18.

An Eigenstructure Assignment
Approach (2)

J e s d s M . d e l a C r u z 1, P a b l o R u i p 6 r e z 2
and Joaquin Aranda 2

A b s t r a c t . In this chapter the Eigenstructure Assignment Method

is applied to the Research Civil Aircraft Model (RCAM) bench-
mark problem. The design is done by making use of the classi-
cal approach which consists of splitting up the controller into two
parts, a longitudinal and a lateral controller, and in using the stan-
dard inner-outer loop control structure. This method allows us
to directly satisfy specifications in terms of transient response and
modes decoupling. However, it does not cope directly with system
uncertainties. In order to cope with robustness, measurements of
multiloop gain and phase margins are used in the choice of a robust
eigenstructure. Both channels make use of a constant gain matrix
in the inner loop and a scalar gain in the outer loop. In spite of the
controller simplicity, good performance and robustness results are
obtained, although robustness may be increased by means of gain
scheduling with respect to airspeed.

18.1 Introduction
The eigenstructure technique has been widely applied to the design of flight
control systems and it is a well-known fact that its use requires an in-depth
knowledge of the system to be controlled [11]. Although this method is not
intended to deal with robustness, many robustification procedures have been
proposed [155]. Here the eigenstructure is chosen so that good multiloop gain
and phase stability margins are obtained.
The chapter is organised in the following way. Section 18.2 describes the
controller structure used. The controller has been decoupled into the longi-
tudinal and the lateral channels. Both have been designed according to an
inner/outer loop control structure. The inner loop controllers are designed
following the eigenstructure method. A constant gain matrix is used in both
channels. The outer loop only uses a scalar gain that is calculated by means
of the root locus method.
1Dpt. Inform~ticay Autom~.tica. Facultadde CienciasFfsicas. UniversidadComplutense.
28040-Madrid. Spain. (Funded by project CICYT TAP94-0832-C02-01).
2Dpt. Inform~ticay Autom£tica. Facultad de Ciencia~. U.N.E.D. 28040-Madrid. Spain.

238
 Section 18.3 deals with the way the RCAM design criteria are translated
into the desired eigenstructure.
Section 18.4 describes the design cycle. The linear model of the plant is
analysed and the most appropriate eigenstructure is chosen. Next, the feed-
back controller is obtained and the performance of the closed-loop system for
the linear models is analysed. Multiloop gain and phase margins are used to
measure robustness and to guide eigenstructure choice.
In section 18.5 the verification of all the design specifications with the non-
linear system is presented. Simulations for the worst possible combinations of
delay, mass and centre of gravity are given.
In section 18.6 the results of the automated evaluation procedure are given.
The theory has been presented in chapter 3. More details, including pro-
gram code for the design written in Matlab can be found in [52]

18.2 S e l e c t i o n of t h e controller a r c h i t e c t u r e
We make use of the classical approach consisting of splitting up the controller
into two parts: a longitudinal and a lateral controller. Both of them have been
designed in two stages: the inner loop and the outer loop. The function of
the inner loop is to make the aircraft easy and pleasant to fly, and it is often
called a stability augmentation system. The outer loop function is to replace
the pilot for certain flight manoeuvres such as maintaining height and speed,
turning onto a specified heading, climbing at a specified rate, etc.
Figure 18.1 shows the structure of the longitudinal controller, and figure
18.2 shows the structure of the lateral controller.

[ (z,q. nx, oz. w v , V A )

Figure 18.1: Longitudinal controller

18.2.1 Longitudinal c o n t r o l l e r
The choice of the signals to be used in a design is based on the analysis of the
system, the specifications and the design method.

Measurement signals
Table 18.1 shows the measurements as used by the longitudinal controller.

239
 ÷
Actv~a¢~

(Y,P,P,r,t,Z)

Figure 18.2: Lateral controller

Inner Loop controller

q Pitch rate
n= Horizontal load factor
nz Vertical load factor
wv z component of inertial velocity
VA Air speed
Outer Loop controller
z z position of aircraft CoG

Table 18.1: Longitudinal measurements used

Although in the inner loop only four measurements are needed to assign
four eigenvalues, two for the short period and two for the phugoid, we make
use of five in order to prevent the slower mode of the actuators (that of the
throttle) from becoming unstable, see [52].
Finally, the outer loop provides altitude tracking by adding a feedback of
the altitude z.

Actuator signals
These signals are the elevator deflection or tailplane deflection ~T, and throttle
position 5TH.

R e f e r e n c e signals

The selection of signals chosen as references has been guided by the specifi-
cations given as design criteria, see §18.3. The selected ones are the reference
velocity wvc and the reference airspeed VAc for the inner loop, and the reference
position z c for the outer loop.

240
Controller structure
The inner loop controller has a static gain matrix acting on the five chosen
measured signals and on the integral of the errors of the commanded variables
w y and VA, in the order here specified. The two integrators result in two addi-
tional states that must be incorporated into the linear model for the controller
design. The outer loop has a proportional action acting on the altitude error.
No integral action is needed here to avoid steady state errors related to altitude
step commands or disturbances, since the altitude dynamics include a pole at
the origin. The output of the outer loop acts as a reference for w v .

18.2.2 Lateral Controller

The lateral controller has a structure analogous to the longitudinal controller.

M e a s u r e m e n t signals
Table 18.2 shows the measurements as used by the lateral controller.

Inner Loop controller

/3 Angle of sideslip
p Roll rate
r Yaw rate
¢ Roll angle
X Inertial track angle
Outer Loop controller
Ay Lateral deviation

Table 18.2: Lateral measurements used

A c t u a t o r signals
These signals are the aileron deflection ~A, and rudder deflection 6a.

Reference signals
Again, the selection of references has been guided by the specifications given as
design criteria, see §18.3. Those selected are the lateral deviation for the outer
loop, and the inertial track angle X c and sideslip angle/3c for the inner loop.
X c is not given as a reference signal but it is obtained from the reference velocity
components u c and v c as a t a n ( v c / u c ) . /3c has a null constant reference value
in order to keep/3 always close to zero.

Controller structure
The inner loop controller has a static gain acting on the five chosen signals
measured, and on the integral of the errors of the commanded variables j3 and

241
X, in the order specified here. The introduction of two integrators results in
two additional states that must be incorporated into the linear model for the
controller design. The outer loop has a proportional action acting on the lateral
error. No integral action is required here to avoid steady state errors relating
to lateral step commands or disturbances, since the lateral dynamics include
a pole at the origin. The output of the outer loop acts as a reference for the
inertial track angle.

18.3 Translation of design criteria into method

dependent objectives
Out of the five classes of design criteria given in chapter 14, for the RCAM de-
sign, the performance criteria are the most significant. These criteria are given
in terms of transient response characteristics to command signals and cross
coupling constraints. The main characteristic of the eigenstructure assignment
method is that it allows the designer to satisfy specifications directly in terms
of transient response and mode decoupling. Then, these are the most suitable
criteria to be used as a guide in the design phase.
Our design is based mainly on these criteria, but we made use of linear and
non-linear closed-loop time responses and robust analysis tools as a guide in
the selection of the eigenstructure.

18.3.1 Performance Criteria

The performance criteria can be classified into two groups: longitudinal and
lateral. We discuss separately for each group the way in which the specifica-
tions have been incorporated into the design.

L o n g i t u d i n a l specifications
There are command response specifications in terms of overshoot, rise time
and settling time for three commanded signals: airspeed, flight path angle and
altitude. These specifications provide a lower limit for the damping ratio and
natural frequency of second order modes and for the time constant of first order
modes coupled with the signals.
The flight path angle 7 is neither available as an output nor as a reference
signal. To cope with this situation we use the relationship sin(7) = - w v / V ,
where V is the total inertial velocity. Therefore, for constant inertial velocity
the flight path angle may be controlled by means of w v . This leads us to an
interpretation of the specifications in terms of commands in w y c . We add a
vertical velocity error integrator state to get good low-frequency tracking.
We add an outer loop for tracking the reference altitude. A signal propor-
tional to the altitude error is used as a reference signal for w y . The proportional

242
gain is calculated using the root locus method in such a way that it fulfils the
transient criteria.
The specification of decoupling between airspeed and altitude may be ob-
tained by decreasing cross-coupling between VA and wv.

Lateral specifications
There are now two command signal specifications (heading angle and lateral
deviation) in terms of transient response characteristics that may be trans-
formed into bounded eigenvalues. The remaining specifications are given in
terms of behaviour in case of engine failure and under disturbances.
The heading angle is a lateral motion state but it is neither available as an
output nor as a reference. Instead, we make use of the inertial track angle that
is an output signal and is related to the heading angle by means of the equation
X = ¢ + j3y. Although there is no reference signal for X, such a signal may
be obtained from the reference velocity components u c and vc by means of
equation Xc = tan-1 (vc/uc). An integrator is introduced to eliminate sideslip
errors, where the command signal for/3 is chosen as 0. Doing so, we cope with
the safety criteria of keeping/3 minimised at all times. An inertial track angle
error integrator is also introduced in order to avoid heading angle steady state
errors.
The lateral deviation is controlled in the outer loop. A signal proportional
to the lateral error is used as a reference signal for the inertial track angle. The
proportional gain is calculated using the root locus method in such a way that
it fulfils the transient criteria.
The engine failure requirements can not be easily interpreted in terms of
eigenvalue and eigenvector specifications and shall not be tested before the
phase of analysis of simulation results. However, the requirement of keeping
sideslip angle to a minimum in case of engine failure can be translated into a
specification of decoupling between the eigenvalues related with the roll motion
and those related with the lateral velocity.

18.3.2 Robustness criteria

The eigenstructure method is not a robust control method and, although many
different algorithms have been proposed to enhance the robustness, no method
will be used here. However, we make use of the stability margins given in chap-
ter 3 to measure the robustness of the feedback system. For a chosen eigen-
structure the stability margins are measured and the eigenstructure changed
to get better stability margins. After a few steps a suitable eigenstructure is
selected.

18.3.3 Other criteria

The ride quality criteria, safety criteria and control activity criteria can not
be incorporated in an ad hoc manner into the eigenstructure method. As done

243
with robustness, they will be analysed in the course of the analysis of results
phase and physical relations between their behaviour and the eigenstructure
achieved will be established in order to cope with them. In the selection of
the eigenvalues we should have in mind to assign the mode values close to the
open-loop aircraft modes to minimise the control activity.

18.4 Design cycle and controller derivation

The design cycle is summarised in the following scheme

1. Analysis of the linear and non-linear model of the plant.

2. Selection of the eigenstructure and gain calculation.
3. Analysis of the open and closed-loop system (linear and non linear
model).
4. Robustification procedure (iterate).
5. Iterate 1 - 4.

As we have seen in the previous section, among all the design criteria only
those of the performance criteria related to the transient response of the system
can be interpreted almost directly in terms of eigenstructure. The rest of them
must be analysed after an eigenstructure has been chosen and the controller
found. From this analysis another eigenstructure will be chosen and so on.
That has been the most consuming time task. A robustification procedure or
any other form of "optimal" solution may help to break the iterative procedure.
We will now explain the method we have used to select an eigenstructure.

• We analysed the design criteria and the coupling of the modes of the
plant with the states, the inputs and the outputs. This analysis was used
as a guide in choosing the outputs for feedback and the eigenstructure.
After selecting the outputs to be used for feedback the integrators were
added to the loop.

• We chose a set of eigenvalues that fulfilled the transient response limits in

accordance with the requirements and were close to the natural aircraft
modes. After that, their associated eigenvectors were chosen to get the
desired decoupling between the modes.

• With the eigenstructure chosen, we computed the feedback gain and anal-
ysed:

- the stability (since one of the non-assigned eigenvalues might become

unstable or badly damped) and the stability margins
- the time response of the system
- the decoupling of the obtained modes

244
 Now the cycle begins. New outputs and/or eigenstructure must be chosen
to improve the results obtained. We have first selected an eigenstructure that
provides acceptable design criteria. We have tested different eigenvalues with
fixed eigenvectors. Once the eigenvalues that give better stability margins have
been chosen, the eigenvectors have been changed to try to improve robustness
and, when necessary, decoupling. After a few steps we convinced ourselves that
the chosen eigenvectors could not be improved.

18.4.1 Longitudinal controller

Longitudinal model

The non-linear model is used to generate linear models for control law design
and to generate non-linear time histories for evaluating control designs. Once
a trim condition is established for the non-linear aircraft model within the sim-
ulation environment, a linear model is generated to capture the perturbational
dynamics around the equilibrium point. The model has been linearised around
the following operating condition: V -- 80 m/s, h = 1000 m, m a s s = 120000
kg, c g x = 0.23 and c g z = 0.1. The aerodynamic model is augmented with first
order actuator models. Two integrated error states are added to the linearised
model, one for w y and another for VA. The number of outputs is now increased
by two and the eigenvalues for the modes of the integrators may be specified.
The maximum allowed transport time delay of 0.10 s is added to the model
with a first order Pad@ approximation.

I n n e r loop controller design

As mentioned in chapter 3 we can specify as many closed-loop eigenvalues as

outputs for feedback used. Therefore, only four measurements are needed to
specify the phugoid and short period modes. However, we used five measure-
ments so that the slower mode of the actuators (that of the throttle) is specified,
in order to avoid it becoming unstable.

Mode Phugoid Short Period Throttle f_ w v -J " V A

Eigenvalues --0.4376 --0.9059 --0.5 --2.0000 --1.9000
=h0.0624z =h0.4388z
q xx ix x x x
0 xl xx x x x
UB lx 00 x 0 x
wB 00 xl x x 0
X2" xx xx x x x
XT H X X X X 1 X X
delay 6T x x x x x x x
delay 6TH x x x x X x X
,~V: O0 xx x 1 0
xx 00 x 0 1

Table 18.3: Desired eigenstructure of the longitudinal closed-loop system

245
 Table 18.3 shows the eigenstructure chosen for the system. The state com-
ponents are given in the first column, where XT is the state corresponding to
the first order tailplane model and XTH the state corresponding to the first
order engine model. The first row shows the desired eigenvalues, and the de-
sired eigenvectors are shown underneath, where the symbol "x" represents the
unspecified elements in the eigenvectors. The resulting gain is:
LonKin__ [0.4755 0.0532 -0.0838 -0.0169 -0.0055 -0.0033 -0.0014]
0.0455 -1.3063 -0.3047 -0.0152 -0.1221 0.0004 -0.0227

Loop stability margins

Figures 18.3 and 18.4 show the singular value plots of the sensitivity function S,
the complementary sensitivity function T and the balanced sensitivity function
S + T at the actuator inputs and at the sensor outputs. Tables 18.4 and 18.5
show the gain and phase margins obtained from the sensitivity functions.
8, T and S.T (NO) ~L T mid ~ T ((IH]
lo' . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
! ! ! :!!!i :::! =:!!!~i !!! ~ :!!! i ! i : : ~:::::::::::" :::~ ' :::: ~:--
!!!!::T:!::!!I! . . . . . . . . ~ ::::::::::::::::::::: . . . . . . . . . . . •. . . . . . . . . . : ======================: .

......... L...i. i_i ~iD.i ......... i...i. ~..~.~i.i2i........ !..! ::. i..i ~,~.:.
i i ~-,~ii.---:: :~ ~'- i
lo" ,o" . : - * : ~ - - ~-.'~.~i~:; : " ....... !

...... :::::::::::::::::::::::
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

" i ' " " i " i "!"i!'!-i ......... !-'!"!"i"!';'~-'!'c-"-" i " "i-!"::'i'i-!';
1o"'
t : ; ~z ~ ~:~i~ .... i :~ ~ ; : ~ ;~:~: ~,$,~ " t it::: i ;:

::::::i:::::;::i:::;:i ;:~:;::::::::i:::::: :: :i:! i:i:;::: ::~:'~;:::i:::i::i::?i::: • . .::'.::: : ........ ~k! :

,o"
lo" m ' lo" ~o' olo " io ' lo" lo*

Figure 18.3: Longitudinal inner Figure 18.4: Longitudinal inner

loop singular values of the input loop singular values of the output
sensitivity functions sensitivity functions

Function F w K,, = 1/~" Gain margin (dB) Phase margin (deg)

S 1.42 0.57 0.70 [-4.6,10.6] -4-41.2
T 1.03 0.09 0.98 [-29.7,5.9] -4-57.8
S + T 1.96 0.48 0.51 [-9.8,9.8] -/-54.1

Table 18.4: Longitudinal inner loop stability margins at the inputs

Function ~" w Km = 1/~" Gain margin(dB) Phase margin (deg)

S 1.47 0.54 0.68 [-4.5,10.0] d:39.9
T 1.00 0.02 1.00 [-54.4,6.0] =1=59.9
S + T 1.96 0.46 0.51 [-9.8,9.8] -4-54.1

Table 18.5: Longitudinal inner loop stability margins at the outputs

Good stability margins are obtained, but we must remember that these
margins are conservative, and even better stability margins should be expected.

246
O u t e r l o o p c o n t r o l l e r des i gn

Choosing LonKo = 0.1027, the slower roots are -0.14 ± 0.14~, having a rise
time of 11 s, which is less than the specified rise time for an altitude command
(15 s), and a settling time of 35.4 s which is well below the required 45 s. The
gain margin for the outer loop is 13 dB at w = 0.35 rad/s and the phase margin
is 63 deg at w = 0.1 rad/s.

18.4.2 Lateral controller

Lateral model

The procedure followed in dealing with the lateral model is analogous to the
one used with the longitudinal model.
A linear model is generated from the same trimmed condition used to obtain
the linearised longitudinal model. The actuator dynamics have been added to
the linear model by augmenting it with corresponding states. Also, we added
the two integrated error states and a 0.10 s delay with a first order Pad~ ap-
proximation.

I n n e r l o o p c o n t r o l l e r des i gn

Table 18.6 shows the eigenstructure chosen. The state components are given in
the first column, where XA is the state corresponding to the first order aileron
deflection model and XR the state corresponding to the first order rudder
deflection model.

Mode D u t c h roll Spiral Roll Sub. Heading J'/3 ,]'X

Eigenvalues - - 0 . 8 4- 0.6~ -0.4 -I -0.8 -1.5 -1.4
P 00 x 1 x 0 x
g lx x 0 x x x
¢ 00 1 x x 0 x
¢ XX x x I X x
~JB xl 0 0 0 x 0
XA XX X X X X X
Xa X X X X X X X
Delay 5, X X X X X X X
D e l a y 5,. XX X X X X X

XX X X X i X

xx x x x x I

Table 18.6: Desired eigenstructure of the lateral closed-loop system

The resulting gain is:

-i
LatKin = [ -3.6246 1.7016 2.9057 3.0480 13.1933 0.6869 2.2288 |

[ -1.5216 -0.0782 2.4251 --0.2268 1.0320 --0.7237 0.1820

J
247
Loop stability margins
The results obtained at the actuator inputs and at the sensor outputs are
summarised in Table 18.7 and Table 18.8, respectively. Figures 18.5 and 18.6
show the singular values plots of the sensitivity functions.

Function ~ w K. = 1/~ Gain margin (dB) Phase margin (deg)

S 1.43 0.65 0.70 [-4.8,10.4] -4-40.8
T 1.18 0.21 0.85 [-16.2,5.3] 4-50.0
S' + T 2.19 0.46 0.46 [--8.6,8.6] 2=49.2

Table 18.7: Lateral inner loop stability margins at the inputs

Function ~ w K,~ = 1 / ~ Gain margin(dB) Phase margin (deg)

S 1.71 0.50 0.59 [-4.0,7.7] 2=34.1
T 1.01 0.04 0.99 [-43.3,6.0] -I-59.5
S -F T 2.19 0.46 0.46 [-8.6,8.6] 4-49.2

Table 18.8: Lateral inner loop stability margins at the outputs

!!!!!!!!!i ......... i!...i!. !!!!i,.!!.

lO" ;" "~ ;"i

• .

i0-~ I0-'
I r ~ t w r , cy (hide)
10= i0 p
,o, o ~ .... i.......... Ioo'
i i !!ii![
m"
....
i'~ io'

Figure 18.5: Lateral inner loop sin- Figure 18.6: Lateral inner loop sin-
gular values of the input sensitivity gular values of the output sensitiv-
functions ity functions

O u t e r loop controller design

Choosing L a t K o = 0.001, the gain margin for the outer loop is 13.9 dB at
w = 0.29 rad/s and the phase margin is 63.9 deg at w = 0.08 rad/s. The
slower roots are -0.11 • 0.10z that have a rise time of 15 s and a setting
time of 44 s. Therefore, the transient specifications for a lateral deviation step
command are met.

248
18.5 Analysis of the resulting controller
The controller was designed by an iterative method considering criteria of per-
formance and robustness in the linear model, without taking into account non
linearities. In this section we present the verification of all the design speci-
fications with the non-linear system. The description of the design criteria is
given in chapter 14.
All the simulations are run using the following configuration (see §18.4):
mass = 120.000 kg, Ax = 0.23 ~, Ay = 0 ~ and Az = 0.1 ~; and the initial
conditions are: VA = 80 m/s, altitude = 1000 m, ¢ = -90 deg, 9' = 0 deg and
= 1.65 deg.

18.5.1 Non-linear simulation observing performance cri-

teria
Figure 18.7 shows the response to lateral deviation, altitude, heading angle,
flight path angle and airspeed commands. It can be seen that all performance
criteria concerning altitude, heading angle, inertial flight path angle, airspeed
and cross coupling from airspeed to altitude are fulfilled. In the lateral deviation
step command a step change of 100 m has an overshoot of 13 %, however in the
linearised model the overshoot is smaller than the specified 5 %, but the lateral
deviation is reduced to 10 % in less than 30 s, according to specifications.

Sbpcf~mlnllOludO S~ ot 3 0 m ~ . i ~ u d a

Stop ol 3 ~ m pl~ Step o; 3 ~g kl G a ~

s~ of ~3 r r d l I . VA $~ o 1 1 3 n ~ . l a V&

Figure 18.7: Response of the non- Figure 18.8: Response of the non-
linear model to command signals linear model to an engine failure

18.5.2 Non-linear simulation of an engine failure

The results are given in Figure 18.8. We can see that: ¢ is less than the specified
10 deg, its steady state deviation does not exceed 5 deg and, when the engine
is restarted, ¢ is reduced to zero with an overshoot that sligtly exceeds the
specification of 50 %; {} is quickly minimised; and ¢ is well below the specified
3 deg/s.
The comfort and safety criteria are also verified: the airspeed is always
higher than 79 m/s, which is much bigger than the specified 62.2 m/s =

249
1.2Vst¢u; the angle of attack a ~ 1.7 deg is less than the specified 12 deg;
and the lateral acceleration ny increases up to 0.1 g, that is less than the
specified 0.2 g.
Finally, the variations in altitude are less than 2 m.

18.5.3 Non-linear simulation observing ride quality and

safety criteria
The specifications for ride quality criteria are a maximum vertical acceleration
and a maximum lateral acceleration. Figure 18.9 shows the lateral acceleration
in a 90 deg turn and the vertical acceleration in a change from 7 = 0 deg to
3' = - 6 deg and from -6 deg to -3 deg. These values correspond to segments
II and III of the Automated Evaluation Procedure (§18.6), respectively. It can
be seen that lateral acceleration n~ is always within the limits and the vertical
acceleration n z slightly surpasses the maximum level.

90 degrees turn gamma rotation to -6 and - 3 degree

O.02j , : : ] 0.05
0 . !,' ....... :i :

-0.02 00
0 20 40 60 80 0 50 IO0
time (sec) time (sec)

Figure 18.9: Ride quality criteria observation

Figure 18.10 depicts the results of the non-linear simulation showing the
safety criteria. The simulations correspond to the landing approach of the
evaluation procedure as used in §18.6, which is a good representation of all
possible flight conditions.

~ 10[ .........
:,so I i
70" : : --
0 200 400 0 200 400
10
~ 2G
~ C ~ 0

o. -2(3
-10
0 200 400 0 200 400
time (see)

Figure 18.10: Safety criteria observation

The safety criteria are fulfilled since the airspeed is always well above
1.2Vst~u = 62.2 m/s; the maximum angle of attack a remains within the limits,
its maximum value is observed during the turn, but is well below the limit of
12 deg; the roll angle ¢ remains also within the limits, but in the turn is near
to the limit of 30 deg; and the sideslip angle ~ is quickly minimised at all times.

250
 Detailed response to an engine failure for airspeed and angle of attack has
been given in the previous section.

18.5.4 Non-linear simulation under moderate turbulence

conditions
We will give an analysis of the non-linear systems under moderate turbulence
conditions. The spectra of the turbulence we used in the simulations are de-
scribed in chapter 14 (§14.2.6).
Figure 18.11 shows actuator behaviour. All the requirements on the mean
of actuators are fulfilled. Moreover, all RMS values are less than the limits (the
specified values are given in parentheses):

- The mean aileron rate is 0.42 deg/s (< 8.25 deg/s) with a RMS of 7.30
deg/s

- The mean tailplane rate is 0.12 deg/s (< 5 deg/s) with a RMS of 1.03
deg/s

- The mean rudder rate is 0.08 deg/s (< 8.25 deg/s) with a RMS of 1.97
deg/s

- The mean throttle rate is 0.09 deg/s (< 0.24 deg/s) with a RMS of 1.50
deg/s

_,1 .,- . . . . . v "1 ~-.s g

@l o u=r4~) Ioo ~ (~) 1 o so Ioo

Figure 18.11: Control activity un- Figure 18.12: Responses under

der moderate turbulence conditions moderate turbulence conditions of
angles ¢, 8, ¢ and X

Figure 18.12 shows in open and closed-loop, the roll angle ¢, the sideslip
angle 8, the heading angle ¢ and the inertial track angle X under these moderate
turbulence conditions. In closed-loop, ¢ always remains within the specified
limit of 5 deg. Table 18.9 shows the RMS values of the errors in open and closed-
loop. The closed-loop values of ¢, ~ and X are less than their corresponding
open-loop values, however the RMS of ¢ is higher in closed-loop than in open-
loop.

251
 ¢ Z ¢ x
Open loop RMS (deg) 2.04 1 . 0 6 0.89 0.97
Closed loop RMS (deg) 0.65 0.96 1 . 4 5 0.28

Table 18.9: RMS of the errors in angles ¢,/3, ¢ and X in open and closed-loop

18.5.5 N o n - l i n e a r simulations observing robustness crite-

ria
Figures 18.13-18.15 show the simulation results obtained at the nominal, min-
imum and maximum values of time delay, mass and centre of gravity, all cases
at the design airspeed of 80 m/s, giving a set of 81 simulations. In the plots,
the angles are given in radians, the displacements in meters and the velocities
in m/s.
Figure 18.13 shows the lateral step response and the altitude step response.
We can see that there is very little control activity, there are almost no over-
shoots and the settling times are well below the specified values.
Figure 18.14 shows the airspeed-altitude cross-coupling. We can see that
the deviations are always within the limits.
Figure 18.15 shows the roll and heading angles at engine failure. In all cases
the performance criteria are fulfilled: the roll angle never exceeds 10 deg and
its steady state deviation does not exceed 5 deg; the heading rate is always well
below the specified 3 deg/s, and the sideslip angle is quickly minimised.
In [52], the simulation results can been found for the inner loop response to
heading and flight path angle steps commands, and for air speed step and wind
step responses. With respect to the inner loop response to heading and flight
path angle steps, in all cases the overshoot is lower than the specified 5 % and
the settling time is less than the specifications. Regarding the airspeed step
and wind step response, all the performance criteria are fulfilled in the airspeed
command. The response to a wind step has three different characteristics which
correspond to the three different values of the mass which have been considered.
In case of minimum and maximum mass values, the deviation in airspeed is
larger than 2.6 m/s after 15 s of the step, and in case of nominal mass value
the specification is fulfilled.
In [52], the simulation results obtained for all possible worst conditions of
time delay, mass and centre of gravity, at a speed of 63.7 m/s and at maximal
flap speed (90 m/s) can also be found.
The worst results are obtained in the engine failure at 63.7 m/s, with min-
imum mass and with Ax = 0.31~. Consequently, a new set of controller gains
was computed using the linearised model at 63.7 m/s velocity, but without
changing the rest of the trimming parameters, and with the same eigenstruc-
ture as before. The resulting controller and simulation results, obtained by
making use of this controller when simulating at 63.7 m/s and for the worst
condition of time delay, mass and centre of gravity, are given in [52].
Similar performance and robustness results to those obtained with the con-
troller designed at the original speed of 80 m/s are obtained, thus gain schedul-

252
 - -. x- z~-.

wren ~1~
w- w_©-. ~- ~_©-.

............, ,;......-i.........~ ,~0: ......... - .................. ; .........

"m~N ml*l
m m m L- P,--

,i ,! ,i o= ....... - ................. ~ ........

!i i
; : : -.o~ ........ i . . . . . . . . . . . . . . . . . . 2. .........

Figure 18.13: Lateral and altitude step response at the design airspeed

-lO

l o i, -
s i

a (z.©) I ~_c)
a2

• ,o . 4....... : .............

....w m2l l...........

l
i
~ L- P,--

o-1

o
...... ~ .............. i"

!
:~~"'°
• .. ..
..~1 ........ ................ ? .........

tl,~ Ill

Figure 18.14: Airspeed-altitude cross coupling at the design airspeed

253
 o .......................... o . : '

-as ' i !

-1

h~. n.j-. ~1- R.J-.

~t , , ,

o~ .~.. ~ ........ ! ........ = .........

o" " i
-0.2

~- AtA-. ~- RtJ*.

d r . R t_l -. u s l m L. n - -

°"I~; .............. ~......... 1 . . . . .

. . . . . . . . . . . . . . . . i ......... 10

Figure 18.15: Roll angle and heading response to right engine failure at the
design airspeed

ing with respect to velocity should be used.

18.6 R e s u l t s of t h e a u t o m a t e d e v a l u a t i o n pro-
cedure
This section presents the results of the evaluation procedure, as proposed in
§14.3.3 consisting of a landing approach.

Segment I
Figure 18.16 shows the performance of the controller in this segment, and it
can be seen that the lateral deviation is always less than 20 m. Therefore, the
controller complies with the corresponding specification. Moreover, the four
plots are almost the same, which means small sensitivity to time delay and to
horizontal centre of gravity variations.

S e g m e n t II
Figure 18.17 gives the behaviour of the model in this manoeuvre. It can be seen
that the trajectory of the model surpasses the bounds marked in the plots but
the lateral deviation never exceeds the maximum value of 200 m and at the end
the lateral deviation is close to zero. The lateral acceleration never surpasses

254
the maximum allowable value (see Figure 18.9). Moreover, the model has a
very smooth turn, fulfils all the performance design criteria and our attempts
to have a trajectory within the bounds diminished the stability margins, so we
accepted it as is.

Segment III
Figure 18.18 represents the behaviour of the model in the descent phase. It
can be seen that the trajectories of the model surpass the bounds marked in
the plots although the vertical deviation never exceeds the maximum value of
20 m and at the end of the segment the deviation is close to zero. In Figure
18.10 we can see that the speed variation is well below the allowed 4 m/s.
Moreover, the model has a very smooth transition during the entire segment,
although the vertical acceleration is a little bit high at some points. We can
see in Figure 18.9 that the vertical acceleration slightly surpasses the maximum
allowed value. This is reflected in the comfort index in Table 18.10. Since the
rest of the design criteria are fulfilled and our attempt to diminish this value
produces worse results, we accepted it.

S e g m e n t IV
Figure 18.19 shows the behaviour of the model in this segment. It can be seen
how the trajectories fall inside the bounds during the entire segment. The rest
of the specifications are fulfilled by the controller.

Numerical results
Table 18.10 summarises the results as obtained by the controller along the
landing approach. For full details see chapter 14. In general the results are
good, except for the comfort criterion in Segment III. The problem with the
comfort has already been explained. It is basically due to a small high level of
the vertical acceleration.
Flrst Ngmlml: top vkDw Second ,,le~Dnt laterat deviations

i
I
i
50

:i
...... ~ ........ -d'~ v / ....... -~
/
I j -IC
i I •
i
I i
i - - -J

-20 -18 -16 -14 -12 -10 8 .-6 -4 -2 1 Z 3 4 5 6 7 6

;daxJ tra~ cllstar~l from p~int 1 pcml

Figure 18.16: Segment I: The effect Figure 18.17: Segment II: Lateral
of engine failure with bounds deviations during the 90 degrees
turn with bounds

255
 F~It~ ~ r w ~ l : ~ u d e devi*lk3r~

~ 102 •
g
!
3 .'~ •

%, -,~ .; ~ -; -~ .~ _~ .~ .~ .;
x-.po*~n (XE) [ ~ !

Figure 18.18: Segment III: Vertical Figure 18.19: Segment IV: Verti-
deviations during the -6 and 3 de- cal deviations during the final ap-
grees glidslope with bounds proach with bounds

Segm. I Segm. II Segm. III Segm. IV Total

Performance 0.0764 0.4964 0.3285 0.1905 0.2730
Perf. Dev. 0.0309 0.0161 0.4926 0.1873 0.1817
Comfort 0.5432 0.7340 1.1808 0.4674 0.7314
Safety 0.0038 0.0382 0.0070 0.0345 0.0209
Power 0.0037 0.0027 0.0150 0.0309 0.0131

Table 18.10: Numerical results of the evaluation procedure

18.7 Conclusions
In this chapter the eigenstructure method has been applied to the RCAM
benchmark problem.
The design was completed by making use of the classical approach. This
consists of splitting up the controller into two parts, a longitudinal and a lateral
controller, and in using a standard inner-outer loop control structure. In every
inner loop the feedback outputs, command signals and integrated outputs have
been chosen guided by the design specifications. For both inner loops a constant
gain feedback matrix has been calculated using the eigenstructure technique.
This method allows the designer to satisfy directly performance criteria given
in terms of damping, settling time and decoupling, but not to cope directly with
system uncertainties. However, the eigenstructure was chosen in an iterative
way, so that good stability margins were obtained. In both outer loops a
constant scalar gain has been used.
When analysing the controller with the non-linear model (§18.5) all the
design criteria are fulfilled, but the maximum vertical acceleration is surpassed
in certain conditions and the RMS of the heading angle error in closed-loop is
greater than in open-loop.
Good robustness results are obtained with respect to variations of the mass,
centre of gravity and transport time delay. However, gain scheduling should
be used with respect to velocity since bad results are obtained with speed

256
variations, mainly for the engine failure case, as explained in §18.5.
The results obtained in the automatic evaluation procedure demonstrate
good good fulfillment of all of the design criteria except the comfort criteria.
In our design this is due to the fact that the vertical acceleration cannot be
diminished without violating other performance criteria.
The selection of a goodeigenstructure follows an iterative process that can be
time consuming. The process could be shortened if some optimisation method
were used. Without any optimisation process, doubts about how good the
controller is will always remain.

257