Академический Документы
Профессиональный Документы
Культура Документы
R.M.P. Goverde
Civil Aircraft Autopilot Design
Using Robust Control
R.M.P. Goverde
January 1996
and thereby giving me the chance to further develop my interest in the applied
mathematics. Finally I thank my colleagues, and especially my room-mates, at the
Statistics, Stochastics, and Operations Research Unit at TUD for a great working
atmosphere. My deepest appreciation goes to Guido te Brake who managed to
survive my company for over 10 hours a (working) day during whole 1994, including
room-mating and car-pooling.
Rob Goverde
Delft, January 1996
Contents
Summary i
Samenvatting (Summary in Dutch) iii
Preface v
1 Introduction 1
1.1 Robust Flight Control : : : : : : : : : : : : : : : : : : : : : : : : : : 1
1.2 The GARTEUR Project : : : : : : : : : : : : : : : : : : : : : : : : : 2
1.3 The RCAM Design Benchmark : : : : : : : : : : : : : : : : : : : : : 3
1.4 Synthesis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4
1.5 The Design Project : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5
1.6 Outline of the Report : : : : : : : : : : : : : : : : : : : : : : : : : : : 5
2 Robust Control Theory 6
2.1 System Descriptions : : : : : : : : : : : : : : : : : : : : : : : : : : : 6
2.1.1 State-Space Realizations : : : : : : : : : : : : : : : : : : : : : 6
2.1.2 Transfer Matrices : : : : : : : : : : : : : : : : : : : : : : : : : 7
2.1.3 Linear Fractional Transformations : : : : : : : : : : : : : : : : 7
2.2 Linear Feedback Control : : : : : : : : : : : : : : : : : : : : : : : : : 9
2.3 H1 Optimal Control : : : : : : : : : : : : : : : : : : : : : : : : : : : 10
2.4 Uncertainty Modelling : : : : : : : : : : : : : : : : : : : : : : : : : : 13
2.4.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13
2.4.2 Interconnections of LFTs : : : : : : : : : : : : : : : : : : : : : 13
2.4.3 Real Parameter Uncertainty : : : : : : : : : : : : : : : : : : : 15
2.5 The Structured Singular Value () : : : : : : : : : : : : : : : : : : : 18
2.6 Analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22
2.7 Synthesis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
2.8 Control System Design Cycle Description : : : : : : : : : : : : : : : : 27
2.9 Additional Reading : : : : : : : : : : : : : : : : : : : : : : : : : : : : 27
3 Control Problem Formulation 29
3.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 29
vii
viii Contents
but also in the sense of controller complexity and controller design time. Especially
the latter is considered to be of the utmost importance for the aircraft industry.
In particular, the designed controller should be robust with respect to variations
in weight and center of gravity position, time delays, nonlinearities, and engine fail-
ure. Moreover, various external conditions such as wind shear and turbulence have
to be taken into account. Also passenger comfort, safety, and power consumption
criteria have to be considered.
The present report deals with the design of the RCAM design benchmark using
synthesis. The design covers the
ight control system for the symmetric (longitudi-
nal) motions of the RCAM aircraft. Robustness with respect to variations in weight
and center of gravity position is considered explicitly. Time delay is not considered
in the present design.
1.4 Synthesis
A modern robust control approach is H1 optimal control theory. Furthermore, a
powerful tool to measure robustness is structured singular value () analysis. A
combination of H1 optimal control theory and analysis results in synthesis
which provides a modern approach to design and analyse robust control systems.
H1 optimal control theory is a highly developed theory which yields control laws
such that even in the worst case scenario the stability and performance requirements
are still met. Of course, the success of nding such a controller depends on the
performance specications. If these are too demanding, no controller can be found.
In this case, the objectives have to be adjusted. An ecient synthesis method has
been developed using state-space algorithms. However, this approach may lead to
conservative results, since it neglects possible knowledge of structure in the uncertain
model [4, 17].
A way to gain better results is to exploit available information about the structure
of the uncertainty. Throughout the system, uncertainties may be present. Instead
of modelling these uncertainties as unstructured, the general uncertainties at the
various components can be taken into account one by one explicitly, and rearranged
to form structured uncertainty at the system level. This uncertainty modelling is
done by making extensive use of the theory of linear fractional transformations
(LFTs) [5, 16].
The in
uence of structured uncertainty on system performance can explicitly
be analysed by applying analysis. Analysis is therefore a powerful tool to
analyse control systems for uncertain systems in a nonconservative way and during
the last decade a lot of research has been done in this area. Unfortunately, no
eective synthesis methods have yet been developed to apply analysis in the
design process of control systems with general uncertainty structure. However, for
a restricted class of modelled uncertainties|bounded complex perturbations of the
system|a -synthesis method has been developed, for which the D-K iteration
1.5. The Design Project 5
approach yields good results despite presently unsolved convergence problems. The
-synthesis problem is solved here by sequentially applying H1 synthesis and
analysis [1, 3, 18].
6
2.1. System Descriptions 7
The system matrices A; B; C; and D are called the state matrix, input matrix, output
matrix, and direct-feedthrough matrix, respectively.
w- z-
u- y-
G
Throughout this report we may refer to a system by either its state-space de-
scription (2.1) or its transfer matrix G.
w- z-
u- y-
P
a. P
w- z-
- P u
u y w z
l
-
u- y-
P
b. Fl(P; l ) c. Fu(P; u )
Figure 2.2: Linear fractional transformations.
P the (linear fractional) coecient matrix, and l and u are called perturbations.
More generally we can in the same way dene LFTs with respect to complex rational
matrices P (s), l(s), and u(s).
The LFTs arise naturally in system theory when describing feedback systems:
Consider a transfer matrix P (s) with inputs w and u, and outputs z and y as in
Figure 2.2.a. Then the closed-loop transfer matrices from w to z of Figure 2.2.b
and from u to y of Figure 2.2.c are the LFTs Fl(P; l) and Fu(P; u ), respectively.
The subscripts l and u refer to whether we are closing the lower or upper loop. We
thus refer to lower LFTs as closed-loop systems from the exogenous input to the
objective, and to upper LFTs as closed-loop systems from the control input to the
measurement.
Transfer functions can also be represented by LFTs. Consider the state-space
realization of a system given as
x_ (t) = Ax(t) + Bu(t)
y(t) = Cx(t) + Du(t):
This can be reformulated
x_ (t) as A B x(t) x(t)
y(t) = C D u(t) = P u(t)
and the transfer matrix is
G(s) = D + C (sI A) 1B
= D + C 1s I (I A 1s I ) 1B
= Fu(P; 1s I ):
2.2. Linear Feedback Control 9
w- z-
- G
u y = w- Fl(G; K ) z -
K
rank conditions in assumption (iii) imply that the feedback connection of the real-
izations for G and K are detectable and stabilizable, corresponding to the rst and
second rank condition respectively. These conditions are necessary for the existence
of solutions to the two Riccati equations that are involved in the solution of the H1
optimization problem.
Theorem 2.1 gives a procedure to check whether the upper bound
on the H1 -
norm is satised. This theorem assumes the following additional conditions that can
be made without loss of generality since any plant can be transformed into a plant
satisfying these conditions [19].
Additional Assumptions
(iv) The system is scaled such that
B 0
D12T 1 T
(C1 D12) = (0 I ) and D D21 = I :
21
2 1
=
- - -
u- - y- u- y-
P2 P1 P
1
r e
- P1
=
-
u
2
?+ y - u-
P
y-
6+
-
- P2
The parallel connection of Fu(P1; 1) and Fu(P2; 2) as in Figure 2.5 is an LFT
Fu (P; ) with as in (2.6) and coecient matrix
0 P 0 P112
1
111
P = @ 0 P211 P212 A:
e r
P121 P221 P122 + P222
u -
+ y-
6
+
-
=
-
u- y-
P
P1
Figure 2.6: An LFT in a feedback loop.
and 00 0 0 s s 1 1
1 2
s1s212 = Fu @@ 1 0 0 A ; 01 0 A :
0 1 0 2
2.4. Uncertainty Modelling 17
The addition of all terms in each varying numerator can be written as a single
LFT using the fact that a parallel connection of LFTs is again an LFT, see Sec-
tion 2.4.2. Since we are using upper LFTs, the nominal (constant) part of each
numerator results in a feedthrough term, which can be incorporated in the P22 term
of the combined LFT. Using the LFTs constructed above for each term of nij (1; 2),
and the results of Section 2.4.2 on parallel connections of LFTs, the resulting com-
bined LFT becomes
00 0 0 0 0 p s 1 0 1 1
02 1 1 0 0 0 C
B
B B
B 0 0 0 0 p01 s2 C C B 0 2 0 0 C C
Fu BB B 0 0 0 0 s s C ; B C C :
B
@@ 0 0 1 0 0 A 0 0 0 C
1 2 C @ 0 0 1 0 A A
1 1 0 1 p01 p02 2
w1 z1
w2 -
- G z2-
-
u y
K
complex; and full complex blocks which can be used to model unmodelled dynamics
(this is not applied in our approach) or to dene the so-called performance block in
order to measure robust performance. This is explained in more detail in Section 2.6.
For simplicity in notation we assume the full complex blocks to be square but
all results are also applicable to nonsquare full complex blocks. The set K 2 C qq
of block-diagonal perturbations is now dened as
K := fdiag[1rIk1 ; . . . ; mr r Ikmr ; 1cIkmr +1 ; . . . ; mc c Ikmr +mc ; C1 ; . . . ; CmC ] :
ir 2 IR; ic 2 C ; Ci 2 C kmr +mc +ikmr +mc+i g: (2.11)
The set K depends on the block structure K that is dened as the m-tuple
K := (k1; . . . ; kmr ; kmr +1; . . . ; kmr +mc ; kmr +mc+1 ; . . . ; km ); (2:12)
which denes the number of blocks, the type of each block, and their dimensions.
For compatibility we require that the sum of all block dimensions equal the total
dimensions of the block-diagonal perturbation, i.e.,
X
m
ki = q:
i=1
Note that a block consisting of a complex scalar only, may be represented as either
a repeated complex scalar block or a full block of dimension one.
We are mainly interested in norm bounded subsets of K and therefore introduce
BK := f 2 K : () 1g: (2:13)
Let for the time being M be a square complex matrix with dimensions q q.
Denition
q q
2.1 (Structured Singular Value) Let K be given. Then for M 2
C the structured singular value (SSV) K (M ) is dened as
( ) 1
K (M ) := min f() : det(I M ) = 0g ; (2:14)
2 K
w - z
M
The gap between R(M ) and (M ) may be arbitrarily large. However these
can be rened using transformations on M that do not aect K (M ) but do aect
R(M ) or (M ). For this we dene the subset QK of K as
QK := fQ 2 K : ir 2 [ 1; 1]; icic = 1; Ci Ci = Ikmr +mc+i g
and the set DK as
DK := fdiag[D1; . . . ; Dmr +mc ; d1Ikmr +mc +1 ; . . . ; dmC Ikm ] :
Di 2 C kiki ; Di = Di > 0; di 2 IR+ g: (2.15)
Then for all D 2 DK and all Q 2 QK we have
K(QM ) = K(M ) = K (DMD 1 );
and the bounds can be rened as
max (QM ) K(M ) Dinf
Q2Q R 2D
(DMD 1 ): (2:16)
K K
In fact the lower bound is proven to be equal to K(M ) but since the maximization
over QK has local maxima which are not global, we can not guarantee to nd the
global maximum and thus from a computational point of view the left-hand side of
(2.16) only yields a lower bound.
The function (DMD 1 ) has global minima only. However, in general the in-
mum is not necessarily equal to K(M ) and thus yields only an upper bound.
For purely complex perturbations (mr = 0) the lower and upper bounds are
usually very tight and moreover in special subcases the upper bound is even equal
to K (M ). However, if mr > 0 the upper bound may be arbitrarily far o. This
is due to the nature of the scaling matrices D which do not explicitly take the real
perturbations into account.
To obtain an alternative upper bound for the case that mr > 0 we rst dene
two more sets of diagonal scaling matrices D^K and G^K as
D^ K := fdiag[D1; . . . ; Dmr +mc ; d1Ikmr +mc+1 ; . . . ; dmC Ikm ] :
Di 2 C kiki ; det(Di) 6= 0; di 2 C ; di 6= 0g (2.17)
and
G^K := fdiag[g1; . . . ; gnr ; On nr ] : gi 2 IRg;
P
with nr := mi=1r ki. An alternative upper bound is then obtained as
( " ^ ^
! # )
K (M ) ^ ^inf^ ^ > 0 : (Iq + G^ 2 ) 4 1 DM D 1
j G^ (Iq + G^ 2) 4 1 :
1
D2DK ;G2GK
(2:18)
This expression is used in the MATLAB -tools toolbox to eciently compute an
upper bound for if real blocks are present. Note that if mr = 0 then G^K is just
the zero matrix and thus G^ = 0. The upper bound then reduces to that in (2.16).
22 Chapter 2. Robust Control Theory
2.6 Analysis
From now on we let M be the transfer matrix of the closed-loop plant
M := Fl(G; K )
and refer to the partition of M corresponding to the perturbation and performance
variables as z M M w
1 11 12 1
z = M M
2 21 22 w : 2
The closed-loop transfer matrix from w2 to z2 is given by the LFT Fu(M; ), see
Figure 2.9. Note that the matrix M22 may be thought of as the nominal transfer
matrix in the absence of perturbations.
Often stability is not the only property of a closed-loop system that must be
robust to perturbations. Typically there are exogenous disturbances acting on the
system (wind gusts, sensor noise, etc.) which result in tracking and regulation errors.
Under perturbations, the eect that these disturbances have on error signals can
greatly increase. In most cases, long before the onset of instability, the closed-loop
performance will degrade to the point of unacceptability. Hence the need for a robust
performance test, which indicates the worst-case level of performance degradation
associated with a given level of perturbations.
The system in Figure 2.9 is said to satisfy robust performance if it is robustly
stable and if
kFu(M; )k1 < 1:
To guarantee that performance is achieved for all perturbed plants we additionally
consider a feedback loop closed between the performance error z2 and the exogenous
input w2 with transfer matrix 0, the so-called performance perturbation. We then
dene the augmented perturbation Ka as
Ka := fdiag[0; ] : 0 2 C m12 p12 ; 2 Kg; (2:20)
see Figure 2.10. Note that we dene 0 as one full complex block.
a
0
w1- z1
w2-
M
z2
Figure 2.10: Augmented perturbation structure for robust performance.
2.7 Synthesis
In the case of purely complex perturbations a -synthesis design procedure is devel-
oped based on the upper bound on K in (2.16). The idea is to design a controller
K such that the robust performance test (2.21) is satised for the upper bound of
, for an appropriate scaling matrix D.
In the case that also real parametric uncertainty is modelled, the procedure
may give conservative results since then the upper bound (2.16) is no longer a good
approximation of . The alternative upper bound (2.18) for the mixed real/complex
case does not have the properties on which the solution to the -synthesis problem
considered below is based. The amount of conservatism can be measured by also
calculating the alternative upper bound (2.18) for the scaled system.
In Section 2.6 we have seen that the synthesis problem reduces to nding an
internally stabilizing controller with transfer matrix K such that
sup K[Fl(G; K )(j!)] < 1: (2:22)
!2IR
In fact, by the Robust Performance Theorem 2.3 we are interested in the case where
the underlying structure of the perturbations is given by Ka as given in (2.20).
In order to accomplish (2.22) we want to nd internal stabilizing controllers that
minimize the left-hand side of this inequality. This design procedure is called
synthesis. By using the upper bound (2.16) on the -synthesis problem can be
formulated as
inf inf kD~ Fl(G; K )D~ 1 k1;
K 2A ~
(2:23)
D(s)2DK
2.7. Synthesis 25
w1 z1
w2 -
- G z2-
-
u y
K
a. General framework.
w1- z1-
w2 -
- G - z2
u y w1 z1
K
- F (G; K )
w2- z2-
l
b. Synthesis. c. Analysis.
2. Holding D~ (s) Fixed. To solve (2.23) we rst consider the case of holding
D~ (s) xed. We absorb the D scaling in the plant to obtain the scaled plant GD ,
see Figure 2.12. This requires nding a state-space realization for D~ (s), and
constructing a state-space model for the system GD . Note that if K stabilizes
G then K also stabilizes GD , and DFl(G; K )D 1 = Fl(GD ; K ). Moreover, if
D~ (s) is stable, minimum-phase, and real-rational then GD is a real-rational
26 Chapter 2. Robust Control Theory
parameter
uncertainty
modelling
?
- selection
weights
?
creation
standard plant
?
H1 synthesis
' $ nonlinear
simulations
& %
?
end
29
30 Chapter 3. Control Problem Formulation
e
control - actuator
noise - wind -- y -
model - aircraft - 6+ -
noise - turbulence
model
references
Figure 3.1: The RCAM system.
The control system design is divided into two separate controller designs. This is
due to the decoupling property of the aircraft dynamics into symmetric and asym-
metric dynamics. Hereby, a longitudinal and a lateral controller can be designed
separately. The nal control system is then obtained by taking the longitudinal and
lateral controllers together.
In this report we consider the design of the longitudinal controller only. All
routines can also be used with respect to lateral controller design, by making some
adjustments.
Figure 3.1 shows the RCAM system. The various models and signals are ex-
plained in the following sections.
Section 3.2 describes the state, measurement, and control variables of the aircraft
for the longitudinal motions. The nonlinear model of the aircraft is not explicitly
introduced in this report. A full description of the nonlinear model can be found
in the RCAM problem formulation and manual [9]. Instead the nonlinear model
is treated as a black box from which an LFT description is derived. However,
in the design project some time has been spent on understanding and checking
the nonlinear model. Some errors and inconsistencies have been reported to the
GARTEUR responsible persons.
Section 3.3 deals with the actuator dynamics. Section 3.4 considers the wind and
turbulence models that are used in the standard plant for controller design. Finally,
Section 3.5 introduces the reference signals that are used in the controller design.
-1
10
-2
10
-3
10
-4
10
-5
10 -2 0 2 4
10 10 10 10
Frequency (rad/s)
0
10
-1
10
-2
10
-3
10 -2 0 2 4
10 10 10 10
Frequency (rad/s)
0
10
-1
10
-2
10
-3
10
-4
10 -2 0 2 4
10 10 10 10
Frequency (rad/s)
Figure 3.4 shows the Dryden velocity spectra for these values of the parameters.
case, with x. The classications for the state matrix are
0v 0 v v1
Aclass = B
Bc 0 0 0C
@v v v vC A
v v v v
which is easily checked to coincide with the given A1 and A5 matrices above. The
classications for the other system matrices are
0v c c v v v v1
B 0 0 0 0 0 0 0 C 0 0 v v
@v c c v v v vC
Bclass = B A ; Cclass = 0 v v v
v 0 0 v v v v
and Dclass contains zeros only, since the direct-feedthrough matrices Di are all 2 7-
dimensional zero-matrices.
A varying entry is thus considered to be a function of the varying parameter.
For the varying entries vectors are dened containing the entries corresponding to
the various points of x. These vectors can be considered to contain function
values corresponding to xed values of x. The vectors are plotted against x
and approximated by polynomials to obtain a description of a varying entry as a
function of all possible values of x. In the case of x as varying parameter, it is
accurate enough to approximate the varying entries with polynomials of order up to
two in the independent variable x. It is also possible to introduce an additional
independent variable such that the polynomial in x ts the function with a certain
uncertainty described by the additional variable.
Figure 4.1 gives an example of a 2nd order polynomial approximation for the
entry a41. The polynomial is given as
a41(p1) = 0:11p21 + 0:168p1 + 77:3
where the independent variable p1 equals x.
The entries of the nominal model corresponding to the varying entries are re-
placed by the computed polynomials in x. We thus obtain a state-space description
that varies with dierent values of x. We can write x as a mean value plus a
perturbation, i.e.,
x = xnom + s;
where
xnom = x +2 x ; s = x 2 x ;
and 2 [ 1; 1]. Here x and x are the upper and lower bound on x, respec-
tively. Substituting this expression for x in the polynomials of the varying entries,
we obtain an LFT description with a real scalar perturbation for each varying entry,
38 Chapter 4. Modelling
77.364
77.362
given datapoints and approximation
77.36
77.358
77.356
77.354
77.352
77.35
77.348
77.346
0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32
p between 1.500e-001 and 3.100e-001.
as explained in Section 2.4.3. All varying entries can thus be considered as LFTs
and the overall system can be considered as an interconnection of LFTs. Using
the properties of LFTs this can then be reformulated as a single LFT having one
repeated real perturbation block. This perturbation block corresponds to the scaled
parameter x. If we would have made use of additional variables to represent uncer-
tainty in the polynomial ts as mentioned above we would have obtained additional
perturbation blocks.
In the uncertainty modelling for the longitudinal controller design 31 of the 66
entries in the system matrices are classied as varying entries. All these varying
entries are approximated by polynomials of order up to 2. The resulting LFT de-
scription for the perturbed system consists of one repeated real perturbation block of
dimension 10. The obtained nominal linear aircraft model has the system matrices
0 0:9825 0 0:0007 0:0161
1
Anom = B
B 1:0000 0 0 0 C;
@ 2:1936 9:7758 0:0325 0:0743 C A
77:3570 0:7675 0:2264 0:6684
0 2:4379 0:2912 0:2912 0:0011 0:0160 0:0007 0:0161 1
Bnom = B
B 0 0 0 0 0 0 0 C
@ 0:1837 9:8100 9:8100 0:0304 0:0742 0:0325 0:0733 C A;
0 6:04785 00:9996 00:02900:2450 0:6614 0:022620 0 0:066800 0 0
Cnom = 0 79:8667 0:0283 0:9996 and Dnom = 0 0 0 0 0 0 0 :
4.2. Parameter Uncertainty Modelling 39
3
10
2
10
1
10
0
10
-1
10
-2
10
-3
10
-4
10
-5
10 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10
2
10
1
10
0
10
-1
10
-2
10
-3
10 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10
Figure 4.2: Frequency response of the nominal aircraft longitudinal model to elevator
input (upper) and throttle input (lower) for 1-dimensional parameter uncertainty.
4.2. Parameter Uncertainty Modelling 41
case the block structure contains more blocks corresponding to these variables).
Finally, the routine ss2lft transforms the state-space description containing
the polynomials to one LFT having a block-structure with two repeated real blocks
(unless the additional variables are used).
For the longitudinal controller design this procedure is not applied. This is
because there are four uncertain parameters of main importance. Note that the 1-
dimensional case is applied for longitudinal controller design to obtain some feeling
for the eects of the various uncertain parameters independently.
The 2-dimensional parameter uncertainty case is here brie
y considered for com-
pleteness. It might be valuable for the lateral controller design. For lateral controller
design the 2-dimensional case might be interesting to check the eects of varying
mass and y only, since they are expected to eect the lateral motions the most.
checked whether all entries are monotonous functions in the varying parameter,
including linear and constant functions.
For the 1-dimensional parameter uncertainty in x of Section 4.2.1 the varying
entries are all monotonous functions in x for the particular set of xed other uncer-
tain parameters. After some experimentation and checking the physical nonlinear
model of the aircraft, it is concluded that for all possible xed values of the other
uncertain parameters, the varying entries also are monotonous functions of x.
For mass and z the same results are obtained. This implies that for mass, x,
and z, only the lower and upper bounds have to be considered, yielding three sets
containing values of a varying parameter of 2 elements.
Analysis of 1-dimensional parameter uncertainty in y gives the following result.
The system-matrix entries are monotonous functions in y for positive y and also
for negative y, but not for the whole range of y. By checking the nonlinear air-
craft model it follows that the sign of y indeed in
uences the behaviour. However,
if the sign is xed the varying behaviour is monotonous. This leads to a set of three
values for y: the lower bound, 0, and the upper bound, since the minimum and
maximum of all entries is obtained at either one of the bounds or at 0.
The pre-analysis using 1-dimensional parameter uncertainty modelling thus yields,
that we only have to trim and linearize the nonlinear model at 3 23 = 24 dierent
permutations of values for the 4 varying parameters. This is a signicant reduction
with respect to the 625 permutations mentioned above.
The second step is to obtain two sets of system matrices containing the minima
and maxima of the entries over all system matrices. As an example, for the RCAM
longitudinal design the lower and upper bounds on the state matrix entries are given
as 0 1:2048 0 0:0015 0:0271
1
B 0:9970
Alower = B 0 0 0 C
@ 6:3842 9:7980 0:0414 0:0568 C A
76:7752 1:2943 0:2532 0:7944
and 0 0:7688 0 0:0003 0:0085
1
Aupper = B
B 1:0000 0 0 0 C
@ 0:7925 9:7201 0:0242 0:0912 C A:
77:7185 0:3942 0:2003 0:5442
The classication of varying entries of the system matrices can be applied simply
using the sets of matrices containing the lower and upper bounds. The classications
for the entries of the state matrix are given as
0v 0 v v1
Bv 0 0 0C
Aclass = B
@v v v vC
A:
v v v v
4.2. Parameter Uncertainty Modelling 43
This coincides with the lower and upper bound matrices given above. The classi-
cations for the other system matrices are
0v v v v v v v1
B 0 0 0 0 0 0 0 C 0 0 v v
@v v v v v v vC
Bclass = B A ; Cclass = 0 v v v ;
v 0 0 v v v v
and Dclass = 027. In total there are 36 entries classied as varying entries. This
is 5 varying entries more compared to the 1-dimensional parameter uncertainty of
x only, as considered in Section 4.2.1. The entries that are classied here (in the
4-dimensional case) as varying are constants with respect to varying x. The a21
entry is varying with respect to y only. The entries b32 and b33 are varying with
respect to mass only, and the entries b12 and b13 are varying in both mass and z.
The other 31 varying entries are varying with respect to all 4 uncertain parameters.
We here treat all varying entries the same for implementation reasons. We
therefore make no distinction between the above mentioned 5 varying entries that
do not vary for all 4 uncertain parameters, and the other 31 varying entries.
An initial nominal linear model is dened as the model consisting of the system
matrices as given by the set of the lower bounds on all system matrix entries. The
next step is to dene independent variables each corresponding to a varying entry
in the system matrices with given lower and upper bounds. Next, the varying
entries are replaced by the corresponding independent variables which can vary
between the lower and upper bound of the particular entries. In this way the system
matrices contain the whole class of linear models induced by the uncertainty. For
the longitudinal controller design 36 entries are classied as varying entries which
yields 36 independent variables to describe the uncertainty.
Again, the varying entries can be seen as a nominal value with a perturbation,
which can be regarded as an LFT. The system can thus be seen as an interconnection
of LFTs which can be reformulated as one single LFT using the properties of LFTs.
The nal perturbation block-structure contains 36 repeated real perturbation blocks
of dimension 1 corresponding to the 36 varying entries. The new nominal linear
system is given by the system matrices
0 0:9868 0 0:0006 0:0178
1
B 0:9985
Anom = B 0 0 0 C
C
@ 2:7958 9:7590 0:0328 0:0740 A;
77:2469 0:8443 0:2268 0:6693
0 2:4487 0:3105 0:3105 0:0013 0:0178 0:0006 0:0178 1
B 0
Bnom = B 0 0 0 0 0 0 C
@ 0:1723 9:8100 9:8100 0:0291 0:0735 0:0327 0:0730 C
A;
6:4723 0 0 0:2421 0:6609 0:2265 0:6689
44 Chapter 4. Modelling
0
Cnom = 0 790:8224 00:9982 0:0364
:0363 0:9981 ;
and Dnom = 027. Figure 4.3 shows the frequency response of the nominal model.
Appendix A gives an overview of the used software for the 4-dimensional param-
eter uncertainty modelling.
3
10
2
10
1
10
0
10
−1
10
−2
10
−3
10
−4
10 −4 −3 −2 −1 0 1 2
10 10 10 10 10 10 10
Nominal aircraft longitudinal model, input 2: Ve (−), we (−−)
3
10
2
10
1
10
0
10
−1
10
−2
10
−3
10 −4 −3 −2 −1 0 1 2
10 10 10 10 10 10 10
Figure 4.3: Frequency response of the nominal aircraft longitudinal model to elevator
input (upper) and throttle input (lower) for 4-dimensional parameter uncertainty.
46 Chapter 4. Modelling
1
10
0
10
-1
10
-2
10
-3
10 -4 -2 0 2 4
10 10 10 10 10
Frequency (rad/s)
-2
10
-3
10 -2 -1 0 1 2
10 10 10 10 10
Frequency (rad/s)
Figure 4.4: The performance weighting functions (upper) and the eort weighting
functions (lower).
48 Chapter 4. Modelling
perturbation perturbation
input output
--
-- Aircraft
noise - Wind weighted
Model - Actuator - eort
er
Weight
noise -Turbulence
Model
-+ ?- - Performance -
r
references Weight weighted
- Actuator errors
Model
control measurement
Controller
10
8
MAGNITUDE
0 -4 -2 0 2 4
10 10 10 10 10
FREQUENCY (rad/s)
3.5
2.5
MU
1.5
0.5
0 -4 -2 0 2 4
10 10 10 10 10
FREQUENCY (rad/s)
Figure 5.1: Magnitude (upper) and (lower) of closed-loop system at the 1st iter-
ation for 1-dimensional parameter uncertainty.
5.1. Robust Controller Design for 1-Dimensional Parameter Uncertainty 51
0
10
-1
10
-2
10 -4 -2 0 2 4
10 10 10 10 10
1) mag data 2) newfit 3) previous D-K
SCALED TRANSFER FUNCTION: OPTIMAL & RATIONAL
4
0 -4 -2 0 2 4
10 10 10 10 10
1) MU UPPER BND 2) UPPER BND WITH RATIONAL FIT
Figure 5.2: Example of 1st order D-scaling t. The unit line is the \previous t".
Iteration 1 2 3 4
Controller order 14 34 46 48
Total D-scale order 0 20 32 34
11.580 1.163 0.287 0.214
peak- 3.738 0.508 0.285 0.204
Table 5.1: Summary of the D-K iteration for longitudinal controller design for 1-
dimensional parameter uncertainty.
52 Chapter 5. Controller Design Process
made. Table 5.1 gives the iteration summary for the longitudinal controller design.
Figure 5.3 shows the magnitude (maximum singular values) plot of the resulting
controller. The magnitude and plot of the closed-loop system is given in Figure 5.4.
The nal value of is 0:204, so robust performance is satised for our performance
criteria.
1
10
0
10
-1
10
-2
10 -4 -2 0 2 4
10 10 10 10 10
Frequency (rad/s)
0
10
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10 -4 -2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 5.3: Frequency response of the longitudinal controller obtained by D-K iter-
ation for 1-dimensional parameter uncertainty.
54 Chapter 5. Controller Design Process
0.2
0.15
Magnitude
0.1
0.05
0 -4 -2 0 2 4
10 10 10 10 10
Frequency (rad/s)
(or 2-dimensional) parameter uncertainty to (pre-) select good estimates for the
weighting functions. This in order to minimize the number of iterations necessary
for tuning in the controller design for higher dimensional parameter uncertainty.
The iteration starts by calculating an H1 controller for the unscaled system.
Figure 5.7 shows the singular value plot of the closed-loop system. The maximum
singular value is 19:876. Second, is computed for the closed-loop system. Its peak
value is 1:945. Figure 5.7 also shows as function of frequency.
Next, 36 D-scalings have to be tted for the second iteration. Five of them need
2nd order ts (the 1st, 3rd, 5th, 13th, and 33rd D-scalings), the other ts are 1st
order ts.
Then an H1 controller is calculated for the new scaled open-loop system, after
which again is computed for the closed-loop system. Five iterations have been
applied of which Table 5.2 gives a summary. Again mainly 1st order D-scaling ts
are applied. In the third iteration 2nd order ts are used for the same D-scalings as
for the second iteration. In the fourth iteration 2nd order ts are used for the same
D-scalings as in the 3rd iteration plus for the 27th D-scaling. In the fth iteration
2nd order ts are used for the 1st, 3rd, 13th, 27th, and 33th D-scalings, and 3rd
order ts are used for the 5th and 20th D-scaling ts.
The nal singular value of the closed-loop system is 0:830, and the nal peak
value of is 0:822, by which robust performance is satised for the selected weighting
functions. Figure 5.8 shows the singular value and plot for the resulting closed-loop
system. The nal controller is shown in Figure 5.9.
5.2. Robust Controller Design for 4-Dimensional Parameter Uncertainty 55
2
10
Log Magnitude
0
10
-2
10
-4
10 -4 -2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
Reduced controller, order 14, output 1 (original ooo )
400
Phase (degrees)
200
-200 -4 -2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
5
10
Log Magnitude
0
10
-5
10
-10
10 -4 -2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
Reduced controller, order 14, output 2 (original ooo )
400
Phase (degrees)
200
-200 -4 -2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
Figure 5.5: Original and reduced order longitudinal controller bode plots for 1-
dimensional parameter uncertainty.
56 Chapter 5. Controller Design Process
0.2
0.15
Magnitude
0.1
0.05
0 -4 -2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 5.6: Plot of for closed-loop system corresponding to the reduced longitu-
dinal controller for 1-dimensional parameter uncertainty.
Iteration 1 2 3 4 5
Controller order 14 96 96 98 104
Total D-scale order 0 82 82 84 90
19.876 1.318 1.001 0.862 0.830
peak- 1.945 1.209 0.965 0.858 0.822
Table 5.2: Summary of the D-K iteration for longitudinal controller design for 4-
dimensional parameter uncertainty.
5.2. Robust Controller Design for 4-Dimensional Parameter Uncertainty 57
15
MAGNITUDE
10
0 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10
FREQUENCY (rad/s)
CLOSED−LOOP MU: CONTROLLER #1
2
1.5
MU
0.5
0 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10
FREQUENCY (rad/s)
Figure 5.7: Magnitude (upper) and (lower) of closed-loop system at the 1st D-K
iteration for 4-dimensional parameter uncertainty.
58 Chapter 5. Controller Design Process
0.8
0.7
0.6
Magnitude
0.5
0.4
0.3
0.2 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 5.8: Magnitude and for longitudinal closed-loop system obtained by D-K
iteration for 4-dimensional parameter uncertainty.
2
10
1
10
0
10
−1
10
−2
10
−3
10 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Magnitude plot of controller, k1, output2: Ve (−), we (−−)
1
10
0
10
−1
10
−2
10
−3
10
−4
10 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 5.9: Frequency response of the longitudinal controller obtained by D-K iter-
ation for 4-dimensional parameter uncertainty.
60 Chapter 5. Controller Design Process
4
10
2
10
Log Magnitude
0
10
−2
10
−4
10 −4 −2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
200
Phase (degrees)
−200
−400 −4 −2 0 2 4
10 10 10 10 10
2
10
Log Magnitude
0
10
−2
10
−4
10 −4 −2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
Reduced controller, order 15, output 2 (original ooo )
300
Phase (degrees)
200
100
−100 −4 −2 0 2 4
10 10 10 10 10
Frequency (radians/sec)
Figure 5.10: Original and reduced order longitudinal controller bode plots for 4-
dimensional parameter uncertainty.
5.2. Robust Controller Design for 4-Dimensional Parameter Uncertainty 61
0.8
0.7
0.6
Magnitude
0.5
0.4
0.3
0.2 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 5.11: Plot of for closed-loop system corresponding to the reduced longitu-
dinal controller for 4-dimensional parameter uncertainty.
Chapter 6
Analysis of the Designed
Controller
6.1 Introduction
There are three ways to analyse the performance of the designed controller. The
rst consists of the tests. This is the subject of Section 6.2. Section 6.3 consid-
ers closed-loop time responses which are ecient tools to analyse the performance
of the designed controller. The third analysis tool is nonlinear simulations. The
GARTEUR Action Group FM(AG08) [9] has developed an automatic evaluation
procedure to test the performance of the controller with respect to the nonlinear
model. However, for this procedure also a lateral controller is needed. For this
reason the nonlinear simulations have not been implemented at this stage.
6.2 Analysis
In Section 5.2 we have designed a longitudinal controller of order 15. The controller
is designed in such a way that of the closed-loop system is smaller than 1 and thus
satises the robust performance criteria. However, in the -synthesis procedure we
had to replace the real perturbation blocks by complex blocks. We therefore also
compute with respect to the original perturbation structure of 36 1-dimensional
(repeated) real pertubation blocks and a complex full performance perturbation
block. Figure 6.1 shows real- of the closed-loop system with the original perturba-
tion block, and complex- of the closed-loop system with the complex perturbation
block-structure as used for the controller design. It is seen that real- and complex-
are not that far o. This is probably due to the fact that the repeated real blocks
are all of dimension 1, and are thus \only" assumed to be complex while the block
dimensions stay the same. The controller designed by the (complex-) -synthesis
approach, is thus also a good robust controller for the original system with the real
perturbation block-structure.
62
6.3. Time Responses 63
0.8
0.7
0.6
Magnitude
0.5
0.4
0.3
0.2
0.1 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 6.1: Robust performance: real- and complex of the longitudinal closed-loop
system of the designed controller.
Since the closed-loop system satises robust performance, the closed-loop system
is also robustly stable. Robust stability is tested by computing with respect to
the original real perturbation block-structure and without the performance pertur-
bation. Figure 6.2 shows the robust stability plot. It is seen that robust stability is
satised.
It follows that the designed controller satises the tests for the original real per-
turbation block-structure. The closed-loop system is thus robust to all uncertainty
in the 4 uncertain parameters for the selected weighting functions.
0.7
0.6
0.5
real−mu
0.4
0.3
0.2
0.1
0 −4 −2 0 2 4
10 10 10 10 10
Frequency (rad/s)
Figure 6.2: Robust stability: real- of the longitudinal closed-loop system of the
designed controller with respect to the parameter uncertainty perturbation structure
only.
corresponding control inputs E and TH are shown in the lower left and right,
respectively. Figure 6.4 shows the response to the same unit step in VC for the
linear model with mass = 150 000 kg and the displacement of the aerodynamic
center from the center of gravity (1:25; 0:2; 1:38). V should be able to track the
speed command VC with a rise time tr < 12 s and settling time ts < 45 s. The plots
show that these criteria are satised.
Figure 6.5 and 6.6 show the response to a unit step in the vertical speed command
wC for the same two linear models as above. Here, also good responses are obtained.
Figure 6.7 and 6.8 show the responses to a step in tailwind of magnitude 13
m=s. There should be no deviation in V larger than 2.6 m=s for more than 5 s. The
gures show that the deviation does not exceed the 2.6 m=s bound. Also no steady
state error due to constant wind is allowed. This criteria is also satised.
6.3. Time Responses 65
1 0.03
0.02
0.01
0.5
0
−0.01
0 −0.02
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
x 10 wpnom, V cmd: delt e wpnom, V cmd: throttle
−3
8 0.02
6 0.015
4 0.01
2 0.005
0 0
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.3: Time response to a unit step in VC for the nominal model.
wp2232, V cmd: Ve, Vcmd wp2232, V cmd: we
0.05
1 0.04
0.03
0.02
0.5
0.01
0 −0.01
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
x 10 wp2232, V cmd: delt e x 10 wp2232, V cmd: throttle
−3 −3
8 20
15
6
10
4
5
2
0
0 −5
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.4: Time response to a unit step in VC for the linear model with
mass=150 000, x = 1:25 m, y = 0:2 m, and z = 1:38 m.
66 Chapter 6. Analysis of the Designed Controller
0.2
1
0.1
0
0.5
−0.1
−0.2 0
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
wpnom, w cmd: delt e wpnom, w cmd: throttle
0.2 0.02
0
0
−0.2
−0.02
−0.4
−0.04
−0.6
−0.8 −0.06
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.5: Time response to a unit step in wC for the nominal model.
wp2232, w cmd: Ve wp2232, w cmd: we, wcmd
0.4 1.5
0.2 1
0 0.5
−0.2 0
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
wp2232, w cmd: delt e wp2232, w cmd: throttle
0.2 0.02
0 0
−0.2 −0.02
−0.4 −0.04
−0.6 −0.06
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.6: Time response to a unit step in wC for the linear model with
mass=150 000, x = 1:25 m, y = 0:2 m, and z = 1:38 m.
6.3. Time Responses 67
10
0.5
0
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
wpnom, uw: delt e wpnom, uw: throttle
0.15 0.04
0.03
0.1
0.02
0.05
0.01
0
0
−0.05 −0.01
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.7: Time response to a step in tailwind of magnitude 13 m=s for the nominal
model.
wp2232, uw: Ve, uw wp2232, uw: we
15 1.5
1
10
0.5
5
0
0 −0.5
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
wp2232, uw: delt e wp2232, uw: throttle
0.1 0.06
0.04
0
0.02
−0.1
0
−0.2 −0.02
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
Figure 6.8: Time response to a step in tailwind of magnitude 13 m=s for the linear
model with mass=150 000, x = 1:25 m, y = 0:2 m, and z = 1:38 m.
Chapter 7
Conclusions and
Recommendations
Uncertain parameters give rise to varying entries in the state-space data. An LFT
description of the aircraft longitudinal model can be obtained using parameter uncer-
tainty modelling. For 1-dimensional and 2-dimensional parameter uncertainty, vary-
ing entries can be t by polynomials in the varying parameter(s). For 4-dimensional
parameter uncertainty, the varying entries of the state-space data are considered as
separate independent parameters with given lower and upper bounds.
Using polynomial ts for the varying entries, as is implemented for the one
uncertain parameter case, leads to a perturbation block with high dimensions for
each uncertain parameter. For the RCAM, an LFT description has been obtained for
1-dimensional parameter variation in x only. The resulting LFT has a perturbation
block-structure of one 10-dimensional repeated real scalar block, corresponding to
perturbations in x.
The alternative modelling method, illustrated by the 4-dimensional parame-
ter uncertainty case, leads to a perturbation block-structure of a number of 1-
dimensional real scalar blocks. The number of these blocks equals the number of
varying entries in the state-space data. For the RCAM, an LFT description with
respect to all 4 uncertain parameters has been obtained. The LFT has a perturba-
tion block-structure of 36 1-dimensional (repeated) real scalar perturbation blocks,
corresponding to 36 varying entries in the state-space data.
To design a robust controller using D-K iteration, the (repeated) real scalar
blocks in the perturbation block-structure have to be replaced by (a number of)
1-dimensional complex full blocks. This is due to the fact that (in the present
MATLAB toolbox) the D-K iteration approach to synthesis can handle perturba-
tion block-structures having full complex blocks only. For the modelling procedure
using polynomial ts, this introduces large conservatism in the uncertainty struc-
ture. A large gap may result between the values of used for synthesis, and the
values of computed for the original real perturbation structure. For the alternative
modelling approach the introduced conservatism is much less, since here the original
68
69
formance with respect to 1-dimensional parameter uncertainty only, for all uncertain
parameters separately. Since, for 1-dimensional parameter uncertainty the pertur-
bation blocks are much smaller and therefore the D-K iteration is much faster.
If necessary, the weighting functions are tuned in these pre-analyses. It is then
hoped that the weighting functions that have good performance for the individual
1-dimensional parameter uncertainty cases, also have good performance for the 4-
dimensional case. For the RCAM longitudinal controller design this approach has
been very eective.
In the longitudinal controller design of the RCAM aircraft in this report, time
delay has not been taken into consideration. Time delay can be implemented by
adding a system based on a Pade approximation to the standard plant. The con-
troller design cycle should then be repeated with respect to this standard plant.
Note that the parameter uncertainty modelling of the aircraft does not have to be
repeated, since the LFT of the aircraft is a separate system in the interconnection
structure of the standard plant.
For the nal control system of the RCAM also a lateral controller has to be
designed. This can be done analogous to the design of the longitudinal controller,
but with respect to another controller design architecture.
References
[1] Balas, G.J., Doyle, J.C., Glover, K., Packard, A.K., Smith, R., -Analysis and
Synthesis Toolbox User's Guide, MUSYN Inc. and The MathWorks Inc., Natick,
1993.
[2] Blight, J.D., Dailey, R.L., Gangsaas, D., \Practical Control Law Design for
Aircraft Using Multivariable Techniques", International Journal of Control,
Vol. 59, No. 1, Special Edition on Aircraft Flight Control, pp. 93-137, 1994.
[3] Doyle, J.C., \Analysis of Feedback Systems with Structured Uncertainties",
IEE Proceedings, Vol. 129, Part. D, No. 6, pp. 242-250, 1982.
[4] Doyle, J.C., Glover, K., Khargonekar, P.P., Francis, B.A., \State-Space So-
lutions to Standard H2 and H1 Control Problems", IEEE Transactions on
Automatic Control, Vol. 34, No. 8, pp. 831-847, 1989.
[5] Doyle, J.C., Packard, A., Zhou, K., \Review of LFTs, LMIs, and ", Proceed-
ings of the 30th IEEE Conference on Decision and Control, Brighton, England,
pp. 1227-1232, 1991.
[6] Fan, M.K.H., Tits, A.L., \Characterization and Ecient Computation of
the Structured Singular Value", IEEE Transactions on Automatic Control,
Vol. AC-31, No. 8, pp. 734-743, 1986.
[7] Fan, M.K.H., Tits, A.L., Doyle, J.C., \Robustness in the Presence of Mixed
Parametric Uncertainty and Unmodeled Dynamics", IEEE Transactions on
Automatic Control, Vol. 36, No. 1, pp. 25-38, 1991.
[8] Francis, B.A., Khargonekar, P.P. (Eds.), Robust Control Theory, Proceedings
of a Workshop that Formed Part of the 1992-93 IMA Program on \Control
Theory", The IMA Volumes in Mathematics and its Applications, Vol. 66,
Springer-Verlag, New York, 1995.
[9] GARTEUR Action Group FM(AG08), Robust Flight Control Design Challenge
Problem Formulation and Manual: the Research Civil Aircraft Model (RCAM),
GARTEUR/TP-088-3, GARTEUR, 1995.
71
72 References
[10] GARTEUR Action Group FM(AG08), Robust Flight Control Design Challenge
Problem Formulation and Manual: the High Incidence Research Model (HIRM),
GARTEUR/TP-088-4, GARTEUR, 1995.
[11] Gerlach, O.H., Vliegeigenschappen I, Lecture Notes Delft University of Tech-
nology, Faculty of Aerospace, Vol. D 26-1 & D 26-2, Delft, 1981.
[12] Glover, K., \A Tutorial on Hankel-Norm Approximation", In: Willems, J.C.
(Ed.), From Data to Model, pp. 26-48, Springer-Verlag, Berlin, 1989.
[13] Glover, K., Doyle, J.C., \State-Space Formulae for All Stabilizing Controllers
That Satisfy an H1 -Norm Bound and Relations to Risk Sensitivity", Systems
& Control Letters, Vol. 11, pp. 167-172, 1988.
[14] Goverde, R.M.P., Robust Flight Control System Design Using H1=-Synthesis,
NLR Technical Report, TR 95356 L, National Aerospace Laboratory NLR,
Amsterdam, 1995.
[15] Goverde, R.M.P., Longitudinal Flight Control Law Design for the RCAM Design
Challenge: the -Synthesis Approach, NLR Technical Report, TR 96063 L,
National Aerospace Laboratory NLR, Amsterdam, 1996.
[16] Lambrechts, P., Terlouw, J., Bennani, S., Steinbuch, M., \Parametric Uncer-
tainty Modeling Using LFTs", Proceedings of the American Control Conference,
San Fransisco, California, pp. 267-272, 1993.
[17] Maciejowski, J.M., Multivariable Feedback Design, Addison-Wesley, Woking-
ham, 1989.
[18] Packard, A., Doyle, J.C., \The Complex Structured Singular Value", Automat-
ica, Vol. 29, No. 1, pp. 71-109, 1993.
[19] Safonov, M.G., Limebeer, D.J.N., Chiang, R.Y., \Simplifying the H 1 The-
ory via Loop-Shifting, Matrix-Pencil and Descriptor Concepts", International
Journal of Control, Vol. 50, No. 6, pp. 2467-2488, 1989.
[20] Schuring, J., Example of H1 =-Synthesis in Robust Aircraft Flight Control
System Design, NLR Contract Report, CR 94127 L, National Aerospace Lab-
oratory NLR, Amsterdam, 1994.
[21] Terlouw, J.C., Lambrechts, P.F., A Matlab Toolbox for Parametric Uncertainty
Modelling, NLR Technical Publication, CR 93455 L, National Aerospace Lab-
oratory NLR, Amsterdam, 1993.
[22] Young, P.M., Newlin, M.P., Doyle, J.C., \Let's Get Real", In: Francis, B.A.,
Khargonekar, P.P. (Eds.), Robust Control Theory, Proceedings of a Workshop
That Formed Part of the 1992-93 IMA Program on \Control Theory", The IMA
References 73
Volumes in Mathematics and its Applications, Vol. 66, pp. 143-173, Springer-
verlag, New York, 1995.
Appendix A
Software Description
A.1 Introduction
This Appendix gives an overview of the software that is used to design a robust
control system for the RCAM. The algorithms are based on the mathematical soft-
ware package MATLAB including the simulation package SIMULINK. In addition,
three MATLAB toolboxes are used. These toolboxes are the Robust Control Toolbox
(used implicitly) and the -Analysis and Synthesis Toolbox (mutools), which are of-
cial MATLAB toolboxes of the MathWorks, Inc., and the Parametric Uncertainty
Modelling (PUM) Toolbox which has been jointly developed at the TUD, the NLR,
and the Philips Research Laboratories NatLab [16, 21].
A controller for the longitudinal dynamics and a controller for the lateral dynam-
ics can be designed separately as a result of the decoupling property of the aircraft
dynamics into longitudinal and lateral dynamics. Hereby, the control system de-
sign is divided into two separate control system designs which are taken together to
obtain the nal control system.
The control system design approach can then again roughly be divided into three
parts. The rst part consists of the uncertainty modelling of the aircraft. Here, a
linear model of the aircraft is obtained in which possible uncertainties are taken into
account. The second part is the actual controller design using the uncertain model
of the aircraft. The nal phase is the simulation phase, in which the control system
is tested to yield good results for the full nonlinear model of the aircraft.
for both longitudinal and lateral controller design. Only some adjustments have
to be made in the routine rcamum1 with respect to the declaration of the varying
parameter.
on systems or to plot system outputs are not included in the table. These routines
include minfo, rifd, spoles, logspace, frsp, sel, and vplot.
A.3. Controller Design 77
General Variables
A state (system) matrix,
B input (system) matrix,
C output (system) matrix,
dci ith repeated complex scalar,
D direct-feedthrough (system) matrix,
D scaling matrix,
D~ (s) real-rational stable minimum-phase approximation to D scaling,
G(s) (open-loop) transfer matrix,
ki (row and column) dimension of the ith perturbation block,
K (s) feedback transfer matrix (compensator),
m1 dimension of exogenous input w(t), w(s),
m2 dimension of control u(t), u(s),
m11 dimension of perturbation input w1(t), w1(s),
m12 dimension of performance input w2(t), w2(s),
mc number of repeated complex scalar blocks,
mC number of full complex blocks,
mr number of repeated real scalar blocks,
80 Appendix B. Nomenclature
Aircraft Variables
Lg turbulence scale length of Dryden spectrum,
m aircraft total mass,
q pitch rate,
uB velocity along x-body axis,
uC reference velocity in x-direction,
v0 trim speed,
vC reference velocity in y-direction,
V total velocity,
VC reference velocity,
Ve velocity error VC V ,
wB velocity along z-body axis,
wC reference velocity in z-direction,
we vertical velocity error wC wV ,
wV velocity along z-vehicle axis,
81
Abbreviations
GARTEUR Group for Aeronautical Research and Technology in EURope,
HIRM High Incidence Research Model,
LFT Linear Fractional Transformation,
NLR Nationaal Lucht- en Ruimtevaartlaboratorium (National Aerospace
Laboratory),
RCAM Research Civil Aircraft Model,
TUD Technische Universiteit Delft (Delft University of Technology),
WBBM Wiskundige Beheers- en Beleidsmodellen.