AlAA 94-3228

Propulsion System Modelling for

Condition Monitoring and Contro ..
Part II, Application to the SSME
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

K. Kolcio and A.J. Helmicki

University of Cincinnati
Cincinnati, OH

30th AIANAS ME/SA WASEE J oi nt

Propulsion Conference
June 27-29,1994 / Indianapolis, IN
For permission to copy or republish, contact the American Institute of Aeronautics and Astronautics
370 L'Enfant Promenade, S.W., Washington, D.C. 20024
 Propulsion System Modelling
for Condition Monitoring and Control:
Part 11, Application to SSME
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

K. Kolcio A. J. Helmicki

Department of Electrical and Computer Engineering

University of Cincinnati
Cincinnati, Ohio 45221-0030, U S A .

Abstract equations of state. The main advantages of such an

approach are the following.
A control oriented method yielding nominal mod-
els for both gaseous and compressible liquid media Extensive use of first principles reduces reliance
propulsion systems has been developed in a com- on empirical data. When it becomes necessary
panion paper. In the sequel, the generic model to resort to empirical data, it is incorporated
equations are validated against the one dimensional in a rigorous m m e r .
wave equation which results from the linearization
, of the original continuity, momentum and energy 0 A nondimensionalization procedure casts sys-
PDEs. The generic model is then applied to the tem equations into an elegant form and gives
SSME HPFP to illustrate the modelling procedure. rise to a dimensionless parameter whose rela-
In addition, it is shown that anomalous conditions tive size determines dynamic order.
can be incorporated using the signal and dynamic
uncertainty type representations to generate mod- The system is initially divided up into com-
els reflecting various engine anomalies. Examples of ponents exploiting the natural boundaries and
anomalous behavior typical of the SSME HPFP in- inputs and outputs. For example, in rodret en-
troduce the signal and uncertainty type structures. gines the combustion chamber and fuel pump
As a result, it is established that the models de- would be two separate components.
veloped using these techniques are low order and
capable of reflecting both nominal and anomalous Spatial discretization transforms the PDE’s to
behavior, making them suitable for control and con- ODE’Sand further compartmentalizes the com-
dition monitoring purposes. ponents into elements, if necessary. The num-
ber of elements depends on the fineness of the
spatial grid.
1 Introduction
!The resulting nominal models are modular and
In the preceding companion paper the authors de- low order which is ideal for control purposes.
veloped a generic model suitable for propulsion sys-
tems. The control oriented modelling approach in- As discussed in the Part I companion paper [13],
troduced in [6,5 , 41 for generating low order nom- it is unclear how to modify nominal models rely-
inal models for air-breathing turbomachinery was ing heavily on empirical data to include anomaloun
extended to include the liquid case via two novel effects. The structure of the nominal models result-
ing fiom the control based approach facilitates the
o@CopYr;ght 19% by Arthur J. Hcknicki. Publkhal by
the Amniean Institute of Aeronautics and Astromutiu, hc. augmentation of anomalous behavior. Extending
/ with permission. the modelling approach to encompass models able

1
 to reflect degraded or anomalous conditions moti- Assume that the liquid is initially motionless and
vated this paper. Such anomalous models in ad- has the fixed properties p,, p,, , C everywhere in
dition to nominal models are important from both the constant area duct. Since the liquid is initially
the condition monitoring and robust control stand- motionless, 3, = 0. Consider a weak perturbation
point. A condition monitoring system detects var- in the fluid properties such that the pressure, den-
ious degraded plant behavior. A robust controller sity and velocity may be written in terms of the
must stabiliie a plant in the face of uncertainty in initial and perturbation quantities
the plant dynamics and extraneous signals perturb-
ing the nominal input and/or output [9].
Some anomalous conditions such as fuel leakage,
for example, can be thought of as additional dy-
namics, or uncertainty. Others such as sensor fail-
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

ure may appear as extraneous signals. If anomalous Substituting the above quantities into Equation 1
conditions are modelled as either uncertain dynam- yields
ics or extraneous signals then some existing system-
theoretic tools can be used to handle these particu-
lar representations [9,14, 151.
This work is organized in the following manner.
First the discretized model equations are validated Because p, and limare constant their derivatives
against the full PDE equations. Second, the generic vanish giving
model developed in Part I is applied to Space Shut-
tle Main Engine (SSME) High Pressure Fuel Pump
(HPFP). Finally, some specific examples that illus-
trate the uncertainty and signal type represent& Expressing pressure as a function of density, p,
tions of anomalous behavior are presented and dis- and entropy, 5, the total derivative can be written
cussed.
as [I11
,
2 Simple Example
Before applying the generic model developed in Part
I [13] to the SSME,the validity of the nondimen- Under the isentropic assumption d5 = 0 and the
sional discretized equations will be examined using above becomes
the following simple example. This example was
chosen so that the PDE equations could solved in
closed form, thereby allowing direct comparison of
the full PDEderived results with those obtained
from the discretized, nondimensional ODES devel- Considering changes in li and in the z direction
oped in Part 1. we have
Starting with the PDE forms of the conservation
of mass and momentum given in [lo, 21, 31, the -=
additional assumptions of no area change and isen-
tropic flow reduce the continuity and momentum Now the speed of sound a is defined (3, 10, 191 as
equations to the following'
ali .
aa := (-)i
ap
The momentum equation is then written in terms of
the perturbation variables and speed of sound thus

'Tkoughovi thir work a distinctioni8madebetween 6

b b v i q units (dimmaionsl) and those void of u n i t a
a
(nondimmaional) From h e n on, the - symbol over the
dimcnrional variables differentiatea thcm from their n o d - If weak perturbations are considered, then the
/
mmriond counterparts. product of two perturbation quantities is negligibly

2
 small, for example, where

{
Ul(2,t) :=
iil[l+
{;I)*] 2n - 3 - 5 < i < 2 n - 1 - 5
a a
otherwise
and

Effecting these approximations, Equations 3 and 4

become
-+ -
ab;
pm-
a6i
=0, (5)
{
U*(z,t):=
iii[l+:;l)*] 2n-3+5
a
2n-l+t
a
af a2
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

otherwise
with n = 1,2,3,. . ..
A plot comparing the wave
equation and the dimensional discretized model ve-
The analysis leading to Equations 5 and 6 is locity at the inlet is shown in Figure l below where
known as small perturbation theory or acoustic a p the duct length L is set to 1.
proximation theory [lo, 31. Equations 5 and 6 can The initial assumptions and approximations
be combined by cross differentiation and elimina- made in forming the most general PDE and then
tion of cross terms to obtain the single
. linear second the additional isentropic and no area change simpli-
order PDE ficatiom are common to the wave equation and the
pi -az-a% = O , model equations. They differ in that the wave equa-
at2 a52
(7)
tion is linear but not discretized while the model
where the 6 notation has been dropped for brevity. equations are nonlinear but discretized. The sec-
Equation 7 is known as the one dimensional wave ond order wave equation evolves from the general
equation [3, 101. The solution to Equation 7 subject PDE’s by linearizing about the perturbations. The
to certain initial conditions (IC) and boundary con- model equations preserve the nonlinearity but are
, ditions (BC) can be found using Laplace methods diacretized by assuming the variables have a linear
[71. distribution in space 1131. In fact, the sinusoidal
Suppose the following experiment is run. Fluid shape of the generic model’s step response is a con-
in a constant area duct of unitlength is initially at sequence of discretization. It can be argued that as
rest so i is zero everywhere, P(0,O) = P(1,O) and the wave equation and the generic model equations
T(0,O)= p(l,O)as shown in Figure 1. A step input are based on different operations, linearization ua
is applied at the outlet end so that at timet = 0+, discretization, we are not any closer in validating
the velocity at the outlet is set to the constant value the generic model against the general PDE. How-
irl. We are interested to see what happens at the ever, the wave equation should closely approximate
5 = 0, or inlet, end. First, the appropriate IC and the PDE behavior in the neighborhood of the oper-
BC are found ating point. The wave equation then can be taken
as closer to the “real thing “. It is not entirely clear
IC: q&o)=iim=o; i(t,O)=O, at this point what exact dfect discretization has on
BC: i(l,q = Cil(i) ; i&(O,q = 0 , the PDE. From Figure 2, it is clear that qualita-
tively the behavior of the generic model equations
where l ( q is the unit step function and the sub- follows that of the wave equation. Indeed showing
script I denotes the partial derivative with respect this was the purpose of the simple example.
to I. The second BC is derived from Equation 5
evaluated at I = 0. Since j(0,i) in fixed then
3 Nominal Model
In this section the generic model, matrix Equa-
tion 9, developed in [13] is applied to the specific
Equation 7 is then solved by the general Laplace example of the SSME HPFP.
methods noted above and by applying the above IC
and BC to get

i(t,i)= U1(2,t)+Uz(z,t)
/

(8)

3
 Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

J
,

4
Figure 1: Duct schematic for wave equation example

Figure 2 Cornpariaon of wave equation and generic model.

 terms either some particular functional forms must
be assumed or maps must be derived through an
, analysis of the SSME steady state operation which
is well understood. In the SSME RTM and the
where SSME MARSYAS models the pressure risc gener-
ated by the pump comes from steady state pump
maps [17,20].Instead of choosing a functional form
for the shaft force the existing HPFP pump map
was folded in [12]. In this case the maps incorpo-
rate the friction term as well.
According to the RLOOOOl document [17] the
pump map, r F P , accepts a nondimensional flow
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

variable input 4 and outputs a corresponding nondi-

mmsional head variable J,. The pump map is de-
fined by $FP := I)Fp(+Fp) . The 4 variable is
x := -4
. a function of the pump fuel flow, DWIFP, density,
C.lK
and pump speed, s ~ p The . pump map will be in-
The SSME HPFP is a three staged centrifugal corporated here in the following manner: first, 4 is
pump. The turbine generates the available torque cast in terms of the dimensional variables, A, ii, d,
to turn the pump shaft. Liquid hydrogen fuel enters and s ~ p Pump
. speed is introduced as another in-
the pump at the inlet volute, courses through the put to the model. Second, the output of the map,
pump stages where the kinetic energy of the flow is $, is redimensionaliied in terms of the pressure dif-
converted to potential energy (pressure), and exits ference across the pump, density and pump speed.
at the outlet volute. The purpose of the pump is to The map input 4 must be constructed from the flow
compress the fuel to a high enough pressure so that variables used in Equation 9. Using the definition
it may enter the combustion chamber. of the flow variable consistent with [17],we have
For simplicity, it is assumed for now that the
~ entire HPFP can be modeled as one element. A
one element component means that k = 1. The
pump inlet pressure, PO,and temperature, TO,and
the Mach number at the pump outlet, M I , serve To construct 4, the variables A and ii must first be
as the model inputs while the pump outlet pres- redimensionalized from the variables At, M I . An
sure, &, and temperature, 7'1, and Mach number s p p comes as an external input, it can be used di-
at the inlet, Mo, become the states and outputs rectly. Note that consistent with the explicit form
of the model. If it is determined that more than of the model developed here, only system inputs
one element is necessmy, each pump stage can be are used. The dimensional 4 is then nondimension-
modeled as an element. The elements would then &ed for use in the map. The nondimensional out-
be concatenated to form the entire HPFP. Increas- put of the map, J, is dimensionaliaed in terms of the
ing the number of elements means that additional pressure across the pump, AP, density and pump
pump maps would be needed for each compressor spud
stage in order to determine the forcing functions.
From a practical standpoint, obtaining these maps la=&,
is not a trivial task. Thus, the classic trade-off be-
where 4 := . The dimensional pressure rise
tween theory and practice is incurred. The same
procedure would be applied to the turbine only us- across the pump is then given by
ing the equations derived for a gaseous fluid. The
HPFP and turbine models together would comprise Fl - Fo = 4hbs;p.
the High Pressure Fuel Turbopump, HPFTP.
Some general assumptions are in order. Flow By inspection of Equation 9 in steady state and
through the pump is assumed adiabatic, with wall the equation above, it is seen that after nondimen-
friction and the pump shaft does work on the fluid. sionalizing the above equation f2 may be obtained
Since there are no valves within the HPFP, the directly as
time derivative of area d r o p out of the nominal
J
model. To determine the shaft and friction forcing f2 = q$ M(pa-To) SFp$;p .
5
 ,
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

Figure 3: SSME HPFP map.

The nominal model for the HPFP is then given by which in turn in steady state and nondimensional
variables implies that

Mi - Mo = 0 .

- [ &-;;
?J*
fl
SF*.%
Consequently, fl drops out as well.
Therefore, under incompressible isentropic con-
stant area flow conditions, only one map fz n e d
Some additional steady state experiments where be found to determine the unknown forcing. The
, input and output temperatures are measured must RTM, DTM and MARSYAS models consider the
be carried out in order to obtain f3. After con- flow through the HPFP as incompressible and adi-
structing fs the fact that at steady state the mass abatic. In light of the above first principles analy-
flow at the input must equal the mas8 flow at the sis, it would seem that additional assumptions of
output is used to relate fi to fz and fa. Thus, one isentropic flow and no area change were applied
additional map f3 must be found to complete the as well. However, steady state data available for
nominal model of the SSME HPFP. the SSME shows that in fact there are temperature
Although it seems that more steady state infor- rises in the ducts leading to and from the HPFP
mation is needed to actually complete the HPFP 1171. Most likely, there is a temperature rise across
nominal model compared to existing models, we the pump as well. Yet, no equations characterizing
now show that the model developed here is the gen- change in temperature are given in the RTM for liq-
eral case while existing models employ some addi- uid flow. Temperature enters the calculations only
tional a s m p t i o n s . Indeed, recall that no initial as- when gaseous flows are assumed. Even if the tem-
sumptions regarding compressibility or friction were perature rises were lumped in the ducts surrounding
made for the models developed in [6,13].Now, sup the HPFP, the model does not provide a means of
pose that incompressible isentropic flow is assumed. calculating these temperature changes.
In this case there is no friction or heat transfer so Apparently there exists the following inconais-
that fw and Q are zero. The energy equation in tency. If incompressible and adiabatic conditions
steady state then reduces to prevail then the pump map accounts for both shaft
force and friction. However, a means to account for
-d?= o . the temperature rise due to the friction is not pro-
d6 vided. If incompressible and isentropic conditions
It can be shown in this case that f3 = 0 [12]. prevail then the pump map accounts solely for the
If there is no area change then the incompressible shaft force and from the isentropic assumption, the
continuity equation implies that inlet to outlet temperature change should be zero.
However, as mentioned above, a rise in temperature
-dir= o , is known to exist.
di
 So it is seen that under the assumptions of in- correction factors, it is crucial to obtain the actual
compressible and isentropic flow, the steady state lengths and areas in order to match both steady
~
maps used in models such as the RTM, are suffi- state and dynamic HPFP behavior. Unfortunately,
cient to complete the nominal HPFP model devel- attempts to glean geometry information from the
oped here. In other words, models like the RTM RTM, DTM and MARSYAS models have failed as
can be viewed as special cases of the more general the information is buried so deeply or combined
model with respect to the assumptions made for the with other parameters that extraction of the necw
forcing functions. sary lengths and areas is impoeaible. Once a work-
We consider now the validity of the incompress- ing model is up and running, it wiIl be difficult to
ible assumption made in the RTM and MARSYAS compare to the DTM HPFP model because it is un-
HPFP models. Taking pressures and tempera- dear how to isolate the HPFP component. Again,
tures from a 109% RPL SSME schematic at a p this is a reflection of the rather vague component
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

proximately the inlet and outlet of the HPFP, and boundaries typical of the DTM. However, compari-
the steady state flow through the HPFP from the son of the simulations derived herein against actual
RLOOOOI document, some rough Mach number and SSME hotfire data are currently underway.
density calculations can be made. The density and The next section discusses how certain anomalous
speed of sound are calculated using the EOS devel- conditions can be modelled using the uncertainty
oped in this work. A 17% decrease in Mach number and signal representations.
from inlet to outlet is observed. Moreover, calcula-
tions show roughly a 45% increase in density from
inlet to outlet. By most engineering standards such 4 Anomalous Model
a large gradient must be taken into account. Ac-
cording to [2], only density changes ofnot more than In order to apply model based condition monitor-

though the Mach numbers are low -

5% warrant the incompressibility assumption. Al-
.05,it must
be remembered that the substance is a lqutd, not a
ing, there must exist a way to incorporate anoma-
lous conditions into the system model [12, 14, 151.
As pointed out in Section 1, the empirical approach
gas, so that the standard practice in gas dynamics of to model development may not yield a model capa-
equating low Mach numbers with incompressibility ble of matching anomalous hot fire tart data. Such
1
no longer applies. models need ‘tuning“ via special “corrective fac-
In many situations, liquids can be considered to tors” to match nominal hot fire test data. A model
be incompressible because they are not subjected tuned to reflect nominal conditions becomes %out
to extreme pressures. In the case of waterhammer, of tune” to anomalous conditions. Typically, the
however, where pressure quickly rises proportion- “corrective factors” are an amalgamation of indi-
ally to the speed of sound [E, 8, 181, compress- vidual system properties. Once formed, however, it
ibility effects are considered. Similarly, in the case is d i c u l t extricate the original properties from the
of the HPFP, LHz undergoes a tremendous pres- factors so that the factors have no physical mean-
sure increase as it flows through the HPFP. From ing. Consequently, in tweaking the factora to match
this analogy, assuming most liquids exhibit simi- steady state, there is no way of discerning which
lar thermodynamic and fluid dynamic behavior, a physical properties are changing.
considerable change in density is to be expected. The control and condition monitoring based ap-
It seems that the more general case which includes proach allows nominal models to be augmented, in
compressibility effects should he used to account for a rigorous manner, with portions describing anoma-
the temperature rise and density change seen across lous behavior. The signal and uncertainty represen-
the SSME HPFP. tations of anomalous conditions are applied here to
In addition to lack of temperature input/output the HPFP.
information, other problems concerning geometric Signal representation involves the injection of an
parametera and model comparison were encoun- extraneous signal at the system input or output
tered. For 2, the effective length of the pump, care characterizing the anomaly. Figure 5 illustrates sig-
must be taken to account for the helical path the nal representation of anomalous conditions. Inject-
fluid follows as it flows through pump. This can be ing a signal of zero magnitude reflects nominal b e
accomplished provided that the axial pump length havior.
is known. The correct inlet and outlet areas must Uncertainty representation treats anomalous con-
also be obtained. Because of the rigorous first prin- ditions as a dynamic block multiplicatively or addi-
4
ciples based modelling method that utilizes no extra tively augmented to the nominal model. Figure 6

7
 . . ..
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

I
0

Figure 4: SSME full power'level schematic

Figure 5: Signal Representation

Figure 6: Uncertainty Representation

8
 depicts the three ways the delta block can be aug- at the inlet volute. Since temperatures and pres-
mented to the nominal model. The delta block con- s u r e are the same then it follows that the densities
tains dynamic representations of anomalous condi- should also be the same. From these assumptions
tions and reflects the uncertainty about the system the following equations hold:
behavior introduced by the anomalies. A zero value
for delta reduces to the nominal model. The choice DWo = D W & iDW,,
of scheme depends on the type of anomalous con-
or by expanding the flow terms,
dition. If an anomaly has some dynamics ssaoci-
ated with it, then the uncertainty representation is h&k- = &ioA&6& - - hiAlk6lk
appropriate. If the anomaly manifests itself as an
O = P&=flk
P
extraneous signal independent of the inputs, then
the signal representation is applicable. The basic To = ?&do=&
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

approach of the two schemes is illustrated below in h = p&=mk

next three sections by some specific cases of HPFP A. = A&.
failure modes taken from SSME failure documenta-
tion [l]. Because the leak can be treated as an orifice, the
well-known orifice equation [16]
4.1 Pump Speed Disturbance
Consider a disturbance, 6 . 7 in ~ the
~ ~pump speed,
6lk /F
-
= cdl -(PO pa) (11)

S p p , so that the actual input to the pump becomes applies. The dimensionless orifice coefficient is de-
fined as
= SFP l

/-
SFP 65FP 9
Cdl :=
where S p p is the nominal speed. As the disturbance
is assumed independent of the input nominal speed,
where c, is the contraction coefficient which de-
it is seen that 6 S ~ pcan be modelled as a signal
pends on the type of orifice. Here we assume a
injected at the input to the nominal model as shown
-I sharp edged orifice. In this caae, c, is taken from
in the left part of Figure 5. Note that j p p appears
[22] to be c. = .62. The pressure, pa,is the ambi-
as one of the external inputs to the steady state
ent pressure present on the other side of the leak.
map, fa. Simplifying the above continuity equation yields

4.2 Fuel Leak i i Q = B & , + B 1 r -Ar

-.
A0
Fuel leaks caused by seal degradation can lead to
problems as serious as complete engine failure. Con- Substituting the orifice Equation 11 into this equa-
sider a damaged seal at the input volute that es- tion, nondimensionalizing and simplifying we have
tablishes another flow path through the leak. The
region around this leak becomes another element in
the model. The dimensional inputs to this element where
consist of the inlet volute pressure just upstream
from the leak, PO,the inlet volute temperaturejust
upstream of the leak, TO, and the inlet flow, DWo. In order to use Equation 12 in conjunction with the
The outputs include the pressure across the inlet nominal model, the derivative with respect to time
volute, F&, the pressure across the leak, &, the must be taken and the result substituted in for Mo.
temperature at the inlet volute, !&, the tempera- The most general case includes both a time vary-
ture acros8 the leak, &, the flow through the leak, ing leak area, imbedded in the discharge coefficient
D W a , and the flow just downstream of the inlet, Cd, and time varying input preeaure, PO. The time
OW&. derivative of Equation 12 is given below
As the split in the flow occurs over a very short
d d d
axial distance, the dynamics of this component can -M&
dt = -Mo
dt
- -Po
be residualized. Furthermore, it is assumed that the 2(PO - Pa) dt
fluid properties (pressure and temperature) just up
/
stream of the leak are equal to the fluid properties

9
 To complete the leaky model, Equation 13 is aug- a constant area duct, the generic model equations
mented to the nominal model by an additional dy- show qualitative fidelity to the linearized continuity,
namic block consisting of momentum and energy PDE’s.
Several important points arise in the application
of the generic model equations to the SSME HPFP.
First, the steady state maps used in some exist-
ing SSME HPFP models are not enough to cover
all the forcing t e r n in the model equations used
here. This is a direct consequence of the fact that
Note that the Mach number associated with the leak the those existing models assume incompresaibility.
is given by MIX= c d d m . Substituting this In this sense, the existing models are special cases
relation into the A equation yields of the models developed in these two works, which
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

mume the most general case of compressible flow.

Furthermore, because the density change across the
pump in roughly calculated to be over 30%, the va-
This anomalous behavior is modelled using the un- lidity of the incompressible assumption in highly
certainty representation. The fundamental parame- questionable. In order to identify another forcing
ter determining the amount of uncertainty is based term for our model, some additional steady state
upon the size of the leak area, AI&. When the input/output temperature data must be obtained.
leak area is zero, Cd -, 0 and the anomalous block, These data are readily available for the turbine por-
A., reduces to zero restoring the nominal model. tion of the HPFTP but not the pump portion. Ge-
The above two examples illustrate how anomalous ometry of the HPFP such as actual area and axial
conditions can be modelled as additional portions length are key for calculating effective values needed
augmented to the nominal model. The complete in for the model. Unfortunately, gleaning these val-
anomalous model becomes the aggregate of the ues from the RTM and DTM has proved impossi-
anomalous portion and the nominal portion. ble because they are buried in the various “correc-
tion factors”. Finally, it may be necessary to model
the entire fuel side of the SSME for comparison to
4.3 Inlet/Outlet Volute Blockage the existing models because extraction of individual
Blockage of inlet or outlet volutes can lead to perf‘or- components from these existing models may not be
m a c e degradation. Clearly, volute blodrages cause feasible.
a decrease in the effective volute areas. Consider a A few examples of how anomalous behavior may
HPFP inlet volute blockage. The inlet area, &, can be incorporated are discussed. Using the FMEA-
be thought of as a parameter that normally remains CIL document [I] as reference, likely failures such
k e d . Assuming that the blodrage builds up over as fuel leaks, loss of pump performance (speed) and
time so that A. keeps decreasing, implies that the volute blockage are cast as either dynamic uncer-
temporal derivative of the area, $Ao, is no longer tainty or exogenous signal type anomalies. By r e p
aero. Therefore additional dynamics due to area resenting anomalies in these two ways, the nominal
change are introduced using the uncertainty repre- models can be easily augmented to generate anoma-
sentation. Some form for $A0 must be assumed. It lous models.
seen that with the appearance of the $A0 term, the The following points are open to further study:
0 matrix comes into play. Setting the $A0 term to
zero restores the nominal model. Effects of discretization procedures,

Use of nominal and anomalous models in the

5 Conclusions and Future design of integrated condition monitoring and
Work control systems,

The control based approach to modelling propul- Validation of nominal and anomolous models
sion systems developed in the Part I companion pa- against SSME hotfire data, and
per for the nominal case is validated, applied to the
SSME HPFP and extended to include the anoma- Investigation of the various trade-0% incurred
lous case. As demonstrated in a simple example in the design of an integrated control and con-
/ wherein the generic model equations are applied to dition monitoring system.

10
 References [14] A. Helmicki, S. Jawed, and K. Kolcio. An
integrated approach to rocket conditon mon-
[l] SSME failure modes and effects analysis and itoring and control. In Fourth Annual Space
critical items. Technical report, Rockwell In- System Rea& Management Technology Con-
ternational, Rocketdyne Division, may 1986. ference, November 1992.
[2] J. Anderson. Intmdudion lo Flight. McGraw- [15] A. Helmicki, F. Kuo, and D. Vallely. Rocket en-
Hill Book Co., 3rd edition, 1989. gine health monitoring and control: Some con-
nectiona and their implications. In Pmeedings
[3] J. Anderson. Modern Compressible Flow. 3rd Annual Health Monitoring Conf. for Space
McGraw-Hill, 1990. Propulsion Syatem, Cincinnati, Ohio,Novem-
her 1991.
[4] 0.Badmus, S. Chowdhury, K. Everker, and
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228

C. Nett. Control-oriented high-frequency tur- [16] H. Merritt. Hydraulic Control System. John
bomachinery modeling: Single-stage compres- WJey and Sons, 1967.
sion system 1D model. In ASME International
Gas Zbrbine and Aemengine Congress and Ez- [17] D. Nguyen. Engine balance and dynamic
position, 1993. ASME 93-GT-385. model. TerMcal report, Rockwell Interna-
tional, Rocketdyne Division, Canoga Pk., Ca,
[5] 0.Badmus, K. Eveker, and C. Nett. Control- 1981.
oriented high-frequency turbomachinery mod-
eling: General 1D model development. In [18] J. Parmakian. Waterhammer Analysis. Pren-
ASME International Gas Turbine and Aem- tice Hall, 1955.
engine Congress and Ezposilwn, 1993. ASME
[19] A. Shapiro. The Dynamics and Tlremody-
93-GT-18.
nomic8 of compressible Fluid plow, volume I.
[a] 0.Badmus, K. Eveker, and C. Nett. Control- Ronald Press Co., 1953.
oriented high-frequency turbomachinery mod-
[20] J. Tiller. MARSYAS SSME model. Technical
eling: Theoretical foundations. In 1992 Joint
report, BCSS, MSFC, Huntsville, Al, 1991.
Propubion Conference, 1993. AIAA paper no.
92-3314. [21] D.Tritton. Physical Pluid Dynamics. Oxford
University Press, 1988.
[7] A. Butkovskiy. Green’s h n c t i o m and Tram-
fer Functiotu Handbook. Ellis Horwood LTD, [22] J. Vennard and R. Street. Ekmentary Fluid
1982. Mechanics. John Wiley and Sons, 1976.
[E] M. Chaudry. Applied Hydraulic ‘Ihann’ents.
Van Norstrand Reinhold Co., 1979.
[9] J. Doyle, B. Francis, and A. Tannenbaum.
Feedback Control Theory. MacMillan, 1992.
[lo] G. Emanuel. Gasdynamics: Theory and Appli-
cationr. Education Series 1. AIAA, 1986.
[ll] G. Emanuel. Advanced Classical Tkrmcdy-
namics. AIAA Education Series, 1987.
[12] A. Helmidri. Aspects of model-based rocket
engine condition monitoring and control. fi-
nal report, Marshall Space Flight Center, May
1994.
[13] A. Helmicki, S. Jawced, and I(. Kolcio. Liq-
uid rocket engine modeling for control and con-
dition monitoring: Part I, theoretical foun-
dations. To be presented at the 1994 Joint
Propulsion Conference.

11
 This article has been cited by:

1. Irem Tumer, Anupa Bajwa. A survey of aircraft engine health monitoring systems . [Citation] [PDF] [PDF Plus]
Downloaded by NORTHWESTERN POLYTECHICAL UNIV. on October 10, 2022 | http://arc.aiaa.org | DOI: 10.2514/6.1994-3228