K. J.

Kutz Simulation of the Secondary Air

T. M. Speer
System of Aero Engines
MTU Motoren- und Turbinen-Union This paper describes a computer program for the simulation of secondary air systems.
Munchen GmbH, Typical flow system elements are presented, such as restrictors, tappings, seals,
Dachauer Strasse 665, vortices, and coverplates. Two-phase flow as occurring in bearing chamber vent
D-80976 Munich 50, systems is briefly discussed. An algorithm is describedfor the solution of the resulting
Federal Republic of Germany nonlinear equations. The validity of the simulation over the engine operation en-
velope is demonstrated by a comparison with test results.

1 Introduction
Moore (1975) reviewed the importance of the secondary air engine. The various flow paths shown in the network diagram
system for the safety, reliability, and performance of an aero in the following are referred to as pipes, although physically
engine and the corresponding design problems. they do not always resemble real pipes.
To meet the increasing demands of engine design, improved The task is to compute the flows in the lines and the pressures
simulation tools are required for the prediction of the air system in the chambers once the boundary pressures and temperatures
behavior over a variety of operating conditions. are given. It may also involve the evaluation of temperatures
This paper reviews a computer program for the simulation in the chambers due to mixing of different air streams or heat
of secondary air systems. It shows the modeling of an air system transfer at hot engine components.
as a network of pipes and chambers and presents the physical In the following section the governing equations for the air
relations governing it. This is followed by a presentation of system network as a whole are presented. This is followed by
typical flow system elements, such as restrictors, tappings, a description of some problems of computing pipe flow in an
seals, vortices, and cover plates. Two-phase flow as occurring air system. Finally, a review is given of the flow elements
in bearing chamber vent systems is briefly discussed. encountered in the network, especially for those typical for a
Thereafter, the algorithm is described for the solution of turbo engine.
the resulting nonlinear equations. This involves the decoupling
of the pipes and several nested levels of iterations, which refer
to the computation of mass flows through flow elements, pipe 3 Governing Equations for the Network
flows, nodal pressures, and chamber temperatures. For each of the N chambers or internal nodes (see Fig. 3)
Furthermore, existence and uniqueness of the solution are the law of conservation of mass applies. For steady-state con-
shown for the fixed temperature case. ditions it takes the form
The validity of the simulation over the engine operation
envelope is demonstrated by a comparison with test results. X>; =
~ (1)
A predecessor of the simulation program presented has been
described by Suntry (1963). This included, however, only a Here Ak denotes the set of all pipes adjacent to node k.
few, simple flow elements and lacked the capability of tem- If mixing temperatures are also to be computed in the system,
perature computation. The same is true for the programs of the law of conservation of energy must be applied to each
Rose (1978) and Miedema and Westgeest (1981) (the latter deals chamber:
only with incompressible flow). 2^(Tf) + Qi = 0. (2)
2 Modeling of an Air System
Q is a possible heat input into a chamber or pipe, e. g., by
An air system of a modern jet engine, which can become convection, which for the sake of simplicity is assumed to be
rather complicated, is shown in Fig. 1. It is best represented added to the chamber.
by a network diagram (Fig. 2). There is a flow exchange be- Unless otherwise stated, each flow temperature T-, equals the
tween the main air stream and the secondary air system at the temperature of the internal or boundary node at its inlet.
boundary nodes. The internal nodes of the network in most In general, mass flow m, of pipe /,
cases physically represent chambers of some volume in the
m,=MPC, TC), (3)
Contributed by the International Gas Turbine Institute and presented at the is a function of all unknowns (and of boundary pressures and
37th International Gas Turbine and Aeroengine Congress and Exposition, Co- temperatures, the dependency upon which is not shown in this
logne, Germany, June 1-4, 1992. Manuscript received by the International Gas
Turbine Institute February 4, 1992. Paper No. 92-GT-68. Associate Technical notation).
Editor: L. S. Langston. Normally, there is no coupling between pipes other than by

Downloaded From: 306 /Vol. 116, APRIL 1994

https://turbomachinery.asmedigitalcollection.asme.org
Transactions of the AS ME
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use
Copyright © 1994 by ASME
 Fig. 1 Air system of a modem jet engine

mass and enthalpy flow, which is taken into account in Eqs. m^fuiP, ij-i< Tu-A 7=1, (6)
(1) and (2). (Some exceptions such as branches and swirl gen-
erators are presented later.) Then, Eq. (3) reduces to a function must hold simultaneously for the unknown values of mass
flow mi and intermediate pressures P ; l P/,«-i-
m,=f,(Pmi,PompT,), (4)
Although the flow phenomena encountered in internal air
where PrN. and POUT- are the inlet and outlet pressures of pipe systems are mostly two or three dimensional, in the scope of
i, respectively; these can be nodal or boundary values. This such a network computer code they have to be treated as one
includes as a special case a fixed prescribed value mh e. g., dimensional. Flow conditions and fluid properties in an area
for a flow that is known by measurement. of interest have, therefore, to be considered as representative
We look for values of PC and TC such that Eqs. (l)-(3) mean values. The flow correlations employed are usually de-
hold. rived by simplified theoretical approaches with empirical cor-
Let pipe / contain n flow elements. We denote the pressure rection coefficients obtained from experiments or, as
at the outlet of they'th element by P,j (for j<n this is the inlet Zimmermann (1990) suggested, CFD computations.
pressure of the subsequent element; further P,> = POUT,) and Altogether, Eqs. (1), (2), and (6) describe a system of 2N+L
•P/,0 = PIN-,- Each element is modeled by a function nonlinear equations for TV chamber pressures and N chamber
m
temperatures, Mpipe flows and L -Mintermediate pressures.
~fij{PiNi POUT> TIN). (5) For actual networks (e. g., L = 260, M= 116, N=50) this cor-
Now, the n equations responds to 360 unknowns.

Nomenclature

A = set of indices of pipes adjacent TC=(TCX, . . • , TCN)T -= vector of unknown chamber

to a node temperatures
c = absolute velocity Tk __= temperature of flow i close to
C„ = discharge coefficient = node k
( w actual/ w ideal TIN -- inlet temperature
D = diameter u -= circumferential velocity
F = vector of mass balance defects V -= velocity
f,g = functions w == relative velocity
h = enthalpy
X - = transformed variable
J = number of flow element € == belongs to (set theory)
J = Jacobian matrix of function F AP == pressure difference
(PQ
k, I = node numbers e --= flow angle
= density
Q.

L = number of flow elements

I = length, thickness f. r == loss coefficient
M = number of pipes/flows Subscripts
Ma = Mach number / == refers to pipe/flow
m = mass flow k, I --= refers to chamber
TV = number of chambers/internal i,j --= refers to jth flow element of
nodes pipe i
n = number of elements in a pipe 0, 1,2 == pipe references of branches
P = pressure 0 - / == pertaining to flow direction
PCk = pressure at chamber k 1,2 == inlet and outlet, respectively
PC= (PCU PCN)T = vector of unknown chamber red == reduced
pressures s, t == static, total
inlet and outlet pressure, re- t, ax -= tangential, axial
spectively
Re Reynolds number Superscripts
Q* heat input into chamber k T --= transposed
T, temperature of flow i — -= refers to new iterative value
TCk temperature of chamber k 0 == initial guesstimate

Journal
Downloaded From: of Turbomachinery
https://turbomachinery.asmedigitalcollection.asme.org
APRIL 1994, Vol. 116/307
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 Straight pipe
Sudden contraction
with blunt edge

- n Sudden contraction
with bevelled edge

- Sudden expansion

Smooth and sharp

bends

O Boundary FBC Forward Bearing Chamber

| | Chamber CBC C e n t r a l Bearing Chamber Elbows

- — Pipe RBC Rear Bearing C h a m b e r
Fig. 2 Air system network for a modern jet engine
n„r T-Pieces

Y-Pieces

Fig. 4 Flow elements with dynamic head losses

nozzles. When a complicated network is analyzed at off-design

conditions, or when effects of failures or seal wear are inves-
tigated, flow reversal can occur in some of the branches. If
such a branch contains one of the above "asymmetric" flow
elements, the appropriate characteristic for reversed flow would
have to be used.
However, reversal of flow in a branch of an air system is
P0UT : usually undesirable and requires rectifying design measures
H r 1n . = r u,

1 " k
anyway. Therefore, the quantitative error caused by the use
of the same characteristic for both flow directions is deemed
acceptable, since only the indication of a flow reversal is im-
Fig. 3 Notation of nodes and pipes portant.

A possible change of temperature over a pipe as a conse-

quence of energy input, e. g., by vortices or heating, is ac- 5 Flow System Elements
counted for by introducing additional: The secondary air system of an engine can be divided into
Tij = gij(PU-u P,j, T,j-t) j=l,...,n (7) two subsystems: the internal system supplying cooling and
for each element, which relate its inlet and outlet temperature. sealing air, and the system for venting of the bearing chambers
This gives additional L unknown intermediate temperatures, to a near ambient pressure.
resulting in 620 unknowns altogether for the above example. The latter usually comprises some internal and external pipe-
These are computed in the course of the calculation of pipe work and thus contains the more "classical" flow elements
flows (cf. Section 7.3) by a variant of the algorithm described. such as pipes, bends, branches (T and Y-pieces), pipe inlets
and outlets.
The total pressure losses of these elements are commonly
4 Problems of Pipe Flow Computation expressed as a multiple of the dynamic head, i.e., by a loss
In a pipe containing many flow elements in series, the pres- coefficient f.
sure normally steadily drops from the inlet to the exit of the For the calculation of these loss coefficients a variety of
pipe. With increasing flow or overall pressure ratio over the information is available in the literature (Ward Smith, 1980;
pipe, choking may occur at some restriction within the flow Miller, 1978; Idel'chick, 1986). The pertinent calculation meth-
path. Choking can also be deliberately caused by a suitably ods are stored in a subroutine library, which can easily be
sized restriction as a means to limit or meter the flow. extended. Figure 4 shows the currently available elements. A
The possibility of choking influences the procedure for com- complication encountered in the vent system is the occurrence
puting flow in pipes with more than one flow element. Whereas of air/oil mixtures; consequently, methods for computing two-
in the subcritical case the flow in a restriction is determined phase flow are required. The approach adopted in the present
by both up- and downstream conditions, only the upstream program is discussed later.
ones set the flow in the choked case. The downstream pressure The elements typical for the internal air system are mainly
then depends only on the conditions at the downstream end restrictors and seals.
of the pipe. A computation procedure was, therefore, selected Restrictors can be divided into tappings, such as holes or
that starts at the end of the pipe and proceeds against the flow slots in a wall parallel to the main flow direction, and rotating
direction. holes. Common to these elements is that the flow is described
Among the flow elements occurring in air system networks, by the gas dynamic flow function and a discharge coefficient
there are some having different loss characteristics when flow Cd. Since the dynamic head generated in the vena contracta is
direction reverses, as e. g., sudden enlargements, T-pieces, not recovered downstream of a restrictor, the static pressure

3 0 8 / V o l . 116, APRIL 1994 Transactions of the AS ME

Downloaded From: https://turbomachinery.asmedigitalcollection.asme.org on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 downstream has to be considered as the total pressure at the Figure 5(a) indicates two things;
8
inlet of the following element. In addition to the geometry, the pressure losses are de-
pendent on the ratios of the velocities in the branches and in
In addition to the above " b a s i c " flow elements, the modeling
of an air system requires further elements. These comprise the mainstream.
elements for the conversion of the conditions of state from an • The pressure losses in the branches are dependent on the
velocity in the mainstream.
absolute to a relative system of reference and for the description
of swirling flow, which is modeled by vortices. In the scope of the main network program, however, flow,
and hence velocity are assumed to depend only on the pressures
Cover plates serve for the supply of rotor blade cooling air.
Although they are composed from the above elements they are at the ends of the pipe,as Eq. (4) shows. The equations have,
worth mentioning, since in addition to mass flow and pressure therefore, been modified as shown in Fig. 5(b). Since the loss
is now referred to the total head of the branch considered,
their description involves also the swirl generated by preceding
elements. each branch can be computed on its own. The loss coefficients
are determined from the result of the preceding iteration loop
In the following, the loss correlations of some of the typical
flow elements mentioned above are highlighted. and kept constant during the actual loop. An example for the
loss coefficients of a T-piece is given in Fig. 6.
5.1 Basic Flow Elements Figure 6 shows that under certain conditions there may be
5.1.1 Branches. Branches, such as T and Y-pieces, are a rise of the total pressure in the mainstream of a T-piece
special flow elements, since their incorporation into the air instead of a pressure loss. This is opposite to the normal char-
system program has a direct impact on the structure of the acteristic of a flow element and may lead to convergence prob-
main iteration process as shown later. They may be used for lems of the main iteration algorithm. The branch must,
either dividing or combining flow. The customary definition therefore, contain additional flow elements that enforce an
to account for the pressure losses of dividing flow and its overall branch characteristic of rising flow with rising pressure
modeling in the air system program is presented in Fig. 5 (for ratio. Such an arrangement, however, exists in practical ap-
combining flow the indices have to be interchanged). plications.
5.1.2 Restrictors. Orifices and Nozzles. Orifices and noz-
zles are used in the air system wherever flow metering or pres-
sure adjustment is required: for example, the metering of
cooling flow to turbine blades or the setting of bearing load
controlling pressures. Furthermore, nozzles provide the re-
quired preswirl for feeding turbine blade cooling systems. For
nozzles, therefore, not only the mass flow has to be computed,
AP but also the flow velocity, which is used as an input to a flow
0*i = ^ i P 0 V 0 / Z W i t h element farther downstream (preswirl nozzle).
C0*i = C 0 - i ( W V V geometry) The discharge coefficient of orifices is influenced by a variety
of factors such as Reynolds number (Miller and Kneisel, 1973),
i = 1, 2 inlet corner radiusing (Benedict et al., 1975), orifice length
a) Definition of Pressure Loss Coef- (Lichtarovicz et al., 1965), and compressibility (Bragg, 1960;
ficient _ Ward Smith, 1979). The above effects, except for compres-
c
0*l sibility, have been combined into a procedure for circular ori-
fices by McGreehan and Schotsch (1988).
0-*- Preswirl nozzles for cover plate systems are often of a very
special shape, and due to the arrangement of the nozzles in a
ring, there are interactions between the jets. There is very little
AP information on Cd values of such nozzles in the literature. The
o,i = ^o,ipivi /2 with Cd's for these configurations are, therefore, mostly determined
T0 +i = ^ i p o / p i ( v o / v i ) 2 by rig tests. Figures l(a, b) present the characteristic of a
stationary orifice showing the effect of geometry and com-
i = 1, 2 pressibility and of a radial preswirl nozzle, respectively.
b) M o d e l l i n g i n t h e Network Tappings and rotating holes. Tappings serve to bleed air
Fig. 5 Branching flow from the engine mainstream to feed the secondary air system.

0.4 1.'
•=01 "-02
0.3 1.2 i

0.2 1.0

0.1 0.8

0. 0.6 +
0. 0.2 0.4 0.6 0.8 1.0
v2/vQ
Fig. 6 Loss coefficients of a T-piece

Downloaded From:Journal of Turbomachinery

https://turbomachinery.asmedigitalcollection.asme.org APRIL 1994,
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use Vol. 1 1 6 / 3 0 9
 0.8-
blS*^Z--—~~
^ ^ _^--^^d^'
— ^

~^_}l
_ _ ^

0 -1-7
3_„

C
d
1/D=10 l.o-
0.7- c
d

\y\ \ o.4 ^—L

0.9-
X \ o.o ^^r-
0.B- , / \ 1.0 ^
1/0 = 1.4 ~^1' 0.0-
0.7-
P
tl/ p s2 0.4

a) Stationary Orifice b) Preswirl Nozzle

1.0-
with Up 1.0

0.8-
C
d
0.6-
0.5
0.4-
without l i p

0.2-

J
0. i i
2 3
<ptrps2>/(pti-psi>
c) Tapping d) Rotating Hole
Fig. 7 Characteristics of restrictors

a) Labyrinth b) Brush Seal c) Carbon Seal

Seal
Fig. 9 Seal types

for the rotating hole the velocities are denoted in the relative
system.
A simple theory for orifices in thin walls shows that the
main parameter controlling Cd is the downstream flow angle
r
tl si t 6, which can be described by the ratio c,/cax in the absolute
Rotating Hole: system or by \u\/wax in the relative system.
When the velocities are expressed by the total and static
pressures of the mainstream, respectively, the branch down-
= tenG stream of the orifice the flow angle can be shown to depend
wax on a pressure parameter (Pn -Pa)/(Pt\ -Ps\)-
sx
Fig. 8 Flow situation at tapping/rotating hole This parameter is very convenient for tappings, which often
operate at values close to 1 (flow angles near 90 deg). Although
derived from incompressible flow relations and for orifices in
The characteristic feature is that the flow is taken through an thin walls, experience shows that it also correlates Cd for tap-
orifice in a wall parallel to the main flow direction. In many pings in thick walls and for mainstream Mach numbers up to
cases orifices are rotating in shafts or disks to convey air from 0.5. Moreover, it also correlates data for such tappings as slots
a stationary into a rotating system, e.g., into the feed system in thick walls with or without a lip. Figure 7(c) presents an
for rotor blade cooling. example for circumferential slots in a thick wall and shows the
In principle the flows for the tapping and the rotating hole beneficial effect of a lip on Cd.
are similar. Figure 8 shows the velocity triangle for a tapping, For rotating holes, where the flow angle may also be small,
or a rotating hole, in a thin wall. The only difference is that the velocity ratio is more convenient, because u is given by the

3 1 0 / V o l . 116, APRIL 1994

Downloaded From: https://turbomachinery.asmedigitalcollection.asme.org Transactions
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use of the ASME
 n/T.
Sea Level Static
5000<Re<12000

10.3 km/Ma=1.0
1- 500<Ro<1300
,1 .2 .3 .4 .5 .6 .7 9 1.0
Vpi
O
3fficient a

Re 2 v / v
N. Fin thickness
ischarge

X Q . 2 5 4 mm
03

0
o .6 1 1 1 1 W/////////////////////.\
V 1.0 1.1 p p
1/ 2
1 2
-
Fig. 11 Effect of Re number on labyrinth seal

100
% 3 f i n labyrinth

80

60
3
O
-H
LL
.02 .04 .06 .08 .10 .14 .16
brush seal
c/s 40 . |
Labyrinth Flow: m = AacpYP1//RT1 (Egli (1935)) triple crush

Fig. 10 Labyrinth characteristic (taken from Zuk, 1976) 20 c o n t a c t i n g carbon seal

2 3
Pressure Ratio
shaft or disk speed. The axial velocity has to be derived from Fig. 12 Comparison between different seal types
the static pressures both up and downstream of the orifice. A
typical characteristic of a rotating hole including the effect of
wall thickness is given on Fig. 1(d). that of labyrinth seals (Fig. 12). These seals can be applied up
to considerably large diameters and thus can replace labyrinth
5.1.3 Seals. Due to their reliability, labyrinth seals (Fig. seals at many locations of the engine. Since there is no theory
9a) are the most frequently used seals in turbomachinery. Be- yet available for the leakage prediction of brush seals, for air
cause of their simplicity, in the main straight through labyrinths system calculations they are usually represented by an equiv-
are selected especially where small seal gaps can be realized. alent labyrinth seal based on test experience.
In areas where superior sealing is required, as at the exit of
high-pressure compressors with larger diameters and hence Carbon seals (Fig. 9c) can be either of the contacting or
larger clearances, stepped labyrinth seals are preferred. noncontacting type. For contacting seals there is no procedure
for leakage prediction available. One has to recourse to ex-
Leakage prediction is based upon the principle of a number perimental characteristics, which have to be fed into the pro-
of restrictions in series. In the chamber between two adjacent gram. For noncontacting seals as clearance seals or
seal fins, the kinetic energy built up in the seal gap is dissipated hydrodynamically balanced seals an approximate theory suit-
by a vortex. Leakage is primarily dependent on the gap width, able for the application in an air system program is available
the number of seal fins, and the discharge coefficient of the for leakage determination (Zuk, 1972).
seal gaps. Since the dissipation of the jet is not 100 percent This theory describes the flow as of the Fanno line type and
efficient, Egli (1935) introduced the concept of the carry over allows leakage calculation for subsonic laminar and turbulent
factor. Above influences are summarized in Fig. 10 from a as well as for choked flow conditions. Carbon seals offer a
paper of Zuk (1976). Usually engine companies have consid- superior sealing quality as shown on Fig. 12. Their application,
erable experience with labyrinth seals and use their own em- however, is limited to relatively small diameters due to limi-
pirical corrections to Egli's formula. An important parameter tations of rubbing speed and pressure differential (contacting
to be taken into account especially at high altitude operation seals) or manufacturing problems (clearance seals). They are,
of an engine is the effect of Reynolds number on Cd, as Fig. therefore, mainly used as bearing chamber seals.
11 shows.
Brush seals (Fig. 9b) are a relatively new seal concept and 5.2 Vortices. Swirling flow is encountered in the vicinity
offer the advantage of being less susceptible to shaft excen- of rotating disks or shafts. Although the actual flow field is
tricity, thermal and centrifugal growth of the rotor compared three dimensional, in the scope of an air system program a
to a labyrinth seal, due to the elasticity of the bristles. The simplified theory must be used. The pressure field in swirling
seal effectiveness depends on the density and the thickness of flow is, therefore, simply described by vortices. Depending on
the packing of the bristles. The sealing quality is superior to the magnitude of the radial flow, and on the starting condi-

Downloaded From:Journal of Turbomachinery

https://turbomachinery.asmedigitalcollection.asme.org APRIL 1994,
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use Vol. 116/311
 Initial Estimates: - PC, m, TC
Rotor Blade RB

1) Nodal Temperatures: PC, m -+ TC

Disk Entry DE

Forced Vortex F 2) Nodal Pressures: TC PC, m

Free Vortex FRV
Preswirl Nozzle 3) Pipe Flows: PIN, POUT -* in
Receiver R
(rot. rest
n=l: 4a) Flow Element:
PIN, POUT, T -* ID

^^7 . n>l: 4b) Flow Element:

^Transition—to Rel.7 Syst POUT, m, T - PIN

TRS
Fig. 13 Simulation of cover plate system

Fig. 15 Nested iterations of the computation process

AP
(kPa)
restrictor to simulate the blade. The flow diagram shows that
20Oi
"1 • n ( 9 / s )
there is a secondary connection, which transmits swirl from
oil 3
'
the preswirl nozzle to the free vortex across a node in the
0 2
system, thus creating an interdependence between two different
branches. Rig tests have proven that this simple network closely
models the behavior of a cover plate system.
100-
6 Two-Phase Flow
The flow in vent pipes normally consists of a mixture of air
and lubricating oil, which is entrained from the bearing cham-
bers (Zimmermann et al., 1991). This oil causes a blockage of
20 40 60 80 the ventlines and thus creates an additional pressure loss. Fig-
ure 14 shows how the pressure loss is increased by the vent-
oil for a typical engine vent line based on engine measurement.
Fig. 14 Effect of oil flow on vent pipe pressure loss The concept of Lockhart and Martinelli (1949) of a two-
phase multiplier is adopted to calculate these two-phase pres-
sure losses. The two-phase multiplier is defined as the ratio of
tions, a free or forced vortex can be assumed. In certain cases, the two-phase pressure loss to the pressure loss obtained if all
for example in coverplate systems, vortices are deliberately the fluid in the pipe were of the same phase. Although in the
generated to increase the pressure. literature the two-phase multiplier is usually referred to the
For solid body vortices between rotating and stationary disks, liquid phase, for the present purpose it is more practical to
the strength of the vortex is described by the core rotation refer it to the gaseous phase, i.e., air, since this is the primary
factor, which relates vortex circumferential speed to the disk working fluid.
speed. This factor is of the order of 40 to 50 percent for a The procedures to derive the two-phase multiplier are all of
rotor-stator configuration. Numerous investigators (Schultz- a semi-empirical nature due to the complexity of the flow. The
Grunow, 1935; Daily and Nece, 1959) have shown this. methods described in the literature are mostly set up for water/
For vortices enclosed by corotating disks, the deviation from water vapor or water/air mixtures. Friedel (1978) has devel-
idealized behavior can be described by a "vortex effectivity," oped a method for computing the pressure loss of pipes based
which is defined as the ratio of actual to theoretical pressure on a large amount of data. With some modifications his method
rise. This figure has either to be found from separate numerical is also suitable for air/oil mixtures as comparisons with engine
calculations or from component rig tests. and rig test evidence has shown (Fischer, 1990).
For pipes, orifices, smooth and sharp bends Zimmermann
5.3 Coverplates. As an example for the application of
et al. (1991) have given procedures to evaluate the two-phase
the aforementioned flow elements and also for the secondary
multipliers. The main problem remaining is the determination
interactions between elements, a coverplate is shown in Fig.
of the oil loading of the vent flow, which for the time being
13. The system provides the cooling air for the turbine blades.
has to be estimated based on past engine test experience.
Its purpose is to reduce the cooling air temperature in the
relative system by strongly preswirling the flow. Seal leakage
is minimized by arrangement of the preswirl nozzles at a low 7 Steady-State Computation
radius. In view of the size of the problem, it does not seem appro-
A cover plate can be modeled by putting the following ele- priate to attempt to solve all the equations simultaneously.
ments in series: preswirl nozzle, rotating restrictor as receiver, Consequently, a hierarchical algorithm is used that exploits
a free an/or a forced vortex to account for the pressure increase the topological structure of the network by decoupling pipe
due to the swirl between the corotating disks, a rotating re- and element equations from each other. A rough outline of
strictor for blade root entry, and another forced vortex and a the nested iterations is given in Fig. 15.

Downloaded From: 3 1 2 / V o l . 116, APRIL 1994

https://turbomachinery.asmedigitalcollection.asme.org Transactions
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use of the AS ME
 For the sake of simplicity, let us assume that the air system occurs in all pipes adjacent to node k. If now there is a fixed
contains neither branches nor swirl generators. mass flow of sufficient magnitude given that escapes from
The algorithm starts with some initial estimates for m, PC, node k, then a solution of the whole problem might not exist.
and TC. This will be discussed in Section 7.6. In all other cases, e.g., if only boundary pressures are pre-
The iteration levels are presented from the top down. scribed and mass flows that enter the system, the existence of
the solution follows from the argument above and the non-
7.1 Computation of Nodal Temperature Values: Iteration singularity of the Jacobian matrix (Ortega and Rheinboldt,
Level 1. As the influence of temperatures on pressures and 1970, p. 137).
mass flows is relatively weak, computation of temperatures is
decoupled from the other unknowns. Let us assume that m, Practical Experience. In practice, the solution was found
PC, and TC satisfy Eqs. (1) and (6) (this is achieved on the whenever its existence could be presupposed from physical
lower iteration levels). If Eq. (2) also holds, the calculation is reasons. For the reduction of mass balance defects by five
stopped. Otherwise we compute new values of enthalpies hj orders of magnitude (starting from initial estimates) the method
from takes about 20 steps. For a large network this corresponds to
J]mfa+Qk = 0 k=l, . . . ,N. (12) 5 s CPU time.
HAk 7.3 Computation of Pipe Flows: Iteration Level 3. Now
As the enthalpies of all flows leaving node k satisfy we consider the computation of mass flow m in pipe / for given
inlet and outlet pressures PIN and POUT and temperature T. If
hj=h{TCk), (13) there is only one element in the pipe, m can be computed
and as enthalpies are given at inlet boundary nodes, this leads directly from the nodal values. This is done on iteration level
to a system of M, corresponding to the number of pipes, linear 4. Otherwise, m and the unknown intermediate pressures Pi,
equation^ for the enthalpies h,. . . . , P„ (we drop the index / for convenience) are found by
From hi, new iterative values TCk for the chamber temper- the following algorithm:
atures are computed from Eq. (13). (A) For given mass flow m, outlet pressure POUT, and tem-
Although there is no theoretical proof of convergence, this perature T compute inlet pressure Pm{m) such from n
algorithm has never failed to meet the desired precision within down to 1 (i. e., in the upstream direction) the equations
a maximum of eight iteration steps.
m=f,j{Pj-U Pj, T) (16)
7.2 Computation of Chamber Pressures: Iteration Level forP,_,.
2. Let the nodal temperatures be fixed. These may be initial SetP / A ,(m)=P 0 .
estimates or iterates from iteration level 1 or, if satisfaction (B) Vary m according to a secant method until PIN(m) meets
of enthalpy balance, Eq. (2), is not desired, they may be fixed PIN-
input values. Nodal pressures PC are computed to solve the
system Computation of Derivatives. The performance of the Hy-
F(PC) = J]mi(PC)=0 k=\ N (14) brid Method depends crucially on the quality of the partial
derivative dFj/dPCj. As these are sums of the derivative dm/
HAk
dPC, the latter have to be evaluated with care. This is done
of nonlinear equations by means of the Hybrid Method of by numerical differentiation within the mass flow computation
Powell (1970), implemented by More et al. (1980). program as only then is it possible to exploit the network
This requires the computation of mass flows mh which enter structure, reuse intermediate results, and adjust the increment
F(PC) according to Eq. (5) for any iterate PC and of the used in computing the finite differences to each pipe.
partial derivatives dFj/dPCj (these form the N-N Jacobian
matrix, J). Accelerating the Computation of "Long"Pipes. The main
computational burden lies in the computation of the mass
Uniqueness of Solution. From consideration of the Ja- flows, especially for long pipes, i.e., those with many elements.
cobian matrix the uniqueness of any solution can be deduced During the early steps of iteration 2 all computed mass flow
in the following way (discussion of its.existence is postponed values of these pipes are stored and combined into a single
to the following paragraph). Let us consider a flow from node pipe flow characteristic in the form of a curve of reduced mass
/ to k. For the mass flow to be uniquely defined, the modeling flow m-jT/PIN versus pressure ratio P//V/POUT- During sub-
of the flow elements must guarantee dm/dPC^O; in fact, dm/ sequent steps, this function
dPCk<0 holds. This derivative dm/dPCk contributes to the
elements Jkk and J/k of the Jacobian (with sign change); likewise m-jT/PIN=mKi(Pm/Pom) (17)
for its fth column. If / (or k) denotes a boundary node, only is evaluated by interpolation for the given nodal values of PIN
the kth (or /th, respectively) diagonal element is influenced by and POUT resulting in values for the mass flow of a pipe with
that flow. Thus, J has negative diagonal elements while all a similar characteristic, Eq. (17), as the original pipe. Partial
others are positive. Further, the inequality derivatives may also be easily computed from Eq. (17).
When convergence is nearly achieved for this simplified
\Jkk\<J]\Jlk\ (15) model, its results serve as rather good starting values for the
l*k
final iteration steps for the original model. Depending on the
holds for k = 1, . . . , N, with strict inequality for all internal proportion of long pipes, this may save up to 40 percent of
nodes that are directly connected to a boundary node by one overall computation time. (A similar feature has been imple-
pipe. mented by Rose, 1978, as was discovered by the authors at the
If there is at least one fixed boundary pressure, / is non- time of writing.)
singular and a so-called Mmatrix. (Otherwise the total problem
7.4 Computation of Flow Elements: Iteration Level 4. As
would be undetermined.) This guarantees that any solution is
unique (Ortega and Rheinboldt, 1970, pp. 55, 141). pipes with one element and those with more elements are treated
differently, one needs two routines to model each element: one
Existence of Solution. For every internal node k the mass to compute m from PIN and P 0 UT> and one to compute
balance defect Fk(PC) tends to infinity as P Q — ° ° . If, how- PIN from m and POUT- This increases program development
ever, PCk approaches zero while all other pressures remain expenditure but saves computation time and increases preci-
unchanged, Fk{PC) decreases to a finite value because choking sion.

Journal of Turbomachinery APRIL 1994, Vol. 116/313

Downloaded From: https://turbomachinery.asmedigitalcollection.asme.org on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 flitltuda Rltitude
© SL O SL
CD
+ 1.5 km + 1 .5 km
A ] 1 .0 km A ] 1.0 km
X 16.8 km X 16.8 km
X 18.3 km - X 18.3 km
• 19.8 km
w
4 C/)
O 19.8 km
f
V) r-f
Q) (0 + CD CO
U Q)

0. en
U CD
D_ CO y
h -p A
•H [-1
•H C •=C CO
(9 CD
•a: o ence
u 9"
cnu.
J'
O
9
en
• y
200 300 200 300

Fan Exit Pressure <kPa> Fan Exit Pressure <kPa;

closed symbols Computer model
open symbols Test results
all pressures are ISA-corrected

Fig. 16 Comparison between computer model and test results for rear bearing chamber sealing air pressures

For almost all types of elements Eq. (4) is implicit and must passes through a number of rotating holes in compressor disks
be solved iteratively for m or PIN. These values are computed and shafts, through vortices, and suffers from some friction
as precisely as possible, thus achieving smooth functions and pressure losses. In the end the air re-enters the mainstream
conserving monotony, which is vital to the performance of the rear of the low-pressure turbine.
method. In Fig. 16 the ISA-corrected sealing air pressures are plotted
versus the corrected fan delivery pressure, which is the feeding
7.5 Coupling Between Pipes. If swirl is transported from pressure of the system. The simulated data (solid symbols)
one element to another or in the presence of branches (cf. consist of a sea level static running line, some near sea level
Section 5.1.1), pipes can still be treated as being decoupled for flight points, and some medium to high altitude flight points,
iteration levels 2 to 4 by the following method: whereas the tested data (open symbols) result from tests in an
The corresponding transported values (swirl and loss coef- altitude test facility over a great part of the flight envelope of
ficients) are "frozen" for a complete execution of iteration the engine. The simulation is based upon performance data
level 2, and they are updated with their thereby computed obtained from an engine synthesis model, boundary conditions
values after that. This necessitates execution of the outer it- obtained from component aerodynamics, and an air system
eration 1 even if no mixing temperatures are to be computed. model derived from the analysis of some SLS test points.
7.6 Estimating Initial Approximations. If necessary, the The figure shows that the internal air system pressures when
program computes initial estimates for internal temperatures ISA corrected can be presented by a single operating line for
and pressures, given their boundary values and at least one working points all over the flight envelope, as indicated by
representative mass flow value. both the simulated and the tested data. The figure further
If the boundary pressures are consistent with the prescribed shows a good agreement between simulation and test evidence.
flow directions, there exist internal pressure values such that This demonstrates that a model, which is well adapted to a
the minimal ratio Pm/Povi over any one pipe is maximized. few test points, can predict the air system behavior all over
The logarithms of these values are the solution of a Linear the flight envelope. It is to be noted, however, that this good
Programming (LP) problem. Maximizing P/N/PQUJ serves to agreement is not only the result of a careful physical modeling
avoid nearly vanishing mass flows and their accompanying of the air system but also of the high quality of the cycle model
high derivatives. If the consistency condition above does not and of component aerodynamics analysis.
hold, the LP problem has no solution.
Given values for all mass flows, initial temperature estimates
are computed in analogy to the procedure of computing new 9 Conclusions
temperature iterates in Section 7.1 within iteration level 1 by An air system program has been set up that can model very
solution of Eq. (12). complex networks. The CPU time for the steady-state calcu-
lation of a typical engine system consisting of a low-pressure
bearing chamber sealing system, an intermediate pressure tur-
8 Comparison With Test Results bine cooling system, and a high-pressure turbine cooling system
To highlight the validity of the computer model as an ex- (with a total of 116 pipes and 50 nodes) is typically 5 s on a
ample the comparison between the simulated and measured NAS XL 100 computer.
rear bearing chamber sealing air pressures of a turbojet engine In addition, selected geometric parameters may automati-
is presented on Fig. 16. The rear bearing chamber has been cally be adjusted to meet prescribed nodal pressures values
selected, because the sealing air, which on this engine is taken (e.g., test results or design targets).
from the fan exit via total head scoops, on its way to the seals The program has turned out to be a valuable tool for the

3 1 4 / V o l . 116, APRIL 1994 Transactions

Downloaded From: https://turbomachinery.asmedigitalcollection.asme.org on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use of the ASME
 design of an engine air system and for the prediction of its Miedema, J. A., and Westgeest, B. Q. M., 1981, "Computation of Mass
behavior throughout the flight envelope. Flow Rates and Pressure Losses in Branched Pipeline Networks," TNO, Rep.
5173202-81-1.
The program has also been used for other networks of turbo Miller, D. S., 1978, "Internal Flow Systems," BHRA Fluid Engineering, Vol.
engines, such as the feed oil system. For this application, the 5.
compressible flow relations have been replaced by those for Miller R. W., and Kneisel, O., 1973, " A Comparison Between Orifice and
incompressible flow. Its application is being extended to cover Flow Nozzle Laboratory Data and Published Coefficients," ASME Paper No.
73-WA/FM-5.
also transient operation of the engine. Moore, A., 1975, "Gas Turbine Engine Internal Air Systems: A Review of
the Requirements and the Problems," ASME Paper No. 75-WA/GT-l.
More, J. J., Garbow, B. S., and Hillstrom, K. E., 1980, "User Guide for
MINPACK-1," Argonne National Laboratory, Argonne, IL.
References Ortega, J. M.,-and Rheinboldt, W. C , 1970, Iterative Solution of Nonlinear
Benedict, R. F., Wyler, J. S., and Brandt, G. B., 1975, "The Effect of Edge Equations in Several Variables, Academic Press, New York.
Sharpness on the Discharge Coefficient of an Orifice," ASME Journal of En- Powell, M. J. D., 1970, " A Hybrid Method for Nonlinear Equations," Nu-
gineering for Power, Vol. 97, No. 4, pp. 576-582. merical Methods for Nonlinear Algebraic Equations, P. Rabinowitz, ed., Gor-
Bragg, S. L., 1960, "Effect of Compressibility on the Discharge Coefficient don and Breach, pp. 87-114.
of Orifices and Nozzles," J. of Mechanical Engineering Science, Vol. 2, No. 1, Rose, J. R., 1978, "FLOWNET: A Computer Program for Calculating Sec-
pp. 35-44. ondary Flow Conditions in a Network of Turbomachinery," NASA TM-73774,
Daily, J. W., andNece, R. E., 1959, "Chamber Dimension Effects on Induced Lewis Research Center, Cleveland, OH.
Flow and Frictional Resistance of Enclosed Rotating Disks," ASME Journal Schultz-Grunow, F., 1935, "Der Reibungswiderstand rotierender Scheiben in
of Basic Engineering, Vol. 82, pp. XX-00. Gehausen," ZAMM, Vol. 15, No. 4, pp. 191-204.
Egli, A., 1935, "The Leakage of Steam Through Labyrinth Seals," Trans. Suntry, H., 1963, "Iterative Airflow Schemes," Computer Program Report
ASME, Vol. 57, No. 3, pp. 115-122. 1204, Rolls-Royce Ltd., Derby, United Kingdom.
Fischer, R., 1990, "Probleme der Zweiphasenstromungen im Sperrluftsystem Ward Smith, A. J., 1979, "Critical Flow Metering: The Characteristics of
von Flugtriebwerken," Inst. f. Luftfahrtantriebe, Univ. Stuttgart, Germany. Cylindrical Nozzles With Sharp Upstream Edges," Int. J. Heat Fluid Flow, Vol.
Friedel, L., 1978, "Druckabfall bei der Stromung von Gas/Dampf-Fliissig- 1, No. 3, pp. 123-132.
keitsgemischen in Rohren," Chem.-Ing.-Tech., Vol. 50, pp. 167-180. Ward Smith, A. J., 1980, Internal Fluid Flow, The Fluid Dynamics of Flow
Idel'chik, I. E., 1986, Handbook of Hydraulic Resistance, 2nd ed., Hemi- in Pipes and Ducts, Clarendon Press, Oxford.
sphere Publishing Corporation, Washington, DC. Zimmermann, H., 1990, "Some Aerodynamic Aspects of Engine Secondary
Lichtarovicz, A., Duggins, R. K., and Markland, E., 1965, "Discharge Coef- Air Systems," ASME Journal of Engineering for Gas Turbines and Power, Vol.
ficients for Incompressible Noncavitating Flow Through Long Orifices," J. 112, pp. 223-228.
Mech. Eng., Vol. 7, pp. 210-219. Zimmermann, H., Kammerer, A., Fischer, R., and Rebhan, D., 1991, "Two-
Lockhart, R. W., and Martinelli, R. C , 1949, "Proposed Correlation of Data Phase Flow Correlations in Air/Oil Systems of Aero-Engines," ASME Paper
for Isothermal Two-Phase Two Component Flow in Pipes," Chem. Engr. Progr., No. 91-GT-54.
Vol. 45, No. 1, pp. 39-47. Zuk, J., 1972, "Fluid Mechanics of Noncontacting Gas Film Seals," Ph.D.
McGreehan, W . F . , andSchotsch, M. J., 1988, "Flow Characteristics of Long Thesis, Case Western Reserve Univ.
Orifices With Rotation and Corner Radiusing," ASME JOURNAL OF TURBO- Zuk, J., 1976, "Fundamentals of Fluid Sealing," Lewis Research Center,
MACHINERY, Vol. 110, pp. 213-217. Cleveland, OH, NASA TN D-8151.

The American Society of

Mechanical Engineers

THE WORLD OF ASME AT YOUR FINGERTIPS

At ASME Information Central, where you are our top priority, we make every effort to
answer your questions and expedite your orders. Our representatives are always ready
to assist you with most any ASME product or service. And now, reaching us is easier
than ever!

TELEPHONE FAX MAIL

800-THE-ASME 201-882-1717 ASME
800-843-2763 or 201-882-5155 22 Law Drive
(USA & CANADA) P.O. Box 2900
95 800-843-2763 Fairiield, New Jersey
(MEXICO)
E-MAIL 07007-2900
201-882-1167 CompuServe
(UNIVERSAL) 73302,1017

Downloaded From:Journal of Turbomachinery

https://turbomachinery.asmedigitalcollection.asme.org APRIL 1994,
on 01/08/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use Vol. 116/315