Sie sind auf Seite 1von 15

Synthesis of Feedback Controls Using Optimization

Theory-An Example”

Summary-This paper illustrates the use of optimization theory lem, a developmentof a suitable linear aircraft model,a
in the synthesisof a linear time-varying feedback control by carrying specification of the performance requirements and con-
out the designof an aircraft landing system. The optimization method
employed is the ParametricExpansion Method.
straints, the formulation of a mathematical error index,
A number of d a e r e n t controls aresynthesized by selecting a brief review of the basic concepts of the Parametric
different functional forms for the weighting factors appearingin the Expansion Method, the synthesis of a number of dif-
error index formulated from the performance requirements. These ferentaircraftlandings)-stems,andfinally a brief
controls are compared by presenting the landing trajectories of the description of one possible mechanization of the most
aircraft.
suitablesystem.
I. IKTRODUCTIOS
11. -AIRCRAFT LAXDING
PROBLEM DEFINITIOS
-4XI- COSTROLproblemsarecharacterized
AM blr the necessityforsatisfyingmultipleper-
formance requirementsand
constraints.In
such cases, the sJ-nthesisof the control by conventional
The landing of an aircraft consists of several phases.
First,theaircraft is guidedtowardtheairportwith
approxirnatelp the correct heading by R D F equipment.
trial-and-error methods may proveexceedingly difficult. \Yithin a few miles of the airport, radio contactis made
I f , in addition, the performance requirements are such with the radio beam of the instrument landing system
that the needfor a time-varying control is indicated, (ILS). In following this beam, the pilot guides the air-
the control engineer is faced with the difficult task of craftalongaglidepathangle of approximately -3”
selecting not onl!- the configuration of the control but toward the runwa!-. Finally, at an altitude of approxi-
also the values of its time-varJ-ing parameters or gains. matell- 100 feet, the flare-out phase of the landing be-
Forsuchproblems,sJmthesismethodsbasedonopti- gins.Duringthis final phase of thelanding,theILS
mizationtheory offer distinctadvantagesovertrial- radio beam is no longer effective for guiding the aircraft
and-errormethods. In particular,theselection of the duetoelectromagneticdisturbances. S o r is the -3”
time-varying parameters or gains is replaced essentially glide path angle particularly desirable from the \ 71ew-

by the selection of constant parameters which are the point of safet\- and comfort. Hence, the pilot must guide
magnitudes of the weighting factors. These weighting theaircraftalongthedesiredflare-path bJ7 making
factorsappear i n an error index formulated from the visual contact with the ground.
performancerequirements.Inaddition,theabsolute The landingproblem described here is concerned on117
stabilitJ- of thesystemisnot a consideration in the with the final phase of the landing, thatis, with the last
selection of the weighting factors, whereas, in the selec- 100 feet of the aircraft’s descent. I t is assumed that the
tion of time-varyinggainsbyconventionalmethods, aircraft is guided to the proper location by air traffic
consideration of absolutestability would assumepri- control, and that its altitude and rate of ascent at the
mary importance. beginning of this flare-out phase may range from 80 t o
This paper is intended to be tutorial i n nature and its 120 feet and - 16 feet/sec to -24 feetisec, respectively.
purpose is to illustrate the use of optimization theory For values outside of this range, i t is assumed that the
in the sbmthesis of a linear time-varying feedback con- aircraft is waved off. Finall!., i t is assumed that only
trol for a processwhich is characterizedby difficult thelongitudinalmotion(motion in averticalplane)
dJmamics and is subjected to multiple performance re- need beconsidered in this final phase of the landing.
quirementsandconstraints.This is accomplished bs- Lateral motion of the aircraft is required primarily to
carrying out the design of an aircraft landing sJ.stem point the aircraft down the runwaJ-. For the most part,
using the Parametric Expansion I l e t h ~ d . l -Included
~ in thislateralmotionisaccomplishedprior tothe final
this paper are a definition of the aircraft landing prob- flare-out phase of the landing.
During the Hare-out, the aircraft may be subjected to
* ReceivedDecember 27, 1962. Thispaper \vas originall>-pre- both stead!. winds and wind gusts. \!‘ind gusts are of
scntcd at the Joint A4utornatic Control Conference. New York Uni- primary importance since they tend to be random. On
versity, Kew York, June 27-29, 1962.
i- General Electric Company, Schenectady, X. I;. the other hand, steady windsparallel to the ground can
C. \V. hlerriam, 111, “Use of a mathematical error criterion in be counteracted bs- merely a stead>--state change in the
t h e design of adaptivecontrols\-stems,” Trans. A I E E , vol. 78
(-4tpZ.and I d . , pt. I I j , pp. 506-512; January, 1960. heading of the aircraft. I n this problem i t is assumed
- C. 11.. Merriam, 111, “Anoptimizationtheory forfeedback that the windgustsare zero. The winddisturbances
control syxem design,’’ In,fonzation md Confiol, x-01. 3 , pp. 32-59;
March, 1960. could be treated, but this would require a discussion of
C. 1V. Merriam, 111, “Synthesis of Xdaptlve Controls,“ Sc.D. the statistical formulation of the Parametric Expansion
dissertation,klassachusettsInstitute of Technology,Cambridge;
May, 1958. Method which is be>-ond the scope of this paper.
90 IEEE T R A N S A C T I O N S ON AUTOMATIC C O N T R O L April

111. EQVATIOKS

\ i
AIRCJUFT OF M O T I O N
4 first step in the synthesis of a control system is the
development of a suitable mathematical description of
the process or plant to be controlled. This description
may take the form of a differential equation of some
order,aset of first-orderdifferentialequations, or a HORIZONTAL
transfer function. In this landing problem, a description
of the aircraft relating the response (output) variables,
/ AIRPLANE
measurable state variables, and control (input) varia- /
/
blesisnecessary. 2
In the past, a great deal of attention hasbeen directed
toward the development of the equations of motion of
an aircraft.j This development proceeds from a consid-
\
eration of theaerodynamicforcesandmoments,and \
from the application of the fundamental laws of
mechanics. The resulting equations are then linearized Fig. 1-Definition of aircraft coordinates and angles.
based on the assumption that the deviation from the
equilibriumflightcondition is small. Inaddition,the
T o complete the description of the aircraft, an equa-
assumption often is made that the glide path angle y is
tion relating pitch rate 0’ and altitude, or a time deriva-
sufficientlysmall so that the smallangleapprosima- tive of altitude. is required. Such a relationship exists
tions, sin y =y and cos y = 1 can be made. This assump- between vertical acceleration and pitch rate and is given
tion is valid in this case due to the landing geometry.
bl-
Finall!,, i t is assumed for this problem that the velocity
I.’ of theaircraft is maintainedessentiallyconstant
during the landing by utilizing throttle control. Thus,
the longitudinal motion of the aircraft is go\.erned en- For this problem, the velocity l r of the aircraft is as-
tirely by the elevator deflection S,(t) and this becomes sumed to be constant during the flare-out a t a value of
the onll- control signal. The use of these assumptions 256 ft/sec. Soting that
leads to the so-called short period equations of motion
of theaircraft.Theseequationscanberewritten in 1
Iz(s) = - /Z”(S)
terms of the following transfer function relating elevator S2
deflection, 6,, and pitch rate, 0’.
the over-all aircraft transfer function relating altitude
and elevator deflection is given by

Throughout this paper, the prime (’j is used to denote the


time derivative of the variable to the left of the prime.
Thus, in (1) the symbole’(s) denotes the Laplace Trans- Thisrelationshipalsocanbewritten in terms of a
form of the pitch rate. The pitch angle 0 and the stability fourth-order linear differential equation as follows
axes x and z of the aircraft aredefined in Fig. 1.
The aircraft parameters K,, T,, w,, and { often are
referred to as follows:
K , =short period gain, Eq. 5 describes the behavior of the aircraft in terms of
T , = path time constant, the four state signals h ( t ) , h’(t), Iz”(t), h”’(t). This fact
w, =short period resonant frequency,
can be demonstrated by drawing from this equation a
block diagram of theaircraftcontainingfourinte-
[=short period damping factor.
grators, with &(t) as the input to the diagram and h(t)
These parameters are assumed to be time invariant and as the output. The outputs of the four integrators are
the numerical values used in this study are: the state signals. The derivative Iz””(t) is not a state
K , = - 0.95 sec-l, signal; it appears as the input to the first integrator in
T, = 2.5 sec, the block diagram.
w, = 1 radian/sec, Because the state signals ultimately appear as meas-
{ = 0.5. ured variables in the feedback portion of the optimum
control system, the state signals selected to describe the
C. D. Perkinsand R. E. Hage,“AirplanePerformanceSta- process must be measurable with available instrumenta-
bility and Control,” John IYiIey and Sons, Inc., S e w York, S . y.,
pp. 374-407; 1949. tion in order to construct the s!-stem. For instance, as
1963 Ellert
Merriam:
and Feedback Controls
Using
Optimization. Theory-an Example 91

indicated above, a complete set of state signals for the


aircraft is h(t), h’(t),R”(t) and h”‘(t). The altitude h ( t )
can be measured wih a radar altimeter, and therate-of-
ascent h’(t) canbemeasuredwithabarometric-rate-
meter. However, the derivative of acceleration h”’(t)
is not readily measurable. Another complete set of state
signals is h ( t ) , h’(t), O ( t ) , O ’ ( t ) . Pitch angle O(t) and pitch
rate O ’ ( t ) are readily measurable with gyros. Hence this
second set of state signals is preferable. T o use this set,
therelationshipsbetween O ’ ( t ) , O ( t ) , k”’(t), and h”(t)
must be utilized in order to obtain from ( 5 ) an equation
in terms of the new state signals. The equations relating
h”(t), h”’(t), O ’ ( t ) , and O ( t ) are found from (2) to be
Fig. 2-Aircraft block diagram in terms of measurable state signals.
v 1
rtf’(t) = - e ( t ) - - lZ’(l) (6)
TS TS
Eq. 12 may be written as asingle equation using matrix
v 1 notationbyletting
~ ’ ( t =) - e’(t) - - h”(t) ( 7)
TS TS
I.,’ 1
Ir””(t) = -O”(t) - - h”’(t).
T, TS
Substituting (6), ( i ) ,and (8) into (5) gives

Eq. 9 also is a valid description of the behavior of the


aircraft.
A block diagram for the aircraft may be drawn from
Lo J
(6) and (9). This is shown in Fig. 2. Since the aircraft T h u s , (12) becomes
description is of fourth order, this diagram contains four
integrators, and the output of each integrator is a state
x’(t) = Ex([) C m ( t ) . +
(1’i)
signal. Some authors refer to this as the “equation of state” of
Incarryingoutthesynthesis of thesystem using the dynamic process. For this aircraft landing problem,
optimizationtheory, it isconvenienttodescribethe i t completely describes the behavior of the aircraft in
behavior of the aircraft in terms of four first-order dif- response to the control signal.I n the more general case,
ferential equations rather than a single higher order dif- an additional term usually appears in this equation to
ferential equation. This is accomplished using (6) and include the effects of disturbances on the response of
(9) by assigning new symbols to the four state signals the process.
and the control signal. Thus, letting
1V. PERFORMAKCE REQUIREMENTS A N D COSSTRAINTS

Xn aircraftlandingsystemissatisfactoryonly if
and
certain prescribed performance requirements and con-
straintsare satisfied.Oftenthesearedescribedin
terms of desiredresponsesignals,desiredcontrol sig-
nals, and in terms of limits on these signals. The follow-
ing requirements and constraints are considered to be
of primary importance and are treated in this problem.
1) T h e desired altitude J t d ( t ) of the aircraft at each
X3’(t) = b32X*(f) + b33X3(t)
instant of time during the landing is described by the
xq’(t) = x3(t). flare-path shown in Fig. 3(a). I t consists of an exponen-
92 IEEE TRAMSACTI0,Y.S O X AUTOLWATIC C O N T R O L April

3) The pitch angle O ( t ) of the aircraft at the desired


touchdown time t = T must lie between 0" and +lo".
T h a t is
oo 5 e(T) 5 + 10". (2 0 )
The lower limit is necessarv to prevent the nose wheel

6ol
\
40

20-t FOR I S 5 t C 20
of an aircraft with a tricycle landing gear from touching
downfirst. The upper limit on the pitch angle is
quiredtopreventthe
first.
tail gearfromtouchingdown
re-

4) During the landing, the angle of attack a ( t ) must


remain below the stall value. For this problem, the stall
value was assumed to be +IS". The aircraft enters the
SECONDS
flare-out phase in equilibrium with an angle of attack
GROUND PLANE
of approximately SO per cent of the stall value. Hence,
the permissible positive increment in the angle of attack
during the flare-out is equal to 20 per cent of the stall

I
5
I
(a!

IO
I
I5

f
20
'
- t

SECONDS
value. Thus, the restrictions on the angle of attack are
cy(/) < 18"
Am(/) < 3.6".
(2 1)
(22)
- IO
5) The elevator, which controls the longitudinal be-
-20s -r havior of the aircraft, is restricted to motion between
DESIRED - 20 FOR 0 5 tL-15 mechanical stops. For this problem i t is assumed that
RATE OF
ASCENT SECOND these stops exist at -35" and +15" (elevator trailing
hi(t)
edgedown).Hence,duringtheflare-out,theelevator
deflection 6,(t) is constrained between these two physi-
( c ) Pitch angle a t touchdown: O " g ( T ) < 10'. cally limiting values. T h a t is,
(d) Xngle of attack: a(t1<18".
(e) Ele\-ator deflection: --3S056,(1)5+1S".
-35" 5 6,(t) + 15'. (23)
Fig. 3-Performance requirements and ronstraints.
For alinearcontroller,theelevator is permittedto
tial function of time followed by a linear function. This attain these limiting values instantaneously but not to
desired path is described by exceed them. That is, for linear operation, the elevator
is not permitted to operate against the mechanical stops
for nonzero periods of time during the landing. In the
s!-nthesis of thisaircraftlandingsl-stem,saturation
effects are not treated.
An exponential-linear path of this form ensures a safe
and comfortable landing. The desired duration of the V. >I.XTHEMATIC=\L IXDEX
ERROR
flare-out is 20 sec,including 5 sec overtherunu-a?-. An important step in the synthesis of a control sys-
This value is appropriate for an aircraft flJ-ing a t 175 tem b\- using optimization theor\- is the formulation of
mph and beginning to flare-outat an altitudeof 100 feet. a mathematical error index. The error index is impor-
2) T h e desired rate of ascent h d ' ( f j of the aircraft is tant because it, to a large degree? determines the nature
given by the time derivativeof (18) and is shown in Fig. of the resulting optimum control. That is, the resulting
3(b). T h a t is, controlmaybelinear,nonlinear,stationary,ortime-
var\.ing depending on the form of the error index. Thus,
the control engineer is able to influence the nature of
the resulting system by the manner i n which he forrnu-
The magnitude of the rate of ascent of the aircraft is lates this index based on the requirements of the prob-
important primaril?- a t touchdown. -1value other than lem at hand. In general, these requirements may include
zero is desirable to prevent the aircraft fromfloating not onlJ- theperformancerequirementsbutalsoany
down the runway and, hence, perhaps overshooting it. restrictions on the form of the optimum control to en-
A very large negative value for the rate of ascent may sure phL-sical realizability.
overstress the landing gear and, thus, is equallJ- unde- -1first step in the formulation of an error index is the
sirable. The value of - 1 foot/sec at touchdown given establishment of an instantaneous error measure which
in (19) is equal to -60 feetiminute which is well below takesintoaccount all of theimportantperformance
the maximum permissible value for a modern aircraft. requirementsandconstraints.Ingeneral,theinstan-
1963 Ellert and Merriam:
Feedback
Controls Using Optimization
Theory-an
Example 93

taneouserrormeasureconsists of thesum of the cumulative effect of this instantaneous measure through-


weightedresponsesignal andcontrolsignalerrors. In out an intervalof time. Hence, the error indexe(t) often
thislandingproblem,theimportantresponsesignal is expressed as the time integral of the error measure
errors are: the deviation of the aircraft altitude from over a suitable future interval of time, t to T , through-
the desired altitude given in (18), the deviation of the out which the system performance isof interest. That is
aircraft rate of ascent from the desired rate of ascent
given in (19), the deviation of the pitch angle a t touch-
down from the desired value of +2" which is within the
S,
e(l) =
T
em(u)du (23

prescribed range given in (20), and the deviation of the where u is a dummy time variable. The lower limit of
angle of attack from its initial valuea t t = O . I n addition, integration a=t is present or real time. For the landing
the pitch rate of the aircraft may be important. The problem,thefutureinterval of time whichis of im-
aircraftequations of motionindicatethatthepitch portance is the time remaining before touchdown occurs.
rate of the aircraft governs the magnitude of the angle Hence, the upper limit of integration u = T is set equal
of attack since the pitch rate term enters the angle of to 20 sec. The error index then can be written,
attackequation as thedrivingfunction.Hence,itis
possible to control the magnitudes of both the angle of
attack and the pitch rate by introducing a weighted
e(O =
s, 2o
+
( d h ( U ) [ h d ( C J ) - JZ(U)]? +h,(u) [kd'(u) - h'(u)]*

pitch-rate-error term into the instantaneous error meas-


ure. Of course, it also is possible to include both pitch
+ +8(CJ)[ed(CJ) +
- e(d12 +s+)[ed'(u) -

rate and angle of attack error terms in thismeasure. + [Se(u)]2fdu. (26)


However, this requires the selection of two weighting
factors instead of one. I n this problem, the use of only The performance requirements indicate that the alti-
the weightedpitch-rate-errorterm mas foundto be tude and rate of ascent errors should be small at the
completely satisfactory for controlling the magnitudes desired touchdown point a = 20 to insure actual touch-
of both the angle of attack and the pitch rate. The de- down very close to this point. Large altitude and rateof
sired pitch rate &'(t) was selected to be zero. T h e final ascent errors a t u = 20 may result in touchdown prior to
term in the instantaneous error measure is the control the startof the runwaqr, or so far down the runway that
signal error, or,in this case, the elevator deflection error. the aircraft cannot be brought to a stop before the end
T o maintaintheelevator deflectionwithinthepre- of the runway. Hence,impulse weighting of these errors
scribed limits i t is convenient to select the value of the a t u = 20 in addition to weighting of the errors during
desired control signal to be zero. Thus, for the landing the flare-out appears to be desirable. Thus, +h(u) and
$ h . ( u ) can be written
problem an appropriate instantaneous error measure is
+
e m ( l ) = +h(f) [ h d ( f ) - J Z ( f ) ] ' 4 h f ( t ) [ h d ' ( t ) - k'(t)12
+
4 h ( U ) = +.i(u) + 4 , T U O ( 2 0 - u) (27)

+ +do +
[e&) - w I 2 + d t ) [ e m - s'(o12 and
+ [se(o12 (2 4) +hh'(C) =).(1+ + 43,TU0(20 - U). (2 8)

where & ( t ) , 4 h ' ( t ) , $ s ( t ) and ds,(t) are the time-varying Here uO(20-u) is an impulse of unit area occurring at
weighting factorswhich indicate the relative importance u =20 and &.T and &.T are the areas of the impulse
of the various terms in the error measure. I n carrying functions. Theselection of the time functions 4+(u)and
out the design of the optimum control, these weighting &(u) is treated in a later paragraph.
factorsmustbeselectedsuchthattheperformance Theperformancerequirementsindicatethatthe
requirements and constraints are satisfied. pitch angle error is important only at the touchdown
As already mentioned, the error measure may assume point. Hence, the functional form of the weighting factor
a variety of forms depending on the nature of the prob- 48(u) should be selected such that only the pitch angle
lem to be solved and the requirements to be met. For error at the desired touchdown point contributes to the
this landing problem, a weighted quadratic error meas- value of the error index. An impulse function exhibits
ure was selected becausei t was felt that thisform would the required form. Thus, 4e(a) can be written
insure satisfactory system behavior. In addition, theuse
M u ) = +2,TU0(20 - u). (29)
of a weighted quadratic error measure in conjunction
with the linearaircraftdescriptionresultsinalinear Since the angle of attack and the pitch rate are im-
controlsystemwithfeedbackandfeed-forwardgains portant throughout the entire landing interval, &(u)
which are not a function of the state of the aircraft but is written .
merelyfunctions of the time-to-go before touchdown.
Such a system is relatively simple to in~plementph!.si- +8'(U) = + l ( U ) . (30)
ca11y and is a practical airborne landing system. The performance requirements suggest the functional
The control engineer is interested not so much in an forms which &(u), &(a), and 41(u)should assume. Since
instantaneous value of the error measure ashe is in the altitude and pitch rate errors are important throughout
94 IEEE T R A N S A C T I O N S OAXp rACiU
lOTI rOTARPOALT I C

the flare-out, it is reasonable to make these weighting


factors constants. On the other hand the rate of ascent
errors are not of primarv interest prior to the start of
therunway.Thus, &(r) shouldbezeroprior to the
s t a r t of the runway and a constant over the runway.
Theresulting expressions for theseu-eightingfactors
are
= d4 (.3 1)

h’ (1)
ddn) = 91. (33)
T h e selection of t h e numerical values of &, + 4 , T , + 3 ,
$3,T, &.T and 41, andthe effects of thesevalueson
the resulting aircraft performance are discussed in Sec- Fig. 4-Block diagram of optimum aircraft landing system.
tion VII.
1,’I. SYSTEM SYNTHESIS vide the values of the time-varying input signal and
feedbackgains of thesystem.Theseequationsare
Having formulated the error index, the control en- derived in the -Appendix and are given in ( S t ) through
gineer endeavors to minimize the value of this index in (71).
such a manner that he obtains information about the Thus, this s?-nthesis procedure provides directly the
configuration andparameters of t h e controlsl-stem
configuration of the optimum s>-stem anda set of equa-
being
sl-nthesized. the
In
Parametric
Expansion tions which may be solved to obtain the system param-
Method, the error index is minimizedwithrespect to eter values.
the components of the control signal x-ector m ( f ) .In
general,thisleadsdirectlytoan expressionfor each VII. AIRCRAFT LANDING
SYSTEMS
control signal in terms of the measurable state signals As a result of the use of the Parametric Expansion
and time-varying coefficients. These coefficients can be Method, the synthesis procedure is reduced to the solu-
considered time-varying gains in the optimum control tion of the K-equations. In general, the solution of these
system.This expression for eachcontrolsignalis re- equations requires the use of an analog or digital com-
ferred to as the control law. In the case of the Para- puter. The systemsdescribed in this paper were synthe-
metric Expansion llethod, this law defines a feedback sized using an I B l I TO4 digitalcomputer.The com-
configuration for the s>.stem. As a result, having derived puter program consists of two major parts. In the first
the control law, the block diagram of the optimum SJ-S- part, the K-equations are solved. In the second part,
tem immediately can be drawn. The application of the the aircraft equationsof motion are solved to obtain the
Parametric Expansion ;\lethod to the aircraft landing actuallandingtrajectories. T h i s twopartprogramis
problem is described in the Appendix. -A more general illustrated in Fig. 5 . The integration algorithm used in
treatment of thissubjectincludingtherigorousde- this stud>- is the Gill modification of the Runge-Kutta
velopment of the principles involved may be found in fourth-orderextrapolation a l g ~ r i t h m . ~However, a
the technical number of algorithms possessing similar properties also
4 s indicated in the Appendix, the use of the Para- couldbeutilized.
metric Expansion 3,Iethod results in the following es- .As indicatedin Fig. 5, thelandingtrajectoriesare
pression for the optimum elevator deflection as a func- computed for three sets of four initial conditions. These
tion of time conditions are tabulatedi n Fig. 6. I t is assumed that the
6,(t) = w,?K,T,[kl(t)- kl,(t)e’(t) - k l , ( t ) s ( t ) initial vertical acceleration h f f ( 0 )is zero and the initial
altitude, vertical velocity, and pitch rate are as given
- k I 3 ( / ) h ’ ( t ) - k14(/jh(i)]. (34) in the table. Using these values, the initial pitch angle is
computed from the aircraft equations of motion.
This expression defines the configuration of the landing
Sets 1 and 3 of the initial conditions correspond to the
system. The block diagram of the system is dran-n with
worst possible combinations of the largest permissible
the aid of (34) and isshown in Fig. 4. The optimum
altitude and rateof ascent errors a t t h ebeginning of the
s>.stemcontainsfourfeedback loops with a measured
flare-out. T h a t is, in set 1 the aircraft is 20 per cent
state signalfedback in each loop. T h e gain i n each
above the desired landing trajectoq- and it is descend-
feedback loop is time-varying. The input to the system
also is time-varying.
T h e use of theParametricExpansionhfethodalso 5. Gill, “X process for the step-by-step integration of differential
equations in an automatic digital computing machine.’ Proc. Catnnb.
results in a set of K-equations which, when solved, pro- Phil. Soc., vol. 47, pt. 1, pp. 96-108; 1951.
1963 Ellert and Aferriam: Feedback Contyols Using Optimization Theo~y-an Example 95

PROGRAMINPUT
A I RCRAFT PARAMETERS
DESIRED RESPONSES
WEIGHING FACTORS
I N I T I ACLO N D I T I O N S

SOLUTION OF K-EQUATIONS
FROM
\ = O TO 1'20
3 BO - 24 - 0.0938 0

1
SOLUTIONOFAIRCRAFTEQUATIONS
Fig. 6-Aircraft initial conditions.
FOR 3 SETS OF I N I T I A LC O N D I T I O N S
FROM reduced to merely the selection of the magnitudes of a
t = O TO t = 20 set of weighting factors.
In this paper, three different setsof weighting factors
are selected to illustrate theeffects on the nature of the
+ resulting system and on the landing trajectories. These

1 OUTPUTPROGRAM I three cases are described in the paragraphs t h a t follow.

sf Altitude Error
I AIRCRAFL
TANDING
SYSTEM
GAINS
TRAJECTORIES I Case I-Constant VLeighting
andElevatorDeflection
In this case, altitude error and elevatordeflection are
Fig. 5-Simplified flow diagram of digital computer program. weighted by a constant over the entire 20-sec landing
interval, and all other weighting factors are zero. T h a t
ing 20 per cent slower than desired. Hence, the aircraft is,
is diverging from the desired trajectory at the largest +,,(u) = +4 = constant (36)
permissiblerate. A similarsituationoccursinset 3 $1 =z $?,T = $ 3 = $ 3 , T = 44.T = 0 - (37)
where the aircraft is below the desired trajectory and is
descending at the maximum permissible rate. In set 2 The error index becomes
the aircraftis on the desired trajectory andis descending
a t t h e r a t eprescribed by the performance requirements. +
e(t) = ~ t B { + ~ [ J z d-u )h ( u ) ] * [ & ( U ) ] ~ } ~ U . (38)
This wide range in initial conditions serves as a severe
test for the aircraft landing system. The value of c$~ is determined by a consideration of the
As described in the -Appendix, the solution of the K - relative importance of altitude errors and elevator de-
equations cannot be carried out until the desired re- flection. -At the beginning of the flare-out, the largest
sponses, aircraft parameters, and weighting factors are positive altitude error is20 feet due to the initial condi-
specified. I n the preceding sections, the desired responses tions. The landing system will attempt to correct this
and the aircraft parameters have already been specified. error by calling for a large positive elevator deflection.
Hence,the design procedurereducestomerelythe The largest permissiblepositiveelevatordeflection is
selection of a set of weighting factors which will result l5", or 0.262 radian. For this case, altitude errors and
in satisfactory system performance for the three sets of elevator deflection are assumed of equal importance in
initial conditions. the performance of the landing system. Therefore, the
There are two important aspects associated with each two terms in the integrand of the error index of (38)
weighting factor 4 ( u ) : its functional form f(c) and its should contribute equally to the value of thisindex.
magnitude a. T h a t is, ~ ( c T ) be written
may Hence, the value of 9 4 may be calculated by equating
these two terms and inserting the above altitude error
+(u) = W U ) . (35)
and elevator deflection values. T h a t is,
The functional formf(a)usually is suggested by the per- +4[2012 = [0.262]? (39)
formance requirements for the system and, hence, only
themagnituderemainstobe selected. Forexample, or
the pitch angle requirement at touchdown has led to
$4 = 0.000171. (40)
the selection of a n impulse function as the functional
form of +~(cT). However, its magnitude &,T is yet to be Theperformance of theaircraft for 4~=0.0001 is
specified. Thus,thesynthesisprocedure ultimatelqr is shown in Fig. 7. This figure contains four graphs. From
96 IEEE T R A h T S A C T I O N S O N A U T OCM
OANTTIRCO L April

+0.3
+0.2 Q3 * 0.0 e 3 , ~~ 0 . 0
klltl

m 2 .0.0 02,r= 0.0


0,'OO o,,,=o.o
0,=0.0001
=:
m,,,=oo
,

I I I I I I

+2 0

Fig. 8-System gains for Case I.

Before proceeding to another error index, it is worth-


Fig. ?-System performance for Case I. while to consider the feedback gains and the input sig-
nal associated with the weighting factors and perform-
top to bottomthese are: ance curves of Fig. 7 . These gains are shown in Fig. 8.
The input signal k , ( t ) appears to be similar in shape to
1) altitude error vs time;
the desired altitude h d ( t ) although substantially reduced
2) rate of ascent error vs time;
in amplitude. The feedback gains k , , ( t ) , k I z ( t ) , k13(t) and
3) angle of attack vs time;
k I 4 ( t )are essentially constant until the last three secs of
4) elevator deflection vs time.
thelandingwhentheygraduallydecreaseto zero.
Each graph, in turn, contains three curves, one for each Hence, just prior to touchdown, the landing system is
of the three sets of initial conditions. I n addition, the operatingessentiallyopenloop.Thisshouldcauseno
values of thepitchangle O(T) a t t = 2 0 secaretabu- concern since it is not possible to alter significantly the
lated at t h e bottom of the figure for the three sets of trajectory of the aircraft during this short interval with
initial conditions. T h e O(T) values correspond, from top the finite control effort available due to the lags in the
to bottom, to the three sets of initial conditions given aircraft.
in Fig. 6. The values of the weighting factors also are
listedonFig. i . Case 11-Impulse W e i g h t i n g of d Ititude
For this figure, eh and e h , are given b!- Error at Touchdown
e,,(/) = [/z(t) - hd(t)](feet) (11) A great deal has been written in the literature about
so-called terminalcontrol systems.6-J Such a system
and attempts to reduce an error to zero a t a future point
eh.(t) = [ ~ z ' ( t )- h'd(f)] (feet,.'sec). (42) in time called the terminal point. X system with this
propertyalsocanbesynthesized using optimization
When t h e aircraft is at an altitudeof 120 feet a t t h e be- theor)- by selecting the weighting factor associated with
ginning of the flare-outl eh=20 feet. The rate of ascent this undesirable future error to be an impulse. Hence,
of the aircraft, for this same condition, is -16 feetisec if theonlyperformancerequirement for thislanding
giving e h , = 4 feet/sec. .Also a t t = 0 , t h e angle of attack system were t o reduce the altitude error at the desired
a is at its equilibrium value of +14.4' (80 per cent of touchdown point to the smallest possible value without
the stall value) for all three sets of initial conditions. exceeding the elevator deflection limits, a suitable error
The set of weighting factors listed in Fig. 7 results in index might be
a system which meets the performance requirements on
altitude, rate of ascent, and elevator deflection. How- { $'e,i"?dn(20 - 0 ) [hd(U) - h(U)]'
ever, the constraints on angle of attack and pitch angle
at touchdown are violated.
Anotheraspect of theperformance is notevident
+ [&(u)]qdu. (43)
from these figures due to the ordinate scale selected for 6 R-I. V. LIatheu-s and C. \V.Steeg, "Terminal Controller Synthe-

eh,. The rate-of-ascent error a t t = 20 varies substantiall!- sis," presented at the Princeton Symp. on Son-Linear Control Sys-
tems, Princeton, S . J.; hIarch 2 6 2 7 , 1956.
with the initial conditions of the aircraft. Hence, the R. C. Booton, Jr., "Optimum design of final-value control sys-
point at which touchdown actually occurs will vary with tems," Proc. PolJtteclznic Institute of Brooklyn S p p . on :Von-Linear
Circuit Analysis, Brooklyn, X. Y . , April 25-27, 1956,Polytechnic
theinitialconditions.This is undesirable as already Institute of Brooklyn, N. Y . ;1957.
mentioned.Therefore, for manyreasonsthesimple E. -4. O'Hearnand R. K. Smyth,"Terminalcontrolsystem
applications,!! IRE TRAXS. os ALTOMATIC COSTROL, vol, -4C-6, pp.
error index of (38) is not suitable for this problem. 142-153; May, 1961.
+2.O,-

c 0.4

I I
0

+PO
bl.O.O aZ,T-o.O
+IO p E G

+0.2
SECONDS
-t
2 4 6 8 10 I2 14 16 18 2 0 ~

Fig. 10-System gains for Case 11.


Fig. 9-System performance for Case 11.

Theresultingperformancefor &,r= 1.0 isshownin


Fig. 9.
The altitude errorat t = 20 is quite small, but, unfor-
tunately, the aircraft crashesfor two sets of initial con- -L,_ _ .
/
-’ 7 1
ditions due to the large altitude errors occurring during
the earlier part of the landing interval. In addition, the
rate of ascent error is undesirably large a t touchdown.
L.
-
0,. 00001 e,;. 0.010
e.. 0.0000s 0,T. 0.001

The system input signal and the feedback gains for


this case are shown in Fig. 10. The input signal k , ( t ) is
zero during the entire landing interval as might be ex-
pected.Thefeedbackgainsare negligibly smalluntil
the aircraft begins to approach the desired touchdown
point. These gains then become quite large in order to
reduce the altitude error tosmall a value a t t = 20. Since
k I 4 ( t )essentially is zero for a large part of the landing
Fig. 11-System performance for Case 111.
interval, the altitude feedback loop in essence is open
during this time and large altitude errors arise, as al-
ready demonstrated. tween the performance obtained and the performance
requirements for thesystem.Thissuggeststhatthe
Case Ill-Constant Weighting of -4ltitude Error, Rate of performance requirements are very difficult to satisfy
Ascent Error ozler theRunway,PitchRate Error, and withthegivenaircraftdynamicsandtheassumed
Elevator Dejection; Inzpulse Weighting at Touckdown of initial conditions. Unfortunately, a great many control
Pitch AngIe Error, Altitude Error, andRate of -4scent problems are characterized by the need to satisfy a set
Errm of veryrestrictiveperformancerequirements. I t is in
The error index for this case is the solution of such problems that optimization theory
+
e(t) = J ~ {: 144 4 4 , w 0 ( 2 0 - u)l[~zd(u) - ~z(u)12
is of particular value.
The input signal and the feedback gains associated
+ +
[4:du) 4n.rzio(20 - u)][hd’(u) - h ’ ( u ) ] z with the performance of Fig. 11 are shown in Fig. 12.
+ 4 ? , T ~ 4-0 d [ e d ( d - eb)12 Thepeaks in thegains k l z ( f ) , K I 3 ( t ) , k14(t) nearthe
+ +
4Jedf(d- e f ( u ) ] 2 [6,(u)]?ldu. (44)
touchdownpointareduetothe use of theimpulse
weighting factors 4?,T, &,T and 44,T.I t is important to
Here,sometrial-and-error is requiredtoarrive a t a note that in this case the input signal k l ( t ) goes negative
suitableset of weightingfactors.Thevalues of the during the final phase of the landing and becomes posi-
weightingfactorsandtheresultingperformanceare tive again just prior to t =20. The negative and then
shown in Fig. 11. This system essentially meets all of positive excursions evidently are required to satisfy the
theperformancerequirements.Theextremelysmall pitch angleconstraint a t touchdown. It appears that
negative pitch angle a t touchdown existing for one set in problems where multiple performance requirements
of initial conditionscould be avoided b>-a slight increase exist, the sl-stem input signal,in essence, is a composite
in the value of 42,~. functionderivedfrom all of theperformancerequire-
Fig. 11 indicates that there is very little margin be- ments.
98 IEEE TRAA7SACTIONS O N ALTTOI1_IA
T I C CONTROL April

kII:l
t o - _ ,
I

+Oo6r /-

-Ot

-0
1 12 4 6 8 10
8
I2
:
14 16 I8 20
SECONDS
-t
L

Fig. 12-System gains for Case 111. Fig. ll--.\ircraft landing trajectories for h(0)= 100 feet.

-40 - SECOXDS

SECONDS
-60/ 2 4 6 A 10 ;1
I
14 16
I
10
'
20
- t
1

1 -
0 2 4 6 8 IO I2 :4 I6 I8 20 Fig. lj--Aircraft landing trajectories for h ( 0 ) = 8 0 feet.
Fig. 13-.lircraft landing trajectories for h(0)= 120 feet.

is necessary toprogramthefeedbackgainsandthe
The behavior of the systems of Cases I , 11, I11 also input signal as functions of time-to-go ( 2 0 4 . In other
maybecompared by displayingtheaircraftlanding words, time-to-go is a quantity that must be available
trajectories for thesecasesonthreeseparategraphs, throughout the landing phase i n order to construct the
one for each set of initial conditions. These trajectories control law specified by the theory. [Tndoubtedly, the
are shown in Figs. 13, 14 and 15 along with the desired measurement of time-to-go is the most difficult measure-
altitude h d ( t ) . ment associated with the physical construction of the
The system of Case I follows the desired flare-path proposed landing sJ-stem. Firstof all, the aircraft veloc-
more closely thandotheothersystems.This is due it>-will not be absolutely constant as hasbeen assumed.
primarilytothegreater weighting of altitudeerrors However, appropriate thrust control should keep per-
throughout the landing intervalin the design of the sys- turbations in the aircraft velocitJ- small enoughin order
tem of Case I. Of course, i t is not evident from these to be considered second order. On the other hand, the
curves alone that other performance requirements are measurement of the total aircraft velocity itself is diffi-
violated by the s>-stem of Case I. cult. This aircraft velocity is considered to be theWIOC-
As already mentioned, the sJ-stem of Case I1 causes it>- of the aircraft with respect to an inertial frame or
the aircraft to crash for tn-o sets of initial conditions rather with respect to the ground. In other u-ords, the
and, therefore, is completely unsatisfactorJ-. aircraftvelocitJ- i n question is groundspeed.Because
T h e s)-stem of Case 111 meets all of the performance landing situations with nonzero winds will occur, ground
requirements but it takes substantially longer than the speed cannot be measured from the aircraftitself. There-
system of Case I to correct the initial condition errors. fore, the success of this landing s y t e m requires the suc-
T h e persistence of the initial altitude errorsfor a longer cessful measurement of groundspeedfromradar. -4
period of time is the price one has to pa!' to satisfy the number of possible radar systems exist. For instance, a
other performance requirements. s>-stemof ground-based radar beacons could be used in
In order to mechanize the aircraft landing system, i t order to estimate the time required to pass over two
fixed points on theearth's surface.Anotherpossibility l.C.'Zov
I MEbSUREO STATE VbRlbBLES

is the use of Doppler radar which is ground based. In


this case, the velocity information would be relayed to
the aircraft. Due to these practical considerations, the
measurement of time-to-go could be obtained from

TZ-
40
(43 Y(,l
J70)

where the time-to-go


- T is defined by
Fig. 16-ItIechanization of aircraft landing system.
7- = 20 - t . (46)
Intheestimation of time-to-go, thequantity d ( t ) is accurate as the desired touchdown point is approached,
defined as the distance from the aircraft to the desired thecontrolsystemdescribed in thispaperstill is a n
point of touchdownontherunway,and V ( t ) is the optimumsystem.Thissystemtreatsthesemeasure-
ground speed of the aircraft. Both this distanced ( t ) and ment errors over a period of time as initial condition
t h i s velocity V ( t ) could be measured from a system of errors occurring at the instant the time-to-go measure-
ground-based radar. ment becomes accurate. Since this system is optimum
It may be desirable to construct theengineering ver- for all initial conditions, itis also optimum for this situ-
sion of thecontrolsystem in terms of analogcom- ation.
ponents.Thereforesome brief commentsconcerning
the analog construction of the control system are pre- VIII. CONCLUSIONS
sented here. T h e block diagram shown in Fig. 16 indi- A number of important advantages to be gained by
cates one possible implementation of the time-var>-ing the use of optimization theory as the basisfor the design
gains appearing in the control s>?stem. If i t is assumed of feedback control systems have been pointed out by
that time-to-go r is available as an angular shaft posi- the example of the aircraft landing system. In particu-
tion,thetime-varyingnlultiplicationscan be realized lar, multiple design considerations associated with diffi-
fromaset of loadedpotentiometers.Suchpotenti- cult dynamic situations readily can be included in the
ometershavemultipletapsandtheappropriate k- system design. The functional forms of the weighting
parameterfunctions of time-to-gocan be constructed factors appearing in the error index are chosen directly
byapplyingappropriatevoltagestothetaps on the fromconsiderations of thetimedomainperformance
potentiometer. The signalon the wiper of such a po- requirements. Except for certain mathematical restric-
tentiometer would then be the appropriate function of tions, the functional forms of theseweightingfactors
the shaft angle T times the excitation voltage applied to can be chosen without consideration of the system sta-
the loaded pot. The summing and gain functions indi- bility per se. However. one of the disadvantages of using
cated in Fig. 16 are achievedfrom standardanalog optimization theory is that,
generallyspeaking, a
computer components. The shaft angleT is achieved by unique set of weighting factors does not exist when a
placing an input signal T t o a position servo. The input set of design considerations is given. In fact, the simpler
signal r couldbe generatedfromthe expressionfor the designproblemfromdynamicconsiderations,the
time-to-go given in (45) by the appropriate division of larger thenumber of sets of acceptableweighting
the distance to the touchdown point b>- the indicated factors.
ground speed of the aircraft. T h e designproblemchosenforthispaperis made
Another possibility for the generation of the signal T difficult because a large number of design specifications
arises under ideal situations where the ground speed is must be met in order to consider the results to be suc-
a known constant and the initial distance to the touch- cessful. I t is difficult to meet all of these design consid-
down point is known. Under these ideal conditions, a erationsbecausethetrajectoryrequirements of flare-
simple clock consisting of an integrator with a constant out are somewhat incompatible with the dynamic char-
input can beused for the generation of time-to-go. Both acteristics of the aircraft. In particular, the trajectory
of these possibilities for the generation of the signal r corrections required for passing from the glide phase to
are indicated in Fig. 16. The system indicated inFig. theflare-outphasearelargeincomparisontothe
16 ishighlysuited for airborneapplicationsbecause amount of control effort allowable during the flare-out
this system is light, cheap, and reliable. Therefore, we portion of the trajectory. Roughly speaking, the amount
canconclude t h a t if themeasurements of thestate of control effort is consideredto be the magnitude of the
variables and also time-to-go are readily available and control signal times the amount of time during which
reasonably accurate, the theoretical results of this es- this control signal can be applied to the aircraft. Opti-
ample form the basis for a practical landing systemt h a t mization theory assumes its most useful purpose under
exhibits high performance. the conditions where the potential response errors are
If the time-to-go measurement is in error during the large relative to the amount of control effort which can
initialportion of the flare-outinterval b u t becomes be applied to the system.
The results of this particular example indicate that It has beenshown in theliteraturethat, for state
the design specifications for the aircraft landing system determined processes, the minimum value of the error
can be met. The digital computer simulation of the air- index is a function only of the measured state x(p) of
craft trajectories shows that under ideal conditions suc- the dpnamic process (the aircraft) and the time p at
cessful automaticaircraftlandingscanbeachieved. which themeasurement is made.’ Thus,functionally
However,because a number of engineeringquestions E = E [x&), p]. &Asa result of this functional relation-
have been neglected in the study presented here, addi- ship, the total derivative of E with respect to p appear-
tional work is required in order to build a practical land- ing in (47) may be written as follows:
ing system from the theoretical results presented thus
far. I n particular, the effects of wind gusts and measure- _d E_ -- d E [ X ( P ) , PI - a E [ x ( / l ) ,PI
ment noise should be included in trajectorl- studies in dP aP
order to ascertain their effect on the performance of the -’* dx, a E [ x ( p ) , p ]
system. These statistical disturbances could be included +E-
r.=l d p ax,
(48)
in a digital computer simulation. However, it mal- be
more efficient to perform these additional studies o n an where :V=4 because the aircraft descriptionis of fourth
analog computer. order. Also, dx,/dp is a time derivative of a state signal
Another area which should be investigated before the which appears in the dynamic process equations given
proposedlandingsystemisconsidered t o bechecked by (12). Thus, (47)becomes
out in an engineeringsense is the area of sensitivity

in the previousresultsareonlyapproximations,and &,(PI {


analysis. In particular, the aircraft parameters assumed min + & ) [ W ) - l2(/l)l2

therefore the sensitivity of the system trajectories to


perturbations around these assumed parameters should
+
+ h f ( p ) [ h d ’ ( p-
) + + o ( ~ ) [ e d (-
d e(p)I2
be investigated. Also the effects of approximations to
thek-parametersshouldbeinvestigatedbecausethe
engineering system may only be able to generate these dE dE
time-varying gains in an approximate sense. This work + +
x2’(p) - s a ’ ( p ) - s4’(p)
d.r* as3
+ (49)
is being done and will be the subject of a later paper.
Eq. (49) states thatwhen the quantity within the braces
APPENDIX
is minimized with respect to &(p), its minimum value is
THEPAFL4METRIC EXPANSION RIETHOD zero. In other words, when the optimum control signal
In this section, the Parametric Expansion >lethod is S,*(p) is substituted into the terms within the braces,
applied to the aircraft landing problem. S o attempt is these terms add up to zero. But, the quantity within
madetojustifytheprocedure followed. Rather,the t h e braces is a minimum when the partial derivative of
readerisreferred to papers in the available technical this quantity with respect to 6,b)is equal to zero. Per-
literaturewhichdevelop, in arigorousfashion,the forming this last operation gives the expression for the
principles employed here. optimum control signal ~ 3 ~ * ( pThis
). expression is
T h e firstconsideration in theminimization of the 611 aE
errorindexisthederivation of theconditionunder 6,*(/l) = - - ~

2 as1
which the minimum exists. This has been carried out
by Bellmaninhisdevelopment of the principles of since terms 1-4, 6 , and 8-10 in (49) are not functions
Dynamic Pr0grarnming.g Theconditionunder which of &(p), and x1’(p) is a function of &(p) as indicated in
the error index of (26) is a minimum is given by (11) and (12).
Substituting (12) and then (SO) into (49) gives

4 h h ) [hd(P) - Jz(dI? ++dP) [kl’(P) - IZ’(d1~


+ + O b ) [ed(P> - e(/*)] +
$8’(11) [ed’b) - ‘9’(p)l2

where u = p is an arbitrary instantof time in the interval


u = t to u = T. Also, E is the minimum value of the error
index and is called the minimum error function. Sote
that the first five terms of (47) are the terms appearing
in the integrand of the error index given in (26).

R . E. Bellman, “Dynamic- Programming,’ Princeton L-niversity


Eq. (51) is a partial differential equation defining the
Press, Princeton, X. J . ; 1957. ~ function E. If thisequationcanbe
m i n i n ~ u nerror
I963 Ellert a n d Aiel-riam: Feedback Con.trols

solvedfor E , thentheoptimumcontrolsignal 8,*(p)


can be found using (50). This is the goal of the control
engineer.
IntheParametricExpansionMethod,thesolution
of (51) is carried out in a linear case (linear process and
weighted quadraticerrorindex)byassumingthe fol-
lowingexpressionfor E:
3.
E = k(P) - 2 Krn(P)&(P)
m=1

+ cc
N A.

KrnP(PL)~*(P).TP(/4 (52)
m=1 p=l

where N = 4 due to the fourth-order representation of


theaircraft. Xlso, k,,(p) =kpm(,u). This expansionis
called a Parametric Expansion because the coefficients
k , ( p ) and k m p ( p )turn out to be parameters of the opti-
mum control system. This is demonstrated in the suc-
ceeding paragraphs.
T h e expression for the optimum control signal given
by (50) can now be expressed in terms of the state sig-
nals and the k-parametersusing (52). Taking the partial
derivative of E with respect to x1 and substituting into
(50) gives
SC*(,u) = Cll[Kl(P) - klI(d.Yl(P) - klBiP)-YY(llo

- k13(p)x3(p) - k14(P)s4(p)]. (53)


Since p may take on the value t , which is present or
real time, (53) defines the control law for the optimum
system to be
6,(t) = w , ? ~ , ~ , [ k l ( -
t ) kll(t)e’(t) - kl2(1)0(t)
- kl3(f)h’(f) - k14(t)~z(t)] (54)

where the expressions for cll and the state signals have
been inserted.Thiscontrollawdefinestheconfigura-
tion of the landing system. The block diagram of the
system is drawn with the aid of (54) and is shown in
Fig. 4. There are a number of important observations
to be made from this block diagram:
The numberof feedback loops is equalto the order
of the process (aircraft).
The quantities fed back are the measurable state
signals.
Thefeedbackgainsaredoublesubscripted k-
parameters.
The feedback gains,for this linear case (linear air-
craftdescriptionandweightedquadraticerror
index), are merely time-varying and not functions
of the state signals.
T h e s)-stem inputsignalis a singlesubscripted
k-parameter.
Thesysteminputsignal, for thislinearcase, is
merelytime-varyingandnotafunction of the
state signals.
The only remaining information requiredby the con-
trol engineer in order to carry out the mechanizationof
102 IEEETRANSACTIONS O N A U T O M A T I C COA'TROL April

INITIAL
K-EPUATIONS VALUES
K

AIRCRAFTPARAMETERS
I I
c
brnnncrnn

DESIREDPERFORMANCE COMWTER
hd*hia e d , 8;

Fig. li-Solution of the R-equations.

I
Ill

FORMULATE
ERROR
INOU
I

I
I
121

STATE
CONOlTlON
FOR
MINIMUM
7
I
I31

DETERMINE
OPTIMUM
CONTROL
S I G W
!I
1
I
I41

EXP4NO MINIMUM
ERROR FUNCTION
IN A SERIES
(FaRAMETRlC
EXPANSION1
I
151

SUSSTITUTE
OPTIMUM CON=
SlGN4L 8 MINIMUM
ERROR F W C T I O N
IHTO THE
CONDITION FOR
-
!I
I
-
I
161

SETEACH
P=O
TO OBTAIN
t-EWATIONS 7
I
L E T 0 'TY
LL TRAUSFDRM
t~EWATIONS
TO OBTAIN
K-EUTIONS
I
I
I
SOLVE
I-EOWTIONS "l'
--K,lr))

KmpIrll

....
...
Ipl- ...
kIT1=k~ITl=kmplTl=O

Fig. 18-Block diagram of the steps involved in the use of the Parametric Expansion Method.

-k2i(P) = b33k?3(P) + bld13(P) + b13KI?(P)


Then,afterhavingtransformedthek&)-equationsinto
+
63&33(P) + k?Q(P) K(q)-equations using (73), the solution of these new
equations can be carried out from 7 = O to 1 =20, as
-~11?~14/~)~13(c() (67) indicated in Fig. 5.
- k?r'(P) = bl&ll(P) +b3?k34(Pcl) The solution of the K-equations is illustrated in Fig.
- cll?kldP)Kle(P) (68) 17. Fig. 17 indicates the quantities which must be speci-
-kA:(P) = 4h'(P) + 2b33k33!P) +
2bl&13(P)
fied before the solution can be carried out. These quan-
titiesappear in the&equations, ( S i ) - ( i l ) . Thus,in-
+
2K31(1*) - C1l2K13'(P) (69) cluded in these equations is all of theinformation
-K3e1(PC1) = b33k3dP) +bl&l4(P) kdP) + pertinent to the problem.
.As has been described in this section, the Parametric
- C112k13(~)k14b) (70) Expansion
3Iethod
provides
directly
configuration
the
--k44'(P) = 4 / , ( P ) - Cll'Kl4?(P)
of the optimum system and a set of equations which
may be solved to obtain the system parameter values.
Theboundaryconditions onthesek-parameterscanTheproceduredescribedhereissummarizedinFig.
be to be 1s.
K(20) = k,(20) = k,,(20) = 0; ??Z) p = 1) 2, 3) 4. (72) NOMEKCLXTURE

T h a t is, the values of the k-parameters are known at the B matrix relating x'(t) and x ( t )
terminal point p = T = 20 rather than at the initial point. bmn an element of the B matrix
Hence, i t is necessary to solve these equations back- C matrix relating x'(t) and m(t)
ward in time. As a result, it is convenient to introduce Cmn
an element of the C matrix
a new dummy variable
time 7,where E minimum value of the error index
e ( t ) , also
called
the
minimum
q = 20 - p, (73) function
error
value of the error index aircraft
short
perioddamping
value of the instantaneous error factor
measure dummy time variable
aircraft altitude aircraft pitch angle at a = T
time derivatives of aircraft alti- aircraft pitch angle
tude aircraft pitch rate
desired aircraft altitude desired aircraft pitch angle
desired aircraft rate of ascent desired aircraft pitch rate
aircraft short period gain dummy time variable
transformedk-parameters using time-to-go before touchdown
n ='l - u,--
weightingfactor for pitchrate
-I
errors
~

k(p), k m ( p ) ,k m p ( p ) k-parametersintheexpansion of
the minimum error function 43 weighting
factor for rate of ascent
errors
m summation index
m ( t ) , m l ( t ) , m d t ) , . . . , m,,(t) processcontrolsignals 44 weighting
factor for altitude
errors
42.T magnitude of the impulse
weight-
m 0) control signal vector
ing factor for pitch angle errors
m*(0 optimum control signal vector
43.T magnitudeimpulse
ofweight-
the
x order of the dynamic process
ing factor for rate of ascent errors
n summation index
summation index 442 magnitude of the impulse
weight-
P ing factor for altitude errors
S complex variable
T a future instant of time '#'h(t) weighting
factor for altitudeerrors
aircraft path time constant $h' (t) weighting
factor for rate of ascent
TS errors
I real time
ZlO(20 -a) unit impulse occurringa t a = 20 49 ( 0 weighting
factor for angle
pitch
errors
17 aircraft total velocity
147 weight of the aircraft 49, (0 weighting
factor for rate
pitch
errors
x ( t ) , xl(t), x 2 ( f ) ,. . . , x,(t) process state signals
w8 aircraft
short period resonant fre-
x@> state
vectorsignal quency
x' ( t ) derivative
time of state signal
vector .~CKNOWLEDGMENT
XI'@), X 2 ' ( t ) , . . . , xn'(t) timederivatives of process The authors are indebted to thefollowing individuals
state signals for theirassistance in definingrealisticrequirements
ff ( 0 aircraft
angle of attack and constraints for this landing problem and forpro-
Y (t) angle
path
aircraft
glide viding the linearaircraftdescription: R. G. Buscher,
44t) angle
change
in of attack from R. Larson (Test Pilot), R.I. F. Marx of the Light Mili-
the equilibrium value tary Electronics Department, and E. J. Smith formerly
Ut) elevator
aircraft deflection of the General Engineering Laboratory.

Das könnte Ihnen auch gefallen