Documente Academic
Documente Profesional
Documente Cultură
29.1 Introduction
The LSDP is an intuitive method for designing fixed gain robust controllers. A
tutorial on 7-/00 loop-shaping is given in Chapter 7. A controller that has been
designed using 7-{oo loop-shaping provides robust stability to coprime factor
uncertainty. Coprime factor uncertainty is a general type of uncertainty much
in the same way that single-input single-output (SISO) gain and phase margins
are. Therefore, when there is little detailed knowledge about the uncertainty
present in a plant the LSDP is a good method for designing robust controllers.
The difference to gain and phase margins is that coprime factor uncertainty
can be used to directly address robustness in systems with multiple feedback
loops, i.e. multivariable systems (see pp. 240-244 in [266]).
The LSDP has been used in a variety of applications and studies. Most
relevant to the HIRM design challenge is the work in [120] where a flight con-
1Cambridge University Engineering Department, Cambridge CB2 1PZ, England
2Cambridge Control Limited, Cambridge CB4 4WZ, England
464
trol law was developed for the DRA Bedford Research Harrier XW175 and
subsequently flight tested. The control law performed well and is the subject
of on-going work. The experience of developing this control law showed that
7-/00 loop-shaping has a number of key attributes which make it particularly
suitable for this type of application. Perhaps the most important attribute
is that the resulting controller can be written as an exact plant observer plus
state feedback. This structure allows gain scheduling, as designs at different
operating points have the same state-space structure and hence the observer
and state feedback gains can be linearly interpolated. The observer also allows
the handling of input limitations (e.g. authority and rate limits) by driving
the observer with achieved plant inputs rather than demanded ones. A third
very important attribute of 7-too loop-shaping controllers is that the observer
structure adds visibility, in that physical units and interpretation can be ap-
plied to the controller coefficients and states. This may be advantageous with
respect to flight clearance. The advantages of the observer structure are fully
exploited for the HIRM control law design.
7/00 loop-shaping has much in common with the design approach currently
used by industry, sometimes referred to as "classical control". The loop-shaping
part of the design procedure is carried out in exactly the same way that classi-
cal design is carried out: inputs and outputs are matched up, and single loop
shaping is carried out to ensure that low frequency gain is large enough, roll-
off at cross-over is not excessive and that sufficient high frequency roll-off is
provided. Once this is done, then the optimal 7-/o0 loop-shaping controller is
synthesised for this so-called "weighted plant", see Section 7.2. Hence, it is
perfectly possible to take a classical design, and to augment it with a corre-
sponding 7-/oo controller which will then modify the feedback structure so as to
allow for the multivariable nature of the system.
When designing using classical control where the system has inherent cross-
coupling (as for example with most yaw - roll augmentation systems), corre-
sponding cross-terms are put into the controller. Design of these terms is not
always straightforward in that their effect on the feedback loops is not ad-
dressed directly, and some iteration may be required. With the LSDP, these
cross-terms can be left to the 7-/00 synthesis part of the design. In the authors'
view, one of the prime motivations for 7-/0o loop-shaping is the potential for de-
sign time reduction, particularly when dealing with multi-input multi-output
systems with strong cross-coupling.
29.2 T h e Controller A r c h i t e c t u r e
The first stage of the design process is to select the control law architecture.
Selection of the architecture is an essentially design method independent task,
and the reasoning used for the presented design is much the same as would be
used for a classical control law. However, "Ho0 loop-shaping and related robust
optimal control methods could have been used to help select the architecture by
looking at the robustness implications of each candidate architecture, e.g. ex-
465
amination of the robustness implications of different canard and taileron blend-
ing schemes would be possible. Limited time available precluded this type of
analysis. If a full control law design for a prototype or production aircraft was
undertaken, control law structure selection with reference to robustness impli-
cations is definitely recommended. In selecting the structure here, robustness
requirements are taken into account in a more heuristic way from knowledge
of the system to be controlled.
All of the actuators available for the design challenge are used with ex-
ception of the differential canards. This is because omitting any of them will
necessarily compromise performance in terms of achievable forces and moments.
The reasons for not using the differential canards are given in Section 29.2.2.
There are four primary feedback loops to design, three rotational ones and
airspeed. Multivariable control allows the designer to design all four simulta-
neously. However, the longitudinal motion - pitch and airspeed - is essentially
decoupled from the lateral loops - yaw and roll. What coupling there is be-
tween lateral and longitudinal motion is due to kinematic cross-coupling and/or
asymmetric aerodynamic forces due to, for example, different flow regimes over
each of the wings. The linearisations provided for wings level, steady flight do
not capture these effects. Benefits from designing on the complete 4-input 4-
output system are likely to be more prominent in a Linear Parameter Varying
(LPV) framework [259] within which these coupling terms could be modelled.
For example, parametric dependence on roll rate could be modelled and hence
designed for. However, given the time constraints, an LPV solution was not
investigated and hence the decision was taken to separate the longitudinal and
lateral control law designs.
Figure 29.1 shows the top level SIMULINK specification of the controller.
The two 7/0o loop-shaping controllers are contained within the lateral and lon-
gitudinal H - i n f i n i t y c o n t r o l l e r blocks. They are implemented in discrete
time observer form and hence have two sets of inputs, the measurements and
the achieved aircraft inputs. The pre-compensator weights, l o n g i t u d i n a l W1
and l a t e r a l W1, contain all the integrators, phase advance terms and roll-off
terms designed in the same way as for a classical control law. They are im-
plemented in a modified Hanus self-conditioned form (see Chapter 7 in [120]
and [110]). This is exactly the structure used for the Harrier control law de-
veloped in [120]. Note the two scaling blocks in the feedback paths. These are
used to trade-off the relative amounts of coupling which are to be tolerated
between outputs, e.g. scaling speed in knots and pitch rate in degrees implies
that a 1 knot variation in airspeed is as equally undesirable as l ° / s coupling
in pitch rate. The o u t p u t s block implements first order high frequency roll-off
filters on the p, q and r measurements. The cut-off frequency is 50 rad/s.
The primary feedback variables used to design the longitudinal controller are
pitch rate q and airspeed V. This choice is straightforward in that these are
the quantities the pilot wishes to control. Use of pitch attitude 8 to stabilise
466
-
the aircraft would require phase advance, which in effect differentiates the mea-
surement over some frequency range. This would produce a noisier signal than
the measured q and a less robust design. Furthermore, the dynamics of the
pitch attitude sensor are slower than those of the q sensor and pitch attitude
can not be used at large roll angles.
The HIRM has tailerons and canards available for longitudinal control.
There are several different strategies which could be used to determine how
to apportion a required pitching moment between the surfaces. Before select-
ing a strategy a number of considerations must be taken into account:
• The canards have much faster dynamics, and hence can be used to higher
frequencies with less phase lag.
467
Two possible schemes are:
• Driving both surfaces in tandem. The inputs can be scaled such that the
demand is a percentage of total travel so that both surfaces saturate at the
same point. One of the motivations for this approach is that rate limiting
is less likely to occur. If both surfaces effect the demanded pitching
moment they both have less far to travel than if one surface had been
used. A disadvantage is that the extra agility of the canards is not being
exploited. Additionally, small high frequency disturbances drive the much
heavier tailerons which may be less energy efficient in terms of required
hydraulic power. A second advantage of this scheme is that a failure of
one of the surfaces still gives a system which generates pitching moments
across the required frequency range, albeit with reduced authority.
Either of the two above strategies could be employed. The first scheme was
chosen for this design. The complementary filter is of the form
s F~ilero~(s) = ~I
F c ~ , ~ r d ( S ) -- s + w I ' s + wf"
These transfer functions are implemented in discrete time. The canard demand
is also normalised with the gain Nc. This gain is such that the gain per unit
demand to pitch rate at open-loop cross-over frequency is the same for both
canard and taileron. Therefore if one surface saturates or rate limits, its de-
mand can be fed directly across to the other surface. All of the limiting and
cross-feeding occurs in the a c t u a t o r demands block, Figure 29.1.
A pitch attitude hold is implemented in the o u t p u t s block in Figure 29.1.
Although not listed as a specification of the control law in Section 27.3.2, some
kind of hold is required in practice. This enables the pilot to go stick-free if he
so wishes. Figure 29.2 shows the p i t c h a t t i t u d e h o l d block, the output of
which is a pitch rate demand. In effect the variable out_l becomes q + 0 . 2 5 8 ~ ,
where 8~,.~ = 8cu~r - 8pr~v. The robustness properties of the closed-loop are not
altered significantly by feeding back q + 0.258. This is because the 8 portion of
the signal does not modify the loop gain at cross-over too much. The attitude
hold is only engaged when the flag input, h o l d f l a g in Figure 29.2, is set to
468
out_l
1 and the roll angle ¢ is smaller than 0.1 rad. A pitch attitude hold is not
desirable at large bank angles.
The construction of the total pitch rate demand, command f i l t e r block in
Figure 29.1, consists of the following terms:
• Incidence limiting term. The excess incidence is turned into a pitch rate
demand that will restore incidence to within the specified limits.
469
zero
0.017
[]
side-slip controller and
demand
alpha
COS
performance in that the side-slip measurement is both more noisy and a slower
measurement than r. These two effects will mean that less bandwidth would
be extracted from the yaw loop. Designing a tight primary feedback controller
using p and r robustly stabilises the aircraft and provides an inner closed-loop
system around which an outer loop side-slip tracking system can be built. The
side-slip controller can be seen in the lower left-hand corner of Figure 29.1.
Its output is a yaw rate demand which enters the inner loop as illustrated in
Figure 29.4.
A yaw rate demand is also cross-fed from lateral stick to effect a velocity
vector roll - the required term is sin a times the lateral stick demand as is
illustrated in Figure 29.4. Omitting this cross-term would leave it to the side-
slip controller to reject the side-slip induced when a roll rate is commanded. As
the side-slip outer loop has a lower bandwidth than the inner loop, excessive
side-slip coupling will occur. The cross-term puts in a fast yaw rate demand
to achieve the velocity vector roll.
T h e HIRM has differential canards and tailerons available for roll control.
However, the canards are very ineffective in roll as they generate smaller forces,
and are located much closer to the centreline of the aircraft. Hence using
canards requires large surface deflections and gives little benefit. Furthermore,
it limits their availability for longitudinal control for which they have definite
470
benefits over the tailerons. By studying the aerodynamics of the HIRM it can
be deduced that the differential canards also have a very significant influence
on the effectiveness of the symmetrical canards and tailerons thus creating a
robustness issue. Hence, only the differential tailerons are used for roll control.
For yaw control, only the rudder is available.
1. Scale the differential taileron and rudder inputs by 1/½PV 2. This nor-
malises the moment generated so as not to vary significantly with flight
condition.
2. Select the lateral states from the linearisation i.e. v, p, r and ¢. Append
this linearisation with the actuator models, full order sensor models, anti-
aliasing filters and computational delay. The full order sensor models are
used since the final controller is model reduced anyway.
3. Scale the outputs to reflect the coupling requirements. The scaling used
is (#-tools and MATLAB commands are used)
471
and roll approximately equal results in poorer robustness. In general,
trying to change the directionality of the plant is not good practice. The
roll rate p is augmented with the roll angle to give the output variable
p +A¢¢. This boosts the low frequency gain, and enables a roll angle
hold to be effected for zero lateral stick demand. The A~ term is removed
during a roll-rate demand. This can be justified (in terms of robust
stability) provided that A~ is chosen such that the open-loop cross-over
is entirely set by the p part of the constructed output variable.
. The desired cross-over frequency for both loops is 10 rad/s. This is the
highest the cross-over frequency can go before robustness margins are
necessarily reduced, due primarily to actuator roll-off. To verify this,
a few design iterations at slightly higher cross-over frequencies can be
carried out, and the resulting achieved robustness margin e, monitored.
Both loops have suitable roll-off rates at cross-over, and so all that is
required is to boost low frequency gain, and add high frequency roll-off
filters. The selection of the appropriate transfer functions is exactly as for
a classical design. Both loops are rolled off with the filter s-W~"
50 The filters
are discretised using a bilinear transformation with frequency warping to
match the filters at 10 rad/s.
Low frequency gain is boosted with the pre-compensator
5. Multiply the two plant inputs with gains kw4 and kw6, an input scaling
which gives the required cross-over frequencies. Plot the singular values
of the shaped 2-input 2-output plant, Figure 29.5. Check that the desired
loop shapes have been achieved.
6. Design the 7/oo loop-shaping controller, Ko~, for the shaped plant. Check
that the resulting robustness margin is sufficient (typically e > 0.3 indi-
472
cates a robust design). Check the step responses. These are shown in
Figures 29.6 and 29.7.
Slop ~ p Slap ~ ¢
4(s + 2)
wl=
s(0.1s + 1)
This has a first order roll-off at 10 rad/s to attenuate side-slip sensor noise, and
additional low frequency gain below 2.0 rad/s. The Bode plot of the shaped
loop is shown in Figure 29.8.
The resulting 7-/o~ controller was model reduced to 4 states by carrying out a
least squares matching of the gain and phase plots of the full order controller.
This can be done using the MATLAB function invfreqz.m. The achieved
robustness margin is e = 0.42. This guarantees a gain margin of at least 2.4,
473
o.e
a.a
2 a 4
~. (s)
and a phase margin of at least 45 ° (see Section 7.3 for the relevant formulae).
Figure 29.9 shows the side-slip step response which meets the specification set
out in Figure 27.16.
By putting Ti > T2 pitch rate overshoot is introduced. The values of 7'1 and
T2 were determined by performing a search over a range of values of Ti and T2,
and then selecting all pairs which give a drop-back of approximately 0.125 s.
The other handling quality criteria were then examined for these pairs, and a
particular solution selected. The values chosen were 7'1 = 0.336 and T2 = 0.144
which gave a drop-back of 0.1 s, an average phase rate of 57°/Hz and ]c = 1.7
Hz. From Figure 27.12, it can be seen that this gives Level 1 performance.
Figure 29.10 shows the time response of a pitch rate demand.
474
I.S
i
1.4 . . . . . . . . . . . . . . ~. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0 0.5 1.5
475
Flight condition Drop-back
Mach 0.20,1000 ft -0.17 s 1.32 -56.0°/Hz
Mach 0.24,20000 ft -0.23 s 1.40 -56.0°/Hz
Mach 0.30,5000 ft -0.10 s 1.58 -56.6°/Hz
Mach 0.40,10000 ft -0.07 s 1.40 -57.0°/Hz
Mach 0.50,15000 ft -0.12 s 1.35 -57.1°/Hz
29.5 Conclusions
A stability augmentation system has been developed for the HIRM using 7/~
loop-shaping. In the authors' view, the design problem was relatively simple,
particularly given the narrow flight envelope of the HIRM. However, the ap-
proach which has been taken will extend in an easy manner to a much wider
flight envelope. Key to this is the observer structure used to implement the
7/~ controller. The control law structure has much in common with a classical
control law, the main difference being the addition of the lateral and longitu-
dinal 7-/~ controllers to the feedback path. The observer structure is also used
to handle actuator authority and rate limiting in a systematic fashion. Ap-
476
Nominal flighL Nichols plot, Mach 0.24, 20000 ft
20
• ..:" . :".. Y.i /,,....' ..-":. '~I
: '"" ::
: ........ " i i '.-: ."
i i ':': '. i ... ...... :
15 _i .:':-..... i..i.:. .." '-. !i ~ .- ',. :-,.i..i.. .-::-. :: -dts
-" "-." .~ ! :"o. '. ii o ~..':-ii .... ..-' '.
• " .'" : : " ," ". ". ,'~ ."l " : : I .". "
- - dtd
10 • "
•: "
.'" ". " :i::
i:~ :" ". " . , .'.' : : ' ~::i ' ? ' ..", - ' ~ .'" II ":"-.!.!.i
:i'll ' ." .'" "'. '-. ":- o dcs
•
--.:...,.:~
: ' "". "
...":-. ~::
". '.~.: " " "" • ~'.
.:.':.. ..",::i:.- ..~.~i i// :".-..
"...'" ~..'"..'" ." "I" ?' '• ." ". -"
:....!"
'. • x dcd
;i : ...-........-
o.. :.:i:..i - I ..'%~ ,:: . "... '.. - ! + dr
!: : ." ..' ". '-. ':-'.'. : " :t -:'Y ." ."'-. ".. ': :
~.": :'...,..
" ." ....' " o... "..:.."~:.-"-:"i
.::.'~,....... ":.~'"".'....."
..~ ~- .. .... .....,..':
: ".!. * thr
" - "": .'":" ".".:'. : ""..'" ].''i'ii'".'"~.~iii: ;:7 .:',. :~'" : "
~-5
:! . ~ = 5 _ + o ~. "+- ; : . - . ~ ' ~,~. . + . . - . . : . . , £ - ¢,,..- .J...'~ .: : : ::
- 1 0 i~""":. .........:: o : : : ....~.... - .~ '.. -~..-...-"-".-"?~
-15
:: " : : O : : : ::
:i ! : : o: : " :~ : : : : ~ : :~
-2(
-350 -300 -250 -200 - 150 - 1O0 -50
Open-Loop Phase (deg)
Figure 29.11: Nichols plot of the nominal plant at Mach 0.24, 20000 ft
G i b s o n criterion, pitch: l i n e a r a n a l y s i s
20
16
10
~ 5
-5
-10
o ii
-180 -160 -140 -120 -100 -60 -60 -40
P h a s e (deg)
477
pw az
0 . 5
_1 I
0 5 10 0 5 10 0 5 10
time (s) time (s)
Va alpha beta
0.4
30 . . . . . . . . . . . . ..........
~25 Ac~-0.0"2
02~............... i ...........
0 5 10 0 10 0 5 10
time (s) time (s) time (s)
q cmd (:its dtd
4
11' i '
2 ............. i ...........
dc~ dcd dr
1
10 . . . . . . . . . . . . . . i. . . . . . . . . . . . . . . o.: , 1
0 5 10 0 5 10 0 5 10
time (s) lime (s) time (s)
oof
50 ........
thrl the2
0
0 10 0 5 10
time (s) lime (a)
478
........... ...
~_2~t ~.-4
0 2 4 8 0 2 4 6 0 2 4
time (s) time (e) ,me (e)
alpha beta
"[ /
Va
2 - •
100r
0 2 4 0 4 6 0 2 4
time (s) Ume(e) time (s)
p cmd dtd
o : i
2o .......... i.......... i . . . . . . .
~ ~-5i i..........
~_ i ...............
0
. . .0. . . . . . .0. . . . . . .~. .
2 4 0 2 4
=[
0
.....
2 4
...... 1
6
time (s) time (s) time (s)
dcs dcd dr
0 2 4 6 0 2 0 2 4 6
time (s) time (s) time (s)
thrl thr2
6
E2o~!'
... t
4 ........ ;.......... : ........
g :
:: ......... :
. . . . .........
. . . . .
0 4 6 0 2 4
time (s) time (s)
479
pw q az
1 0.05~ : ;
o.5 ............ i ................
-0.06
-9.6L i I
1 /
0 10
-0.10 5 10 0 10
time (s) time(s) time (s)
Va alpha beta
161,4~
10
61.2 7"1f i //
............. : . . . . . . . . . . .
161
160.8 69
0 10 0 10 0 10
time (s) UmeIs) time (s)
betcmd dts did
10
-6.6 ............... ::. . . . . . . . . . . .
~ 5 ...........................
-6.7
0~ ~ _ 1 0 ~ ........
0 5 10 0 5 10 0 5 10
time (s) time (s) time (s)
dcs dcd dr
1
or 1
0.1
tO . . . . . . . . . . . . . . . . . . .
0.05
ca
. . . . . . . . . . . . .
0
-0.5
-I [ . . . . . . . . . . . . .
-0.05 -5 ' ' r
0 5 10 0 5 10 0 5 10
time (s) time (s) time (s)
thrl thr2
14
12 . . . . . . . . . . . . . . : - - 12 . . . . . . . . . . . . . . . . . . . . . .
g, gt
0 10
61 0
. . . . . . . . . . .
5
. . . . . . . . . . . . . .
10
time (s) time (s)
480
10 x 10 -'0 pw 0.5 - q * -9 . az. .
..... ,o
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
time (s) time (s) time (s)
0 10 20 30 40 0 10 20 30 40 0 '~0 20 30 40
time (s) time (s) time (s)
~rl ~r2
100 - 100 - *
v 0 0
0 10 20 30 40 0 10 20 30 40
time (s) time (s)
F i g u r e 29.16: A 50 m / s s p e e d c h a n g e
481
propriate handling of these non-linearities is essential for achieving the desired
handling qualities, and ensuring sensible apportioning of control power.
7-/oQ loop-shaping is a fairly intuitive controller design method that can be
picked up in a relatively short amount of time, particularly from someone with
a flight control background. The design strategy employed bypassed the step
of trying to express in detail, the time response requirements in the frequency
domain (a drawback of frequency domain based methods).
The order of the control law is not seen as an issue as regards implementa-
tion provided that the complexity can be justified in terms of desired robustness
and performance. Experimentation with model reduction showed that the lat-
eral 7Qo controller could be reduced to 10 states and the longitudinal to 8
states, both with minimal change to the robustness and performance. Bal-
anced truncation of a coprime factorisation of the controller was used for this.
However, some care would be required when model reducing a gain scheduled
design, as the physical interpretation of the model reduced states must be the
same for all designs to allow gain scheduling.
The type of uncertainty (coprime factor uncertainty) adopted, although
quite general, did not prove to be too conservative. Scheduling was avoided
partly because the open-loop HIRM was scaled with dynamic pressure, a well
known technique within industry.
The paradigm of 7-/00 loop-shaping is extremely powerful. There are other
extensions which have not been demonstrated here due to lack of time available
for the project. One exciting new area is that of self-scheduled design meth-
ods whereby a parameter dependent controller is synthesised in one step for a
parameter dependent plant. For the HIRM, the objective would be to find a
controller dependent on airspeed given a speed dependent model of the HIRM.
The synthesis of the controller relies on solving a set of linear matrix inequal-
ities for which there are numerous algorithms available. In the authors' view,
self-scheduled methods and linear parameter varying (LPV) plant descriptions
will provide very powerful and relevant tools for the aerospace industry. In
particular, they provide a framework in which to address aerodynamic non-
linearities and rate dependent effects. ~ r t h e r investigation of the best control
structure is also a possibility. Designing a single controller for roll, pitch and
yaw would be worth carrying out to see what the cross-terms between lateral
and longitudinal feedback loops are as a function of flight condition. This might
include looking at non-steady flight conditions such as a non-zero roll rate.
For new types of controllers to be used by industry, there are two major
pre-requisites. Firstly benefits need to be quantified so as to justify a change in
approach. Secondly, once the benefits have been established, the flight clear-
ance aspects need to be addressed. The HIRM design challenge was not set up
to address these issues directly. Its focus was to demonstrate the methods to
industry, and to highlight what the advantages might be. As regards potential
benefits of the method, there are two main ones worth mentioning. Firstly, for
complex multi-input multi-output systems, the ~oo loop-shaping method pro-
vides a way of synthesising a controller which reflects the cross-coupling in the
plant. As such, better performance and robustness may be obtained since se-
482
lecting controller diagonal terms with non-multivariable methods can be a trial
and error task. Secondly, the 7-/oo loop-shaping approach has great potential
to reduce design time. This is partly a result of its ability to handle com-
plex multivariable systems, and partly that given the nature of the robustness
optimisation, it is hard to design a bad controller.
Flight clearance of this type of control law may be easier than for many
other non-classical design methods. In the main, this is because the loop-
shaping aspect has direct connections to classical design. The main difference
to a classical design is the addition of the multivariable 7/00 feedback controller
element. However, the 7-/oo feedback element is implemented in an observer
form for which the structure is clear, and for which interpretation of the physical
units of individual gains are clear. However, this is still (in the UK and probably
many other countries) a new approach to have state-space elements within
the control law as opposed to SISO transfer functions. However, this will
probably not be an insurmountable problem, particularly given the numerical
and efficiency advantages of using state-space implementations. The next step
therefore has to be to quantify benefits in some meaningful manner.
483