tu J, Yeoh GH, Liu C. Computational fluid dynamics: a practical approach. 1st
ed. United States of America: Elsevier Inc.; 2008.
[13] CD-adapco. User guide: STAR-CCMþ version 8.02. 2013.
[14] Maharaj A. Experimental and numerical studies of air preheater leakage. MSc
Dissertation (Eng). Johannesburg: University of the Witwatersrand; 2014.
http://www.wits.ac.za.