Cinco Casos de Control Digital

Case Study 1: Satellite Control
CS1.1 Description
Space technology is one of the main achievements of the twentieth century. It has evolved
extraordinarily since the very irst satellites, the Soviet Sputnik (October 1957) and the
U.S. Explorer (January 1958). Our modern society depends greatly on satellites. They have
become the key instruments for telecommunications, airplane navigation, GPS systems,
meteorology, scientiic research, cellphones, TV, internet, etc.
Satellites can operate from low Earth orbits (LEO) around 300 km altitude, to geo-
stationary Earth orbits (GEO) about 42,000 km, and beyond the Earth gravity, toward
the exploration of other planets. Satellites can be very light, from just a few kilograms
(nano-satellites), or quite heavy, with more than 1000 kg (large satellites). In most cases,
they have solar panels to harvest energy from the Sun—see Figure CS1.1. This is usu-
ally possible if they are located not farther than the Asteroid belt, located between the
orbits of the planets Mars and Jupiter. The physical laws that describe the dynamics and
motion of a satellite are well known. The control system, and in particular the so-called
attitude control system (ACS), is one of the key components of every satellite. For addi-
tional information about spacecraft dynamics and control, see Reference 363, written by
Marcel Sidi, the irst PhD student of Professor Isaac Horowitz. Incidentally, notice that
the QFT-bounds are also known as the Horowitz–Sidi bounds.
This case study CS1 presents the design of an ACS for a satellite with lexible solar
panels. The control system has three simultaneous objectives for every plant within the
model uncertainty: (1) to regulate the angle of motion of the satellite, (2) to reject unpredict-
able disturbances, and (3) to minimize the solar panel vibrations.
We propose a classical 2DOF control system that includes a feedback compensator G(s)
and a preilter F(s), as shown in Figure CS1.3. In the following sections, we model and ana-
lyze the dynamics of the satellite, propose a set of stability and performance speciications,
design the controllers, and validate them in both the frequency and time domains.
CS1.2 Plant Model

Figure CS1.2 shows a model diagram for a satellite with two solar panels. It includes the
variables and parameters that describe the dynamics of the system. We consider the satel-
lite as a rigid central body with two equal and lexible solar panels, each one modeled as a
lumped mass at the end of a lexible beam.
The variable we want to control is θ(t), which is the angle of motion of the central body, in
[rad]. The satellite is equipped with reaction wheels that provide a symmetric torque Tm(t)
301
302 Case Study 1: Satellite Control
FIGURE CS1.1
Hispasat. A telecommunication satellite with solar panels.
about the axis of rotation, in [Nm]. In addition, yp(t) is the deformation of the solar panel,
in [m]. The main parameters of the system are
• J0, the moment of inertia of the central body, (kg m2)

• m, the mass of each solar panel, as a mass-point at the end of the panel, (kg)
• lp, the length of each solar panel, from center of the mass of the central body, (m)
• J = J0 + 2 m lp2, the moment of inertia of the entire system, (kg m2)
• Kp, the panel translational stiffness coeficient, (N/m)
• Bp, the panel translational damping coeficient, (Ns/m)
• t, the time (s).
y1
θ(t)
yp(t)
Bp
Kp
Tm(t)
m
x1
J0 m
Kp
lp Bp
yp(t)
FIGURE CS1.2
Satellite with solar panels. Model description.
Case Study 1: Satellite Control 303
We start the modeling with the general equation of motion of a mechanical system,
which is
M qɺɺ + C qɺ + K q = Q (qɺ , q, u, t) (CS1.1)
where M, C, and K are the mass, damping, and stiffness matrices, Q is the matrix of inputs,
q is the vector of generalized coordinates, and u is the vector of inputs. Also, the general
form of the Euler–Lagrange equation is
d  ∂L  ∂L ∂Dd
  − + = Qi , i = 1, 2, … , n and L = Ek − Ep (CS1.2)
dt  ∂qɺ i  ∂qi ∂qɺ i
where qi are the generalized coordinates or degrees of freedom of the system, Qi the gen-
eralized forces applied to each subsystem “i,” Ek and Ep the kinetic and potential energies,
Dd the dissipator co-content, and L the Lagrangian.236
The generalized coordinates for the satellite with two solar panels are the angle of
motion of the central body and the deformation of each solar panel, which are, respec-
tively: q1 = θ(t), and q2 = yp(t)—see Figure CS1.2. The system contains the kinetic, potential,
and dissipative functions described by Equations CS1.3 through CS1.5. Notice that we con-
sider a linear dissipator, which means that the Dd expression or dissipator co-content for
the Lagrange equations equals one-half of the power being absorbed by the dissipator.236
1 ɺ2 1 2
Ek = J 0θ + 2  m yɺ p + lp θɺ 
( ) (CS1.3)
2  2 
1 
Ep = 2 K p y p2  (CS1.4)
 2 
1 
Dd = 2 Bp yɺ p2  (CS1.5)
 2 
The generalized force applied about the axis of rotation of the central body, i = 1, is the
torque: Q1 = Tm(t). Using Equations CS1.3 through CS1.5, the terms for the Euler–Lagrange
equation are
 ∂Ek 
 
d  ∂L  d  ∂θɺ   J 0 + 2mlp
2
2mlp   θɺɺ 
  =  = = Mqɺɺ (CS1.6)
dt  ∂qɺ i  dt  ∂Ek   2mlp 2m   yɺɺp 
 
 ∂yɺ p 
 ∂Ep 
 
∂L ∂Ep  ∂θ   0 0  θ 
− = = =    = Kq (CS1.7)
∂qi ∂qi  ∂Ep   0 2K p   y p 
  
 ∂y p 
 ∂Dd 
 
∂Dd  ∂θɺ   0 0   θɺ 
= =  = Cqɺ (CS1.8)
∂qɺ i  ∂Dd   0 2Bp   yɺ p 
   
 ∂yɺ p 
1 0 Tm 
Q=    = Ru (CS1.9)
0 0  0 

Rearranging the equation of motion Equation CS1.1 as
qɺɺ = −M −1C qɺ − M −1K q + M −1R u (CS1.10)
and using also Equations CS1.6 through CS1.9, we ind the state space description of the
system ( xɺ = Ax + Bu ; y = Cx) with the expressions,
−M −1C −M −1K   M −1R 

xɺ =  x + 
 0 u

 I 0    (CS1.11)
 
y = [0|I ]x
with I = [1 0; 0 1], 0 = [0 0; 0 0], and,
 2lpBp 2lp K p 
0 0 
 J0 J0 
 
 −Bp ( J 0 + 2mlp2 ) −K p ( J 0 + 2mlp2 ) 
A =  0 0  (CS1.12)
 m J0 m J0 
1 0 0 0 
 
 
 0 1 0 0 
 1 
 0
 J0 
 
 −lp 
B= 0 (CS1.13)
 J0 
 
 0 0
 
 0 0

0 0 1 0
C=  (CS1.14)
0 0 0 1

 qɺ  T
x =   = θɺ yɺ p θ y p  (CS1.15)
 q  
 
yp(s)
p21(s)
do(s)
+
r(s) Tm(s) θ(s)
F(s) G(s) p11(s)
+ +
–
+
H(s)
+
n(s)
FIGURE CS1.3
ACS for satellite with solar panels.
T
u = Tm 0 (CS1.16)
y = [θ y p ]T (CS1.17)
Now, applying P(s) = C(sI−A)−1B for y(s) = P(s)u(s), the transfer function p11(s), from the
torque Tm to the angular movement θ of the satellite, is
θ(s) 1
= p11(s) =   2 2
(s2 + (Bp /m)s + (K p /m))
Tm (s)  J 0  s (s + (Bp J/(m J 0 ))s + (K p J/(m J 0 ))) (CS1.18)
with J = J 0 + 2m lp2
and the transfer function p21(s), from the torque Tm to the deformation of the solar panel yp,
is—see also Figure CS1.3
y p (s) 1 −lp

= p21(s) =   2
Tm (s)  
 J 0  (s + (Bp J/(m J 0 ))s + (K p J/(m
m J 0 ))) (CS1.19)
with J = J 0 + 2 m lp2
CS1.2.1 Parametric Uncertainty

For this study, we consider a medium size satellite with the parameters suggested in
Reference 363. In particular, the mass of each panel is m = 20 kg, the length of each panel,
from the center of the mass of the satellite to the end of the panel lp = 2 m, and the panel
translational stiffness coeficient Kp = 320 N/m. The moment of inertia of the entire system
follows the expression:J = J 0 + 2mlp2 = J 0 + 160.
Many satellites also have a signiicant mass of fuel to carry out orbit maneuvers with
thrusters. This mass of fuel is typically up to a 40% of the mass of the central body. As
this mass changes during the lifetime of the satellite, we include the corresponding uncer-
tainty in the moment of inertia J0 of the system. In addition, we include uncertainty in Bp,
which is the panel translational damping coeficient for the lexible panels, as it is quite
dificult to measure with enough precision. Thus, the transfer functions p11(s) and p21(s)
described in Equations CS1.18 and CS1.19 become
θ(s) 1
= p11(s) =   2 2
(s2 + (Bp /20)s + 16)
 J 0  s (s + (Bp ( J 0 + 160)/(20 J 0 ))s + (16( J 0 + 160)/J 0 )) (CS1.20)
Tm (s)
y p (s) 1 −2
= p21(s) =   2 (CS1.21)
Tm (s)  
 J 0  (s + (Bp ( J 0 + 160)/(20 J 0 ))s + (16( J 0 + 160)/J 0 ))
with: J0 ∈ [432, 720] kg m2, Bp ∈ [0.24, 0.48] Ns/m.
CS1.2.2 Frequencies
Figure CS1.5 shows the Bode diagram of the plant p11(s), including also the parametric
uncertainty. Based on this picture, we select an array of frequencies so that
w = [0.001 0.005 0.01 0.02 0.05 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2 3 4 4.68 5 10 15] rad/s
(CS1.22)
CS1.2.3 Nominal Plant

For the nominal plants p11,0(s) and p21,0(s), we select J0 = 432 and Bp = 0.24, resulting in the
transfer functions
θ(s) (s2 + 0.012s + 16)

= p11,0 (s) = 0.002315 2 2 (CS1.23)
Tm (s) 0 s (s + 0.0164s + 21..93)
y p (s) −0.00463
= p21,0 (s) = 2 (CS1.24)
Tm (s) 0 (s + 0.0164s + 21.93)
CS1.2.4 QFT Templates

Figure CS1.4 shows the QFT templates for this case. They are calculated using the plant
model and parametric uncertainty deined with Equation CS1.20, and the array of frequen-
cies of interest deined in Equation CS1.22.
CS1.3 Preliminary Analysis

Before starting the design of the controller, we conduct a brief preliminary analysis of the
dynamics of the satellite. Figure CS1.5 shows the Bode diagram and the NC of p11(s) with
four cases within the uncertainty:
Case p11-1(s): J0 = 720, Bp = 0.48; Case p11-2(s): J0 = 720, Bp = 0.32

Case p11-3(s): J0 = 720, Bp = 0.24; Case p11-4(s): J0 = 432, Bp = 0.48
QFTCT
–55 0.9
–60 1
–65
Open-loop magnitude (dB)
–70 5
2
–75
–80
3
–85
–90
–95 10
–100
15
–105
–180 –179.5 –179 –178.5 –178
Open-loop phase (deg)
FIGURE CS1.4
QFT templates for p11(s).
As we can see, the system p11(s) has a double integrator. This means that there is initially
no damping. The system is at the very limit of stability. If a disturbance enters into the
system, it will oscillate forever, as we can expect for any body in orbit without a control
system.
In addition, there is a severe resonance mode around ω = 4.50 rad/s, with an extremely
low damping—see peak in the Bode diagram or loop in the Nichols chart (Figure CS1.5).
This resonance is a consequence of the lexible panels. As a result, the bandwidth of the
control system is limited due to this resonance mode.
Moreover, this resonance can also vary with the parametric uncertainty. Thus, the natu-
ral frequency and damping are, respectively, ωn ∈ [4.42, 4.68] rad/s and ζ ∈ [0.0017, 0.0037].
As a consequence, this variation of the parameters could eventually change the natural
frequency and increase the peak of the resonance, making the system unstable.
(a) Bode diagram (b) Nichols chart
–20 20 –1 dB
3 dB
–3 dB
Magnitude (dB)
–40 6 dB
p11-4(s) 0 –6 dB
–60 –12 dB
p11-1(s) p11-4(s) –20 dB
Open-loop gain (dB)
–80 –20
p11-2(s)
–100 p11-2(s) –40 dB
p11-3(s) –40
–120
–60 –60 dB
0
–80 –80 dB
Phase (deg)
–45 p11-1(s)
–90 –100 –100 dB
–135 –120 dB
–120 p11-3(s)
–180
10–1 100 101 102 –180 –135 –90 –45 0
Frequency (rad/s) Open-loop phase (deg)
FIGURE CS1.5
Frequency-domain p11-i(s). (a) Bode diagram and (b) Nichols chart.
Bode diagram
Magnitude (dB)
–40 p21-4(s)
–60
–80
p21-1(s) p21-2(s) p21-3(s)
–100
180
Phase (deg)
135
90
45
0
100 101
Frequency (rad/s)
FIGURE CS1.6
Bode diagram of p21-i(s).
Similarly, Figure CS1.6 shows the Bode diagram of the plant p21_i(s) for the same four
cases. This plant describes the effect of the torque command Tm on the solar panel dis-
placement yp. Again, the peak at ωn ∈ [4.42, 4.68] rad/s is the resonance mode of the solar
panels.
CS1.4 Control Specifications

Taking into account the dynamics of the satellite and the operator objectives and limita-
tions, we deine the following robust control speciications—with H(s) = 1:
• Type 1: Stability speciication
p11(jω )G(jω )
T1(jω ) = ≤ δ1(ω) = Ws = 1.93
1 + p11(jω )G(jω )
(CS1.25)
ω ∈ [0.001 0.005 0.01 0.02 0.05 0.1
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2 3 4 4.68 5 10 15] rad/s
which is equivalent to PM = 30.03°, GM = 3.63 dB—see Equations 2.30 and 2.31.
• Type 3: Sensitivity or disturbances at plant output speciication—see Figure CS1.7a
θ( jω ) 1 (s/ad )
T3 ( jω) = = ≤ δ3 (ω ) = ; ad = 0.2
do ( jω) 1 + p11( jω )G( jω ) (s/ad ) + 1 (CS1.26)
ω ∈ [0.001 0.005 0.01 0.02 0.05 0.1] rad/s
(a) Bode diagram (b) Step response

1.2
0 δ6-up(s)
δ3(s) 1
–5
Magnitude (dB)
0.8
Amplitude
–10
0.6
–15
δ6-lo(s)
0.4
–20
0.2
–25
0
10–2 10–1 100 0 10 20 30 40 50 60 70 80
Frequency (rad/s) Time (s)
FIGURE CS1.7
Control speciications for p11(s). (a) Sensitivity or disturbance rejection at the output of the plant: δ3. (b) Reference
tracking: δ6-up, δ6-lo.
• Type 6: Reference tracking speciication—see Figure CS1.7b
p11( jω )G( jω )
δ6-lo (ω ) < T6 (jω ) = F( jω ) ≤ δ6-up (ω )
1 + p11( jω )G( jω )
(CS1.27)
ω ∈ [0.001 0.005 0.01 0.02 0.05 0.1
0.2 0.3 0.4 0.5 0.6 0.7 0..8 0.9 1] rad/s
1
δ6 -lo (s) = 2 ; aL = 0.13 (CS1.28)
[(s/aL ) + 1]
[(s/aU ) + 1] 1.25 aU
δ6 -up (s) = ; aU = 0.1; ζ = 0.8; ω n = (CS1.29)
(s/ωn )2 + (2ζ s/ωn ) + 1 ζ
 
• Type k1: External disturbances do(s) over yp(s) via p21(s) × G(s) and control loop
y p ( jω ) p21( jω ) G( jω )
Tk 1( jω ) = = ≤ δk 1(ω ) = 0.20
do ( jω ) 1 + p11( jω ) G( jω ) (CS1.30)
ω ∈ [4 4.68 5] rad/s
CS1.5 Controller Design

CS1.5.1 QFT-Bounds
The QFT-bounds integrate the dynamics of the plant, the model uncertainty, and the control
speciications at each frequency of interest. Figure CS1.8a–d shows the bounds for stability,
(a) QFTCT (b) QFTCT

50
40
45
35 0.001 0.005
40

30
From 0.001 to 2
25 35
0.5 4.68 0.01
20 30
10 0.02
15 25
15
10 20 0.05
5
3 15
0
4 10
–5 0.1
5
–10
–220–200–180–160–140–120–100 –80 –60 –40 –350 –300 –250 –200 –150 –100 –50 0
Open-loop phase (deg) Open-loop phase (deg)
(c) QFTCT (d) QFTCT

80
0.001 0
70
–10
60
0.005 5
–20
50
40 0.01 –30
30 –40
0.02
4.68
20 –50
0.05 4
10 –60
0
0.1 –70
0.2
–10
–350 –300 –250 –200 –150 –100 –50 0 –350 –300 –250 –200 –150 –100 –50 0
FIGURE CS1.8
(a) Stability bounds, (b) sensitivity bounds, (c) reference tracking bounds, and (d) solar panel vibration attenu-
ation bounds.
sensitivity, reference tracking, and solar panel vibration attenuation, respectively. Figure
CS1.9 presents the intersection of all these bounds at each frequency. As we have a bound
solution for each frequency, the selected control speciications for the satellite are compatible.
CS1.5.2 Loop-Shaping—G(s)
The design of the feedback controller G(s) is carried out on the Nichols chart (NC). It is
done by adding poles and zeros until the nominal loop, deined as L0(jω) = p11,0(jω)G(jω),
meets the QFT-bounds presented in Figure CS1.9.
We start the design of G(s) by adding a low frequency zero (z1 = 0.1) to move L0(s) from
−180° line to the right, and pass the stability bounds through the right. Then, we increase
the controller gain to k = 20. Afterward, to meet the dashed-line bound at 4.68 rad/s, we
add a notch ilter (s2 + 2 ζa ωn s + ω n2)/(s2 + 2 ζb ωn s + ω n2 ), with a natural frequency of
ωn = 4.68 rad/s, a damping numerator coeficient of ζa = 0.07, and a damping denominator
coeficient of ζb = 1.
QFTCT
80
60 0.005 0.001
Open-loop magnitude (dB) 40
20
0
0.01 0.05
0.02 5 0.1
–20 0.2 to 3 and 10, 15
–40
4 4.68
–60
–80
–350 –300 –250 –200 –150 –100 –50 0
FIGURE CS1.9
QFT-bounds. Intersection of bounds.
This gives a depth of 20 × log10(ζa/ζb) dB = –23.1 dB, which is enough to place L0(4.68)
below the B(4.68) dashed-line bound. Finally, we add two additional poles at a higher fre-
quency (p1 = p2 = 1), to ilter out high frequency noise. The inal controller G(s) is shown in
Equation CS1.31 and the loop-shaping in Figure CS1.10.
20 ((s/0.1) + 1) (s2 + 3.066s + 21.9)

G(s) = (CS1.31)
(s + 1)2 (s2 + 43.80s + 21.9)
CS1.5.3 Prefilter Design—F(s)

Taking into account the controller G(s) deined in Equation CS1.31, the reference track-
ing speciications presented in Equations CS1.27 through CS1.29, and the plant model
described in Equation CS1.20, we design a preilter F(s) to assure that all the input/output
functions p11(s)G(s)F(s)/[1 + p11(s)G(s)] from low frequencies to 1 rad/s (which is before the
lexible mode) are inside the band deined by the limits δ6-up(ω) and δ6-lo(ω). The preilter
is shown in Equation CS1.32, and the input/output functions and limits in Figure CS1.11.
((s/0.5) + 1)
F(s) = (CS1.32)
((s/0.1) + 1)
CS1.6 Analysis and Validation

The analysis of the closed-loop stability in the frequency domain is shown in Figure
CS1.12a. The dashed-line is the stability speciication Ws, deined in Equation CS1.25.
QFTCT
0.001 0.01
80
0.02
60
0.005 0.05 0.1
40 0.2
Open-loop gain (dB)
20
0
0.9 0.3
1
–20 0.6 0.5 0.4
–40 5 0.7
2 4.68
3 0.8
–60
–80 4
15 L0(s)
10
–100
–300 –250 –200 –150 –100
FIGURE CS1.10
QFT-bounds and G(s) design—loop-shaping.
The solid-line represents the worst case of all the possible functions p11G/(1 + p11G) at
each frequency and due to the model uncertainty. The control system meets the stabil-
ity speciication (the solid-line is below the dashed-line Ws) for all the plants within the
uncertainty.
QFTCT
20
0 δ6-up
–20
Magnitude (dB)
P G F/(1 + P G)
–40
δ6-lo
–60
–80
–100
10–3 10–2 10–1 100 101
Frequency (rad/s)
FIGURE CS1.11
Preilter F(s) for reference tracking specs: δ6-lo(ω) ≤ |T6| ≤ δ6-up(ω).
(a) QFTCT (b) QFTCT

10 10
0 0
δ3
–10
–10
Ws
–20
–20
Magnitude (dB)
Magnitude (dB)
–30 Worst cases of

Worst cases of “1/(1+p11G)”
–30
“p11G/(1+p11G)”
–40
–40 0.8
15 –50
0.7
–50
0.6 –60
0.5 10 0.001
–60 –70
0.4 4 0.01 0.05
–70 0.001 0.01 0.1 0.3 0.9 3 4.68 –80 0.005 0.02 0.1
0.005 0.02 0.05 0.2 1 2 5
–80 –90
10–3 10–2 10–1 100 101 10–3 10–2 10–1 100 101
Frequency (rad/s) Frequency (rad/s)
(c) QFTCT
–20
δk1
–40
Magnitude (dB)
–60 4 5
Worst cases of
“p21G/(1+p11G)” 4.68
–80
–100
–120
100 101
Frequency (rad/s)
FIGURE CS1.12
Frequency-domain analysis: (a) stability, (b) sensitivity or disturbance rejection at plant output, and (c) panel
vibration attenuation.
The analysis of the robust sensitivity speciication, or rejection of disturbances at the

output of the plant, is shown in Figure CS1.12b. The dashed-line is the sensitivity specii-
cation δ3(ω), deined in Equation CS1.26. The solid-line represents the worst case of all the
possible functions 1/(1 + p11G) at each frequency and due to the model uncertainty. The
control system meets the sensitivity speciication in all cases (the solid-line is below the
dashed-line δ3) from 0 to 0.1 rad/s.
The panel vibration analysis in the frequency domain is shown in Figure CS1.12c. It
represents the attenuation of the effect of disturbances do(s) over the panel movement yp(s)
through p21(s) × G(s) and the control loop—see also Figure CS1.3. The dashed-line is the
speciication δk1(ω) deined in Equation CS1.30. The solid-line represents the worst case of
all the possible functions p21G/(1 + p11G) at each frequency due to the model uncertainty.
(a) QFTCT (b) QFTCT

1.2
δ6-up 1
1 0.8
0.6
0.8
0.4
δ6-lo
0.6 0.2
0
0.4
–0.2
0.2
–0.4
0 –0.6
0 10 20 30 40 50 60 70 80 0 5 10 15 20 25 30 35 40 45
Time (s) Time (s)
FIGURE CS1.13
Time-domain analysis. (a) Reference tracking: δ6-lo and δ6-up specs, and θ(t) of 200 cases of p11GF/(1 + p11G) to a
unitary step reference r(t). (b) Sensitivity: θ(t) of 200 cases of 1/(1 + p11G) to a unitary step disturbance do(t).
Once more, the control system meets the speciication in all cases (solid-line is below
dashed-line δk1).
The time-domain analysis of the reference tracking speciication is shown in Figure
CS1.13a. The igure shows the limits δ6-up(ω) and δ6-lo(ω)—see Equations CS1.29 and CS1.28,
and the time responses of the angle θ(t) to a unitary step reference r(t). We simulate 200
cases of the closed-loop transfer function p11GF/(1 + p11G)—see Figure CS1.3 with H(s) = 1.
The control system meets the speciication (is between the upper and lower limits) in all
cases.
In addition, the time-domain analysis of the sensitivity is shown in Figure CS1.13b. The
igure presents the time responses of the angle θ(t) to a unitary step disturbance do(t)—see
also Figure CS1.3 with H(s) = 1. We simulate 200 cases of 1/(1 + p11G). The control system
achieves a good disturbance rejection in all cases.
In addition, Figure CS1.14 shows the solar panels oscillation when a unitary step distur-
bance input do(s) is introduced into the system. The disturbance do(s) enters at the output of
the plant p11(s), travels the control loop until Tm(s), and inally goes through p21(s) to yp(s)—
see the block diagram in Figure CS1.3. We study two cases. The irst one, represented in
Figure CS1.14a, uses the complete controller G(s) described in Equation CS1.31. The second
case, represented in Figure CS1.14b, uses this controller without the notch ilter.
The solar panels oscillation is signiicantly reduced when we add this notch ilter in
the feedback controller G(s), that is, the complete Equation CS1.31. Figure CS1.14a shows
the yp(t) results with the complete G(s), and Figure CS1.14b without the notch ilter. Both
analysis are performed for the same cases within the plant uncertainty. The igures have
the same scale.
Finally, Figure CS1.15 zooms in one case of the simulations of Figure CS1.14a and b. As
we can see, the frequency of the oscillation in the igure (15 peaks in 20 s) is the natural
frequency of the solar panel oscillation: 0.75 Hz (4.68 rad/s). The results of the complete
controller G(s) with the notch ilter are represented by the solid-line—Equation CS1.31,
and the results without the notch ilter by the dashed-line. As anticipated, the complete
(a) QFTCT (b) QFTCT

0.02 0.02
0.015 0.015
0.01 0.01
0.005 0.005
0 0
–0.005 –0.005
–0.01 –0.01
–0.015 –0.015
–0.02 –0.02
–0.025 –0.025
–0.03 –0.03
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (s) Time (s)
FIGURE CS1.14
Solar panels oscillation yb(t). Simulation of 25 plants. (a) G(s) with the notch ilter—see Equation CS1.31. (b) G(s)
without the notch ilter.
QFTCT
0.02
Solar panel natural frequency
0.015 ~0.75 Hz = ~15 peaks in 20 s
0.01
0.005
–0.005
–0.01
Case G without notch filter
–0.015
Case G with notch filter
–0.02
–0.025
0 5 10 15 20 25 30
Time (s)
FIGURE CS1.15
Zooming in. Solar panel oscillation y b(t). G(s) with notch ilter (solid-line), and without notch ilter (dashed-line).
controller G(s) with the notch ilter obtains a much better peak to peak reduction of the
solar panels vibration.
CS1.7 Summary
This case study has successfully designed a robust QFT Attitude Control System for a sat-
ellite with lexible solar panels. The design considers the model uncertainty introduced by
the fuel consumption and the imprecise knowledge of the damping coeficients. It accom-
plishes four simultaneous control objectives:
• Stability
• Reference tracking and regulation of the satellite angle
• Rejection of unpredictable disturbances
• Minimization of the solar panel vibrations
Case Study 2: Wind Turbine Control
CS2.1 Description
Wind turbines are complex systems with large and lexible structures that work under
unpredictable environmental conditions and a variable electrical grid. The eficiency and
reliability of a wind turbine depend on how the control system deals with the nonlin-
ear characteristics, high model uncertainty, stability limitations, energy capture maximi-
zation, and load and mechanical fatigue reduction. These objectives require the control
system to coordinate numerous variables in real time, including the blade pitch angles,
torques, active and reactive powers, rotor speed, yaw orientation, temperatures, currents,
voltages, and power factors.
This case study presents the design of a pitch control system to regulate the rotor
speed of a variable-speed pitch-controlled wind turbine in Region 3 (above-rated region,
12 < wind speed < 17 m/s, see Figure CS2.19), while attenuating simultaneously the
tower and blades vibration. The machine is composed of a gearless drive-train, a multi-
pole synchronous generator, and a full-power converter. It is also called a Type-4 wind
turbine. As an example, Figure CS2.1 shows a picture of the irst prototype of the TWT-
1.65, a 1.65 MW Type-4 wind turbine designed by the author and the MTOI team in the
period 1998–2011.
The four objectives pursued in this case study are: (1) to control the rotor speed Ωr by
pitching simultaneously the angles β of the turbine blades—also called collective pitch,
(2) to reject the effect of unpredictable wind disturbances, (3) to minimize the tower fore-
aft oscillations, and (4) to attenuate the blades lap-wise vibrations.
For these objectives, we propose a classical 2DOF control system that includes a feedback
compensator G(s) and a preilter F(s), as shown in Figure CS2.5. In the following sections
we model and analyze the dynamics of the wind turbine, propose a set of stability and
performance speciications, design the controllers, and validate them in both frequency
and time domains.
CS2.2 Plant Model

Figure CS2.2 shows the model diagram for the Type-4 wind turbine. The model consid-
ers four degrees of freedom: the rotor rotation θr(t), the generator rotation θg(t), the lap-
wise blade bending yb(t), and the fore-aft tower bending yt(t). The main variables and
parameters of the system are the following:
317
318 Case Study 2: Wind Turbine Control
FIGURE CS2.1
TWT-1.65 variable-speed pitch-controlled wind turbine, MTOI, 2001.
yb
β
Bb
rb Shaft
Blade
FT Kb Rotor excitation
Generator
rp mb Nacelle
Wind
ν1
Grid
Ta mh θr Bs θg Tg
DC P, Q
Hub
Ks Jg Transformer
Jr mn
Rotor mb Bt yt
Full-power
Kb converter
Kt
mt
Bb
yb β
Tower
Foundation
FIGURE CS2.2
Variable-speed pitch-controlled gearless wind turbine with full-power converter (Type-4 wind turbine).
Model description.
Inputs
• v1 undisturbed upstream wind speed, [m/s]
• FT thrust force applied by the wind on the rotor, [N]
• Ta aerodynamic torque applied by the wind on the rotor, [Nm]
• Tg antagonistic electrical torque applied by the rectiier/generator, [Nm]
• β pitch angle of the blades, [rad]
Case Study 2: Wind Turbine Control 319
• Tgd desired electrical torque, output of torque controller, [Nm]

• βd desired pitch angle, output of pitch controller, [rad]
Outputs
• yt fore-aft displacement of the tower at the nacelle level, [m]
• yb lap-wise displacement of the tip of the blade, [m]
• θr rotor angular position, [rad]
• θg generator angular position, [rad]
• Ωr = θɺr rotor angular speed, [rad/s]
• Ωg = θɺg generator angular speed, [rad/s]
• P active power at the output of the inverter, [W]
• Q reactive power at the output of the inverter, [VA]
Parameters
• ρ density of the air, [kg/m3]
• N number of blades, [−]
• rb radius of the blade, [m]
• rp distance from the center of the rotor to the center of pressure, or point where the
equivalent lumped force FT is applied. rp = (2/3) rb , [m]
• Jg moment of inertia of the generator, at Ωg, [kg m2]
• Jh moment of inertia of the hub, at Ωr , [kg m2]
• Jb moment of inertia of one blade, about root, at Ωr , [kg m2]
• Jr = Jh + N Jb , moment of inertia of the rotor, at Ωr , [kg m2]
• mb mass of one blade, [kg]
• mh mass of the hub, [kg]
• mn mass of the nacelle, [kg]
• mt mass of the tower, [kg]
• Kt tower translational fore-aft stiffness coeficient, [N/m]
• Bt tower translational fore-aft damping coeficient, [Ns/m]
• Kb blade translational lap-wise stiffness coeficient, [N/m]
• Bb blade translational lap-wise damping coeficient, [Ns/m]
• Ks shaft torsional stiffness coeficient, [Nm]
• Bs shaft torsional damping coeficient, [Nms]
• m1 = mn + mh + (1/3)mt + Nmb. Equivalent mass at the tip of the tower, for fore-aft
yt movement, [kg]
• m2 = J b /rb2 . Equivalent mass at the tip of the blade, for lap-wise yb movement, [kg]
As in case study CS1, we start with the general equation of motion for the mechanical
system, which is
M qɺɺ + C qɺ + K q = Q (qɺ , q, u, t) (CS2.1)

where M , C, and K are the Mass, Damping, and Stiffness matrices, Q the matrix of inputs,
q the vector of generalized coordinates, and u the vector of inputs. Also, we consider the
general form of the Euler–Lagrange equation,
d  ∂L  ∂L ∂Dd
  − + = Qi , i = 1, 2, … , n and L = Ek − Ep (CS2.2)
eralized forces applied to each subsystem “i,” Ek the kinetic energy, Ep the potential energy,
The generalized coordinates for the wind turbine are the linear displacement of the tower
yt(t), the linear displacement of the blades yb(t), the rotor angle θr(t) and the generator angle
θg(t), which are respectively: q1 = yt(t), q2 = yb(t), q3 = θr(t), and q4 = θg(t)—see Figure CS2.2. The
system contains the kinetic, potential, and dissipative functions described by Equations CS2.3
through CS2.5. Notice that we consider linear dissipators, which means that the Dd expression
or dissipator co-content equals one-half of the power being absorbed by the dissipators.236
1 1 1 1
Ek = m1 yɺ t2 + N m2 ( yɺ t + yɺ b )2 + J r θɺr2 + J g θɺg2 (CS2.3)
2 2 2 2
1 1 1
Ep = Kt yt2 + N K b yb2 + K s (θr − θ g )2 (CS2.4)
2 2 2
1 1 1
Dd = Bt yɺ t2 + N Bb yɺ b2 + Bs (θɺr − θɺg )2 (CS2.5)
2 2 2
The generalized forces applied to the wind turbine are the thrust force FT(t) and aero-
dynamic torque Ta(t), both applied by the wind to the rotor, and the antagonistic electrical
torque Tg(t), applied by the power electronics and electrical generator. Now, using Equations
CS2.3 through CS2.5, the terms for the Euler–Lagrange equation are
 ∂Ek 
 
 ∂yɺ t 
 
 ∂Ek   m1 + N m2 N m2 0 0   yɺɺt 
   
d  ∂L  d  ∂yɺ b   N m2 N m2 0 0   yɺɺb  (CS2.6)
  =  =   = M qɺɺ
dt  ∂qɺ i  dt  ∂Ek   0 0 Jr 0   θɺɺr 
 ∂θɺ   0 0 0 J g   θɺɺg 
 r   
 ∂Ek 
 ɺ 
 ∂θg 
 ∂Ep 
 
 ∂yt 
 
 ∂Ep   Kt 0 0 0   yt 
    
∂L ∂Ep  ∂yb   0 NK b 0 0   yb  (CS2.7)
− = = = =Kq
∂qi ∂qi  ∂Ep   0 0 Ks −K s   θr 
   K s   θg 
 ∂θr   0 0 −K s
 ∂Ep 
 
 ∂θg 
 
 ∂Dd 
 
 ∂yɺ t 
 
 ∂Dd   Bt 0 0 0   yɺ t 
   
∂Dd  ∂yɺ b   0 NBb 0 0   yɺ b 
=    = C qɺ
∂qɺ i = −Bs   θɺr 
(CS2.8)
 ∂Dd   0 0 Bs
 
 ∂θɺr   0 0 −Bs Bs   θɺg 
    
 ∂Dd 
 
 ∂θɺg 
 
1 0 0
   FT 
1 0 0   
Q =   Ta = R u (CS2.9)
0 1 0   
Tg
0
 0 −1  
and applying Equations CS2.6 through CS2.9, we ind a state space description of the
system ( xɺ = Ax + Bu; y = Cx) with the expressions,
−M −1C −M −1K   M −1R 

xɺ =  x + 
 0 u

 I 0  (CS2.11)
  
y = [ I | 0]x
with I = [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1], 0 = [0 0 0 0; 0 0 0 0; 0 0 0 ; 0 0 0 0], and,
 Bt N Bb Kt N Kb 
− 0 0 − 0 0 
 m1 m1 m1 m1 
 
 Bt −Bb (m1 + N m2 ) Kt −K b (m1 + N m2 ) 
 0 0 0 0 
 m1 m1 m2 m1 m1 m2 
 
 −Bs Bs −K s Ks 
 0 0 0 0 
 Jr Jr Jr Jr 
A =  
Bs −Bs Ks −K s 
 0 0 0 0 
 Jg Jg Jg Jg 
 
 1 0 0 0 0 0 0 0 

 0 1 0 0 0 0 0 0 

 0 0 1 0 0 0 0 0 

 
 0 0 0 1 0 0 0 0 
(CS2.12)
 0 0 0 
 
 m1 + N m2 1 
 − 0 0 
 N m1 m2 m1 
 
 1 
 0 0 
 Jr
 
B =  0 0
−1  (CS2.13)
 J g 

 0 0 0 

 
 0 0 0 
 
 0 0 0 
 
 0 0 0 
1 0 0 0 0 0 0 0
 
0 1 0 0 0 0 0 0
C =  (CS2.14)
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0

 qɺ 
x =   = [ yɺ t yɺ b θɺr θɺg yt y b θr θ g ]T (CS2.15)
 q
 
u = [FT Ta Tg ]T (CS2.16)
y = [ yɺ t yɺ b θɺr θɺg ]T (CS2.17)
The thrust force FT and aerodynamic torque Ta applied by the wind to the rotor are pro-
portional to the rotor area (πrb2 ) , the density of the air ρ, and the square and cubic of the
wind velocity v1, respectively, as described by Equations CS2.18 and CS2.19, so that
1
FT = ρπ rb2CT (λ , β)v12 (CS2.18)
2
ρπrb2CP (λ , β )v13
Ta = (CS2.19)
2Ωr
where Cp is the power coeficient and CT the torque coeficient. Both coeficients are a
nonlinear function of the tip-speed ratio λ = Ωrrp/v1 and the pitch angle β, as shown in
Figure CS2.3. For more information see the author’s book, Wind Energy Systems: Control
Engineering Design, Reference 7.
According to these expressions and the characteristics of Cp and CT, the inputs FT and
Ta of Equation CS2.16 depend on v1, β, and Ωr in a nonlinear way. Notice that Ωr = θɺr . Now,
linearizing these equations around a working point (v10, β0, Ωr0), and ignoring the bias
0.5
β = 0° Pitch angle β
0.45
increase
0.4 β = 1°
β = 2°
0.35
Region 3 β = 3°
0.3 12 to 17 m/s
working points
Cp
0.25
0.2
0.15
0.1
0.05
β = 22°
0
0 2 4 6 8 10 12 14 16
Tip-speed ratio
FIGURE CS2.3
C p /λ curves for pitch angles 0° ≤ β ≤ 22° and working points.
components, the inputs FT and Ta are described by a transfer matrix whose elements are
just gains, so that
Ωr(s)
 FT (s)  K FΩ K FV K Fβ   
 =   v1(s)  (CS2.20)
 Ta (s)  KTΩ
   KTV KTβ   

 β(s) 
where the gains are calculated by using the CT and Cp curves (Figure CS2.3) and Equations
CS2.18 and CS2.19, so that
∂FT (t) 1 ∂CT 2

K FΩ = = ρπrb2 v10 (CS2.21)
∂Ωr(t) 0 2 ∂Ωr 0
∂FT (t) 1  ∂C 
K FV = = ρπ rb2  T v10 2
+ 2v10CT 0  (CS2.22)
∂v1(t) 0 2  ∂v1 0 
∂FT (t) 1 ∂CT 2

K Fβ = = ρπ rb2 v10 (CS2.23)
∂β(t) 0 2 ∂β 0
∂Ta (t) 1  ∂C p 1 1  3
KTΩ = = ρπrb2  − Cp 0 2  v10 (CS2.24)
∂Ωr (t) 0 2  ∂Ωr 0 Ωr 0 Ωr 0 
∂Ta (t) 1 1  ∂Cp 3 2 


KTV = = ρπrb2  v10 + 3Cp 0v10  (CS2.25)
∂v1(t) 0 2 Ωr 0  ∂v1 0 
∂Ta (t) 1 1 ∂C p 3
KTβ = = ρπrb2 v10 (CS2.26)
∂β(t) 0 2 Ωr 0 ∂β 0
Then, adding the electrical torque, we have
Ωr(s)
 M −1R 
 FT (s)  K FΩ K FV K Fβ 0  
     v1(s) 
  u = B  T (s) = B  K KTV KT β 
0    (CS2.27)
 0   a   TΩ β ( s) 
  T (s)  0 0 0 1  
 g   
Tg (s)
On the other hand, the transfer functions of the actuators are
β(s) = Aβ (s)βd (s) (CS2.28)
Tg (s) = AT (s)Tgd (s) (CS2.29)
where βd is the demanded blade pitch angle and Tgd the demanded electrical torque, both
calculated by the control system, and where Aβ(s) and AT(s) are the transfer functions from
the control signals (βd, Tgd) to the actual value of the actuators (β, Tg). Using these expres-
sions with Equation CS2.27, we have
 Ωr (s)   Ωr(s) 
 M −1R 
 K FΩ K FV K Fβ Aβ (s) 0    
   v1(s)   v1(s) 
  u = B K KT β Aβ (s) 0      
 0   TΩ KTV
β ( s)  = BK  βd (s)  (CS2.30)
   AT (s) 
d   
 0 0 0  T (s)
Tgd (s)  gd 
Now, putting all the Ωr = θɺr variables together, that is, adding the irst column of BK to
the third column of A {i.e., A(1:8,3) = A(1:8,3) + BK(1:8,1)}, and removing it from BK, {i.e.,
BK(:,1) = []}, the new vector of inputs is
u = [v1 βd Tgd ]T (CS2.31)
and applying P(s) = C(sI − A)−1 BK for y(s) = P(s)u(s), the transfer matrix from the three
independent inputs to the four outputs is
 yɺ t (s) 
   v1(s) 
 yɺ b (s)   
  = P4×3 (s)  βd (s)  (CS2.32)
 Ωr (s)   
  T (s)
Ω (s) 
 gd 
 g 
The third row of P4×3(s) contains the transfer functions of the rotor speed Ωr(s) from the
wind speed v1(s), the demanded blade pitch angle βd(s), and the demanded electrical torque
Tgd(s), respectively, so that
v1(s) KTV
+
(Jg s2 + Bss + Ks)
β(s) +
βd(s) Aβ(s) KTβ + 1 Ωr(s)
+ ∆(s)
Tg(s)
Tgd(s) AT(s) [–(Bss + Ks)]
KTΩ
Depending on
working point (v10, β0, Ωr0)
FIGURE CS2.4
Rotor speed linear transfer functions, p31(s), p32(s), p33(s), from the wind speed, demanded pitch angle, and
electrical torque, respectively. The parameters KTV, KTβ, and KTΩ have uncertainty to absorb the nonlinear
characteristics.
Ωr(s) = p31(s)v1(s) + p32 (s)βd (s) + p33 (s)Tgd (s) (CS2.33)
with
p31(s) =
( J g s2 + Bss + K s ) KTV (CS2.34)
∆(s)
p32 (s) =
( J g s2 + Bss + K s ) KTβ Aβ (s) (CS2.35)
∆(s)
−(Bs s + K s )AT (s)

p33 (s) = (CS2.36)
∆(s)
∆(s) = J g J r s3 + (Bs J g + Bs J r − J g KTΩ )s2 + ( J g K s − Bs KTΩ + J r K s )s − KTΩ K s (CS2.37)
Figure CS2.4 shows graphically the rotor speed signal Ωr(s) as a function of the wind
speed v1(s), the demanded blade pitch angle βd(s), and the demanded electrical torque Tgd(s)
according to Equations CS2.33 through CS2.37. In addition, Figure CS2.5 shows the com-
plete block diagram representation of Equation CS2.32 and the rotor speed/pitch control
system, with the feedback controller G(s) and the preilter F(s) to be designed in the next
sections.
CS2.2.1 Parametric Uncertainty

The parameters for the model of the wind turbine considered in this case study are shown
in Table CS2.1. They are taken from the NREL 5-MW baseline wind turbine described in
Reference 364, and from the author’s book, Reference 7.
the parameters KTΩ, KTV, and KTβ have been calculated from Figure CS2.3 and accord-
NO T E :
ing to Equations CS2.20 through CS2.26, and for a scenario of wind velocity between 12
and 17 m/s. See Figure CS2.19 and Reference 7 for more details.
v1(s) p11(s) v1(s)

+ . p21(s)
yt(s) + .
p12(s) βd(s) yb(s)
+ p22(s)
+ +
p13(s) +
Tgd(s) p23(s)
p31(s) do(s) v1(s)

+ p41(s)
r(s) βd(s) Ωr(s) +
F(s) G(s) p32(s) βd(s) Ωg(s)
+ + p42(s)
– Pitch control p33(s) + +
Tgd(s) Tgd(s) p43(s) +
Torque
control
+
H(s)
+
n(s)
FIGURE CS2.5
Complete block diagram representation of Equation CS2.32 and wind turbine rotor speed/pitch control system.
TABLE CS2.1
Wind Turbine Parameters
Power to the grid [W]—rated 5.0 × 106 ρ (air) [kg/m3] 1.225
Eficiency shaft and generator 94.5% Eficiency converter 98.5%
Tg.max to generator [Nm]—rated 400,6111 h [m]—tower height 87.6
Ωr_nom [rad/s] 1.2671 N (number of blades) 3
rb (blade radius) [m] 61.5 rp [m] 41
Jg [kg m2]—generator 5,025,497 Jh [kg m2]—hub 115,926
Jb [kg m2]—one blade 11,776,047 Jr [kg m2]—rotor 35,444,067
mb [kg]—one blade 17,740 mh [kg]—hub 56,780
mn [kg]—nacelle 240,000 mt [kg]—tower 347,460
m1 [kg] = mn + mh + (1/3)mt + Nmb 465,820 m2 [kg] = J b /rp2 3113.5
Ks [Nm]—shaft 76,387,512 Bs [Nms]—shaft 6,215,000
Kt [N/m]—tower 1,930,491 Bt [Ns/m]—tower 18,965
Kb [N/m]—one blade 54,586 Bb [Ns/m]—one blade 124.5
KTβ min [Nm/deg] −4.1439 × 105 KTβ max [Nm/deg] −2.7894 × 104
KTΩ min [Nm/(rad/s)] −3.3095 × 10 6 KTΩ max [Nm/(rad/s)] −8.8646 × 105
KTV min [Nm/(m/s)] 5.0115 × 10 5 KTV max [Nm/(m/s)] 6.1698 × 105
Aβ(s) 1.0 AT(s) 1.0
The wind turbine studied here presents three lexible modes: one for the shaft tor-
sion, another for the tower fore-aft movement, and another for the blade lap-wise
movement. These displacements are described in Figure CS2.2. Using the param-
eters of Table CS2.1, the natural frequency and damping coeficients of these three
movements are
Ks Bs
ωn.shaft = = 3.8987 rad/s (0.6205 Hz), ζ shaft = = 0.1586
Jg 2 Ks J g
QFTCT
0 0.005
0.05
0.1
–20
0.5 0.001
–40 1 0.01
2.04
–60
10
5 4.19
50
3
–80
–100
100
–270 –260 –250 –240 –230 –220 –210 –200 –190 –180 –170
FIGURE CS2.6
Kt Bt
ωn.tower . fore.aft = = 2.0358 rad/s (0.3240 Hz), ζ tower . fore.aft = = 0.01
m1 2 Kt m1
Kb Bb
ωn.blade. flapwise = = 4.1871 rad/s (0.6664 Hz), ζ blade . flapwise = = 0.004775
m2 2 K b m2
This leads to the following expressions:
Ωr (s) (5.025497 ×106 s2 + 6, 215, 000s + 7.63875×107 )KTV (CS2.38)

= p31(s) =
v1(s) 1.78124 ×1014 s3 + (2.51518 ×1014 − 5.025497 ×106 KTΩ )s2 + ⋯
+ (3.09137 ×1015 − 6 , 215 , 000KTΩ )s − (7.63875×107 KTΩ )
Ωr(s) (5.025497 ×106 s2 + 6, 215, 000s + 7.63875×107 )KT β (CS2.39)

= p32 (s) =
βd (s) 1.78124 ×1014 s3 + (2.51518 ×1014 − 5.025497 ×106 KTΩ )s2 + ⋯
+ (3.09137 ×1015 − 6, 215, 000KTΩ )s − (7.63875×107 KTΩ )
(a) Bode diagram (b) Nichols chart

10 3 dB
6 dB
Magnitude (dB)
–20 p32-0(s)
0 p32-1(s)
–40 –10
Open-loop gain (dB)

p32-1(s)
–20
–60
–30 p32-0(s)
180
p32-1(s) –40
Phase (deg)
p32-0(s) –50
135
–60
–70
90
10–3 10–2 10–1 100 101 90 135 180 225
FIGURE CS2.7
p32-0(s) and p32-1(s). (a) Bode diagram and (b) Nichols chart. See shaft resonant mode at ωn.shaft = 3.90 rad/s.
with the parametric uncertainty in the aerodynamic coeficients,
KTβ ∈ [−41.439, −2.7894] × 104 Nm/deg,

KTΩ ∈ [−33.095, −8.8646] × 105 Nm/(rad/s),
KTV ∈ [5.0115, 6.1698] × 105 Nm/(m/s).
CS2.2.2 Frequencies
Figures CS2.7 and CS2.8 show the Bode diagrams for the transfer functions p32(s), p12(s),
and p22(s). We select the array of frequencies of interest based on this information—see
Equation CS2.40. Notice that we include frequencies that represent the resonances (peaks)
that appear in the diagrams, and which represent the lexible modes of the wind turbine
for shaft torsion, and tower fore-aft and blade lap-wise displacements.
ω = [0.001 0.005 0.01 0.05 0.1 0.5 1 2.04 3 4.19 5 10 50 100] rad/s
(CS2.40)
(a) Bode diagram (b) Bode diagram

20
–20
Magnitude (dB)
Magnitude (dB)
0
–40 p12-0(s) p22-0(s)
–20
–60
p12-1(s) –40
–80
–60 p22-1(s)
–100
180 180
Phase (deg)
Phase (deg)
90 p12-0(s) 135 p22-0(s)

0 90
p22-1(s)
–90 p12-1(s) 45
–180 0
100 101 100 101

FIGURE CS2.8
Bode diagrams of (a) p12-0(s) and p12-1(s), (b) p22-0(s), and p22-1(s). See tower and blade modes at ωn.tower = 2.04 rad/s
and ωn.blade = 4.19 rad/s.

Equations CS2.41 and CS2.42 represent the nominal plants p32-0(s) and p31-0(s), respectively.
They are based on the following selection of parameters within the uncertainty:
KTβ0 = −41.439 × 104 Nm/deg,

KTΩ0 = –33.095 × 105 Nm/(rad/s), and
KTV0 = 5.0115 × 105 Nm/(m/s), resulting the next transfer functions
Ωr(s) 0.01414s2 + 0.01749s + 0.2149

= p31.0 (s) = 3 (CS2.41)
v1(s) 0 s + 1.505s2 + 17.47 s + 1.419
Ωr (s) −0.01169s2 − 0.01446s − 0.1777

= p32 ,0 (s) = 3 (CS2.42)
βd (s) 0 s + 1.505s2 + 17.47 s + 1.419

Figure CS2.6 shows the QFT templates for p32(s), which is the main plant of the control loop.
They are calculated using the model and uncertainty deined after Equation CS2.39, and
the array of frequencies of interest deined in Equation CS2.40.

Before starting the design of the controller, we conduct a brief preliminary analysis of
the dynamics of the wind turbine. Figure CS2.7 shows the Bode diagram and the Nichols
chart of p32(s) with two extreme cases within the uncertainty:
Case 0: p32,0(s). KTΩ = −33.095 × 105, KTβ = −41.439 × 104, KTV = 5.0115 × 105
Case 1: p32,1(s). KTΩ = −2.7894 × 105, KTβ = −8.8646 × 104, KTV = 6.1698 × 105
The plant p32(s) describes the effect of the pitch command βd on the rotor velocity Ωr.
As we can see, the system has no integrators (system type 0) and has a negative gain.
The uncertainty affects signiicantly the magnitude of the gain. It increases by a factor of
2.5 as we go from case 0 (v1 = 12 m/s) to case 1 (v1 = 17 m/s). Numerically, dcgain ∈ [−0.125,
−0.318]. This variation is due to the nonlinear changes of the aerodynamics between 12
and 17 m/s, respectively.
In addition, there is a resonance mode at ω = 3.90 rad/s. This is represented by the peak
in the Bode diagram and the loop in the Nichols chart. The damping coeficient is about
ζ = 0.16. This resonance mode is a consequence of the shaft torsion. As a result, the band-
width of the control system will have to be below this resonance frequency to reduce the
excitation of this shaft mode.
Furthermore, Figure CS2.8a shows the Bode diagram of the plant p12(s) for both cases,
0 and 1. This plant describes the effect of the pitch command βd on the fore-aft tower
movement yt. The irst peak in the Bode diagram, at ω = 2.04 rad/s, is the resonance mode
related to the fore-aft tower movement, which is particularly undamped, with a damping
coeficient of about ζ = 0.01. The second peak, at ω = 4.19 rad/s, is the resonance mode
related to the lap-wise blade movement. It is also extremely undamped, with a damping
coeficient of about ζ = 0.004775.
Similarly, Figure CS2.8b shows the Bode diagram of the plant p22(s) for both cases, 0 and
1. This plant describes the effect of the pitch command βd on the lap-wise blade displace-
ment yb. Again, the irst peak, at ω = 2.04 rad/s, is the resonance mode of the tower, and the
second peak, at ω = 4.19 rad/s, the resonance mode of the blade.
As we can see, both movements tower fore-aft and blade lap-wise are coupled. The two
resonance modes appear in both transfer functions, p12(s) and p22(s), from the pitch com-
mand βd to the tower and blade movements, yt and yb, respectively. They are also much
more severe than the shaft torsion mode. The bandwidth of the control system will have
to be below these two resonance modes to reduce the movement of the tower and blades
at these frequencies.

Taking into account the dynamics of the wind turbine, the operator objectives and limita-
tions, we deine the following robust control speciications—with H(s) = 1:
p32 ( jω )G( jω )
|T1( jω )|= ≤ δ1(ω ) = Ws = 1.46
1 + p32 ( jω )G( jω )
ω ∈ [0.001 0.005 0.01 0.05 0.1 0.5 1 2.04 3 4.19 5 10 50 100] rad/s
(CS2.43)

1.2
0 δ3(s) = δk1(s) 1
–5
Magnitude (dB)
0.8
Amplitude
–10 δ6–up(s)
0.6
–15
0.4 δ6–lo(s)
–20
–25 0.2
0
10–2 10–1 100 0 10 20 30 40 50 60
FIGURE CS2.9
Control speciications. (a) Disturbance rejection speciications: δ3 = δ k1 and (b) reference tracking: δ6-up, δ6-lo.
 s 
 
Ωr ( jω ) 1  ad 
|T3 ( jω )|= = ≤ δ3 (ω ) = ; ad = 0.2
 
do ( jω ) 1 + p32 ( jω )G( jω )  s  + 1
 ad  (CS2.44)
ω ∈ [0.001 0.005 0.01 0.05 0.1] rad/s
• Type k1: Wind disturbances v1(s) over Ωr(s) via p31(s) and loop—also Figure CS2.9a
 s 
 
Ωr( jω ) p31( jω )  ak 1 
|Tk 1( jω )|= = ≤ δk 1 ( ω ) = ; ak 1 = 0.2
v1( jω ) 1 + p32 ( jω )G( jω )  s 
  + 1
 ak 1  (CS2.45)
ω ∈ [0.001 0.005 0.01 0.05 0.1] rad/s
p32 ( jω )G( jω )
δ6 -lo (ω ) <|T6 ( jω )|= F( jω ) ≤ δ6 -up (ω )
1 + p32 ( jω )G( jω ) (CS2.46)
ω ∈ [0.001 0.005 0.01 0.05 0.1 0.5] rad/s
1 (CS2.47)
δ6 -lo (s) = ; aL = 0.22
 s   2
  + 1
 aL  
 
 s  
  + 1
 aU   1.25aU
δ6 -up (s) =   ; aU = 0.09; ζ = 0.8; ωn = (CS2.48)
 s 2  2ζ s   ζ
  +   
 ω   ω  + 1
 n n 
• Type k2: Wind disturbances v1(s) over yt(s) via p12(s) × G(s) × p31(s)—Figure CS2.10
yt ( jω ) p ( jω )p31( jω )G( jω ) 1
|Tk 2 ( jω )|= = 12 ≤ δk 2 (ω ) =
v1( jω ) 1 + p32 ( jω )G( jω ) 5s + 1 (CS2.49)
ω ∈ [2.04 4.19] rad/s
• Type k3: Wind disturbances v1(s) over yb(s) via p22(s) × G(s) × p31(s) and loop
yb ( jω ) p ( jω )p31( jω )G( jω )
|Tk 3 ( jω )|= = 22 ≤ δk 3 (ω ) = 0.85
v1( jω ) 1 + p32 ( jω )G( jω ) (CS2.50)
ω ∈ [2.04 4.19] rad/s
Bode diagram
5
–5
2.04 rad/s
–10
Magnitude (dB)
4.19 rad/s
–15
–20
–25
–30
–35
10–2 10–1 100 101
Frequency (rad/s)
FIGURE CS2.10
Control speciications. Type k2: attenuation of the effect of wind disturbances v1(s) over the tower yt(s) through
p12(s) × G(s) × p31(s) and control loop.

CS2.5.1 QFT Bounds
As mentioned, the QFT bounds take into account the plant dynamics, model, uncertainty,
and control speciications. Figure CS2.11a–f, show the bounds for stability, sensitivity, ref-
erence tracking, wind disturbance rejection, tower fore-aft oscillation attenuation, and
blade lap-wise vibration reduction, respectively. Figure CS2.12 presents the intersection
of all these six bounds at each frequency. As we have a bound solution for each frequency,
the selected control speciications for the wind turbine are compatible.
CS2.5.2 Loop Shaping—G(s)

The design of the feedback controller G(s) is carried out on the Nichols chart. It is done by
adding poles and zeros until the nominal loop, deined as L0(jω) = p32,0(jω)G(jω), meets the
QFT bounds presented in Figure CS2.12.
We start the design of G(s) by adding a negative sign, as the plant has a negative DC gain.
Then, we add an integrator to obtain zero steady-state error for step reference inputs, and
also to place L0(s) at the same phase of the indentation of the low-frequency bound B(0.001).
After this, we increase the magnitude of the controller gain to k = 35, to put L0(0.001) above
the B(0.001) bound. Next, we add a low-frequency zero (z1 = 0.11), to move L0(s) to the right,
and to pass the stability bounds (“circles”) through the right.
Afterward, to meet the dashed-line bound at 2.04 rad/s, we add a notch ilter
(s2 + 2ζ aωns + ωn2 )/(s2 + 2ζbωns + ωn2 ), with a natural frequency of ωn = 2.04 rad/s, a damp-
ing numerator coeficient of ζa = 0.2, and a damping denominator coeficient of ζb = 1.
(a) QFTCT (b) QFTCT

40
0.1 70
30 0.001
2.04 0.005
60

3 1 0.5
20 4.19
5
50
10
50 0.01
10
100 40 0.05
0 30
0.005 0.1
–10 0.01 20
0.05
0.001
10
–20
–240 –220 –200 –180 –160 –140 –120 –100 –350 –300 –250 –200 –150 –100 –50 0
(c) QFTCT (d) QFTCT
120 60
0.001
50
100 0.001 0.005
0.005 40
80
30
0.01
60 20
0.01 0.05
0.05 10 0.1
40
0
20 0.1
–10
0.5
0 –20 –350
–350 –300 –250 –200 –150 –100 –50 0 –300 –250 –200 –150 –100 –50 0
(e) QFTCT (f ) QFTCT
15 20
10
15
4.19 2.04
5
10
5
–5
2.04
4.19
0
–10
–15 –5
–350 –300 –250 –200 –150 –100 –50 0 –350 –300 –250 –200 –150 –100 –50 0
FIGURE CS2.11
QFT bounds. (a) Stability bounds, (b) sensitivity bounds, (c) reference tracking bounds, (d) wind disturbance
rejection bounds, (e) fore-aft tower vibration attenuation bounds, and (f) blade lap-wise vibration reduction
bounds.
QFTCT
120
0.001
100
0.005
80
60
0.01 0.05
40
0.1 0.5
20
4.19 100 50
10
0 1
2.04 3
5
–20
–350 –300 –250 –200 –150 –100 –50 0
FIGURE CS2.12
QFT bounds. Intersection of bounds.
This gives a depth of 20 × log10(ζa/ζb)dB = −13.98 dB, which is enough to place L0(2.04)
below the B(2.04) dashed-line bound.
Finally, we add one more additional pole at high frequency (p1 = 40), to ilter out high-
frequency noise. The inal controller G(s) is shown in Equation CS2.51 and the loop
shaping in Figure CS2.13.
 s  
−35  + 1
 0.11   (s2 + 0.816s + 4.1616)
G(s) = (CS2.51)
 s   (s2 + 4.08s + 4.1616)
s  + 1
 40  
CS2.5.3 Prefilter Design—F(s)

With the controller G(s) just designed—see Equation CS2.51, the reference tracking specii-
cations presented in Equations CS2.46 through CS2.48, and the plant model p32(s) described
in Equation CS2.39, we design a preilter F(s) to assure that all the input/output functions
p32(s)G(s)F(s)/[1 + p32(s)G(s)], from low frequencies up to 0.5 rad/s are inside the band deined
by the limits δ6-up(ω) and δ6-lo(ω). The designed preilter is shown in Equation CS2.52, and the
input/output functions and limits in Figure CS2.14.
 s  
  + 1

 2.5   (CS2.52)
F(s) =
 s  
  + 1
 0.2  
QFTCT
120
L0(s)
100
0.001
80 0.005
60
Open-loop gain (dB)
0.05 0.01
40
0.1 0.5
20
1
5
0
–20
4.19
2.04 3
–40 50
100 10
–60
–220 –200 –180 –160 –140 –120 –100 –80 –60 –40
FIGURE CS2.13
QFT bounds and G(s) design—loop shaping.
QFTCT
0
δ6-up
–10
–20
Magnitude (dB)
–30
p32GF/(1 + p32G)
–40
–50
–60
δ6-lo
–70
10–2 10–1 100 101

Frequency (rad/s)
FIGURE CS2.14

The analysis of the closed-loop stability in the frequency domain is shown in
Figure CS2.15a. The dashed line is the stability speciication Ws deined in Equation
CS2.43. The solid line represents the worst case of all the possible functions p32G/
(1 + p32G) at each frequency due to the model uncertainty. The control system meets the
stability speciication (the solid line is below the dashed line Ws) for all the plants within
the uncertainty.
The analysis of the robust sensitivity speciication, or rejection of disturbances at the
output of the plant, is shown in Figure CS2.15b. The dashed line is the sensitivity specii-
cation δ3(ω) deined in Equation CS2.44. The solid line represents the worst case of all the
possible functions 1/(1 + p32G) at each frequency due to the model uncertainty. The control
system meets the sensitivity speciication in all cases (the solid line is below the dashed
line δ3) from 0 to 0.1 rad/s.
The analysis of the robust wind disturbance rejection in the frequency domain is
shown in Figure CS2.15c. The dashed line is the speciication δk1(ω) deined in Equation
CS2.45. The solid line represents the worst case of all the possible functions p31/(1 + p32G)
at each frequency due to the model uncertainty. Again, the control system meets the
speciication for all the plants within the uncertainty (the solid line is below the dashed
line δk1).
The fore-aft tower vibration analysis in the frequency domain is shown in Figure CS2.15d.
It represents the attenuation of the effect of wind disturbances v1(s) over the tower move-
ment yt(s) through p12(s) × G(s) × p31(s) and the control loop—see also Figure CS2.5. The
dashed line is the speciication δk2(ω) deined in Equation CS2.49. The solid line represents
the worst case of all possible functions p12p31G/(1 + p32G) at each frequency due to the
model uncertainty. Once more, the control system meets the speciication in all cases (solid
line is below dashed line δk2).
The lap-wise blade vibration analysis in the frequency domain is shown in
Figure CS2.15e. It represents the attenuation of the effect of wind disturbances v1(s) over
the blades movement yb(s) through p22(s) × G(s) × p31(s) and the control loop—see Figure
CS2.5. The dashed line is the speciication δk3(ω) deined in Equation CS2.50. The solid line
represents the worst case of all possible functions p22p31G/(1 + p32G) at each frequency due
to the model uncertainty. The control system meets the speciication in all cases (solid line
is below dashed line δk3).
The time-domain analysis of the reference tracking speciication is presented in
Figure CS2.16a. The igure shows the limits δ6-up(ω) and δ6-lo(ω)—see Equations CS2.48
and CS2.47, and the time responses of the rotor velocity Ωr(t) to a unitary step reference
r(t). We simulate 100 cases of the closed-loop transfer function p32GF/(1 + p32G)—see
Figure CS2.5, with H(s) = 1 and the electrical torque constant at Tgd = Tgdmax = 4,006,111 Nm,
and Figure CS2.19, Region 3, above-rated. The results are given in per unit, the unit being
Ωr_nom = 1.2671 rad/s = 12.1 rpm. The control system meets the speciication (is between
the upper and lower limits) in all cases.
Additionally, the time-domain analysis of the wind disturbance rejection is shown in
Figure CS2.16b. The igure presents the time responses of the rotor velocity Ωr(t) to a uni-
tary step wind speed disturbance v1(t). We simulate 100 cases within the uncertainty of
p31/(1 + p32G), also with H(s) = 1 and Tgd = Tgdmax = 4,006,111 Nm. Now, the zero level is the
non-perturbed Ωr_nom = 1.2671 rad/s = 12.1 rpm. The control system achieves a good dis-
turbance rejection in all cases.
(a) QFTCT (b) QFTCT

5
5
0 0
Ws –5
–5 Worst cases of –10 δ3
“p32G/(1+p32G)”
Magnitude (dB)
Magnitude (dB)
–15
–10
–20
50 Worst cases of
–15 –25 “1/(1+p32G)”
0.005
0.001 0.01 0.1 1 3 100 –30
0.001
–20 0.005 0.05 0.5 –35
2.04 5 0.01 0.1
–40
–25 4.19 10 0.05
–45
10–3 10–2 10–1 100 101 102 10–3 10–2 10–1 100
(c) QFTCT (d) QFTCT
0 0
δ k1 δ k2
–10
–10
Worst cases of
“p31/(1+p32G)” –20
–20 –30
Magnitude (dB)
Magnitude (dB)
–40
–30
–50
Worst cases of
–40 –60
“p12p31G/(1+p32G)”
–70
–50
–80
–60 –90
0.005 0.05 2.04 4.19
–100
–70 0.001 0.01 0.1
10–3 10–2 10–1 100 101 10–1 100 101
QFTCT
(e)
0
–5
δ k3
–10
–15
Magnitude (dB)
–20
–25
–30

“p22 p31G/(1+p32G)”
–40
–45 2.04 4.19
–50
100 101
Frequency (rad/s)
FIGURE CS2.15
Frequency-domain analysis: (a) stability, (b) sensitivity, (c) wind disturbance rejection, (d) fore-aft tower vibra-
tion, and (e) lap-wise blades vibration.
(a) QFTCT (b) QFTCT

1.2 0.035
δ6-up
0.03
1
0.025
0.8 0.02
δ6-lo
0.015
0.6
0.01
0.4 0.005
0
0.2
–0.005
0 –0.01
0 5 10 15 20 25 30 35 40 45 50 0 10 20 30 40 50 60 70
Time (s) Time (s)
FIGURE CS2.16
Time domain. (a) Reference tracking Ωr(s) = p32GF/(1 + p32G)r(s) and (b) wind disturbance rejection Ωr(s) =
p31/(1 + p32G)v1(s).
Additionally, Figure CS2.17 shows the tower fore-aft oscillation when a unitary step
wind velocity input v1(s) is introduced into the system. The disturbance v1(s) enters through
p31(s), travels the pitch control loop until βd(s), and inally goes through p12(s) to yt(s)—see
the block diagram in Figure CS2.5. We study two cases. The irst one, represented in
Figure CS2.17a, uses the complete controller described in Equation CS2.51. The second
case, represented in Figure CS2.17b, uses this controller without the notch ilter.
(a) QFTCT (b) QFTCT

0.09 0.09
0.08 0.08
0.07 0.07
0.06 0.06
0.05 0.05
0.04 0.04
0.03 0.03
0.02 0.02
0.01 0.01
0 0
0 50 100 150 0 50 100 150
Time (s) Time (s)
FIGURE CS2.17
Tower fore-aft oscillation, yt(t). Simulation of 100 plants. (a) G(s) with the notch ilter—see Equation CS2.51 and
(b) G(s) without the notch ilter.
(a) QFTCT (b) QFTCT

0.08
Tower fore-aft natural freq. = 0.32 Hz = 0.895 Blade flap-wise natural freq. = 0.6664 Hz =
~20 peaks in 60 s ~13.3 peaks in 20 s
0.89
Case G without notch filter
0.078
Case G without notch filter 0.885
0.88
0.076
0.875
0.074 0.87
0.865
0.072 0.86
Case G with notch filter 0.855

0.07 Case G with notch filter
0.85
40 50 60 70 80 90 100 50 52 54 56 58 60 62 64 66 68 70
Time (s) Time (s)
FIGURE CS2.18
Zooming in. (a) Tower fore-aft oscillation yt and (b) blade lap-wise oscillation y b. G(s) with notch ilter (solid
line) and without (dashed line).
The tower oscillation is signiicantly reduced when we add the notch ilter in the feed-
back controller G(s), that is, the complete Equation CS2.51. Figure CS2.17a shows the yt(t)
results with the complete G(s), and Figure CS2.17b without the notch ilter. Both analyses
are performed for the same 100 cases within the plant uncertainty. The igures have the
same scale.
Figure CS2.18a zooms in one case of the simulations of Figure CS2.17a and b. As we
can see, the frequency of the oscillation in the igure (20 peaks in 60 s) is the natural fre-
quency of the tower fore-aft oscillation: 0.320 Hz. The results of yt(t) using the complete
controller G(s) are represented by the solid line—Equation CS2.51, and the results of G(s)
without the notch ilter by the dashed line. As anticipated, the complete controller G(s)
with the notch ilter obtains a much better peak to peak reduction of the tower fore-aft
oscillation.
Finally, Figure CS2.18b presents the attenuation of the blade lap-wise vibration yb(t) at
its resonance frequency of 0.6664 Hz (13.3 peaks in 20 s), also with the controller G(s) with
and without the notch ilter. In this case, the step input disturbance v1(s) enters through
p31(s), travels the pitch control loop until βd(s), and goes through p22(s) to yb(s)—see also the
block diagram in Figure CS2.5. As in the tower vibration case, the complete controller
G(s) with the notch ilter obtains a better peak to peak reduction of the blade lap-wise
oscillation.
CS2.7 Extension to Higher Wind Velocities

This case study has designed the wind turbine pitch control system for the irst part of the
above-rated region, also called Region 3—see Figure CS2.19, range of wind speed v1 from
v1 = vr ∼ 12 m/s to v1 = 17 m/s.
Rated
Power (P)
power
Below rated—power optimization Above rated—power limitation

Pr
Region 4
This case study Extended mode:
Region 1 12 < v1 < 17 m/s load limitation,
partial power
Torque control: Region 3
maximum Region 2
aerodynamic Pitch control:
Transition : stability,
efficiency
good disturbance
efficiency, rejection
smooth
transients
vcut-in v12 vr v34 vcut-off Wind

speed (v1)
FIGURE CS2.19
Wind turbine power curve: Active power P versus wind speed v1.
The proposed controller G(s) can be easily extended to the entire Region 3, from vr to v34
in Figure CS2.19. As the aerodynamics changes signiicantly with the wind velocity, the
parameters KTΩ, KTV, and KTβ also change considerably—see Figure CS2.3 and Equations
CS2.20 through CS2.26. As a result, the gains and one pole of the denominators of p31(s),
p32(s), and p33(s) change with the wind speed—see Equations CS2.38 and CS2.39. A quan-
titative analysis of these variations is in the author’s book, Wind Energy Systems: Control
Engineering Design, Chapter 12, Reference 7.
To deal with this problem, the controller G(s), Equation CS2.51, had to modify its gain
and zero in real time, as a function of the desired pitch angle βd, so that
k = –35× f1(βd ) (CS2.53)
z1 = 0.11× f 2 (βd ) (CS2.54)
As the magnitude of the gain of the plant Ωr(s)/βd(s) = p32(s) increases with the wind
speed increase, the correcting function f1(βd) typically decreases with the pitch angle βd to
compensate this effect. Also, as one pole of the plant p32(s) becomes faster with the wind
speed increase, the correcting function f2(βd) typically increases with the pitch angle to
make the controller zero faster as well.
The design, stability analysis, and validation of this extended control strategy can be
easily done according to the methodology presented in Chapters 6 and 7.
CS2.8 Summary
This case study has successfully designed a robust QFT control system for a variable-speed
pitch-controlled gearless wind turbine. The design considers the model nonlinearities
and uncertainty introduced by the aerodynamics. It accomplishes ive simultaneous con-

trol objectives:
• Stability
• Tracking and regulation of the rotor speed by pitching the angle of the blades
• Rejection of unpredictable wind disturbances
• Minimization of the tower fore-aft oscillations and
• Attenuation of the blades lap-wise vibrations
Case Study 3: Wastewater
Treatment Plant Control
CS3.1 Description
Human activity is seriously increasing the amount of nitrogen and phosphorus in the
environment. When the concentration of nitrogen or phosphorus is too high, the water
becomes polluted. This affects streams, rivers, lakes, and ground water, often resulting
in human health issues, environmental destruction, and economic impact. Fortunately,
the environmental policies and standards on water pollution have become increasingly
stringent during the last few decades. Wastewater treatment plants (WWTP) with activated
sludge processes (ASP) play a vital role in removing harmful organic matter, nitrogen, and
phosphorus from domestic and industrial wastewater.
The high cost of WWTPs and the stringent water quality standards justify the effort
to design advanced control strategies to manage ASP with maximum eficiency and
minimum cost. The large variety of biological mechanisms and processes involved, their
nonlinearity and multi-input multi-output (MIMO) characteristics, and the uncertainties
in the composition and low of the inluent challenge the design of control solutions.
This Case Study CS3 presents the design of a MIMO QFT robust control system for
an ASP-WWTP that simultaneously reduces the concentration of ammonia (NH4) and
nitrates (NO3) in the plant efluent to meet the environmental standards. As an example,
Figure CS3.1 shows a picture of the Crispijana-Vitoria plant, a municipal water treatment
plant for nitrogen removal, able to deal with a 5000 m3/h inlow, for which the author with
the CEIT Research Centre designed an advanced robust control system in the 90s.192,208,212
In the following sections, we model and analyze the dynamics of an ASP-WWTP,
analyze the MIMO characteristics, propose a set of robust stability and performance
speciications, design SISO and MIMO control solutions, and validate them in both the
frequency and time domains.
CS3.2 Plant Model

Figure CS3.2 shows the plant layout of a municipal ASP-WWTP that reduces the
concentration of nitrogen (NH4 and NO3) and phosphorus (PO4) in water. In this case study,
we propose the design of a MIMO controller for simultaneous removal of NH4 and NO3.
The inluent low (QIN) is the wastewater coming from the city, typically with a high
concentration of nitrogen, that is NH4 and NO3. The plant objective is to reduce these con-
centrations in the efluent low (QOUT). The WWTP hydraulic coniguration is composed of
343
344 Case Study 3: Wastewater Treatment Plant Control
FIGURE CS3.1
WWTP, Crispijana, Spain. Inlow 5000 m3/h.
4NO3– + 5CH2O + 4H+→ 2N2 + 5CO2 + 7H2O

XMeOH NH4+ + 2O2 → NO3– + H2O + 2H+
QIN
N2 gas QRI
QOUT
Anaerobic Anoxic Aerobic Settler
Tank Tank Tank
QW
u1(s) = KLa
u2(s) = Rext
FIGURE CS3.2
ASP-WWTP coniguration.
three biological tanks in series (Anaerobic, Anoxic, and Aerobic) and one settler for efluent
clariication and sludge thickening—see Figure CS3.2.
The purpose of the anaerobic tank (irst tank) is to promote phosphorous removal
through growth of PAO bacteria, which accumulate phosphorus in approximately 35%
of their weight, compared with regular bacteria which accumulate only 2%. Additionally,
an external feed of ferric hydroxide (X MeOH) is provided to ensure eficient phosphorus
removal. In this way, the phosphates not biologically removed by the PAO bacteria are
coagulated through a chemical reaction.
In the anoxic tank (second tank), heterotrophic bacteria are responsible for the denitrii-
cation. These bacteria biodegrade the organic matter, which at the same time turns NO3
recycled from the aerobic tank (QRI) into nitrogen (N2) gas. This reaction takes place in
an anaerobic environment where the bacteria responsible for denitriication respire with
nitrate instead of oxygen. The process, called denitriication, can be summarized by the
formula:
4NO−3 + 5CH 2O + 4H+ → 2N 2 + 5CO 2 + 7H 2O (CS3.1)
NO3 that enters the previous expression is both coming from the inluent QRI and also
the product of a nitriication process taking place in the aerobic tank (third tank) by means
of nitrifying bacteria and the injection of air (u1 = KLa). In other words, NH4 is oxidized
Case Study 3: Wastewater Treatment Plant Control 345
to NO3 with an aeration system in the aerobic tank. The process, called nitriication, can be
summarized by the formula:
NH+4 + 2O 2 → NO−3 + H 2O + 2H+ (CS3.2)
Additionally, in the settler the activated sludge is thickened, so that the clariied
supernatant overlows into the efluent. At the same time, the activated sludge is recycled
to the anaerobic tank (u2 = Rext) to maintain a high biomass concentration in the reactors.
As a result, NH4 and NO3 that enter into the WWTP are removed from the water using
two biological processes: nitriication in the 3rd tank and denitriication in the 2nd tank,
converting them inally into N2 gas. The case is a 2 × 2 MIMO plant, being the plant
outputs (sensors) and actuators respectively:
• y1(s) = NH4, or concentration of ammonia in the efluent QOUT

• y2(s) = NO3, or concentration of nitrates in the efluent QOUT
• u1(s) = KLa, or air low injected in the aerobic tank
• u2(s) = Rext, or external sludge recirculation low, from settler to anaerobic tank
Other variables of the plant are the internal recirculation low QRI, the waste sludge low
QW and the ferric hydroxide added in anaerobic tank X MeOH. In this problem, we consider
these three variables working at a ixed rate.
We use the International Water Association (IWA) ASM2d nonlinear model for our
ASP-WWTP. We linearize the system around the working points, take the effect of the
nonlinearities as parametric uncertainty, and identify and calibrate a resulting 2 × 2
MIMO model, as shown in Equation CS3.3,212
 NH 4 (s)  p ( s) pw12 (s)  K L a(s)

  = Pw (s) =  w11   (CS3.3)
 NO 3 (s)  pw 21(s) pw 22 (s)  Rext (s)
  
where
k11 k12 ((s/z12.1 ) + 1)((s/z12.2 ) + 1)

pw11(s) = ; pw12 (s) =
 s  ( 12 ) + 1) (s/ωn12 )2 + (2ζ12 /ωn12 )s + 1
( s/ a
 + 1
 a11 
k 21 k 22
pw 21(s) = ; pw 22 (s) =
 s   s 
 + 1  + 1
 a21   a22 
and where the parameters are
• k11 ∈ [−0.045, −0.035]; a11 ∈ [6.25 × 10−5, 8.25 × 10−5].

• k12 = −6.239 × 10−6; z12.1 = 7.534 × 10−4; z12.2 = −3.17 × 10−5; a12 = 8.04 × 10−5; ωn12 =
4.58 × 10−4; ζ12 = 0.8493.
• k21 = 0.0464; a21 = 1.008 × 10−4.
• k22 ∈ [−2.2 × 10−5, −1.8 × 10−5]; a22 ∈ [1.57 × 10−4, 1.77 × 10−4].
Bode diagram
2
1
0
–1
Magnitude (dB)
–2
–3
–4
–5
Bandwidth (–3 dB)
–6
–7
–8
10–6 10–5 10–4

Frequency (rad/s)
FIGURE CS3.3
Bode diagram for daily-average low-pass ilter f LP(s), Equation CS3.4.
Additionally, environmental regulations and operation procedures require the control

of WWTPs following daily-average measurements of the output concentrations. For this
reason, we multiply each pwij(s) plant by a low-pass ilter f LP(s) with a unitary dc-gain and
a 2π/(24 × 60 × 60) = 7.27 × 10−5 rad/s bandwidth (as a day is 24 h, with 60 min/h and
60 s/min)—see Figure CS3.3. In this way, the daily-averaged plant models, now called
pij(s), are
 f LP (s)
P(s) = Pw (s)  ,
 f LP (s)
 
with pij (s) = pwij (s) f LP (s) ; i = 1, 2; j = 1, 2
(CS3.4)
1
with f LP (s) =
 2  
 s  2  
 1.1×10−4  +  1.1×10−4  s + 1
 
CS3.2.1 Frequencies
Figure CS3.4 shows the Bode diagram of the 2 × 2 MIMO plant P(s) = [pij(s)], i = 1, 2, j = 1, 2.
Based on this picture, we select the array of frequencies of interest, so that
ω = [1×10−7 5×10−7 1×10−6 5×10−6 1×10−5 2×10−5 5×10−5 1×10−4 2×10−4

(CS3.5)
5×10−4 1×10−3 2×10−3 5×10−3 ] rad/s

The nominal plant is selected at the operating point of the WWTP, which is described by
the mean values in the intervals of uncertainty in Equations CS3.3.
Bode diagram
From: In(1) From: In(2)
–50
To: out(1)
p12(s)
–100 p11(s)
–150
180
Magnitude (dB); phase (deg)
To: out(1)
–180
To: out(2)
–50 p22(s)
–100 p21(s)
–150
180
To: out(2)
–180
10–6 10–5 10–4 10–3 10–6 10–5 10–4 10–3

Frequency (rad/s)
FIGURE CS3.4
Bode diagram of P(s) = [pij(s)], i = 1,2, j = 1,2.

Before starting the design of the controllers, we conduct a brief preliminary analysis of the
dynamics of the WWTP. As just mentioned, Figure CS3.4 shows the Bode diagram of the
2 × 2 MIMO plant P(s) = [pij(s)], i = 1, 2, j = 1, 2, including also the parametric uncertainty.
As expected, the plant elements pij(s) have no integrators (system type 0) and include neg-
ative (k11, k12, and k22) and positive (k21) dc gains. Also, the phases indicate that there is a non-
minimum phase zero component. This is analyzed in-depth with the relative gain array.
Figure CS3.5 presents the relative gain analysis (RGA) over the frequencies of interest
of the nominal 2 × 2 MIMO plant P0(s). These graphical results are also presented numeri-
cally for very low frequency or ω = 0 rad/s in Equation CS3.6, and for very high frequency
or ω = ∞ rad/s in Equation CS3.7.
u1 u2

 0.7414 0.2586  y1 (CS3.6)
Λ(ω=0 ) =  
 0.2586 0.7414  y 2
  
u1 u2

−0.8898 1.8898   y1 (CS3.7)
Λ(ω=∞) =  
 1.8898 −0.8898  y 2

3 4
2
2
1
λ11
λ12
0 0
–1
–2
–2
–3 –4
10–6 10–5 10–4 10–3 10–2 10–6 10–5 10–4 10–3 10–2
4 3
2
2
1
λ21
λ22
0 0
–1
–2
–2
–4 –3
10–6 10–5 10–4 10–3 10–2 10–6 10–5 10–4 10–3 10–2
FIGURE CS3.5
RGA of P0(s) with frequency: λij, i = 1,2 and j = 1,2.
By looking at the matrix Λ(ω=0), Equation CS3.6, we select the input–output pairing for the
2 × 2 MIMO system as: (KLa, NH4) and (Rext, NO3), that is, (u1, y1) and (u2, y2).
Also, comparing the two matrices Λ(ω=0) and Λ(ω=∞), Equations CS3.6 and CS3.7, we see
that the signs of λ11, λ12, λ21, and λ22 change. As we discussed in Chapter 8, Section 8.3.3,
this means that the MIMO system has a non-minimum phase (nmp) zero. By inspection of
the transfer matrix, we ind the nmp zero at: z1 = –2 × 10 –4 rad/s, which is the frequency of
the discontinuity in the diagrams of Figure CS3.5. As we will see next, this nmp zero will
limit the bandwidth of the control system.

Taking into account the dynamics of the WWTP and the operator objectives and limita-
tions, we deine the following robust control speciications. They are the same for both
channels, and include stability, output disturbance rejection, and reference tracking
objectives as presented below:
pii ( jω) gii ( jω )

T1(jω ) = ≤ δ1(ω ) = Ws = 1.66, i = 1, 2
1 + pii ( jω ) gii ( jω )
(CS3.8)
ω ∈ [1×10−7 5×10−7 1×10−6 5×10−6 1×10−5 2×10−5 5×10−5
1×10−4 2×10−4 5×10−4 1×10−3 2×10−3 5×10−3 ] rad/s

1.2
0 δup(s)
δop(s) 1
–5
Magnitude (dB)
0.8
Amplitude
–10
0.6
–15
0.4 δlo(s)
–20
0.2
–25
0
10–6 10–5 10–4 10–3 0 0.5 1 1.5 2 2.5
Frequency (rad/s) Time (s) ×105
FIGURE CS3.6
Control speciications for all channels. (a) Disturbance rejection at the output of the plant: δ op. (b) Reference
tracking: δ up, δlo.

 s 
 
y(jω) 1  ad 
T3 (jω) = = ≤ δop (ω ) = ; ad = 2×10−5 ,
do (jω ) 1 + pii ( jω) gii ( jω)  s  (CS3.9)
  + 1
 ad 
i = 1, 2. ω ∈ [1×10−7 5×10−7 1×10−6 5×10−6 1×10−5 2×10−5 ] rad/s

pii ( jω ) gii ( jω )
δlo (ω ) ≤ f ii ( jω ) ≤ δup (ω ), i = 1, 2 (CS3.10)
1 + pii ( jω ) gii ( jω )
1
δlo (s) = ; aL = 2×10−5
 s   2 (CS3.11)
  + 1
 aL  
 
 s  
  + 1
 aU   1.25aU
δup (s) =   ; aU = 2×10−5 ; ζ = 0.8 ; ωn = (CS3.12)
 s 2  2ζ s   ζ
  +   
 ω   ω  + 1
 n n 
ω∈ [1×10−7 5×10−7 1×10−6 5×10−6 1×10−5 2×10−5 5×10−5 1×10−4 ] rad/s

This section presents two control solutions for the WWTP. The irst one is composed of
two independent SISO PI controllers, and is implemented according to the structure in
r1(s) – u1(s) y1(s)

i p11(s)
f11(s) g11(s)
p12(s)
p21(s)
r2(s) u2(s) y2(s)

i
f22(s) g22 (s) p22(s)
–
FIGURE CS3.7
SISO QFT control system for the WWTP.
Figure CS3.7. The second solution applies the MIMO QFT Method 2 presented in Chapter 8,
Sections 8.6 and 8.9, and is implemented according to the diagram of Figure CS3.14.
CS3.5.1 Independent SISO QFT Control

Controller Structure
Figure CS3.7 shows the SISO QFT controller structure for the WWTP. It is composed of
a 2 × 2 diagonal matrix Gi (s) =  g11
i i
(s) 0; 0 g 22 (s) and a 2 × 2 diagonal preilter matrix
F(s) = [f11(s) 0; 0 f22(s)].
Design of 2 × 2 Diagonal Matrix Gi(s) Controller and Preilter F(s)

The SISO controller, or 2 × 2 diagonal controller Gi, is composed of two independent
i i i
controllers, g11 (s) and g 22 (s). The element g11 (s) is designed by means of the standard
SISO QFT loop-shaping technique, for the plant p11(s), and for the control speciications
described in Equations CS3.8 through CS3.12. The QFT bounds and the loop shaping
i i
for g11 (s) are shown in Figure CS3.8. The expression found for the controller g11 (s) is a
PI so that
 1 
−0.0006  s + 1

 3 ×10−5

i
g11 (s) = (CS3.13)
s
i
The f11(s) preilter element is designed for the plant p11(s) and the controller g11 (s).
Figure CS3.9 shows the design. The expression found for the preilter f11(s) is
1 (CS3.14)
f11(s) =
 1 
 s + 1
 3.2×10−5 
Figure CS3.10 shows the analysis of the disturbance rejection at the output of the plant
and the reference tracking speciications. The analysis is made using the plant p11(s) with
i
uncertainty, the controller g11 (s) and the preilter f11(s). When considering no coupling
(p12 = p21 = 0), the irst channel of the SISO control system meets the speciications for all
the p11(s) plants within the uncertainty.
QFTCT
1e-7 rad/s
60
40
Open-loop gain (dB)
20
–20
L011(s) = p11(s) g11i(s)
–40
–60
–260 –240 –220 –200 –180 –160 –140 –120 –100 –80 –60
FIGURE CS3.8
i i
Loop shaping of controller g11 (s). L0_11(s) = p11(s) g11 (s).
QFTCT
5
δup
0
–5
–10
Magnitude (dB)
–15
p11 g11i f11/(1 + p11 g11i)
–20
–25
–30
δlo
–35
–40
10–7 10–6 10–5 10–4
Frequency (rad/s)
FIGURE CS3.9
i
Design of preilter f11(s) for p11(s) and g11 (s).
(a) QFTCT (b) QFTCT

10 1.2
δop δup
0 1
–10 0.8
Magnitude (dB)
Worst case at each frequency

–20 0.6
δlo
–30 0.4
0.2
–40
up to 2e-5 rad/s 0
–50
10–7 10–6 10–5 10–4 10–3 0 0.5 1 1.5 2
FIGURE CS3.10
i
Analysis of controller g11 (s) and preilter f11(s) for p11(s). (a) Disturbance rejection at plant output: δ op. (b) Reference
tracking: δ up, δ lo.
i
The element g 22 (s) is also designed by means of the standard SISO QFT loop-shaping
technique, for the plant p22(s), and for the control speciications described in Equations
i
CS3.8 through CS3.12. The QFT bounds and the loop shaping for g 22 (s) are shown in Figure
i
CS3.11. The expression found for the controller g 22 (s) is also a PI, so that
 1 
−1.5  s + 1
i

 4.5×10−5
 (CS3.15)
g 22 (s) =
s
QFTCT
80
60
1e-7 rad/s
Open-loop gain (dB)
40
20
–20
L022(s) = p22(s) g22i(s)
–40
–260 –240 –220 –200 –180 –160 –140 –120 –100 –80 –60
FIGURE CS3.11
i i
Loop shaping of controller g 22 (s) . L0_22(s) = p22(s) g 22 (s) .
QFTCT
5
δup
0
–5
–10
Magnitude (dB)
–15
p22 g22i f22/(1 + p22 g22i)
–20
–25
–30 δlo
–35
–40
10–7 10–6 10–5 10–4
Frequency (rad/s)
FIGURE CS3.12
i
Design of preilter f22(s) for p22(s) and g 22 (s) .
i
The f22(s) preilter element is designed for the plant p22(s) and the controller g 22 (s).
Figure CS3.12 shows the design. The expression for the preilter f22(s) is
1 (CS3.16)
f 22 (s) =
 1 
 s + 1
 3.2×10−5 
and the reference tracking speciications, both for the plant p22(s) with the controller
i
g 22 (s) and the preilter f22(s). When considering no coupling (p12 = p21 = 0), the second
channel of the SISO control system meets the speciications for all the p22(s) plants within
the uncertainty.
CS3.5.2 MIMO QFT Control, Method 2

In the next subsections, we follow the low chart and steps presented in Figure 8.31 for the
design of a MIMO QFT controller according to Method 2. See also Sections 8.6 and 8.9 for
more details.
Step A: Controller Structure

Figure CS3.14 shows the MIMO QFT Method-2 controller structure for the WWTP. It is
composed of a 2 × 2 full matrix Gα, a 2 × 2 diagonal matrix Gβ and a 2 × 2 diagonal pre-
ilter matrix F.
(a) QFTCT (b) QFTCT

10 1.2
δop δup
0 1
–10 0.8
Magnitude (dB)

–20 0.6
δlo
–30 0.4
0.2
–40
Up to 2e-5 rad/s 0
–50
10–7 10–6 10–5 10–4 10–3 0 0.5 1 1.5 2
FIGURE CS3.13
i
Analysis of controller g 22 (s) and preilter f 22(s) for p22(s). (a) Disturbance rejection at plant output: δ op.
(b) Reference tracking: δ up, δlo.
r1(s) – β α
u1(s) y1(s)
f11(s) g11(s) g11 (s) p11(s)
α
g12(s) p12(s)
α
g21(s) p21(s)
r2(s) β
u2(s) y2(s)
f22(s) g22(s) α
g22(s) p22(s)
–
FIGURE CS3.14
MIMO QFT control system for the WWTP. Method 2.
Step B: Design of 2 × 2 Full Matrix Gα Controller

As discussed in Section 8.6.2, the fully populated matrix controller G is composed of two
matrices: G = GαGβ , so that
 g11 α
(s) α
g12 (s)  g11
β
(s) 0 
G = GαGβ =  α   (CS3.17)
 g 21(s)
α
g 22 (s)  0 g (s)
β
22
The main objective of the pre-compensator Gα is to diagonalize the plant P as much as

possible. As discussed in Sections 8.6 and 8.9, the expression used to calculate Gα is based on
 g11α
(s) α
(s)  * *
(s)  p11(s) 0 
Gα (s) =  α
g12
= P −1
( s) Pdiag ( s) =
 p11(s) p12
  (CS3.18)
 21(s)
g g 22 (s)
α  p21
*
 (s) p22 (s)  0
*
p22 (s)
where the plant matrix P, the corresponding inverse P–1, and the diagonal Pdiag are selected
so that the expression of the extended matrix Px = P Gα presents the closest form to a
diagonal matrix, nulling the off-diagonal terms as much as possible.
Pin v Pdiag (1:end).tfm(1,1) and g11-α
Magnitude (dB)
–2
–4
–6
–8 α
g11
–10
180
Phase (deg)
90
α
g11
0
–90
10–5 10–4
Frequency (rad/s)
FIGURE CS3.15
α *
Controller element g11 (s) and p11 (s)× p11 (s).
* * *
The Bode diagrams for the four expressions p11 (s)× p11(s), p12 (s)× p22 (s), p21 (s)× p11(s),
*
and p22 (s)× p22 (s) , including all the model uncertainty, are shown in Figures CS3.15
α α
through CS3.18, respectively. The controller elements g11 (s), g12 (s), gα21(s), and gα22 (s) are
calculated as the transfer function that matches the mean value of the respective Bode
diagram at low frequencies and then ilters out the dynamics before the nmp zero at
−2 × 10 –4 rad/s—see also Equations CS3.19 through CS3.22, respectively.
–65
Magnitude (dB)
–70
α
g12
–75
180
Phase (deg)
135
α
g12
90
10–5 10–4 10–3
Frequency (rad/s)
FIGURE CS3.16
α ∗
Controller element g12 (s) and p12 (s)× p22 (s).
62
Magnitude (dB)
60
58
α
g12
56
180
Phase (deg)
90
α
g12
0
–90
10–5 10–4 10–3
Frequency (rad/s)
FIGURE CS3.17
*
Controller element gα21 (s) and p21 (s)× p11 (s) .

0
Magnitude (dB)
–2
–4
–6
α
–8 g22
180
Phase (deg)
90
α
g22
0
–90
10–5 10–4 10–3
Frequency (rad/s)
FIGURE CS3.18
*
Controller element gα22 (s) and p22 (s)× p22 (s) .
α 0.9439
g11 (s) = (CS3.19)
7143s + 1
α −0.0001858
g12 (s) = (CS3.20)
7143s + 1
1382
gα21(s) = (CS3.21)
7143s + 1
0.9439
gα22 (s) = (CS3.22)
7143s + 1
β
Step C.1.1: Design of the Diagonal Controller g11 ( s)
β
The element g11(s) is designed by means of the standard SISO QFT loop-shaping technique,
for the inverse of the extended equivalent plant qx11(s) = [p11x*e]1−1, and for the control
speciications described in Equations (CS3.8 through CS3.12). According to the iterative
expression Equation 8.97, the extended equivalent plant is
x*e x*
p11 (s) = p11 (s) (CS3.23)
1
β
and the plant to be controlled by g11 (s)
1 (CS3.24)
qx11(s) = x*e
p11 (s)
1
The MATLAB code that calculates the plant qx11(s) for all the cases within the paramet-
ric uncertainty is similar to the one included in Appendix 8, Example 8.1, and Section 8.9
β
(QFT MIMO Method 2). The QFT bounds and the loop shaping for g11 (s) are shown in
β
Figure CS3.19. The expression found for the controller g11(s) is
 1  1  1  1 
−0.0008  s + 1 s + 1 s + 1 s + 1
 5×10−5  9×10−5  0.00014  0.0003 
β
g (s) =
11
(CS3.25)
2
 1  1   1 
s  s + 1 s + 1  s + 1
 0.0009  0.04   0.2 
The design also fulils the two stability conditions:

β
a. Lx0_11(s) = qx11(s) g11 (s) satisies the Nyquist encirclement condition and
β
b. There are no RHP pole-zero cancellations between qx11(s) and g11 (s).
This is checked following to the methodology presented in Chapter 3, Section 3.4,

with the QFT Control Toolbox (QFTCT), Controller design window, File, Check stability
option.
QFTCT
90
80
70
1e-7 rad/s
60
Open-loop gain (dB)
50
40
Lx011(s) = qx11(s) g11b(s)
30
20
10
0
–10
–240 –220 –200 –180 –160 –140 –120 –100 –80 –60
FIGURE CS3.19
β β
Loop shaping of controller g11 (s). L x0_11(s) = qx11(s) g11 (s).
Step C.1.2: Design of the Diagonal Preilter f11(s)

The f11(s) preilter element is designed for the equivalent plant qx11(s) and the diagonal
β
controller g11 (s) designed in the previous Step C.1.1. Figure CS3.20 shows the design.
The expression found for the preilter f11(s) is
1 (CS3.26)
f11(s) =
 1 
 s + 1
 3.2×10−5 
and the reference tracking speciications for the equivalent plant qx11(s) and with the
β
controller g11 (s) and the preilter f11(s). The control system meets the speciications with
qx11(s) in all cases within the uncertainty.
β
Step C.2.1: Design of the Diagonal Controller g 22 (s)
β
The element g 22 (s) is designed by means of the standard SISO QFT loop-shaping tech-
nique, for the inverse of the extended equivalent plant qx22(s) = [p22x*e]2−1, and for the control
speciications described in Equations CS3.8 through CS3.12. According to the iterative
expression Equation 8.97, the extended equivalent plant is
x* x*
x*e x*
p21 (s) p12 (s)
1 1
p (s) = p (s) −
22 22 x* β (CS3.27)
2 1 p11 (s) + g11 (s)
1
β
and the plant to be controlled by g 22 (s)
QFTCT
5
–5
δup
–10
Magnitude (dB)
–15
–20
–25
qx11 g11b f11/(1 + qx11 g11b)
–30
–35 δlo
–40
–45
10–7 10–6 10–5 10–4 10–3
Frequency (rad/s)
FIGURE CS3.20
β
Design of preilter f11(s) for qx11(s) and g11 (s).
(a) QFTCT (b) QFTCT

10
1.2
δup
0 δop
1
–10
Magnitude (dB)
0.8
–20 0.6 δlo
–30 0.4
0.2
–40
Up to 2e-5 rad/s 0
–50
10–7 10–6 10–5 10–4 10–3 0 0.5 1 1.5 2
FIGURE CS3.21
β
Analysis of controller g11 (s) and preilter f11(s) for qx11(s). (a) Disturbance rejection at plant output: δ op.
(b) Reference tracking: δ up, δ lo.
1
qx 22 (s) = x*e
(CS3.28)
p22 (s)
2
The MATLAB code that calculates the plant qx22(s) for all the cases within the paramet-
ric uncertainty is similar to the one included in Appendix 8, Example 8.1, and Section 8.9
β
(QFT MIMO Method 2). The QFT bounds and the loop shaping for g 22 (s) are shown in
β
Figure CS3.22. The expression for the controller g 22 (s) is
 1 3
−0.52  s + 1 
 5×10−5 
β
g 22 (s) = (CS3.29)
 1  1 2
s  s + 1 s + 1
 0.001  0.0013 
The design also fulils the two stability conditions:

β
a. Lx0_22(s) = qx22(s) g 22 (s) satisies the Nyquist encirclement condition, and
β
b. There are no RHP pole-zero cancellations between qx22(s) and g 22 (s).
This is checked following to the methodology presented in Chapter 3, Section 3.4, with
the QFT Control Toolbox (QFTCT), Controller design window, File, Check stability option.
Step C.2.2: Design of the Diagonal Preilter f22(s)

The f22(s) preilter element is designed for the equivalent plant qx22(s) and the diagonal
β
controller g 22 (s) designed in the previous Step C.2.1. Figure CS3.23 shows the design.
The expression found for the preilter f22(s) is
QFTCT
80
1e-7 rad/s
60
Open-loop gain (dB)
40
20
–20
Lx022(s) = qx22(s) g22b(s)
–40
–250 –200 –150 –100

FIGURE CS3.22
β β
Loop shaping of controller g 22 (s) . L x0_22(s) = qx22(s) g 22 (s) .
QFTCT
δup
0
–5
–10
Magnitude (dB)
–15
–20
–25 qx22 g22b f22/(1 + qx22 g22b)
–30
–35 δlo
–40
10–7 10–6 10–5 10–4 10–3

Frequency (rad/s)
FIGURE CS3.23
β
Design of preilter f22(s) for qx22(s) and g 22 (s) .
1 (CS3.30)
f 22 (s) =
 1 
 s + 1
 3.2×10−5 
and the reference tracking speciications for the equivalent plant qx22(s) and with the con-
β
troller g 22 (s) and the preilter f22(s). The control system meets the speciications with qx22(s)
in all cases within the uncertainty.
(a) Disturbance rejection at plant output: δop. (b) Reference tracking: δup, δlo.
Step D: Final Checks

D.1. The design also fulils the other additional two stability conditions:
• No Smith-McMillan pole-zero cancellations occur between P(s) and G(s), and

• No Smith-McMillan pole-zero cancellations occur in |P*(s) + G(s)|.
D.2. The inal system P(s)G(s) does not have any additional RHP transmission zero.
(a) QFTCT (b) QFTCT

10
1.2
δup
δop
0
1
–10 0.8
Magnitude (dB)
–20 0.6
δlo
–30 0.4
–40 0.2
Up to 2e-5 rad/s
–50 0
10–7 10–6 10–5 10–4 10–3 0 0.5 1 1.5 2
FIGURE CS3.24
β
Analysis of controller g 22 (s) and preilter f 22(s) for qx22(s). (a) Disturbance rejection at plant output: δ op. (b) Reference
tracking: δ up, δ lo.

This section presents the time-domain analysis of the controllers designed in this
case study. The analysis includes the simulation of the complete MIMO ASP-WWTP with
both the SISO solution (Section CS3.5.1) and the MIMO QFT solution (Section CS3.5.2).
Figure CS3.25 shows the time-domain simulation of the two outputs, y1(s) = NH4 and
y2(s) = NO3, of the MIMO nominal plant P0(s) = [p11(s) p12(s); p21(s) p22(s)], Equations CS3.3
and CS3.4.
We introduce two reference inputs, r1(s) and r2(s), and two external disturbances, d1(s) and
d2(s) in the system. The reference inputs start at a polluted water level and change to the
level required by the standards, that is NH4 r1(s) from 2 g/m3 to 1 g/m3 at 2 × 106 s and NO3
r2(s) from 8 g/m3 to 4 g/m3 at 3 × 106 s—see dashed lines in the igure.
The external disturbances introduce a signiicant step increase of the concentration of
NH4 and NO3, respectively, in the water, with d1(s) from 0 to 0.5 g/m3 at 4 × 106 s and d2(s)
from 0 to 2 g/m3 at 5 × 106 s.
The results with the SISO controllers are represented by the dotted lines. They use the
i i
control structure shown in Figure CS3.7, and the expressions for g11 (s), g 22 (s), f11(s), and
f22(s) in Equations CS3.13, CS3.15, CS3.14, and CS3.16, respectively.
The results with the MIMO QFT controller are represented by the solid lines. They use the
control structure shown in Figure CS3.14, and the expressions for G = G α Gβ in Equations
CS3.19 through CS3.22, CS3.25, and CS3.29, and for F = [f11(s) 0; 0 f22(s)] in Equations CS3.26
and CS3.30.
The independent SISO controllers achieve a good performance. The MIMO control-
ler improves the results, reducing signiicantly the effect of the coupling between the
control loops.
(a) 2.5
d1 (step)
2
MIMO control MIMO control
1.5
r1, y1
1
r1 (step)
0.5
SISO control SISO control
0
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
×106
(b) 9
SISO control
8
d2 (step)
7
MIMO control
r2, y2
6
SISO control
5
4
r2 (step) MIMO control
3
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Time (s) ×106
FIGURE CS3.25
MIMO WWTP. SISO controllers (dotted line), MIMO controller (solid line), references (dashed line).
(a) 1st channel NH4, (b) 2nd channel NO3.
CS3.7 Summary
This case study has designed a robust 2 × 2 MIMO QFT control system for a wastewa-
ter treatment plant (WWTP) with an activated sludge process (ASP) that simultaneously
reduces the concentration of ammonia (NH4) and nitrates (NO3) in the plant efluent. The
design considers the model uncertainty and MIMO loop interaction. It accomplishes four
simultaneous control objectives:
• Stability
• Reference tracking
• Rejection of unpredictable disturbances and
• Minimization of loop interaction
Case Study 4: Radio-Telescope Control
CS4.1 Description
Radio astronomy is a young scientiic ield. It was born by serendipity in 1931, when Karl
Jansky, a physicist and radio engineer at Bell Telephone Laboratories, built an antenna to
study the properties of 20.5 MHz radio waves for use in transatlantic telephone services.
While recording signals from all directions, Jansky found a unique radiation repeated
every 23 h and 56 min, which is the period of the Earth’s rotation relative to the stars
(sidereal day), instead of 24 h (solar day). Based on this unexpected observation, Jansky
concluded that the radio signals came from the Milky Way, and radio astronomy was
born.366
Since then, many radio telescopes have been built, and thousands of new astronomical
discoveries have changed our understanding of the universe. One of the most prominent
facilities is the Robert C. Byrd Green Bank Telescope, or GBT (WV). It is the world’s largest
single-dish fully steerable radio telescope, operating at meter to millimeter wavelengths
(0.1–116 GHz)—see Figure CS4.1. It has an enormous 100 × 110 m elliptical collecting area,
an unblocked aperture, a 148 m structure, and an excellent surface accuracy with over 2200
actuators.
The telescope is driven in the azimuth axis by 16 motors, each of 30 hp, coupled together
in four trucks. In the elevation axis, there are eight motors, each of 40 hp. All motors are
preloaded to remove backlash. The telescope has an off-axis geometric focus, with a feed
arm and a sub-relector, as shown in Figure CS4.2.
This case study presents the design of the velocity and position cascade control loops
for the azimuth axis of a radio telescope similar to GBT. Coincidentally, the author has
worked with National Radio Astronomy Observatory (NRAO) engineers on the design of
advanced control solutions for the azimuth, elevation, and sub-relector servo systems of
GBT for a number of years.
In the following sections, we model and analyze the dynamics of the radio telescope,
propose a set of stability and performance speciications, design the velocity and position
cascade control systems, improve the results with a nonlinear dynamic control (NDC)
strategy, and validate them in both frequency and time domains.
CS4.2 Plant Model

Figure CS4.2 shows the diagram for the azimuth axis of the radio telescope. The model
considers four degrees of freedom: the angular displacement of the dish θd(t), the angular
displacement of the central base of the structure θb(t), the angular displacement of each
365
366 Case Study 4: Radio-Telescope Control
FIGURE CS4.1
Green Bank Telescope (GBT), NRAO, West Virginia.
Sub-reflector
Feed arm
Radio-frequency waves
Wind Tw
Jd
Upper structure
θd Main dish
Ks Bs Truck 4
Truck 3 θb
rb
Lower structure Bb
Km Wheel
Jb Kb Jw
Motor
Truck 2 and Jm rw
Arm gearbox T Bm
m
θm θw Truck 1
Jt
Azimuth
track θt
FIGURE CS4.2
Radio-telescope diagram. Azimuth model description.
Case Study 4: Radio-Telescope Control 367
truck system θt(t), and the angular displacement of each motor θm(t). The main variables
and parameters of the system are the following:
Inputs
• Tw: aerodynamic torque applied by the wind on the dish and structure, [Nm]
• Tm equivalent torque applied by the four motors of each truck, [Nm]
Outputs
• θd: dish angular position, [rad]
• θb: base (lower structure) angular position, [rad]
• θt: truck angular position, [rad]
• θw: truck wheel angular position, θw = θt (rb/rw) [rad]
• θm: motor shaft angular position at the gearbox output, [rad]
Parameters
• N: number of truck systems. Each truck is composed of one arm and a drive-train
system (with one equivalent motor, gearbox, brake, shaft, and pinion), [-]
• rb: radius of the arm between the lower structure and each truck, [m]
• rw: radius of the equivalent truck wheel, [m]
• R: radius rate, R = rb/rw, [-]
• Jd: moment of inertia of dish, feed arm, and upper structure, at θd, [kg m2]
• Jb: moment of inertia of lower structure, at θb, [kg m2]
• Jt: moment of inertia of each truck, at θt, [kg m2]
• Jw: moment of inertia of each equivalent truck wheel, at θw, [kg m2]
• Jm: moment of inertia of each equivalent motor with its brake, gearbox, and shaft,
at θm, [kg m2]
• Ks: upper structure torsional stiffness coeficient, [Nm]
• Bs: upper structure torsional damping coeficient, [Nms]
• Kb: one arm torsional stiffness coeficient, [Nm]
• Bb: one arm torsional damping coeficient, [Nms]
• Km: one motor shaft torsional stiffness coeficient, [Nm]
• Bm: one motor shaft torsional damping coeficient, [Nms]
As in previous cases, we start with the general equation of motion for the mechanical
system, which is
M qɺɺ + C qɺ + Kq = Q (qɺ , q, u, t) (CS4.1)
where M, C, and K are the Mass, Damping, and Stiffness matrices, Q the matrix of inputs,
q the vector of generalized coordinates, and u the vector of inputs. Also, we consider the
general form of the Euler–Lagrange equation,
d  ∂L  ∂L ∂Dd
  − + = Qi , i = 1, 2,..., n and L = Ek − Ep (CS4.2)
eralized forces applied to each subsystem “i”, Ek and Ep the kinetic and potential energies,
The generalized coordinates for the radio telescope are the angular displacement of the
dish θd(t), the angular displacement of the central base of the structure θb(t), the angu-
lar displacement of each truck system θt(t), and the angular displacement of each motor
θm(t), which are, respectively: q1 = θd(t), q2 = θb(t), q3 = θt(t), and q4 = θm(t)—see Figure CS4.2.
The azimuth-axis displacement contains the kinetic, potential, and dissipative functions
described by Equations CS4.3 through CS4.5. Notice that we consider linear dissipators,
which means that the Dd expression or dissipator co-content equals one-half of the power
being absorbed by the dissipators.236
1 ɺ2 1 ɺ2 1 1 1
Ek = J d θd + J b θb + N Jt θɺt2 + N J w (Rθɺt )2 + N J m θɺm2 (CS4.3)
2 2 2 2 2
1 1 1
Ep = K s (θd − θb )2 + N K b (θb − θt )2 + N K m (R θt − θm )2 (CS4.4)
2 2 2
1 1 1
Dd = Bs (θɺd − θɺb )2 + N Bb (θɺb − θɺt )2 + N Bm (R θɺt − θɺm )2 (CS4.5)
2 2 2
The generalized forces applied to the radio telescope are the aerodynamic torque applied
by the wind on the dish and upper structure Tw(t), and the torque applied by the N trucks,
N Tm(t). Now, using Equations CS4.3 through CS4.5, the terms for the Euler–Lagrange
equation are
 ∂Ek 
 ɺ 
 ∂θd 
   ɺɺ 
 ∂Ek   J d 0 0 0   θd 
   
d  ∂L  d  ∂θɺb   0 Jb 0 0   θɺɺb 
  =  =    = M qɺɺ (CS4.6)
dt  ∂qɺ i  dt  ∂Ek   0 0 N ( Jt + R2 J w ) 0   θɺɺt 
 ɺ    
 ∂θt   0 0 0 N J m  θɺɺ 
    m 
 ∂Ek 
 ɺ 
 ∂θm 
 ∂Ep 
 
 ∂θd 
 
 ∂Ep   K s −K s 0 0   θd 
    
∂L ∂Ep  ∂θb  −K s Ks + N Kb −N K b 0   θb 
  =K q
− = = = (CS4.7)
∂qi ∂qi  ∂Ep   0 −N K b N (K b + R2 K m ) −N R K m   θt 
  
 ∂θt   0 0 −N R K m N K m  θm 
  
 ∂Ep 
 
 ∂θm 
 ∂D 
 ɺd 
 ∂θd 
    θɺd 
 ∂Dd   Bs −Bs 0 0
 
    θɺb 
∂Dd  ∂θɺb  −Bs Bs + N Bb −N Bb 0
   = C qɺ
= = (CS4.8)
∂qɺ i  ∂Dd   0 −N Bb N (Bb + R2 Bm ) −N R Bm   θɺt 
 ɺ    
 ∂θt   0 0 −N R Bm N Bm  θɺ 
    m 
 ∂Dd 
 ɺ 
 ∂θm 
1 0
 
0 0  Tw 
Q =   =Ru (CS4.9)
0 0  Tm 
0 N 

and using Equations CS4.6 through CS4.9, we ind a state space description of the system
( xɺ = Ax + Bu;y = Cx ) with the expressions
−M −1C −M −1K   M −1R 

xɺ =   x +  u

 I 0 
  0  (CS4.11)
y=Cx
with I = [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1], 0 = [0 0 0 0; 0 0 0 0; 0 0 0 ; 0 0 0 0], and,
 Bs Bs Ks Ks 
− 0 0 − 0 0 
 Jd Jd Jd Jd 
 
 Bs −(Bs + N Bb ) N Bb K s −(K s + N K b ) N Kb 
 0 0 
 Jb Jb Jb Jb Jb Jb 
 
 Bb 2
−(Bb + R Bm ) RBm Kb K b + R K m ) RK m 
−(K 2
 0 0
 J w R2 + Jt J w R2 + Jt J w R2 + Jt J w R2 + Jt J w R2 + Jt J w R2 + Jt 
A=
 0 R Bm −Bm R Km −K m 
 0 0 0 
 Jm Jm Jm Jm 
 1 0 0 0 0 0 0 0 
 
 0 1 0 0 0 0 0 0 
 
 
 0 0 1 0 0 0 0 0 
 
 0 0 0 1 0 0 0 0 
(CS4.12)
1 
 0
 Jd 
 
0 0
 
0 0 

 1 
B =  0 (CS4.13)
 J m 
0 0 

0 0 

 
0 0
 
 0 0 
0 0 0 1 0 0 0 0
C=  (CS4.14)
0 0 0 0 1 0 0 0

 qɺ 
x =   = [θɺd θɺb θɺt θɺm θd θb θt θm ]T (CS4.15)
 q
 
u = [Tw Tm ]T (CS4.16)
y = [θɺm θd ]T (CS4.17)
Now, applying P(s) = C(sI − A)−1 B for y(s) = P(s)u(s), the transfer matrix from the two
independent inputs to the two outputs of interest is
θɺm (s) Tw (s)

   
 θ (s)  = P2×2 (s) Tm (s) (CS4.18)
 d   
where the velocity of the motors dθm(s)/dt = θ’m(s) and the dish angular position θd(s), from
both the wind torque Tw(s) and the motor torque Tm(s), are
θɺm (s) = p11(s) Tw (s) + p12 (s) Tm (s) (CS4.19)
θd (s) = p21(s) Tw (s) + p22 (s) Tm (s) (CS4.20)
Figure CS4.3 graphically shows the motor speed θ’m(s), position θm(s), and dish dis-
placement θd(s) as a function of the motor and wind torques, Tm(s) and Tw(s), according to
Equations CS4.19 and CS4.20. In addition, the igure also shows the cascade control servo
system, with the velocity controller Gv(s) and the position controllers, Gp(s) and Fp(s), all to
be designed in the next sections.
p21(s)
+
θd(s)
p22(s)
+
θ dRef (s)×R =
θ mRef (s) θ ′mRef (s) Tw(s) p11(s)
+
Tm(s) θ ′m(s) 1 θm(s)
Fp(s) Gp(s) Gν (s) M(s) p12(s)
+ + + s
– Position – Velocity
control control +
Hν (s)
+
+ nν (s)
Hp(s)
+
np(s)
FIGURE CS4.3
Block diagram representation of the radio-telescope dynamics and velocity/position control servo systems.
Parameters and Parametric Uncertainty

The parameters for the model of the radio telescope considered in this case study are
shown in Table CS4.1. They are based on the results published in Reference 365 about the
NRAO-GBT. The parameters with uncertainty are the inertia of the dish and upper struc-
ture, which changes with the variation of the elevation-axis angle, and the three damping
coeficients, so that
J d ∈ [ 86.7 , 117.3 ]×106 kg m 2 (CS4.21)
Bs ∈  54.8, 74.2 ×106 Ns/m
Bb ∈ [ 2.66, 3.59]×10 5 Ns / m
Bm ∈ [ 233.67 , 316.14 ] Ns/m
TABLE CS4.1
Radio Telescope Parameters (Nominal Case)
Radio-telescope weight [kg] 3.85×106 Ld (dish dimensions) [m] 100×110
Sensitivity radio freq. [GHz] 0.1 to 116 h (total height) [m] 148
N—number of azimuth tracks 4 R—radius ratio rb / rw 50
Jd [kg m2]—dish, f.a., up.struct. 102×106 Jb [kg m2]—lower structure 34.3×106
2
Jw [kg m ]—equiv wheel, truck 426 2
Jm [kg m ]—equiv. motor, truck 517.56
Jt [kg m2]—equiv truck system 3.43×106 M(s)—motor dynamics 1
Ks [N/m]—upper structure 1.9731×109 Bs [Ns/m]—upper structure 4.4862×106
Kb [N/m]—lower structure 94.9×10 6 Bb [Ns/m]—lower structure 3.1249×105
Km [N/m]—motor shaft 1.6851×10 6 Bm [Ns/m]—motor shaft 274.89
Hv—feedback dynamics 1 Hp—feedback dynamics 1
Bode diagram
–50
Magnitude (dB)
–100
–150 p12-0(s)
p11-0(s)
–200
p21-0(s)
–250 p22-0(s)
180
p12-0(s)
0
Phase (deg)
–180
–360 p11-0(s)
p21-0(s)
–540
p22-0(s)
–720
10–1 100 101 102

Frequency (rad/s)
FIGURE CS4.4
Bode diagram, nominal plants: p11-0(s), p12-0(s), p21-0(s), and p22-0(s).
Frequencies
Figures CS4.4 through CS4.6 show the Bode diagrams for the transfer functions p11(s), p12(s),
p21(s), and p22(s). We select an array of frequencies of interest based on this information—see
Equation CS4.22. Notice that we include frequencies that represent the resonances (peaks)
that appear in the diagrams, and which represent the lexible modes of the radio tele-
scope for upper structure and feed-arm torsion, lower structure and truck-arm torsion,
and motor shaft torsion.
ω = [0.01 0.05 0.1 0.5 1 1.5 2 3 4 5 10 50 60 65 70 100 500] rad/s (CS4.22)
Bode diagram Nichols chart

20 1 dB0.5 dB –1 dB
3 dB –3 dB
Magnitude (dB)
–50
p12-0(s) 0 6 dB –6 dB
–12 dB
–20 –20 dB
–100 p12-0(s)
Open-loop gain (dB)
–40 –40 dB
–150 –60 –60 dB
–80 –80 dB
0 –100 –100 dB
Phase (deg)
–45 –120 –120 dB

–90 –140 –140 dB
–135 –160 dB
–160
–180 –180 dB
–180
10–1 100 101 102 103 –225 –180 –135 –90 –45 0
FIGURE CS4.5
Bode diagram and Nichols diagram of p12-0(s).
Bode diagram
–40
–50
–60 p12(s)
Magnitude (dB) –70

–80
–90
–100
–110
–120
–130
100 101 102

Frequency (rad/s)
FIGURE CS4.6
Bode diagram (magnitude) of p12(s) with ±15% of uncertainty on Jd, Bs, Bb, and Bm.
Nominal Plant
The nominal plant for all, p11-0(s), p12-0(s), p21-0(s), and p22-0(s), is selected according to the fol-
lowing parameters within the uncertainty:
J d = 102×106 kg m 2 (CS4.23)
Bs = 4.4862×106 Ns/ m
Bb = 3.1249×10 5 Ns/m
Bm = 274.89 Ns/m

Before starting the design of the controllers, we conduct a preliminary analysis of the
dynamics of the radio telescope. Figure CS4.4 shows the Bode diagram of the nominal four
plants: p11-0(s), p12-0(s), p21-0(s), and p22-0(s).
The plants p11(s) and p12(s) describe, respectively, the effect of torque applied by the wind
Tw(s) and the torque applied by the motor Tm(s) on the motor speed θ’m(s). Both transfer
functions are system-type 1, that is, they have one integrator.
The plants p21(s) and p22(s) describe, respectively, the effect of torque applied by the wind
Tw(s) and the torque applied by the motor Tm(s) on the dish position θd(s). Both transfer func-
tions are system-type 2, that is, they have two integrators.
All the plants contain three resonance modes—see Figures CS4.4. The irst one, at
ω = 4.1 rad/s (0.65 Hz), represents the vibration of the dish, feed arm, and upper
structure in the azimuth direction (θd). The second mode, at ω = 9.35 rad/s (1.49 Hz),
represents the vibration of the lower structure and arms in the azimuth direction (θb),
and the third mode, at ω = 64.8 rad/s (10.31 Hz), represents the vibration of the shaft of
the motors (θm).
The main plant to be controlled is p12(s), which is the transfer function from the truck
equivalent motor torque to the motor velocity θ’m(s)/Tm(s)—see Figure CS4.3. The resonance
modes of p12(s) limit the potential bandwidth of the control system, which will be below
these modes to reduce the vibration of the dish and feed arm at these frequencies—see
Figures CS4.5 and CS4.6.
CS4.4 Azimuth Axis. Velocity Control

Figure CS4.3 shows the cascade control diagram for the telescope azimuth-axis servo sys-
tem. It is composed of two control loops: an inner loop, which is the velocity (rate) control
system, and an outer loop, which is the position control system. In this section, we design
the inner controller: Gv(s).

Figure CS4.7 shows the QFT templates for p12(s), which is the main plant of the velocity
control loop. They are calculated using the model and uncertainty deined in Equations
CS4.18 and CS4.21, and the array of frequencies deined in Equation CS4.22.
QFTCT
–40
0.01
–50
0.05
–60 65
4 0.1
–70
5
60 70
–80 3 0.5
1 10
–90 50
100
2
–100 1.5
–110 500
1.5 1.5
–120
–300 –250 –200 –150 –100 –50

FIGURE CS4.7
CS4.4.2 Control Specifications

Taking into account the dynamics of the radio telescope, the astronomers’ objectives
and the system limitations, we deine the following robust control speciications—with
Hv(s) = 1:
Type 1: Stability speciication
p12 (jω ) Gv (jω )

T1(jω ) = ≤ δ1(ω ) = Ws = 1.46
1 + p12 (jω ) Gv (jω ) (CS4.24)
ω ∈ [0.01 0.05 0.1 0.5 1 1.5 2 3 4 5 10 50 60 65 70 1000 500] rad/s
Type k1: Wind disturbances Tw(s) over θ’m(s) via p11(s)—Figures CS4.3 and CS4.8.
θɺm (jω ) p11(jω )

Tk 1(jω ) = = ≤ δk 1(ω ) (CS4.25)
Tw (jω ) 1 + p12 (jω ) Gv (jω )
(s ak 1 )
δk 1(ω ) = ; ak 1 = 3 ×10 5 ; ω ∈ [0.01 0.05 0.1 0.5] rad/s
(s ak 1 ) + 1
CS4.4.3 Controller Design

QFT Bounds
As mentioned, the QFT bounds take into account the plant dynamics, model, uncer-
tainty, and control speciications. Figures CS4.9a and b show the bounds for stability and
wind disturbance rejection calculated from Equations CS4.24 and CS4.25, respectively.
Figure CS4.10 presents the intersection of the two bounds at each frequency. As we have
Bode diagram
0
–20
–40
–60
Magnitude (dB)
–80 δk1(s)
–100
–120
–140
–160
–180
–200
10–3 10–2 10–1 100 101 102 103
Frequency (rad/s)
FIGURE CS4.8
Attenuation of wind disturbances Tw(s) over θ’m(s)—Type k1.
(a) QFTCT (b) QFTCT

30 70
1.5
25 60

20 1 0.01
5 50
0.05
15 0.5 40
10 0.1
30
5 65
0.01 20
0 0.05
10 4
–5 50 10
0.1
60 3 0.5
–10 70 0
100 500 2
–15 –10
–350 –300 –250 –200 –150 –100 –50 0 –350 –300 –250 –200 –150 –100 –50 0
FIGURE CS4.9
QFT bounds. (a) Stability and (b) wind disturbance rejection.
QFTCT
60
0.01 Lv 0(s)
40 0.05
0.1
Open-loop gain (dB)
20
0.5 65
0
4 5
–20 1
2 3
50 70 1.5 10
–40
100
–60
500
–350 –300 –250 –200 –150 –100 –50 0

FIGURE CS4.10
QFT bounds and Gv(s) design—loop shaping.
a bound solution for each frequency, these control speciications for the radio telescope
are compatible.
Loop Shaping—Gv(s)
The design of the feedback controller Gv(s) is carried out on the Nichols chart. It is done
by adding poles and zeros until the nominal loop, deined as Lv0(jω) = p12,0(jω)Gv(jω), meets
the QFT bounds presented in Figure CS4.10.
We start the design of Gv(s) by adding an integrator to obtain zero steady-state error
for ramp reference inputs. After this, we increase the magnitude of the controller gain
to k = 1900, to put Lv0(0.01) above the B(0.01) bound. Then, we add a low-frequency zero
(z1 = 0.2), to move Lv0(s) to the right, and to pass the stability bounds (“circles”) through
the right.
Afterward, to attenuate the effect of the irst lexible mode at 4 rad/s, we add a notch
ilter (s2 + 2 ζa ωns + ωn2)/(s2 + 2 ζb ωn s + ωn2), with a natural frequency of ωn = 4 rad/s,
a damping numerator coeficient of ζa = 0.1, and a damping denominator coeficient of
ζb = 1. This gives a notch depth of 20 × log10(ζa/ζb) dB = –20 dB.
frequency noise. The inal expression for Gv(s) is the PI with a low-pass ilter and a notch
ilter controller shown in Equation CS4.26. The loop shaping is shown in Figure CS4.10.
1900 ((s/0.2) + 1) (s2 + 0.8 s + 16)

Gv (s) = (CS4.26)
s ((s/5) + 1) (s2 + 8 s + 16)
CS4.4.4 Analysis and Validation of Gv(s)

The analysis of the closed-loop stability in the frequency domain is shown in Figure
CS4.11a. The dashed line is the stability speciication Ws deined in Equation CS4.24. The
solid line represents the worst case of all the possible functions p12Gv/(1 + p12Gv) at each fre-
quency due to the model uncertainty. The control system meets the stability speciication
(the solid line is below the dashed line Ws) for all the plants within the uncertainty, and
including the lexible modes at 4.1, 9.35, and 64.8 rad/s.
The analysis of the robust wind disturbance rejection in the frequency domain is shown
in Figure CS4.11b. The dashed line is the speciication δk1(ω) deined in Equation CS4.25.
The solid line represents the worst case of all the possible functions p11/(1 + p12Gv) at each
frequency due to the model uncertainty. The control system meets the speciication for all
the plants within the uncertainty (the solid line is below the dashed line δk1).
(a) QFTCT (b) QFTCT

10 –50
0
–100
δk1
–10 Ws
–150
Magnitude (dB)
Magnitude (dB)
–20
Worst cases of
“p12Gv / (1 + p12Gv)’’
–30 –200
Worst cases of
–40 “p11/(1 + p12Gv)’’
–250
–50
500
100 –300
–60 0.01 0.01
0.05 0.5 1 2 4 10 60 70 0.05 0.5
0.1 1.5 3 5 50 65 0.1
–70 –350
10–2 10–1 100 101 102 10–2 10–1 100 101 102 103
FIGURE CS4.11
Frequency-domain analysis. (a) Stability and (b) wind disturbance rejection.
Additionally, Figures CS4.20b and CS4.32a show the time-domain simulation of the
velocity loop under different circumstances. In all cases, the controller Gv(s) achieves a
good performance according to the required speciications.
CS4.5 Azimuth Axis—Position Control

This section presents the design of the position control system of the telescope azimuth-
axis servo system. The following sections propose a solution for the 2DOF controller, Gp(s)
and Fp(s), of the outer loop. See again Figure CS4.3 with the cascade control diagram. We
use the velocity controller Gv(s) designed in the previous section for the inner loop.
CS4.5.1 Plant
The plant pp(s) to be controlled by the position controller Gp(s) is composed of the inner veloc-
ity loop and the additional integrator, as shown in Figure CS4.3. In this way, the plant pp(s) is
p12 (s) M(s) Gv (s)  1 

pp (s) =   (CS4.27)
1 + p12 (s) M(s) Gv (s) H v (s)  s 

Figure CS4.12 shows the QFT templates for pp(s)—Equation CS4.27. They are calculated
using the pp(s) plant model, the uncertainty deined in Equation CS4.21, and the array of
frequencies of interest selected in Equation CS4.22.
QFTCT
40 0.01
20 0.05
0.1
0 0.5
1
–20 4
1.5 5 3
–40 2
10 65
–60 60
70
50
–80
100
–100
–120 500
–140
–250 –200 –150 –100
FIGURE CS4.12
QFT templates for pp(s).
CS4.5.3 Control Specifications

Taking into account the dynamics of the radio telescope, the astronomers’ objectives, and
the system limitations, we deine the following robust control speciications for the posi-
tion control system—with Hp(s) = 1:
pp (jω ) Gp (jω )
T1(jω ) = ≤ δ1(ω ) = Ws = 1.46
1 + pp (jω ) Gp (jω ) (CS4.28)
ω ∈ [0.01 0.05 0.1 0.5 1 1.5 2 3 4 5 10 50 60 65 70 100 500] rad/s
Type 3: Sensitivity or disturbances at plant output speciication—see Figure CS4.13a.
θm (jω ) 1 (s/ad )
T3 (jω ) = = ≤ δ3 (ω ) = ; ad = 0.5
do (jω ) 1 + pp (jω ) Gp (jω ) (s/ad ) + 1 (CS4.29)
ω ∈ [0.01 0.05 0.1 0.5] rad/s
Type 6: Reference tracking speciication—see Figure CS4.13b.
pp (jω ) Gp (jω )
δ6 _lo (ω ) < T6 (jω ) = F(jω) ≤ δ6 _up (ω )
1 + pp (jω ) Gp (jω ) (CS4.30)
ω ∈ [0.01 0.05 0.1 0.5 1] rad/s
1 (CS4.31)
δ6 _lo (s) = ; aL = 0.292
(s aL ) + 1 2
 

1.2
0
1
–5 δ3(s)
Magnitude (dB)
–10 0.8
Amplitude
δ6-up(s)
–15 0.6
–20 0.4
δ6-lo(s)
–25
0.2
–30
0
10–2 10–1 100 0 5 10 15 20 25 30 35
FIGURE CS4.13
Control speciications. (a) Disturbance rejection speciications: δ3 and (b) reference tracking: δ6-up, δ6-lo.
[(s/aU ) + 1] 1.25 aU
δ6 _up (s) = ; aU = 0.215 ; ζ = 0.8 ; ω n = (CS4.32)
(s/ω n )2 + (2 ζ s/ω n ) + 1 ζ
 
CS4.5.4 Controller Design

QFT Bounds
As mentioned, the QFT bounds take into account the plant dynamics, model, uncertainty,
and control speciications. Figures CS4.14a through c, show the bounds for stability, sensi-
tivity, and reference tracking for the position loop, see Equations CS4.28 through CS4.32.
Figure CS4.14d presents the intersection of the three bounds at each frequency of interest.
As we have a bound solution for each frequency, the selected control speciications for the
position loop are compatible.
(a) QFTCT (b) QFTCT

30 35
1.5
25 30
0.01 0.05
25
20 1
20
15 65
15
10 0.01
0.1 0.05 0.5 10
5 0.1 0.5
5 5
10 4
0 50 0
60
–5 70 –5
100 500 2 3
–10 –10
–350 –300 –250 –200 –150 –100 –50 0 –350 –300 –250 –200 –150 –100 –50 0
(c) QFTCT (d) QFTCT

35
4 30 0.01 1.5
25 0.05
3
1 20
1
2 15
0.5 65
10 0.1
5
1 10 0.5
5
0.1 50
0 60
0
70 4
0.05 –5 2
0.01 100 3
–1 –10 500
–190 –185 –180 –175 –170 –165 –160 –155 –350 –300 –250 –200 –150 –100 –50 0
FIGURE CS4.14
QFT bounds. (a) Stability, (b) sensitivity, (c) reference tracking, and (d) intersection of bounds.
Loop Shaping—Gp(s)
The design of the feedback controller Gp(s) is carried out on the Nichols chart. It is done by
adding poles and zeros until the nominal loop, deined as Lp0(jω) = pp,0(jω)Gp(jω), meets the
QFT bounds presented in Figure CS4.14d.
We start the design of Gp(s) by increasing the magnitude of the controller gain to
k = 0.5, to put Lp0(0.01) above the B(0.01) bound. Next, we add a low-frequency zero
(z1 = 0.6), to move Lp0(s) to the right, and to pass the stability bounds (“circles”) through
the right.
frequency noise. The inal expression for Gp(s) is the lead controller shown in Equation
CS4.33. The loop shaping is shown in Figure CS4.15.
0.5 ((s/0.6) + 1)
Gp (s) = (CS4.33)
((s/25) + 1)
Preilter Design—Fp(s)
With the controller Gp(s) just designed—see Equation CS4.33, the reference tracking specii-
cations presented in Equations CS4.30 through CS4.32, and the plant model pp(s) described
in Equation CS4.27, we design a preilter Fp(s) to assure that all the input/output functions
pp(s)Gp(s)Fp(s)/[1 + pp(s)Gp(s)], from low frequencies up to 1 rad/s are inside the band deined
by the limits δ6-up(ω) and δ6-lo(ω). The designed preilter is shown in Equation CS4.34, and
the input/output functions and limits in Figure CS4.16.
QFTCT
40
0.01 0.05
20
0.5 0.1
0
Open-loop gain (dB)
1 3
4 65
–20 5
70
1.5 2
10 60
–40
100 50
–60
Lp0(s)
–80
–350 –300 –250 –200 –150 –100 –50 0
FIGURE CS4.15
QFT bounds and Gp(s) design—loop shaping.
QFTCT
10
δ6-up
–10
Magnitude (dB)
–20
ppGpF/(1 + ppGp)
–30
δ6-lo
–40
–50
–60
10–2 10–1 100
Frequency (rad/s)
FIGURE CS4.16
((s/0.3) + 1)
Fp (s) = 2 (CS4.34)
((s/0.38) + 1)
CS4.5.5 Analysis and Validation of G p(s) and Fp(s)

The analysis of the closed-loop stability of the position loop in the frequency domain
is shown in Figure CS4.17a. The dashed line is the stability speciication Ws deined in
Equation CS4.28. The solid line represents the worst case of all the possible functions ppGp/
(1 + ppGp) at each frequency due to the model uncertainty. The control system meets the
stability speciication (the solid line is below the dashed line Ws) for all the plants within
the uncertainty.
The analysis of the robust output disturbance rejection of the position loop in the fre-
quency domain is shown in Figure CS4.17b. The dashed line is the speciication δ3(ω)
deined in Equation CS4.29. The solid line represents the worst case of all the possible
functions 1/(1 + ppGp) at each frequency due to the model uncertainty. The control system
meets the speciication for all the plants within the uncertainty (the solid line is below the
dashed line δ3 from 0 to 0.1 rad/s).
The time-domain analysis of the reference tracking speciication of the position loop is
presented in Figure CS4.18a. The igure shows the limits δ6-up(ω) and δ6-lo(ω)—see Equations
CS4.31 and CS4.32, and the time responses of the motor position θm(t) to a unitary step refer-
ence θmRef(t). We simulate multiple cases of the closed-loop transfer function ppGpFp/(1 + ppGp)
within the uncertainty—see Figure CS4.3, with Hp(s) = 1. The results are given per unit. The
control system meets the speciication (is between the upper and lower limits) in all cases.
(a) QFTCT (b) QFTCT

20 5
δ3
0 0
Ws –5
–20
Magnitude (dB)
Magnitude (dB)
Worst cases of –10

“ppGp/(1 + ppGp)”
–40 –15 Worst cases of
“1/(1 + ppGp)”
–20
–60
500
–25 0.01 0.05 0.1 0.5
–80 100
0.01 0.05 0.5 1 2 4 10 60 70
–30
0.1 1.5 3 5 50 65
–100
10–2 10–1 100 101 102 10–2 10–1 100
FIGURE CS4.17
Frequency-domain analysis. (a) Stability and (b) sensitivity.
Additionally, the time-domain analysis of the output disturbance rejection is shown in

Figure CS4.18b. The igure presents the time responses of the motor position θm(t) to a
unitary disturbance at the output of the plant. We simulate multiple cases of 1/(1 + ppGp)
within the uncertainty, also with Hp(s) = 1. The zero level is the non-perturbed case. The
control system achieves a good disturbance rejection in all cases.
(a) QFTCT (b) QFTCT

1.2 1
d6-up
1 0.8
0.8 0.6
0.6 0.4
d6-lo
0.4 0.2
0.2 0
0
0 5 10 15 20 25 30 35 0 5 10 15 20
Time (s) Time (s)
FIGURE CS4.18
Time domain. (a) Reference tracking θm = ppGpFp/(1 + ppGp) θmRef and (b) output disturbance rejection θm =
1/(1 + ppGp) do.
CS4.6 Simulation: Position and Velocity Loops

This section presents the time-domain simulation of the complete cascade control sys-
tem presented in Figure CS4.3, including the velocity controller Gv(s) in the inner loop—
Equations CS4.26, and the position controller Gp(s) and Fp(s) in the outer loop—Equations
CS4.33 and CS4.34. We simulate the nominal plants—see Equation CS4.23, and include
the velocity limitation of the azimuth axis, which is 0.67 deg/s or 0.0117 rad/s in terms of
the dish movement dθd/dt, or 33.52 deg/s or 0.585 rad/s in terms of the motor movement
dθm/dt—see Figure CS4.19.
The motor reference input θmRef changes as a step at t = 1 s, from 0° to 750°, which in
terms of the dish (θdRef) means from 0° to 15°.
Also, at t = 60 s we introduce a step wind disturbance TW from 0 to 3 × 105 Nm, and
keep this value until the end of the simulation. This torque TW is equivalent to a wind gust
of about v = 6.12 m/s, and considers the worst case scenario where the wind affects only
half of the surface of the dish (taken as a circular dish), and where the aerodynamic thrust
coeficient is CT = 0.1. That is to say
1  π r2  2 
Tw = FT rp = ρair  dish  CT v 2  rdish  (CS4.35)
2 
 2   
 3 
where FT is the thrust force applied by the wind on the dish, and rp the distance from the
center of the dish to the center of pressure, or point where the equivalent lumped force FT
is applied, rp = (2/3) rdish. Also, the density of the air is ρair = 1.225 kg/m3, and the radius of
the dish rdish = 50 m.
Figure CS4.20a shows the results of the position loop controlling the motor angle θm(t)
and Figure CS4.20b the corresponding velocity loop controlling the motor velocity dθm(t)/dt
at the same time. The two igures are in degrees of the motor, and conirm a good perfor-
mance in both reference tracking and wind disturbance rejection.
Additionally Figures CS4.21a and b show, respectively, the complete movement and the
details of the dish position in the same simulation. The peak of the excursion of the dish
angle θd(t) under the wind gust of 6.12 m/s is 0.35°. Also, notice that although the steady-
state error of the motor “θdRef –θm/R” is zero, the dish θd(t) ends up with a permanent error
of 0.06° due to the constant force applied by the wind.
The effect of the wind in the dish angle is a limiting factor that compromises the qual-
ity of astronomical observations. In order to improve the results of the wind disturbance
rejection achieved here, the next section proposes a new design based on NDC methodol-
ogy introduced in Chapter 7.
v
umax
Slope = 1
u
umin
FIGURE CS4.19
Velocity saturation. In dish: 0.6704 deg/s = 0.0117 rad/s. In motor: 33.52 deg/s = 0.585 rad/s. Slope = 1.
umax = +0.585, umin = –0.585.
(a) Position loop: θmRef (t), θm(t) (b) Velocity loop: dθmRef (t)/dt, dθm(t)/dt
900 50
θmRef (t)
800 40
700
30
600 dθm(t)/dt
deg/s (motor)
20
deg (motor)
500
400 θm(t) 10
300 0
200 –10
100 dθmRef (t)/dt
–20
0
–30
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (s) Time (s)
FIGURE CS4.20
Simulation with Gp(s), Fp(s), and Gv(s). (a) Position loop, θm(t). (b) Velocity loop, dθm(t)/dt.
(a) θdRef (t), θd (t), θm(t)/R (b) θdRef (t), θd (t), θm(t)/R
18
θdRef (t) 15.4
16 θd (t)
θd (t)
15.3
14
12 θm(t)/R 15.2
deg (dish)
deg (dish)
10 15.1
8
15
6
14.9 θm(t)/R
4 θdRef (t)
2 14.8
0 14.7
0 10 20 30 40 50 60 70 80 90 100 60 65 70 75 80 85 90 95 100
Time (s) Time (s)
FIGURE CS4.21
Simulation with Gp(s), Fp(s), and Gv(s). Dish position θd(t) and motor θm(t)/R. (a) Full signals and (b) zoom at t = 60–100 s.
CS4.7 Improving with NDC

This section presents a nonlinear dynamic control (NDC) solution for the position loop of
the azimuth axis of the radio telescope. The design is based on the theory and examples
developed in Chapter 7. In the next sub-sections, we follow the 5-step methodology dis-
cussed for the NDC—see Section 7.5.
Step A: Controller Structure

The NDC structure proposed for the position control system is shown in Figure CS4.22. It
consists of two parallel channels, Gp1(s) N1 and Gp2(s) N2. The irst channel Gp1(s) N1 is the
aggressive one, and the second channel Gp2(s) N2 the moderate one. We take Gp2(s) = Gp(s),
which is the position controller already designed in the previous section. Also, we keep
p21(s)
+ θd(s)
p22(s)
θdRef (s)×R = +
uN1(s) θ′mRef (s)
θmRef (s) N1 Gp1(s) Tw(s)
+ Tm(s)
e (s) + uNDC(s) θm(s)
Fp(s) pp(s)
+ N^0
- + + N0 Velocity
N2 Gp2 (s) uA(s) - Saturation loop
uN2(s)
A(s)
+
+
NDC Hp(s)
+
np(s)
FIGURE CS4.22
NDC block diagram. Hp(s) = 1.
the same preilter Fp(s) designed previously. The structure also includes an anti-wind-up
nonlinear inner loop for the irst channel, with a linear transfer function A(s) and a satura-
tion model N̂ 0 . Notice that it is not necessary any anti-wind-up loop for the second chan-
nel, as Gp2 does not include any integrator—see Equation CS4.33.
The nonlinear functions N1 and N2 are deined in Figure CS4.23—see also Figures 7.19
and 7.20, respectively, where N1 = N01 is deined as in Equations 7.17 and 7.18, and N2 = N02
as in Equations 7.19 and 7.20.
The parameters selected for these two nonlinearities are: δh = 0.75 and δl = 0.60. Notice
that N01 + N02 = 1. As a summary, the system has four nonlinearities: the actuator satura-
tion N0, the actuator saturation model N̂ 0 and the two control channels N1 and N2.
Step B.1: Control Elements for First Channel, Gp1(s)—Aggressive

The control speciications for the position controller Gp1(s) are shown in Equations CS4.36
and CS4.37. They are
pp (jω) Gp1(jω)
T1(jω) = ≤ δ1(ω ) = Ws = 4.0
1 + pp (jω ) Gp1(jω ) (CS4.36)
ω ∈ [0.01 0.05 0.1 0.5 1 1.5 2 3 4 5 10 50 60 65 70 1000 500] rad/s
Type 3: Sensitivity or disturbances at plant output speciication
θm (jω ) 1 (s ad ) (CS4.37)
T3 (jω ) = = ≤ δ3 (ω ) = ; ad = 3.3333
do (jω ) 1 + pp (jω ) Gp1(jω ) (s ad ) + 1
ω ∈ [0.01 0.05 0.1 0.5] rad/s
As the main objective for the NDC is to achieve a much better wind disturbance rejec-
tion, these two speciications for Gp1(s) are more aggressive than the ones deined for
(a) (b)
Output Output
δh δh
δl δl
–δh –δl –δh –δl
δl Input δl δh Input
–δl –δl
N01 N02
–δh –δh
FIGURE CS4.23
Nonlinear elements for the NDC: (a) N1 = N01. (b) N2 = N02.
Gp2(s) = Gp(s) in the previous section—see Equation CS4.29. Figure CS4.24 compares the
disturbance rejection speciications required for Gp1(s) and Gp2(s).
The QFT bounds are calculated taking into account the pp(s) transfer function—Equation
CS4.27, the parameter uncertainty given by Equation CS4.21, the array of frequencies of
interest—Equation CS4.22, and the stability and output disturbance rejection speciica-
tions given by Equations CS4.36 and CS4.37. Figures CS4.25a and b show the QFT bounds
of each speciication, and Figure CS4.26 the intersection of bounds, which is compatible at
each frequency.
To meet the speciications, we design a sixth order structure for the Gp1(s) controller, as
shown in Equation CS4.38. The QFT bounds and the loop shaping of Lp0(s) = pp0(s)Gp1(s) are
shown in Figure CS4.26.
0.04 ((s/0.134) + 1)((s/0.25) + 1)((s/0.3) + 1)((s/0.33) + 1) (s2 + 0.13 s + 4225)

Gp1(s) =
s2 ((s/50) + 1)((s/70) + 1) (s2 + 130 s + 4225)
(CS4.38)
Bode diagram
10
0
d3(s) for Gp2(s)
Magnitude (dB)
–10
–20
–30
d3(s) for Gp1(s)
–40
–50
10–3 10–2 10–1 100 101 102
Frequency (rad/s)
FIGURE CS4.24
Comparison disturbance rejection speciications for Gp1(s) and Gp2(s).
(a) QFTCT (b) QFTCT
50
15

1.5 0.01 0.05
40
1
10
65 30
0.01
5 0.1
0.1 0.05 0.5 20
5
10 4
0
50 10 0.5
60
70 3
–5 100 0
500 2
–240 –220 –200 –180 –160 –140 –120 350 –300 –250 –200 –150 –100 –50 0
FIGURE CS4.25
QFT bounds for Gp1(s). (a) Stability and (b) sensitivity.
The analysis of the closed-loop stability of the position loop in the frequency domain
is shown in Figure CS4.27a. The dashed line is the stability speciication Ws deined in
Equation CS4.36. The solid line represents the worst case of all the possible functions
ppGp1/(1 + ppGp1) at each frequency due to the model uncertainty. The control system meets
the stability speciication (the solid line is below the dashed line Ws) for all the plants
within the uncertainty.
The analysis of the robust output disturbance rejection of the position loop in the fre-
quency domain is shown in Figure CS4.27b. The dashed line is the speciication δ3(ω)
QFTCT
0.01 Lp0(s)
80
0.05
60
0.1 10
40
Gain dB
3 4
20 0.5
70
2
0
100 65 50
–20 60
500
–300 –250 –200 –150 –100 –50

Frequency (rad/s)
FIGURE CS4.26
QFT bounds and Gp1(s) design—loop shaping.
(a) QFTCT (b) QFTCT

15
0
10
δ3
Ws –10
5
–20
0
–30
Magnitude (dB)
Magnitude (dB)
–5
–40
“ppGp1/(1 + ppGp1)” –50
–15
Worst cases of
–60 “1/(1 + ppGp1)”
–20
–70
–25
100 0.01 0.05 0.1 0.5
–80
–30 0.01 0.05 0.5 1 2 4 10 60 70
0.1 50 65 –90
–35 1.5 3 5 500
10–2 10–1 100 101 102 10–2 10–1 100
FIGURE CS4.27
Frequency-domain analysis, Gp1(s). (a) Stability and (b) sensitivity.
deined in Equation CS4.37. The solid line represents the worst case of all the possible
functions 1/(1 + ppGp1) at each frequency due to the model uncertainty. The control system
meets the speciication for all the plants within the uncertainty (the solid line is below the
dashed line δ3).
Step B.2: Control Elements for Second Channel, Gp2(s) & Preilter Fp(s)—Moderate
The controller proposed for the second channel (moderate) is the same expression we used
in Section CS4.6, Equation CS4.33,
0.5 ((s/0.6) + 1)
Gp 2 (s) = Gp (s) = (CS4.39)
((s/25) + 1)
The purpose of implementing two channels, Gp1(s) and Gp2(s), is to improve the system
performance beyond the linear limitations. The element Gp1(s) has been designed as an
active/aggressive controller to work with small errors (see N1 = N01) and the element Gp2(s)
as a moderate controller to work with large errors (see N2 = N02). The Bode diagram shown
in Figure CS4.28 compares the two controllers, the aggressive Gp1(s) and the moderate
Gp2(s). Notice that the magnitude of Gp1(s) is much larger for almost all frequencies, except
at ω = 65 rad/s (motor shaft resonance), where a notch ilter reduces Gp1(s) to the Gp2(s)
level. Also, at low frequency, two integrators make Gp1(s) much more active rejecting exter-
nal disturbances.
Finally, we keep the same preilter designed in Section CS4.6, Equation CS4.34, for the
NDC structure. Again, the expression for this preilter is
((s/0.3) + 1)
Fp (s) = 2 (CS4.40)
((s/0.38) + 1)
Gp1(–.), Gp2(–)
100
Magnitude (dB)
Gp1(s)
50
0
Gp2(s)
–50
180
90
Phase (deg)
–90 Gp1(s) Gp2(s)
–180
10–2 10–1 100 101 102 103
Frequency (rad/s)
FIGURE CS4.28
Controllers: aggressive Gp1(s), moderate Gp2(s).
Step C: Inner Loop and Control Element A(s)

The anti-wind-up inner loop is necessary for the irst channel, as Gp1(s) includes two inte-
grators. At the same time, the second channel does not need this anti-wind-up inner loop,
as Gp2(s) does not have any integrator. Then, the expression for A(s) is
ka
A(s) = , being:
s
→ case only Gp1 : k a = 1
→ case only Gp 2 : k a = 0 (CS4.41)
for Gp1 , k a = 0.020

→ case NDC : 
 for Gp 2 , k a = 0
Step D: Isolines—Stability Analysis for the Complete System

The describing functions DFs for each nonlinearity, both in the controller and in the plant,
are given by the following expressions:
• N0 = N05, actuator saturation—Equations 7.25 and 7.26.

• N̂ 0 = N 0 = N 05 , controller actuator saturation model—Equations 7.25 and 7.26.
• N1 = N01, controller irst channel nonlinearity—Equations 7.17 and 7.18.
• N2 = N02 , controller second channel nonlinearity—Equations 7.19 and 7.20.
The isolines are calculated cascading the linear and nonlinear elements. They include
the DF of each nonlinearity at their speciic position in the loop, from the input (right of the
expression) to the output (left of the expression), as expressed by Equation CS4.42
NDC. Isolines
80 L
10(s) Isolines
QFT stability 1e-3 < E < 1e2
60 bounds for
Ws = 4.0
40
20
Open-loop (dB)
–20
–40
–60
–80
–250 –200 –150 –100 –50 0 50
FIGURE CS4.29
Isolines in the Nichols chart, for error E from 10 –3 to 102, ω from 10 –4 to 102 rad/s, and QFT stability bounds for
Ws = 4.0 (or PM = 14.3 deg).
 1  
L = pp (s)DFN 0  Gp1(s)DFN 1 + Gp 2 (s)DFN 2  (CS4.42)

 1 + A( s)DF Nɵ 0  
The algorithm that calculates the isolines for every amplitude E of the input and fre-
quency ω is similar to the one included in Appendix 7. Figure CS4.29 represents in the
Nichols chart the isolines L—Equation CS4.42, for the input errors 10 –3 ≤ E ≤ 102 and
for the frequencies 10 –4 ≤ ω ≤ 102 rad/s. The igure is calculated for the nominal plant—
Equation CS4.23. It also shows the QFT stability bounds found for this nominal plant and
Ws = 4.0 (or PM = 14.3°). As there is no RHP poles and the isolines do not enter into the
QFT bounds, the NDC system is stable for all the cases within the model uncertainty. The
method gives suficient conditions for stability—see also Chapter 3.
Step E: Simulations and Discussion—Position and Velocity Loops

Finally, Figures CS4.30 through CS4.33 show the results of the time-domain simulation
of the complete cascade control system with the NDC presented in Figure CS4.22, includ-
ing the inner velocity loop shown in Figure CS4.3. We simulate the nominal plants—
see Equation CS4.23, and include the velocity limitation of the azimuth axis, which is
±0.585 rad/s in terms of the motor movement dθm/dt—see Figure CS4.19. The expressions
for the controllers are
• Inner loop.
Velocity controller Gv(s): Equation CS4.26
• Outer loop.
Preilter Fp(s): Equation CS4.40
Channel 1. Gp1(s): Equation CS4.38, N1: Equations 7.17 and 7.18, δh = 0.75, δl = 0.60
(a) Position loop: θmRef (t), θm(t) (b) Position loop: θmRef (t), θm(t)
1200 780 θmRef (t)
θm(t) with Gp1 θm(t) with Gp1
θmRef (t)
1000 770 θm(t) with Gp2
800 760
deg (motor)
deg (motor)
750
600
740
400 θm(t) with Gp2
730
θm(t) with NDC θm(t) with NDC
200 720
0 710
0 10 20 30 40 50 60 70 80 90 100 30 40 50 60 70 80
Time (s) Time (s)
FIGURE CS4.30
Simulation, motor angle θm(t). Position control loop. Strategies: aggressive Gp1, moderate Gp2, and NDC solutions.
(a) Complete simulation and (b) zoom of the disturbance rejection part.
Channel 2. Gp2(s): Equation CS4.39, N2: Equations 7.19 and 7.20, δh = 0.75, δl = 0.60
Anti-wind-up. A(s): Equation CS4.41, N̂ 0 : Equations 7.25 and 7.26
We repeat the same simulation conditions of Section CS4.7. The motor reference input
θmRef changes as a step at t = 1 s, from 0° to 750°, which in terms of the dish (θdRef) means from
0° to 15°. Also, at t = 60 s we introduce a step wind disturbance TW from 0 to 3 × 105 Nm,
and keep it until the end of the simulation.
Figure CS4.30 shows the simulation of the position loop—motor angle θm(t), including
the results with only the aggressive controller Gp1(s)—dashed line, with only the moderate
controller Gp2(s)—dashed-dotted line, and with the NDC solution—solid line. The control-
lers Gp2(s) and NDC perform better than Gp1(s) on reference tracking—see Figure CS4.30a.
Also, as we intended, the NDC performs better than the original Gp2(s) on disturbance
rejection, as shown in Figure CS4.30b.
(a) θdRef (t), θd(t), θm(t)/R (b) θdRef (t), θd(t), θm(t)/R
18
θdRef (t) 15.2
16 θd(t)
14 15.1
12 15
deg (dish)
deg (dish)
10 θd(t) 14.9 θm(t)/R

8
14.8 θdRef (t)
6 θm(t)/R
4 14.7
2 14.6
0 14.5
0 10 20 30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
Time (s) Time (s)
FIGURE CS4.31
Simulation, dish angle θd(t). NDC solution. (a) Complete simulation and (b) zoom of the disturbance rejection
part.
(a) (b) Error(t) [–]

Velocity loop: dθmRef (t)/dt, dθm(t)/dt 100
Position error
deg (motor)
50 50
dθm(t)/dt
40 0
30 –50
0 10 20 30 40 50 60 70 80
deg/s (motor)
20
Time (s)
10
0 (c) OutN1 (-), OutN2 (--), uA(-.)
–10 40
deg (motor)
OutN2
–20 OutN1
dθmRef (t)/dt 20
–30
0
uA
0 10 20 30 40 50 60 70 80 90 100
0 10 20 30 40 50 60 70 80
Time (s)
Time (s)
FIGURE CS4.32
Simulation NDC combining Gp1 and Gp2. (a) Motor velocity loop dθm(t)/dt, (b) motor angle error, and (c) switching
signals uN1, uN2, and uA.
Additionally, Figures CS4.31a and b show, respectively, the complete movement and the
details of the dish position with the NDC solution. Also with the NDC solution, Figure
CS4.32a presents the corresponding velocity loop controlling the motor velocity dθm(t)/dt
simultaneously, and Figure CS4.32b the motor angle error e(t) and switching signals uN1(t),
uN2(t), uA(t)—see signals in Figure CS4.22.
Finally, Figure CS4.33 shows the details of the dish angle θd(t) and motor angle θm(t)/R
in dish degrees when we use the original moderate controller Gp2(s)—dashed-dotted line,
and the NDC solution—solid line. The peak of the excursion of the dish angle θd(t) under
the wind gust of 6.12 m/s is reduced from the 0.35° with Gp2(s) to 0.13° with NDC. Also,
notice that although the steady-state error of the motor “θdRef–θm/R” is always zero, the
θdRef (t), θd(t), θm(t)/R

15.4
θd(t), Gp2
15.3 θd(t), NDC
15.2
deg (dish)
15.1
15
14.9 θm(t)/R, Gp2

θdRef (t)
14.8 θm(t)/R, NDC
14.7
60 65 70 75 80 85 90 95 100
Time (s)
FIGURE CS4.33
Disturbance rejection, dish angle θd(t) and motor angle θm(t)/R. Strategies: Gp2 (dashed-dotted line) and NDC
(solid line).
dish θd(t) ends up with a permanent error of 0.06° due to the constant force applied by the
wind in both cases, and also because the feedback sensor used in this case study is θm(t)
and not θd(t).
CS4.8 Summary
This case study has designed two robust QFT control solutions for the velocity and posi-
tion loops of a radio-telescope servo system. The irst solution is based on the classical QFT
methodology and the second one proposes an NDC strategy. Both designs accomplish ive
simultaneous control objectives:
• Stability
• Tracking of the azimuth-axis telescope position
• Regulation of the azimuth-axis telescope velocity
• Rejection of unpredictable wind disturbances
• Reduction of dish and feed-arm vibration
Case Study 5: Attitude and Position Control of
Spacecraft Telescopes with Flexible Appendages
CS5.1 Introduction
With high priority science objectives to break the current barriers of our understanding
of the universe, and dealing with weight limitations of launch vehicles for cost-effective
access to space, new NASA and ESA missions involve both formation lying technology
and satellites with large lexible structures. See for instance the Terrestrial Planet Finder,
Stellar and Planet Imager, Life Finder, Darwin, Lisa, and Proba-3 missions, etc.
Control of spacecraft with large lexible structures and very demanding astronomical
speciications involves signiicant dificulties due to the combination of lexible modes
with small damping, model uncertainty, and coupling among the inputs and outputs.
This case study deals with the design of a MIMO QFT robust control strategy to
simultaneously regulate the position and attitude of a spacecraft telescope with large lex-
ible appendages. The spacecraft is part of a multiple formation lying constellation of ESA
cornerstone project: the Darwin mission (Figure CS5.1).211,216
The scientiic objectives of this mission require high-performance control speciications,
including micrometer accuracy for position and milli-arc-second precision for attitude,
high disturbance rejection properties, loop-coupling attenuation, and low controller com-
plexity and order. The dynamics of the spacecraft presents a 6-inputs × 6-outputs MIMO
plant, with 36 transfer functions of 50th order, signiicant model uncertainty, and high-
loop interactions introduced by the lexible modes of the low-stiffness appendages.
The results presented in this case study are part of a project developed by the author for
the international benchmark proposed by ESA for the design of the control system of the
Darwin mission. The design introduced here is based on the MIMO QFT methodology
(Chapter 8), and got the best performance in that international competition.
CS5.2 System Description and Modeling

The Darwin mission consists of one master satellite (central hub) and three to six telescopes
lying in formation (Figure CS5.1). They will operate together to analyze the atmosphere
of remote planets through appropriate spectroscopy techniques. The mission will employ
nulling interferometry to detect dim planets close to bright stars. The infrared light col-
lected by the free-lying telescopes will be recombined inside the hub-satellite in such a
way that the light from the central star will be cancelled out by destructive interference,
allowing the much fainter planets to stand out. That interferometry technology requires
very accurate and stable positioning of the spacecraft in the constellation, which puts high
395
396 Case Study 5: Attitude and Position Control of Spacecraft Telescopes
FIGURE CS5.1
Spacecraft telescopes with large lexible appendages lying in formation. (Courtesy ESA.)
demands on the attitude and position control system. Instead of an orbit around the Earth,
the mission will be placed further away, at a distance of 1.5 million kilometers from Earth,
in the opposite direction from the Sun (Earth–Sun Lagrangian Point L2).
Each telescope lyer is cylindrically shaped (2 m diameter, 2 m height) and weighs
500 kg. In order to protect the instrument from sunlight, it is equipped with a sunshield
modeled with six large lexible beams (4 m long and 7 kg) attached to the rigid structure
(see Figure CS5.2; beam end-point coordinates in brackets).
For every beam, two different frequencies for the irst modes along Y and Z beam axes are
considered. Their frequency can vary from 0.05 to 0.5 Hz, with a nominal value of 0.1 Hz,
and their damping can vary from 0.1% to 1%, with a nominal value of 0.5%. As regards
spacecraft mass and inertia, the corresponding uncertainty around their nominal value
is of 5%.
Based on this description, and using a mechanical modeling formulation for multiple
lexible appendages of a rigid body spacecraft, the open-loop transfer function matrix
representation of the lyer is given in Equation CS5.1 and Figure CS5.3, where x, y, z are the
2m
Rigid body
[–2.5; –5√3/2; –1] m

[–2.5; –5√3/2; –1] m
Zcont
Xcont
2m CoM
Beams
1m
Ycont
[5; 0; –1] m [–5; 0; –1] m
[2.5; 5√3/2; –1] m [2.5; 5√3/2; –1] m

4m 1m
FIGURE CS5.2
Spacecraft description. (Adapted from Garcia-Sanz, M. et al. 2008. J. ASME J. Dyn. Syst. Meas. Control, 130,
011006-1–011006-15.)
Case Study 5: Attitude and Position Control of Spacecraft Telescopes 397
From: In(1) From: In(2) From: In(3) From: In(4) From: In(5) From: In(6)
To: Out(3) To: Out(2) To: Out(1)
0
–50
–100
–150
0
–50
–100
–150
0
–50
–100
–150
To: Out(6) To: Out(5) To: Out(4)
0
–50
–100
–150
0
–50
–100
–150
0
–50
–100
–150
10–2 100 101 10–2 100 101 10–2 100 101 10–2 100 101 10–2 100 101 10–2 100 101
FIGURE CS5.3
Spacecraft model dynamics. (Adapted from Garcia-Sanz, M. et al. 2008. J. ASME J. Dyn. Syst. Meas. Control, 130,
011006-1–011006-15.)
position coordinates; ϕ, θ, ψ are the corresponding attitude angles; Fx, Fy, Fz are the force
inputs; Tϕ, Tθ, Tψ are the torque inputs; and where each pij(s), i, j = 1,…,6, is a 50th order
Laplace transfer function with uncertainty.
 x(s)   p11(s) p12 (s) p13 (s) p14 (s) p15 (s) p16 (s)  Fx (s) 
    
 y(s)   p21(s) p22 (s) p23 (s) p24 (s) p25 (s) p26 (s)  Fy (s) 
  
 z(s)   p31 (s) p32 (s) p33 (s) p34 (s) p35 (s) p36 (s)  Fz (s) 
  
 φ(s)  =  p (s) p42 (s) p43 (s) p44 (s) p45 (s) p46 (s)  Tφ (s)
(CS5.1)
   41
 θ(s)   p (s) p52 (s) p53 (s) p54 (s) p55 (s) p56 (s)  Tθ (s) 
   51
    
ψ (s)  p61 (s) p62 (s) p63 (s) p64 (s) p65 (s) p66 (s) Tψ (s)
The Bode diagram of the plant (Figure CS5.3) shows the dynamics of the 36 matrix ele-
ments. Each of them and the MIMO system (matrix) itself are minimum phase. The lex-
ible modes introduced by the appendages (second-order dipoles) affect all the elements
around the frequencies ω = [0.19, 10] rad/s. The diagonal elements pii(s), i = 1,…,6, and
the elements p15(s), p51(s), p24(s), and p42(s) are double integrators with additional lexible
modes.

The main objective of the spacecraft control system is to meet a set of astronomical require-
ments, which need to keep the lying telescope pointing at both the observed space target
and the central hub-satellite. This set of speciications leads to some additional engineer-
ing requirements (bandwidth, saturation limits, noise rejection, etc.) and also needs some
complementary control requirements (stability, low loop interaction, low controller com-
plexity and order, etc.). Speciically, the requirements are
1. Astronomical speciications:
a. Position accuracy: maximum absolute error ≤1 µm, and standard deviation
≤0.33 µm for x, y, z axes.
b. Pointing accuracy: maximum absolute error ≤25 mas, and standard deviation
≤8.5 mas for ϕ, θ, ψ axes.
2. Engineering speciications:
a. Bandwidth: ∼0.01 Hz for all axes.
b. Saturation limits: Fx, Fy, Fz maximum force = 150 µN, Tϕ, Tθ, Tψ maximum
torque = 150 µNm.
c. High frequency noise rejection: high roll-off after the bandwidth.
3. Control speciications:
a. Loop interaction: minimum.
b. Rejection of lexible modes: maximum.
c. Controller complexity and order: minimum.
To achieve these goals, the astronomical, engineering, and control speciications are
translated into frequency-domain requirements. This translation into the frequency
domain is based on control-ratio models,259 and takes into account the expected exter-
nal disturbances on the spacecraft lexible modes and the coupling among loops. As a
result, four speciications are deined to calculate the QFT bounds: Spec.1: Robust stability;
Spec.2: Robust sensitivity; Spec.3: Robust disturbance rejection at plant input; and Spec.4:
Robust control effort attenuation. The set of frequencies of interest is within the range:
ω = [6.28 ⋅ 10−4, 62.8] rad/s.
Spec.1: Robust Stability

This speciication, shown in Equation CS5.2, is stated to guarantee a robust stable
control. All the required values, displayed loop-by-loop in Equations CS5.3 and CS5.4,
imply at least 1.54 (3.75 dB) gain margin and at least 49.25° phase margin. The speciica-
tion corresponds not only to the closed-loop transfer function yi(s)/ri(s), but also to trans-
fer functions yi(s)/ni(s) and ui(s)/di(s)—see Equations 2.35 and 2.36. Hence, this condition
additionally imposes the requirements on sensor noise attenuation, disturbance rejection
at plant input, and lexible modes attenuation.
 pii* e (s) −1 gii (s)

  i
−1 ≤ δ1(ω ) (CS5.2)
1 +  pii* e (s) i gii (s)
−1
Note that  pii* e (s) is the inverse of the equivalent plant—see Equation 8.86, which cor-
 i
responds to the diagonal plants pii(s).
Loops 1, 2, and 3 : δ1(ω ) = Ws = 1.85; ∀ω (CS5.3)
0.1687
Loops 4, 5, and 6 : δ1(ω ) = ; ∀ω (CS5.4)
s2 + 0.4s + 0.0912
Spec.2: Sensitivity Reduction

The main objective of this speciication, Equations CS5.5 and CS5.6, is sensor noise attenu-
ation and reduction of the effect of the parameter uncertainty on the closed-loop transfer
function—see Equation 2.37. It corresponds to the ei(s)/ni(s) and “[dtii(s)/tii(s)]/[dpii(s)/pii(s)]”
transfer functions.
1
−1 ≤ δ2 (ω ) (CS5.5)
1 +  pii* e (s) i gii (s)
All loops : δ 2 (ω ) = 2; ∀ω (CS5.6)
Spec.3: Rejection of Disturbances at Plant Input

Solar pressure perturbation and gravity gradient are considered to affect at plant input in
the form of both force and torque. The purpose of this speciication, Equation CS5.7, which
corresponds to ei(s)/di(s) and yi(s)/di(s) transfer functions—see Equation 2.38, is to attenuate
the effect of plant input disturbances on the control error and the output signal. Thus, a
high gain is required in the low-frequency band, Equations CS5.8 through CS5.10. Besides,
since di(s) also represents the lexible modes, special attention is paid to their frequency
range to accomplish the attitude requirements.
 pii* e (s) −1
  i ≤ δ3 (ω )
−1 (CS5.7)
1 +  pii* e (s) i gii (s)
0.21553(s + 0.385)
Loops 1 and 2 : δ 3 (ω ) = ; ∀ω (CS5.8)
(s + 0.307 )(s + 6.18)(s2 + 0.4s + 0.0912)
0.313 (s − 0.01705)(s2 + 0.009974 s + 5.104 ⋅ 10−5 )

Loop 3 : δ3 (ω )= ; ∀ω (CS5.9)
(s − 0.01813)(s2 + 0.02554 s + 0.0004754)
Loops 4, 5, and 6 : δ3 (ω )
(s + 0.2)(s + 0.186)(s + 0.2044)(s + 0.003892)(s2 + 0.06014s + 0.02736)
= ; ∀ω
(s + 0.007333)(s + 0.445)(s2 + 0.07904s + 0.000326)(s2 + 0.2352s + 0.0981)
(CS5.10)
Spec.4: Control Effort Reduction

Because of saturation limits, control signal movements should be kept reasonably small
despite disturbances—see Equation 2.39. This speciication, Equation CS5.11, corresponds
to ui(s)/ni(s) transfer function and is depicted in Equations CS5.12 through CS5.14.
gii (s)
−1 ≤ δ 4 (ω ) (CS5.11)
1 +  p (s) gii (s)
*e
ii
i
Loops 1 and 2 : δ4 (ω ) = 557.1(s + 5) ; ∀ω

(CS5.12)
(s2 + 3.23 s + 6.5)
106.9210(s + 0.55)(s2 + 0.04s + 0.13)

Loop 3 : δ 4 (ω ) = ; ∀ω (CS5.13)
(s + 1.4)2 (s2 + 0.1227 s + 0.097 )
4.026(s2 − 0.1854s + 0.203)(s2 + 0.04s + 0.54)

Loops 4, 5, and 6 : δ 4 (ω ) = ; ∀ω (CS5.14)
(s2 + 0.305s + 0.056)(s2 + 0.115s + 0.095)
Reducing Coupling Effects as much as Possible

The coupling effects from other axes are considered as part of the disturbances acting at
the input of the equivalent SISO plant. The design of the non-diagonal elements of the
matrix compensator intends also to minimize the off-diagonal elements of the coupling
matrices—see Equations 8.71 through 8.73.
CS5.4 Control System Design

The MIMO QFT method 1, explained in Sections 8.5 and 8.8, is applied here to design the
6 × 6 robust control system for the spacecraft telescope described in Section CS5.2, and
with the performance speciications deined in Section CS5.3.
Step A: Input–Output Pairing and Loop Ordering

An example of RGA, calculated from low frequency (steady state) up to 0.19 rad/s, and for
all cases within the uncertainty, is shown in Equation CS5.15. According to it, the pairing is
selected through the main diagonal of the matrix, which contains positive RGA elements.
In addition, due to the very demanding speciications, the RGA elements λ15, λ24, λ42, and
λ51 are considered relevant as well.
 1.0064 0 0 0 −0.0064 0
 0 1.0064 0 −0.0064 0 0

 0 0 1 0 0 0
RGA( ω=6.28 ⋅ 10−4 rad/s ) =  0 −0.0064 0 1.0064 0 0 (CS5.15)

 −0.0064 0 0 0 1.0064 0
 0 0 0 0 0 1

Also, in accordance with this RGA results and taking into account the requirement
of minimum controller complexity and order (Section CS5.3, Speciication 3c), we select
a controller structure with only six diagonal elements and two additional off-diagonal
elements—see Equation CS5.16.
 g11(s) 0 0 0 0 0 
 0 g 22 (s) 0 0 0 0 

 0 0 g 33 (s) 0 0 0 
G(s) =  0 g 42 (s) 0 g 44 (s) 0 0  (CS5.16)

 g 51(s) 0 0 0 g 55 (s) 0 
 0 0 0 0 0 g66 (s)

From this, four independent control design problems are adopted, two SISO controllers,
[g33(s)] and [g66(s)], and two 2 × 2 MIMO controllers, [g11(s) 0; g51(s) g55(s)] and [g22(s) 0; g42(s)
g44(s)]. The SISO problems are considered as a classical SISO QFT problem, while the two
2 × 2 MIMO subsystems are studied through the non-diagonal MIMO QFT method 1—see
Sections 8.5 and 8.8.
Step B0: Design of Diagonal Controllers gkk(s), k = 3, 6—SISO Cases

The controllers g33(s) and g66(s) are independently designed by using classical SISO QFT to
satisfy the performance speciications stated in Section CS5.3 for every plant within the
uncertainty. The corresponding QFT bounds and the nominal open-loop transfer func-
tions Lii(s) = pii(s) gii(s), i = 3, 6, are plotted on the Nichols charts shown in Figure CS5.4.
The expressions for g33(s) and g66(s) are shown in Equations CS5.24 and CS5.27, respectively.
For the two 2 × 2 MIMO problems, we follow the low chart presented in Figure 8.19,
MIMO QFT Method 1. As we do not have reference tracking speciications, we do not need
preilters and just follow the Steps B.1.1, B.1.3, B.2.1, and B.2.3.
Step B.1.1 (1st): Design of Diagonal Controller g11(s)—First MIMO Problem

The compensator g11(s) is designed according to the non-diagonal MIMO QFT method 1,
*e *
explained in Section 8.5, for the inverse of the equivalent plant [ p11 (s)]1 = p11 (s) .
(a) Loop-shaping L33 (b) Loop-shaping L66

L33 ( j 6.3 10–4) B66 ( j 6.3 10–4)
100 B33 ( j 6.3 10–4) L66 (j 6.3 10–4)
L33 ( j 4.3 10–3) 50 B66 ( j 4.3 10–3) B66 ( j 0.26) L66 ( j 4.3 10–3)
B33 ( j 4.3 10–3)
B66 ( j 0.2) L66 ( j 0.03)
B66 ( j 0.33) L66 ( j 0.26)
50 B33 ( j 0.2) L33 ( j 0.03)
Open-loop gain (dB)
Open-loop gain (dB)
L33 ( j 0.2) 0 B66 ( j 0.03)

B33 ( j 0.03)
B33 ( j 0.26) L66 ( j 0.33)
0 B33 ( j 0.33) L33 (j 0.26) L66 (j 0.2) L66 ( j 2.72)
L33 ( j 0.33) –50 B66 ( j 2.72)

B33 ( j 1.61) B66 ( j 3.53)
–50 B33 ( j 3.53) L33 ( j 1.61) B66 ( j 15)
L33 ( j 3.53) –100
B33 ( j 15) B66 ( j 62.82) L66 ( j 3.53)
–100 L33 ( j 15)
B33 (j 62.82)
L33 ( j 62.82) –150 L66 ( j 62.82) L66 ( j 15)
–360 –315 –270 –225 –180 –135 –90 –45 0 –360 –315 –270 –225 –180 –135 –90 –45 0
FIGURE CS5.4
Loop shaping: (a) L33(s) = p33(s) g33(s) and (b) L66(s) = p66(s) g66(s). (Adapted from Garcia-Sanz, M. et al. 2008.
J. ASME J. Dyn. Syst. Meas. Control, 130, 011006-1–011006-15.)
100 B11 (j 6.3 10–4) B55 (j 6.3 10–4)

L11 (j 6.3 10–4) L55 (j 6.3 10–4)
80 B11 (j 4.3 10–3) L11 (j 4.3 10–3) 50 B55 (j 4.3 10–3) L55 (j 4.3 10–3)
60 L11 (j 0.03) B55 (j 0.26) L55 (j 0.03) L55 (j 0.26)

L11 (j 0.2) L11 (j 0.26)
Open-loop gain (dB)

Open-loop gain (dB)
B11 (j 0.26) B55 (j 0.2)

40 B11 (j 0.03)
B55 (j 0.03)
0
B11 (j 0.2) L11 ( j 0.33)
20 B55 (j 0.33)
B11 (j 0.33) L55 (j 0.33) L55 (j 0.2)
B55 (j 0.43)
0
B55 (j 3.53)
–20 B11 (j 1.61) –50 L55 (j 0.43)
B11 (j 3.53) B55 (j 0.73) L (j 0.73) L55 (j 3.53)
–40 55
L11 (j 1.61) B55 (j 15)
–60 L11 (j 3.53)
B11 (j 15) –100
–80 L11 (j 15)
L55 (j 15)
–100 B11 (j 62.82) L11 (j 62.82)
–150
–360 –315 –270 –225 –180 –135 –90 –45 0 –360 –315 –270 –225 –180 –135 –90 –45 0
FIGURE CS5.5
−1 −1
Loop shaping: (a) L11 (s) =  p11
*e
(s) g11 (s) and (b) L55 (s) =  p55
*e
(s) g 55 (s). (Adapted from Garcia-Sanz, M. et al.
1 2
2008. J. ASME J. Dyn. Syst. Meas. Control, 130, 011006-1–011006-15.)
Figure CS5.5a presents the bounds and achieved loop shaping. The expression for g11(s) is
shown in Equation CS5.21.
Step B.1.3 (1st): Design of Non-Diagonal Controller g51(s)—First MIMO Problem

The non-diagonal compensator g51(s) is designed to minimize the (5,1) element of the cou-
pling matrix in the case of disturbance rejection at plant input, which gives the following
expression:
opt *N
g 51 (s) = −p51 (s) (CS5.17)
*
where N denotes the middle plant that interpolates the expression [−p51 (s)] from 0 to
−1
10 rad/s, as shown in Figure CS5.6a. The expression for g51(s) is shown in Equation CS5.22.
(a) Magnitude plot of g51(s) (b) Magnitude plot of g42(s)

150
100
100
50
50
Magnitude (dB)
Magnitude (dB)
0
0
–50 –50
–100 –100
–150
10–4 10–3 10–2 10–1 100 101 102 10–3 10–2 10–1 100 101 102
FIGURE CS5.6
∗ ∗
(a) Magnitude plot of [−p51 (s)] with uncertainty and g51(s)—solid line and (b) magnitude plot of [−p42 (s)] with
uncertainty and g42(s)—solid line. (Adapted from Garcia-Sanz, M. et al. 2008. J. ASME J. Dyn. Syst. Meas. Control,
130, 011006-1–011006-15.)
Step B.2.1 (1st): Design of Diagonal Controller g55(s)—First MIMO Problem

*e
explained in Section 8.5, for the inverse of the equivalent plant [ p55 (s)]2 , which is
*e
p (s)2 =
*e
 p55
 (s) 1 −  p51 s 1  (
* e ( ) +  g 51 ( s)  *e 
 1  p15 (s) 1 )( )
55
*e
 p11  (
 (s) 1 + [ g11(s)]1 )
Figure CS5.5b presents the bounds and achieved loop shaping. The expression for g55(s)
is shown in Equation CS5.26.
Step B.2.3 (1st): Design of Non-Diagonal Controller g15(s)—First MIMO Problem

As mentioned, the non-diagonal compensator g15(s) is chosen as null to simplify the
controller complexity, according to the speciication 3c in Section CS5.3 and the RGA
results:
opt
g15 (s) = 0 (CS5.18)
The second MIMO problem is presented in the following steps. It consists of the design
of the elements g22(s), g42(s), g44(s), and g24(s), which are equivalently performed as in the
previous Steps B.1.1, B.1.3, B.2.1, and B.2.3, respectively.
Step B.1.1 (2nd): Design of Diagonal Controller g22(s)—Second MIMO Problem

The compensator g22(s) is designed according to the non-diagonal MIMO QFT method
1, explained in Section 8.5, for the inverse of the equivalent plant  p22
*e
(s) = p22
*
(s).
1
Figure CS5.7a presents the bounds and achieved loop shaping. The expression for g22(s) is
shown in Equation CS5.21.
100 B22 ( j 6.3 10–4) B44 ( j 6.3 10–4)

L22 (j 6.3 10–4) L44 ( j 6.3 10–4)
80 B22 ( j 4.3 10–3) L22 ( j 4.3 10–3) 50 B44 ( j 4.3 10–3) L44 ( j 4.3 10–3)
60 L22 ( j 0.03) B44(j 0.2) L44 ( j 0.03) L44 ( j 0.26)
B44 ( j 0.26)
Open-loop gain (db)
L22 ( j 0.2)
Open-loop gain (db)
40 B22 (j 0.03)
0 B44 ( j 0.03)
20 L22 (j 0.33) B44 ( j 0.33)
B22 (j 1.61) L44 ( j 0.2)
B22 (j 0.33) L44 ( j 0.33)
0 B22 (j 0.2)
–20 –50 B44 ( j 0.73)
–40 B22 ( j 3.53) B44 ( j 3.53) L44 ( j 0.73)

L22 ( j 1.61)
–60 B22 ( j 15) L22 ( j 3.53) –100 B44 ( j 15)
–80 L22 ( j 15)

L44 ( j 15) L44 ( j 3.53)
–100 L22 ( j 62.82) B22 ( j 62.82)
–150
–360 –315 –270 –225 –180 –135 –90 –45 0 –360 –315 –270 –225 –180 –135 –90 –45 0
FIGURE CS5.7
−1 −1
Loop shaping (a) L22 (s) =  p22
*e
(s) g 22 (s) and (b) L44 (s) =  p44
*e
(s) g 44 (s). (Adapted from Garcia-Sanz, M. et al. 2008.
 1  2
J. ASME J. Dyn. Syst. Meas. Control, 130, 011006-1–011006-15.)
Step B.1.3 (2nd): Design of Non-Diagonal Controller g42(s)—Second MIMO Problem

The non-diagonal compensator g42(s) is designed to minimize the (4,2) element of
the coupling matrix in the case of disturbance rejection at plant input, which gives
the following expression:
opt *N
g 42 (s) = −p42 (s) (CS5.19)
where N denotes the middle plant that interpolates the expression [−p∗42 (s)] from 0 to
10−1 rad/s, as shown in Figure CS5.6b. The expression for g42(s) is shown in Equation CS5.22.
Step B.2.1 (2nd): Design of Diagonal Controller g44(s)—Second MIMO Problem

explained in Section 8.5, for the inverse of the equivalent plant  p44
*e
(s) , which is
2
*e
p (s)2 =
*e
 p44  (
 *e   )(
  p* e 
 (s) 1 −  p42 (s) 1 +  g 42 (s) 1  24 (s) 1 )
44
(
*e
 p22 
 (s) + [ g 22 (s)]1
1
)
Figure CS5.7b presents the bounds and achieved loop shaping. The expression for g44(s)
is shown in Equation CS5.25.
Step B.2.3 (2nd): Design of Non-Diagonal Controller g24(s)—Second MIMO Problem

As in Step B.2.3 (1st), the non-diagonal compensator g24(s) is chosen as null to simplify
the controller complexity, according to the speciication 3c in Section CS5.3 and the RGA
results:
opt
g 24 (s) = 0 (CS5.20)
Steps B.k.2: Design of Preilters f kk(s), k = 1, 2,…,6

There is not preilter required in this example, because we do not have reference tracking
speciications. Then, fkk(s) = 1, k = 1, 2,…,6.
Controller Expressions
The notation adopted below for the transfer function expressions of the controllers denotes
the steady state gain as a constant without parenthesis; simple poles and zeros as (ω), which
corresponds to (s/ω + 1) in the denominator and numerator respectively; poles and zeros
at the origin as (0); conjugate poles and zeros as [ζ; ωn], with ((s/ωn)2 + (2ζ/ωn)s + 1) in the
denominator and numerator respectively; n-multiplicity of poles and zeros as an exponent
()n. Following this notation, the MIMO QFT controller designed in this section consists of
the following elements:
{31.5(0.6194)(0.2138)(0.1663)(0.1649)}
g11(s) = g 22 (s) = (CS5.21)
{(0.666)(0.4982)(0.07526)[0.676; 1.479]}
g 51(s) = −g 42 (s) = {42.4(0)2 }/{(0.3)3 } (CS5.22)
g15 (s) = g 24 (s) = 0 (CS5.23)
{125(0.13)(0.057)[0.07 019; 0.3565][1; 0.02]}

g 33 (s) = (CS5.24)
{(1.48)(0..7875)(0.2)(0.004)(0.00246)[0.18; 0.314]}
{2.242(0.03412)[0.08644; 0.7114][0.1131; 0.3414][0.1145; 0.2604][0.008792; 0.2593]

[0.7 ; 0.0052][1; 0.0007]}
g 44 (s) = 2
{(0.9776)(0.8)(0.0005) [0.2451; 0.27 08][0.297 ; 0.2673][−0.0005; 0.254]
[0.14; 0.252][0.7 ; 0.0045]}
(CS5.25)
{2.242(0.13)(0.03)[0.1079; 0.7 099][0.07 069; 0.341][0.03; 0.2593][0.7 ; 0.0052][1; 0.0007]}

g 55 (s) =
{(0.9776)(0.8)(0.12)(0.0005)2 [0.2451; 0.27 08][−0.0008; 0.254][0.3; 0.25][0.7 ; 0.0045]}
(CS5.26)
{2.242 (0.02584)[0.08644; 0.7114][0.1131; 0.3414][0.1145; 0.2604][0.008792; 0.2593]

[0.7 ; 0.0052][1; 0.0007]}
g66 (s) =
{(0.9776)(0.8)(0.0005)2 [0.2451; 0.27 08][0.297 ; 0.2673][−0.0007 ; 0.254]
[0.1687 ; 0.241][0.7 ; 0.0045]}
(CS5.27)
CS5.5 Simulation and Validation

Time domain simulations were performed for 300 plants in the ESA spacecraft telescope
benchmark simulator—see Figure CS5.8. The plant models were selected randomly within
the space of uncertainty using a Monte Carlo technique. The same plants were used for all
the participants in the benchmark.
The performance achieved for the position and attitude control by the non-diagonal
MIMO QFT was excellent, meeting all the required speciications—see Section CS5.3.
Moreover, the MIMO QFT obtained the best results in the benchmark—see maximum
and standard deviation results in Table CS5.1, improving by two orders of magnitude the
results obtained by the second classiied, that used a H-ininity control technique.
At the same time, while the second classiied required controller structures with full-
matrices of 36 elements of 42nd order, the non-diagonal MIMO QFT design consisted of
only eight compensators with simple structures, all from 3rd to 14th order—see Equations
CS5.21 through CS5.27, dividing by more than 20 the number of operations per second
needed for the real-time control algorithm—see Table CS5.2.
Fx x
Fy y
Fz z
Disturbance
Tϕ ϕ
force and torque
Commanded Tθ θ
External disturbances
position and Position
Tψ ψ
attitude + + + and attitude
G(s) FEEP
– thruster
Controller Actuator
Darwin flyer 6DOF
dynamics
Position and
attitude
measurement
Sensor
FIGURE CS5.8
ESA telescope-type spacecraft simulator. (Adapted from Garcia-Sanz, M. et al. 2008. J. ASME J. Dyn. Syst. Meas.
Control, 130, 011006-1–011006-15.)
TABLE CS5.1
Time Simulation Performance with Controllers
Non-Diagonal H-Ininity
Spec. MIMO QFT Controller Controller
Maximum Position Error X <1 µm 0.0131 0.0293
0.0816 0.511
Maximum Position Error Y <1 µm 0.0120 0.0299
0.0120 0.0299
Maximum Position Error Z <1 µm 0.0288 0.0292
0.0288 0.0292
Maximum Attitude Error X <25.5 mas 25.27 25.95
25.27 25.95
Maximum Attitude Error Y <25.5 mas 22.91 23.21
22.55 28.91
Maximum Attitude Error Z <25.5 mas 21.15 22.84
21.15 22.84
Standard Deviation of Position Error X <0.33 µm 0.00275 0.00686
0.0511 0.341
Standard Deviation of Position Error Y <0.33 µm 0.00265 0.00722
0.00265 0.00722
Standard Deviation of Position Error Z <0.33 µm 0.00668 0.00691
0.00668 0.00691
Standard Deviation of Attitude Error X <8.5 mas 5.57 5.68
5.57 5.68
Standard Deviation of Attitude Error Y <8.5 mas 5.76 6.01
5.80 8.23
Standard Deviation of Attitude Error Z <8.5 mas 4.83 5.00
4.83 5.00
Note: At each speciication (each row), the best result (QFT or H-ininity) is in bold.
TABLE CS5.2
Number of Operations per Second Required by Controllers
Controller # Multips/s # Sums/s
Non-diagonal MIMO QFT 130 124
H-ininity 2994 2988
CS5.6 Summary
This case study demonstrated the feasibility of the sequential non-diagonal MIMO robust
QFT control strategies to simultaneously regulate the position and attitude of a 6 × 6
spacecraft telescope with large lexible appendages. The spacecraft was part of a multiple
formation lying constellation of a ESA cornerstone mission. The controller satisfactory
met the astronomical, engineering, and control requirements.

Cinco Casos de Control Digital

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cinco Casos de Control Digital

Uploaded by

Copyright:

Available Formats

Case Study 1: Satellite Control

CS1.2 Plant Model

• J0, the moment of inertia of the central body, (kg m2)

M qɺɺ + C qɺ + K q = Q (qɺ , q, u, t) (CS1.1)

Rearranging the equation of motion Equation CS1.1 as

qɺɺ = −M −1C qɺ − M −1K q + M −1R u (CS1.10)

−M −1C −M −1K   M −1R 

with I = [1 0; 0 1], 0 = [0 0; 0 0], and,

y p (s) 1 −lp

CS1.2.1 Parametric Uncertainty

with: J0 ∈ [432, 720] kg m2, Bp ∈ [0.24, 0.48] Ns/m.

CS1.2.3 Nominal Plant

θ(s) (s2 + 0.012s + 16)

CS1.2.4 QFT Templates

CS1.3 Preliminary Analysis

Case p11-1(s): J0 = 720, Bp = 0.48; Case p11-2(s): J0 = 720, Bp = 0.32

CS1.4 Control Specifications

• Type 1: Stability speciication

which is equivalent to PM = 30.03°, GM = 3.63 dB—see Equations 2.30 and 2.31.

• Type 3: Sensitivity or disturbances at plant output speciication—see Figure CS1.7a

(a) Bode diagram (b) Step response

• Type 6: Reference tracking speciication—see Figure CS1.7b

CS1.5 Controller Design

(a) QFTCT (b) QFTCT

Open-loop magnitude (dB)

(c) QFTCT (d) QFTCT

Open-loop magnitude (dB)

Open-loop magnitude (dB) 40

20 ((s/0.1) + 1) (s2 + 3.066s + 21.9)

CS1.5.3 Prefilter Design—F(s)

CS1.6 Analysis and Validation

(a) QFTCT (b) QFTCT

–30 Worst cases of

The analysis of the robust sensitivity speciication, or rejection of disturbances at the

(a) QFTCT (b) QFTCT

(a) QFTCT (b) QFTCT

CS2.2 Plant Model

• Tgd desired electrical torque, output of torque controller, [Nm]

M qɺɺ + C qɺ + K q = Q (qɺ , q, u, t) (CS2.1)

Rearranging the equation of motion Equation CS2.1 as

qɺɺ = −M −1C qɺ − M −1K q + M −1R u (CS2.10)

−M −1C −M −1K   M −1R 

with I = [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1], 0 = [0 0 0 0; 0 0 0 0; 0 0 0 ; 0 0 0 0], and,

y = [ yɺ t yɺ b θɺr θɺg ]T (CS2.17)

∂FT (t) 1 ∂CT 2

∂FT (t) 1 ∂CT 2

∂Ta (t) 1 1  ∂Cp 3 2 

Then, adding the electrical torque, we have

On the other hand, the transfer functions of the actuators are

β(s) = Aβ (s)βd (s) (CS2.28)

Tg (s) = AT (s)Tgd (s) (CS2.29)

u = [v1 βd Tgd ]T (CS2.31)

Ωr(s) = p31(s)v1(s) + p32 (s)βd (s) + p33 (s)Tgd (s) (CS2.33)

−(Bs s + K s )AT (s)

∆(s) = J g J r s3 + (Bs J g + Bs J r − J g KTΩ )s2 + ( J g K s − Bs KTΩ + J r K s )s − KTΩ K s (CS2.37)

CS2.2.1 Parametric Uncertainty

v1(s) p11(s) v1(s)

p31(s) do(s) v1(s)

This leads to the following expressions:

Ωr (s) (5.025497 ×106 s2 + 6, 215, 000s + 7.63875×107 )KTV (CS2.38)

Ωr(s) (5.025497 ×106 s2 + 6, 215, 000s + 7.63875×107 )KT β (CS2.39)

(a) Bode diagram (b) Nichols chart

Open-loop gain (dB)