Professional Documents
Culture Documents
CH 31
CH 31
Johan Markerink i
31.1 Introduction
This chapter reflects the design of a controller for the manual control augmen-
tation problem of the High Incidence Research Model (HIRM) as defined in
chapter 27. This work served as the Delft University of Technology (DUT) en-
try for the HIRM design challenge and was performed as a graduation project
by the author. The method used for this benchmark is 7-/oo-, p-synthesis; one
of the robust control methodologies recently developed. Using this method,
controllers for both longitudinal and lateral-directional aircraft motions are
503
designed. During this design, the controller architecture and the method de-
pendent objectives are chosen to comply with the HIRM design criteria as far
as possible.
The structure of this chapter is as follows: In section 31.2 the layout of the
chosen controller architecture and a description of the internal architecture is
given. Section 31.3 will be dealing with the translation of the HIRM design
criteria into the structure of a general 7/oo-, /z-synthesis design. This results
in the set-up of an interconnection structure in which the desired handling
qualities play a major role. Section 31.4 contains a description of the design
cycle and a motivation for some of the choices made during this design cycle.
In section 31.5 the resulting linear controllers will be analyzed in both the
frequency and the time domain. In section 31.6 the nonlinear results of the
method independent automated evaluation software will be reviewed. Finally,
in section 31.7, the conclusions and lessons learned are presented.
31.2 T h e S e l e c t i o n of t h e Controller A r c h i t e c -
ture
This section deals with the layout of the controller architecture used in the
HIRM design problem. Investigation of the linearized models provided with the
HIRM software at different points in the flight envelope, revealed that there is
little interaction between the longitudinal and lateral-directional aircraft mo-
tions. Therefore the/z-controller designs for the two sets of aircraft motions are
performed separately, resulting in two/z-controllers and two separate controller
blocks. For scheduling purposes, both blocks have an additional input. This
input is used to scale the commanded control surface deflection with the inverse
of the dynamic pressure. The scheduling parameter is constructed using:
6900
scheduling f a c t o r =
This means that the flight condition with a dynamic pressure of 6900 N/m 2
can be seen as nominal, requiring no scaling. For other flight conditions, the
change in control surface effectiveness due to the change in dynamic pressure
is cancelled by the scheduling factor. The stick gains are chosen to command
12.5°/s pitch rate and 90°/s roll rate at the maximum stick forces.
504
dynamic preemure
~J~eduJing
az_meaeured
~
pitch rate cmd I)
q_measured q-limi~ng
throlilldemand
V_measumd
V tdm
O_mealured
O_,r~m_ _ l ;1 ,~-1
lin(O)-¢n(O ~') Gain
505
roll rat~ cmd
p_measumd j
dynamic pressure
sd~ed~ing
506
frequency [rad/s] damping
9.8927e-006 1.0000e+000
5.0908e-003 1.0000e+000
8.6871e-002 1.0000e+000
1.1542e-001 1.0000e+000
4.5883e-001 1.0000e+000
3.9647e+000 1.0000e+000
1.7474e+001 4.9664e-001
1.9973e+001 1.0000e+000
5.2606e+001 2.4637e-001
9.4169e+001 5.6592e-001
2.2386e÷002 7.0502e-001
-~* 4 o ~ "* 4 D s 4 4 • •
10 10 1o 0 ~0 1o 10 1o lO 10
tO I ~o i Io L
• longitudinal: 15 states
• lateral-directional: 16 states
This adds up to a total of 31 states for the #-controllers. With other, more ad-
vanced model reduction methods, this order could possibly be reduced further.
To cope with some of the nonlinear effects, three additional filters were used in
the controller adding 4 states. The total number of controller states is there]ore
35. The longitudinal and lateral-directional p-controllers are stable dynamic
systems. Table 31.1 gives the frequency and damping data of the longitudinal
#-controller. To further enhance the visibility, the frequency responses from
inputs to outputs are plotted in figure 31.3.
507
31.3 The Translation of the H I R M Design Cri-
teria into #-Objectives
In this section the HIRM design criteria as described in section 27.3 will be
translated into objectives and criteria that can be used with an 7~o~-, /z-
synthesis design procedure. To include the handling qualities criteria into the
#-design, reference models with ideal handling qualities are used. In section
31.3.1, these reference models will be derived for both the longitudinal and
the lateral-directional aircraft motions. Section 31.3.2 deals with the way in
which the robustness requirements are incorporated into the design. In section
31.3.3, the general interconnection structure including all the dynamics, filters
and weighting functions needed to perform the design, is constructed.
508
• Speed control overshoot criterion: overshoot should be Iess than 3%.
To fulfill these requirements, a first order ideal speed command reference model
with a time constant of 1 second is chosen.
V 1 1
Vc l+ws l+s
By diagonally augmenting the ideal Pitch Rate Demand system and speed
command system, one longitudinal reference model is generated. This model
will be used in the set-up of the longitudinal interconnection structure used for
the ~/oo-, #-synthesis design procedure.
T h e lateral-directional reference m o d e l
The control strategy for the lateral-directional aircraft motions of the HIRM
call for a velocity vector Roll Rate Demand (RRD) system, commanded by
lateral stick deflection and a sideslip demand system, commanded by rudder
pedal deflection.
The HIRM criteria on the Roll Rate Demand system are:
• Roll mode time constant criterion: This time constant should be ~-p ~ 0.4
sec.
• Roll attitude frequency response criterion: This criterion consists of a Nichols
plot from roll rate command (or lateral stick deflection) to roll angle. The
resulting graph should lie in the region labelled with "good response" in
figure 27.14.
To comply with the first criterion a first order model with the specified time
constant is chosen. The first order model was unable to generate sufficient
high frequency lag to stay within the boundaries. For this reason, a first order
lowpass filter was added to provide the phase lag. The result is the following
roll mode reference model:
p 1 1 1 1
Pc 1 + rps 1 + rings 1 + 0.4s 1 + 0.07s
For the sideslip demand system there is only one criterion:
• Sideslip step response criterion: The response of sideslip to a step input in
sideslip command should lie within the boundaries of figure 27.16.
The model used for the sideslip demand system is that of a simple first order
transfer function:
/~ 1 1
/9c 1 + T~s 1 + 0.5774s
The two lateral-directional reference models are diagonally augmented to one
another, yielding one lateral-directional reference model. This model is then
used in the set-up of the interconnection structure used for the 7/oo-, #-synthesis
design procedure.
509
* Mach: 0.15 to 0.5
• Altitude: 100 to 20000 ft.
In order to take the model variations over the flight envelope into account,
the linear system dynamics in various operating points are parameterized as a
function of the dynamic pressure. The method used in this design, is to model
the HIRM as a Linear Parametrically Varying (LPV) system. Although the
linear representations of the nonlinear HIRM model are functions of several
variables, making them dependent on flight condition and flight envelope, the
leading parameter is the dynamic pressure (1 = 1/2p V 2. All aerodynamic vari-
ables in the state-space model, vary linearly with (1. With the construction
of the HIRM LPV-model, this linear dependence is modelled via additional
inputs and outputs on the HIRM model. This means that in the #-synthesis,
the variations of the model through the flight envelope are taken into account
through this varying parameter. For a more extensive review of the LPV model
and the construction of this model in the HIRM design see the design report
[163].
In the HIRM manual, two additional kinds of modelling errors are described:
errors on the aerodynamic moment derivatives and errors on the total moment
coefficients. All these modelling errors are incorporated into the LPV HIRM
model as uncertainties on the individual matrix elements. These parametric
uncertainties are modelled using a real, structured, additive uncertainty repre-
sentation. As a result of this, the models are augmented with additional inputs
and outputs through which these uncertainties act.
The gain and phase requirements dictate that the closed-loop system should
not become unstable when adding additional gain and phase offsets at the input
of each actuator demand. The definition of these requirements exactly fits the
multiplicative uncertainty block used in this design, because the Am-block acts
at the same point in the loop and adds its gain and phase offsets. See figure
31.4.
510
el
noise ~ +I +
~ wno,se~-~. ~-~"
structure and the number and type of inputs and outputs is known.
Waet The multiplicative robustness weight. This weighting function scales
the multiplicative perturbation-matrix Am. If Gnom is the nominal
plant, the set of plants characterized by this uncertainty representation
is given by:
GA = G,~om(I + mmWdel)
At any frequency ~, the magnitude of Wdel(w) can be interpreted as
the percentage of uncertainty in the model at that frequency. For
example: a Wdet value of 0.5 represents a 50% modelling error at that
frequency.
A~ The additive perturbation matrix A~ is used to model the real, struc-
tured, additive uncertainties to the individual state-space matrix ele-
ments.
A,~ The multiplicative perturbation matrix Am represents the input un-
certainty in the model.
Aq8 The perturbation matrix Aqs is used to model the Linear Paramet-
rically Varying (LPV) nature of the HIRM-system. It introduces the
linear dependence of the A- and B-matrix elements on the dynamic
pressure into the system.
ACTS This element represents the actuator dynamics as described in the
HIRM problem definition. The input is actuator demand and the out-
puts are actuator deflection and rate. In the case of multiple actuators,
these are diagonally augmented to obtain one actuator system.
W~c~8 The actuator weight. This weighting function scales the admissible
actuator deflections and rates to unity. In the simplest case, Wact~
would consist of the inverse of the maximum actuator outputs, result-
ing in an error (output of Wacts) of unity, if the maximum deflection
was reached.
HI R M The linear, state-space model of the aircraft.
511
Wp The performance weight. This weighting function scales the perfor-
mance error to unity. The input of Wp is the tracking error between
the response of the reference model and the closed-loop system. This
weighting can be seen as the definition of the allowable deviations
from the reference model response. Because the output errors are
scaled to unity, these deviations have to stay below the value of the
inverse of Wp at every frequency. If at a certain frequency Wp=50,
this means that the closed-loop response may deviate 1/50 = 2% from
the reference model.
SENS This element represents the dynamics of the sensors used. Also the
anti-aliasing, notch and averaging filters described in the HIRM prob-
lem definition, are incorporated into SENS. As with the actuator dy-
namics, multiple systems are augmented diagonally into SENS.
Wnoise The noise weight. This weighting function scales the unity-intensity
noise entering the system. Because it represents high frequency sensor
noise, high frequency filters are used in Wnoise.
Breaking the loops at the controller K and at the perturbation A-blocks and
collecting the remaining systems into one system results in the generalized
plant P. The inputs and outputs of this generalized plant are depicted in figure
31.5. The information of all systems and functions and the interconnections
between them, present in the generalized plant, is used by 7/~-,/z-synthesis to
design a robust controller. The robustness and performance properties of this
controller are related directly to the weighting functions used in the intercon-
nection structure.
___.._•Za
_____~el~e
a{,,o/_~ e2
U
512
The general design cycle consists of the following steps:
1) Define the design problem: the design problem definition states the overall
design problem to be solved. In this case it consists of the design of a
control augmentation system that provides a high angle of attack fighter
aircraft with satisfactory handling qualities. This design problem definition
is treated in chapter 27.
2) Translate the design problem into a general interconnection structure. This
structure defines the basic layout of plant, controller and ideal model inter-
connection.
3) Formulate criteria: a set of numerically defined criteria is used as specifica-
tion of the overall design goals and to evaluate the resulting design. These
criteria are defined in section 27.4.
4) Translate handling qualities criteria into reference models: for aircraft con-
troller design problems in which handling qualities play a major role, like
the HIRM, a reference model is an ideal method of translating the time do-
main criteria into criteria that can be used in an 7-/0o-,/~-synthesis design.
The set-up of the reference models with ideal handling qualities is treated
in section 31.3.1.
5) Translate robustness criteria into system perturbations: the controller has
to provide sufficient stability and performance over a range of flight condi-
tions and has to be robust against several model uncertainties of the plant.
To accommodate these criteria, the controller will be designed for a set
of plants. This set consists of a nominal plant with several perturbations
acting on it, to model the various uncertainties. The construction of these
plant models and perturbations for the HIRM is treated in section 31.3.2.
6) Extend the general interconnection structure: several criteria can be im-
plemented directly into the design process by adding elements, weighting
functions or inputs and outputs to the general interconnection structure.
The result of this process can be found in figure 31.4.
7) Choose the weighting functions: after the interconnection structure is de-
fined, the only variable elements are the weighting functions. The choice
and construction of these functions is the greater part of the design process.
Usually numerous redesigns are necessary to find a combination of weight-
ing functions that result in a controller which complies with the criteria.
Weighting functions are in a sense, the "knobs" the designer can turn to
shape the resulting controller and closed-loop response.
8) Construct the actual interconnection structure and resulting generalized
plant: this step involves writing a computer program that implements the
connection of all the models, filters and weighting functions. The output of
this program is a linear model of the generalized plant with the input/output
definition of figure 31.5.
9) Perform a/,-synthesis: using the generalized plant resulting from the previ-
ous step, the/z-synthesis procedure designs the actual controller. It consists
of a D-K iteration scheme which involves an ~/oo-synthesis, a #-analysis and
a fitting of scaling functions.
513
10) Analyse the closed-loop system: using frequency domain techniques such as
Singular Value analysis and p-analysis, the closed-loop system can be ana-
lyzed. Also, a variety of time domain techniques can be used to investigate
compliance of the resulting controller with the design criteria. If the design
criteria, frequency or time domain, are not met, one has to go back to step
7), change the weighting functions and redesign the controller. This process
continues until all criteria are met or no more improvement is achieved.
11) Reduce the order of the controller: the synthesized controller can have quite
a large number of states. To reduce the order of the controller a balanced
realization of the controller is created. This balances the observability and
controllability grammians and orders the states accordingly [50]. Next a
truncation can be used to remove (nearly) unobservable and uncontrollable
modes. In this design, a number of high frequency modes that were incom-
patible with the HIRM's assumed 80 Hz Flight Control Computer (FCC)
iteration rate were removed also.
12) Analyse the closed-loop system: using the same techniques as in step 10),
the closed-loop system with the reduced controller can be analyzed. If re-
duction deteriorates the controller performance, one has to go back to step
11) and truncate less states of the original controller. If the truncated con-
troller exhibits no performance differences with the original controller, one
can go back to 11) and try to truncate the controller further. The controller
order reduction ends when this balance between controller performance and
controller order has been found. A number of these analysis results for the
resulting controllers will be presented in section 31.5.
13) Construct the controller architecture: the ~-controllers are part of the com-
plete controller block. This block also consists of input and output connec-
tions, signal additions and subtractions and additional filters and functions.
In order to cope with nonlinearities in the model or in the criteria, some non-
linear and limiting blocks may also be present. The set-up of this controller
architecture is treated in section 31.2.
14) Test compliance of the controller with the design criteria stated in 3): using
the full nonlinear model, the automated evaluation software checks if the
controller meets these criteria. This final analysis includes several nonlinear
time simulations, robustness and performance checks and other tests. Some
results of the evaluation can be found in section 31.6.
In the numerical execution of this design cycle the matrix manipulation soft-
ware package Matlab is used. In particular the add-on toolbox for analysis
and synthesis of robust control systems #-tools (mutools) is used extensively.
Besides the functions provided by mutools, additional programs written by the
author in Matlabs scripting language, are used to implement certain elements
of the design cycle.
There is little intermediate analysis and redesign effort. Because the anal-
ysis part is incorporated into the design cycle, it requires no extra effort. Both
514
frequency and time domain analysis techniques are used in this design cycle.
The frequency domain analysis is strongly related to the used design meth-
ods: 7{oo-, #-synthesis. These analysis techniques can therefore be used to
see whether the design objectives in terms of minimizing closed-loop gains are
achieved. The linear time domain analysis has more connections with the ac-
tual criteria, in particular, the criteria concerning handling qualities. If the
choice of the weighting functions and the set-up of the interconnection struc-
ture is done ideally, the frequency and time domain analysis results coincide.
Unfortunately this may prove to be very difficult to achieve. The most reli-
able method of testing compliance with the design objectives and criteria uses
nonlinear time simulations. Once the controller architecture is present, in-
termediate linear/~-controllers can be incorporated into this architecture and
analyzed this way.
For a redesign after an aircraft design change, the part of the design cycle
starting with step 7) has to be reperformed. The choice of new weighting
functions results in the loop from step 7) to step 10). After the criteria are
met, the order of the new controller can be reduced with step 11) and 12).
The effort related to this redesign depends on the scale of the change. Certain
weighting functions may remain the same, while others may need considerable
changing. If the aircraft dynamics are changed, the set-up of the HIRM model
in step 5) also needs to be reperformed resulting in a new LPV HIRM model.
IO' Io I
10 "~
f
1o I
"~1o °
1o"
Io"
515
In the design cycle used in the HIRM design, the design objectives were to
minimize the robust performance level as indicated by the frequency #-plot.
According to the theory, if robust performance is satisfied, Nominal Stability,
robust stability and nominal performance are also satisfied, robust performance
is achieved if the/z-plot stays below unity over the entire frequency range. It
is not uncommon to find that in complex designs this goal cannot be achieved.
With its many different and demanding criteria, design flight envelope and
several model variations, this is also the case with the HIRM design. This does
not mean that the design will result in bad controllers. Remember that the
acceptability of the control solution does not depend on # alone. Nevertheless,
the discrepancy in the theoretical definition of achieving design objectives with
a #-value below unity and the possibility of not achieving this value is felt to
be a weak point of/~-theory. One criterion that had to be achieved however,
was that of robust stability. The controllers had to guarantee stability for the
complete set of model perturbations, as indicated by a robust stability value
below unity at all frequencies.
As an example of the use of weighting functions figures 31.6 and 31.7 give a
robustness and performance weight of the longitudinal HIRM design.
Wael For the taileron demand signal, the robustness weight is a first order filter
with a 0.1% uncertainty at low frequencies and 10% uncertainty at high
frequencies. The motivation behind this choice is that the uncertainty of
the HIRM model is expected to increase with frequency.
Wp The longitudinal performance weight consists of a system with three di-
agonal entries: a weighting function for the pitch rate error, for the pitch
angle error and for the speed error. These three weights are performance
weights because they scale the error between the closed-loop response and
the ideal reference model response. The plot of the weighting functions
in figure 31.7 can be interpreted as the inverse of the allowable error. The
pitch rate error and pitch angle error weighting functions are first order
low-pass filters. This puts the emphasis on steady state tracking and lim-
its the bandwidth in which the closed-loop system has to follow the ideal
models. The q-weight emphasizes that at low frequencies errors may not
exceed 10%. To achieve some form of pitch attitude hold and pitch atti-
tude tracking in the final closed-loop system, the pitch angle error is also
used as a performance output. The pitch angle weight limits the pitch
angle error to 0.4%. The airspeed weighting function has a large value
of 200 at high frequency, limiting errors to 0.5%. This is to achieve good
tracking of the transient part of the first order ideal model. The weight is
increased to 1000 at low frequencies to ensure steady state airspeed hold
with minimal errors.
516
0 , . . . . .
~ t Performarce:
NominaPerformance
l
517
slightly larger than the nominal performance plot. An interesting feature of this
last plot is that because of the numerical difficulty in the #-calculation with
the real valued perturbations used in this analysis, the upper and lower bounds
of # show a small gap. Similar results were achieved in the lateral-directional
design, as can be seen in figure 31.9.
-a G4 O4
o2
O2
0
p-error
~2
-(14
1 2 3 4 5 S 7 s s 10 1 2 : 4 s 6 7 s 9
time [see] time [sec]
Figure 31.10: Pitch command lin- Figure 31.11: Roll command linear
ear time response time response
31.6 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
In the automated evaluation procedure, the controller was subjected to a se-
ries of nonlinear simulations to verify compliance with the HIRM requirements.
The overall impression was that the controller showed good performance and
robustness characteristics. The controller satisfied the criteria concerning dis-
turbance rejection, structural coupling and limited noise and turbulence control
518
activity. Gibson criteria and departure protection also looked good. Only the
lateral channel multivariable uncertainty criterion was not satisfied completely.
As an example two figures from the nonlinear simulation assessment manoeu-
vres are plotted.
In figure 31.12 the response is given for a pitch rate demand, at M -- 0.3
and h = 5000 ft. This plot shows that the pitch response achieved corresponds
to the ideal handling qualities reference model. The airspeed varies because of
the increased pitch attitude while the controller tries to compensate for that
with an increase in thrust.
In figure 31.13 the response is given for a roll rate demand, at M = 0.5 and
h = 15000 ft with parametric uncertainties and measurement errors present.
Despite the added uncertainties the roll rate response follows the ideal roll
model. The transients in the roll rate and sideslip responses of this figure are
caused by the 2° measurement error on ]3. As a result of the instantaneous
application of this error at t--0 seconds, the controller generates fast control
surface deflections to try to force the measured sideslip back to zero.
519
• A thorough understanding of the theory is needed to effectively design and
analyse controllers.
• A large and difficult part of the design cycle consists of choosing and chang-
ing the weighting functions. This effort can be reduced by experience and
a good physical understanding of the possibilities and limitations of the
systems at hand.
• The resulting controllers may have a large number of states needing good
model reduction techniques.
• It is not always feasible to keep on tuning the weighting functions to force
the #-value below unity for all frequencies. Often, design objectives are too
complicated to be expressed exactly as first or even second order weighting
functions and good designs can be made with larger p-values.
• It is however, sensible to search for a set of weighting functions that results
in a controller which satisfies robust stability. This way, use is made of
one of the strong points of °Aoo-, p-theory, namely: to be able to guarantee
stability for a given set of plant models.
• The linear controllers designed with 7-/c0-, p-synthesis are not able to cope
with large nonlinearities in the model and the design requirements. A lim-
ited number of additional filters, functions and nonlinear blocks can be
added to the controller architecture to deal with these nonlinearities.
All in all, this design has clearly demonstrated that 7-/~-, p-synthesis has capa-
bilities which can be used to solve a realistic nonlinear aircraft robust control
law design problem. It is expected that 7-/~-, p -theory will prove to be a valu-
able tool to the European Aircraft Manufactures for the design and analysis of
Flight Control Systems.
520
x10-~ pw
::
~'°tA
. i
5 . . . . . . . . . . . . . . . . . . . . . . . . . . . .
-5 ~
0 5 10 0 10 0 5 10
time(s) time (s) ~rne (s)
i ......
1001 ...... ~"i'"'/ ......
~'~1150
~ . . . . .2. . . . . . . . . .0. . . . . . . . . .
99.5 -i- ...............
0 5 10 0 5 10 0 5 10
time (s) ume (s) time (s)
q cmd dts x 104 did
4 ............. i ...............
-5 ............... i . . . . . . . . . . . . .
• ......... ! ...........
-10 . . . . . . . '. . . . . . . . . . . . . . .
-15 -21 i
0 5 10 0 10 0 5 10
time (s) time (s) ~me (s)
0 10 0 5 10 0 5 10
time (s) time (s) ~me (s)
thrl thr2
1 0 0 ~
g~ so
0 0~
0 10 0 5 10
time (s) time (s)
Figure 31.12: Responses to a pitch rate c o m m a n d at Mach 0.3 altitude 5000 ft.
521
pw q
5 . az
50 -5 ............
! -1 .... !
0 2 4 6 0 2 4 6
: l0 2
-Ji4..........
time (s) time (s) time (s)
Va alpha beta
162"51 i 6
~2
0 4 0 4 6 0 2 4 6
time (s) time (s) time (s)
60 ....................
0 2 4 6 0 2 4 6 0 2 6
time (s) tlme (s) time (s)
dcs dcd fir
5 - -
_ ....
-4 -15
0 2 4 6 0 4 6 0 4 6
time (s) time (s) time (s)
thrl thr2
10
0 2 4 0 2 4 6
time Is) ~me (s)
Assessm. man.: roll rate demand, Mach 0.5, 15000 ft, param, unc.
Figure 31.13: Responses to a roll rate c o m m a n d at Mach 0.5 altitude 15000 ft.
with parametric uncertainties and measurement errors
522