Giovanni Gozzini

Aerospace Control Systems

– Exercises –

May 6, 2024

Politecnico di Milano
Preface

Copyright ©2022
Giovanni Gozzini, Marco Lovera
Dipartimento di Scienze e Tecnologie Aerospaziali, Politecnico di Milano

This document is subjected to copyright by the authors and the Department of

Aerospace Science and Technology of Politecnico di Milano. Its reproduction and
diffusion is free provided it complies with Italian copyright rules and not for profit.
The partial reproduction and the use for instructional purposes are allowed. The
document cannot be modified without or against the consent of the authors, which
is binding. This copyright notice cannot be removed.

This book represents the unoffical collection of the exercise lectures of the course
of Aerospace Control Systems by Professor Marco Lovera, for students of the master
in Space Engineering and Aeronautical Engineering.
Beyond its usefulness, reading these notes does not achieve the effectiveness of
following the exercise lectures in person.
Use this book carefully and with critical sense because it may contain errors.

Milano,
2022 May
Giovanni Gozzini

v
Contents

1 Exercise classroom 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Example of static performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Solution of Linear Matrix Inequality (LMI) . . . . . . . . . . . . . . . . . . . . . 2
1.3 Additional exercise: Example of static performance . . . . . . . . . . . . . . 3
1.4 Additional exercise: Example of dynamic performance . . . . . . . . . . . 3
1.5 Additional exercise: 𝐻∞ norm for LTI as solution of LMI problem . 3
1.5.1 LMI for a given 𝛾 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5.2 LMI with optimal 𝛾 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Exercise classroom 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Rigid body detumbling problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Exercise classroom 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1 Example of loop shaping design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Eigenvalue Assignment: Problem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2.1 Eigenvalue Assignment: Problem 2 . . . . . . . . . . . . . . . . . . . . . 8

4 Exercise classroom 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1 LQR design example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 LQG design example: double integrator . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3 Additional Exercise: Guaranteed margins for LQG regulators . . . . . . 10

5 Exercise classroom 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.1 Uncertain models: Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Uncertain models: Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.3 Mixed sensitivity synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

vii
Acronyms

AS Asymptotically Stable
CCF Canonical Controllability Form
EA Eigenvalue Assignment
GAS Globally Asymptotically Stable
LMI Linear Matrix Inequality
LQG Linear Quadratic Gaussian
LQR Linear Quadratic Regulator
LTI Linear Time Invariant
MIMO Multiple-Input Multiple-Output
PID Proportional Integral Derivative
SISO Single-Input Single-Output

ix
Chapter 1
Exercise classroom 1

1.1 Example of static performance

Consider the Single-Input Single-Output (SISO) system described by the block

diagram in Figure 1.1, with

5(𝑠 + 1) 1
𝑅(𝑠) = 𝐺 (𝑠) =
𝑠 (𝑠 + 1) (0.1𝑠 + 1) 2
compute:
(a) the error and the output at steady state for a step set-point input 𝑦 0 (𝑡) = step(𝑡),
(b) the error and the output at steady state for a ramp disturbance input 𝑑 (𝑡) =
ramp(𝑡),
(c) the error and the output at steady state for a step noise input 𝑛(𝑡) = step(𝑡),
(d) the ”steady state” error for a sinusoidal disturbance input 𝑑 (𝑡) = sin(Ω𝑡) with
Ω = 1 rad/s,
(e) the ”steady state” error for a sinusoidal noise input 𝑛(𝑡) = sin(Ω𝑡) with Ω =
80 rad/s.

𝑛(𝑡 ) 𝑑 (𝑡 )
𝑦 0 (𝑡 ) − + 𝑦 (𝑡 )
𝑢(𝑡 )
𝑅 (𝑠) 𝐺 (𝑠)
+ + +
−

Fig. 1.1 Block diagram of a SISO control loop where 𝑦 0 (𝑡 ) is the reference signal, 𝑛(𝑡 ) and 𝑑 (𝑡 )
are disturbance at the input and at the output, respectively, and 𝑦 (𝑡 ) is the output.

1
2 1 Exercise classroom 1

1.2 Solution of LMI

Study the stability of the following Linear Time Invariant (LTI) system solving
LMI problem.
−1 2 3 

𝐴 =  0 −2 1 

0
 0 −5
1.5 Additional exercise: 𝐻∞ norm for LTI as solution of LMI problem 3

1.3 Additional exercise: Example of static performance

Consider the SISO system described by the block diagram in Figure 1.1, with
𝑠+1 10
𝑅(𝑠) = 𝐺 (𝑠) =
𝑠 (𝑠 + 1) 2
verify that:
(a) the error at steady state for a step set-point input 𝑦 0 (𝑡) = step(𝑡) is 0,
(b) the error at steady state for a sinusoidal disturbance input 𝑑 (𝑡) = sin(Ω𝑡) with
Ω = 1 rad/s is less or equal than 0.2,
(c) the settling time is less or equal than 8 s,
(d) the percentage overshoot is less or equal than 60 %,
Are the performance of this controller good? If not, try to improve it.

1.4 Additional exercise: Example of dynamic performance

Consider the SISO system described by the block diagram in Figure 1.1, with

5(𝑠 + 1) 1
𝑅(𝑠) = 𝐺 (𝑠) =
𝑠 (𝑠 + 1) (0.1𝑠 + 1) 2

compute for a step set-point input 𝑦 0 (𝑡) = step(𝑡):

(a) the settling time,
(b) the maximum percentage overshoot.

1.5 Additional exercise: 𝑯∞ norm for LTI as solution of LMI

problem

1.5.1 LMI for a given 𝜸

Given    
0 1 0  
𝐴= 𝐵= 𝐶= 1 0
−1 −0.02 1
1
and 𝛾 = 60 find 𝑃 that solves 𝑃 𝐴 + 𝐴⊤ 𝑃 + 𝛾2
𝑃𝐵⊤ 𝐵𝑃 + 𝐶 ⊤ 𝐶 ≤ 0.
4 1 Exercise classroom 1

1.5.2 LMI with optimal 𝜸

Given    
0 1 0  
𝐴= 𝐵= 𝐶= 1 0
−1 −0.02 1
1
and find 𝑃 that solves 𝑃 𝐴 + 𝐴⊤ 𝑃 + 𝛾2
𝑃𝐵⊤ 𝐵𝑃 + 𝐶 ⊤ 𝐶 ≤ 0.
Chapter 2
Exercise classroom 2

2.1 Rigid body detumbling problem

Following separation of the spacecraft from the launcher, the problem to bring the
satellite angular dynamics close to rest is addressed. This must be achieved rapidly,
reliably and using as little resources as possible. Assuming ideal sensors for angular
rate and actuators to torque the body in three dimension, the Euler’s equation for
angular rigid body dynamics is considered:

𝐼 𝜔¤ = 𝑆(𝜔)𝐼𝜔 + 𝑇 (2.1)

where 𝜔 ∈ R 3 is the angular rate in body frame, 𝐼 ∈ R 3×3 is the inertia matrix,
𝑇 ∈ R 3 is the control torque and
 0
 𝜔 𝑧 −𝜔 𝑦 
𝑆(𝜔) = −𝜔 𝑧
 0 𝜔 𝑥  . (2.2)
 𝜔 𝑦 −𝜔 𝑥 0 

The problem is formulated as follows: at time 𝑡 = 0 the body has an initial angular
rate 𝜔(0) = 𝜔0 , we look for a control law which applies torques to the body based
on angular rate measurements 𝑇 = 𝑓 (𝜔) such that 𝜔 = 0 is an Asymptotically
Stable (AS) equilibrium, i.e., it is stable and lim𝑡→∞ 𝜔(𝑡) = 0.
Consider Euler’s equation (2.1) together with the feedback controller

𝑇 = −𝑘𝜔, 𝑘 > 0 (2.3)

which gives the closed-loop system

𝐼 𝜔¤ = 𝑆(𝜔)𝐼𝜔 − 𝑘𝜔. (2.4)

We need:
(a) to check if 𝜔 = 0 is an equilibrium for the system,

5
6 2 Exercise classroom 2

(b) to find a way to prove that the equilibrium is AS,

(c) to estimate if possible the region of attraction.
Detumbling performance for different controllers
We want to investigate how long does it take to stop the tumbling motion.
In the following, three different simulations are performed with MATLAB
Simulink and Simscape toolbox with:
5 10 0 0 
   
𝜔(0) = 5 , 𝐼 =  0 20 0  .
5  0 0 30
   
The three test cases are:
(a) 𝑘 scalar,
(b) 𝑘 ∝ 𝐼,
(c) symmetric saturation.
Chapter 3
Exercise classroom 3

3.1 Example of loop shaping design

Consider the SISO system described by the block diagram in Figure 1.1, with
10
𝐺 (𝑠) =
(𝑠 + 1) 2

design a regulator 𝑅(𝑠) such that the following requirements are met:
(a) the error at steady state for a step set-point input 𝑦 0 (𝑡) = step(𝑡) is 0,
(b) the percentage overshoot is less or equal than 20%,
(c) the settling time is less or equal than 5.5 s.
If needed modify the regulator in such a way that:
(d) the error at steady state for a sinusoidal disturbance input 𝑑 (𝑡) = sin(Ω𝑡) with
Ω = 1 rad/s is less or equal than 0.1.

3.2 Eigenvalue Assignment: Problem 1

Consider the system

     
0 1 0   𝑥1
𝐴= , 𝐵= , 𝑢 = 𝐾𝑥 + 𝑣 = 𝑘 0 𝑘 1
−1 −1 1 𝑥2

find 𝐾 such that the eigenvalues of the closed-loop are

Λ𝑜 = {−3, −4}

7
8 3 Exercise classroom 3

3.2.1 Eigenvalue Assignment: Problem 2

Consider the system

     
1 −9 7 2   𝑥1
𝐴= , 𝐵= , 𝑢 = 𝐾𝑥 + 𝑣 = 𝑘 0 𝑘 1
2 −15 13 4 𝑥2

find 𝐾 such that the eigenvalues of the closed-loop are

Λ𝑜 = {−5, −6}
Chapter 4
Exercise classroom 4

4.1 LQR design example

Given
   
0 1 0
𝑥¤ = 𝐴𝑥 + 𝐵𝑢 = 𝑥+ 𝑢
0 −1 1
∫ ∞  
10
𝑥 ⊤ 𝑄𝑥 + 𝑢 ⊤ 𝑅𝑢 𝑑𝑡, 𝑄 =


𝐽= and 𝑅=𝑟
0 00

(a) Find the optimal controller as a function of 𝑟,

(b) Compute the closed-loop characteristic polynomial as a function of 𝑟,
(c) Study the root locus as a function of 𝑟.

4.2 LQG design example: double integrator

Given the double integrator with position 𝑥1 and velocity 𝑥2

(
𝑥¤1 = 𝑥2
.
𝑥¤2 = 𝑢

(a) Compute the Linear Quadratic Regulator (LQR) with 𝑄 = 𝐼 and 𝑅 = 𝑟 𝑐 .

(b) Compute the closed-loop characteristic polynomial as a function of 𝑟 𝑐 and study
the stability.
(c) Study the root locus as a function of 𝑟 𝑐 .
(d) Study the damping and the step response as a function of 𝑟 𝑐 .
(e) Design an optimal state estimator from position measurement, i.e., Linear
Quadratic Gaussian (LQG), with 𝑄 = 𝐼 and 𝑅 = 𝑟 𝑜 .
(f) Find the state matrix of the complete closed-loop system.

9
10 4 Exercise classroom 4

4.3 Additional Exercise: Guaranteed margins for LQG

regulators

There are none.

In this example we analyse the paper written by John C. Doyle ”Guaranteed
margins for LQG regulators” in IEEE Transactions on automatic control, vol. 23, n.
4, 1978. In this paper Doyle demonstrated with a counterexample that there not exist
any guaranteed margins (robustness) for the full LQG (Kalman filter in the loop)
regulator.
Consider the following
        
𝑥¤1 1 1 𝑥1 0 1
= + 𝑢+ 𝑤
𝑥¤2 0 1 𝑥2 1 1
 
  𝑥1
𝑦= 1 0 +𝑣
𝑥2

where (𝑥1 , 𝑥2 ), 𝑢 and 𝑦 denote the usual states, control input and measured output, and
where 𝑤 and 𝑣 are Gaussian white noises with intensities 𝜎 > 0 and 1, respectively.
Let performance integral have weights
 
1 
𝑄 𝑐 = 𝑞𝐶𝑞 𝐶𝑞⊤ = 𝑞

1 1 , 𝑞>0
1
 
⊤ 1  
𝑄 𝑜 = 𝜎𝐶 𝜎 𝐶 𝜎 = 𝜎 1 1 , 𝜎>0
1
𝑅𝑐 = 1
𝑅𝑜 = 1.
Chapter 5
Exercise classroom 5

5.1 Uncertain models: Example 1

Construct 𝑊 (𝑠) for an uncertain system given by the transfer function

𝛾 𝜔2𝑛
𝑃(𝑠) =
𝜏𝑠 + 1 𝑠 + 2𝜉𝜔 𝑛 𝑠 + 𝜔2𝑛
2

with 𝛾 = 2 with uncertainty of ±30%, 𝜏 = 1 s with uncertainty of ±30%, 𝜔 𝑛 = 50

rad/s without uncertainty and 𝜉 = 0.25 with uncertainty of ±70%.

5.2 Uncertain models: Example 2

Construct 𝑊 (𝑠) for an uncertain system given by the transfer function

𝜇 1
𝑃(𝑠) =
𝜏𝑠 + 1 𝑠 + 1
with 𝜇 = 10 with uncertainty of ±20%, 𝜏 = 0.1 s with uncertainty of ±30%.
Moreover construct a regulator and verify graphically the condition for robust sta-
bility.

11
12 5 Exercise classroom 5

5.3 Mixed sensitivity synthesis

Linear model of the quadrotor lateral dynamics

A state space representation of the lateral dynamics of a quadrotor is given

𝑥¤ = 𝐴𝑥 + 𝐵𝑢
𝑦 = 𝐶𝑥 + 𝐷𝑢

where
𝑣   
  𝑝
𝑥 =  𝑝  , 𝑢 = 𝛿𝑙𝑎𝑡 , 𝑦=
𝜙 𝜙
 
 𝑌𝑣 𝑌 𝑝 𝑔  𝑌 𝛿     
    0 1 0 0
𝐴 =  𝐿 𝑣 𝐿 𝑝 0 , 𝐵 =  𝐿 𝛿  , 𝐶 =
    , 𝐷= .
 0 1 0 0 0 0 1 0
   
First of all we can define all the parameters,
Y_v_nom = -0.264; % [1/s] -> 4.837 %
Y_p_nom = 0; % [m/(s rad)]-> -
L_v_nom = -7.349; % [rad s/m] -> 4.927 %
L_p_nom = 0; % [1/s] -> -
Y_d_nom = 9.568; % [m/sˆ2] -> 4.647 %
L_d_nom = 1079.339; % [rad/sˆ2] -> 2.762 %
Uncertainties are given in terms of standard deviations 𝜎 (provided as percentage of
corresponding nominal values) assuming a Gaussian density for each parameter.

In this example the controller for the lateral dynamics of a quadrotor is tuned.

The controller used for the lateral dynamics of the quadrotor is composed by a
P - PID cascade. Where the outer proportional controller for the roll angle is

Fig. 5.1 Closed-loop of the quadrotor lateral dynamics.

𝑝 0 = 𝐾 𝑃, 𝜙 𝑒 𝜙
5.3 Mixed sensitivity synthesis 13

and the inner PID controller for the roll rate is

𝐾 𝐼, 𝑝 𝑠𝐾 𝐷, 𝑝
𝛿𝑙𝑎𝑡 = 𝐾 𝑃, 𝑝 ( 𝑝 0 − 𝑝) + ( 𝑝 0 − 𝑝) + (−𝑝) .
𝑠 1 + 𝑠𝑇 𝑓

The following steps are taken:

1. Build a design model and the corresponding uncertain model of the quadrotor
lateral dynamics;
2. Analyse nominal and uncertain models in terms of poles, zeros and frequency
response function;
3. Build a design model for the feedback system (P - PID);
4. Nominal design requirements: based on nominal performance requirements
write the weights to be used in the 𝐻∞ synthesis;
5. Tune the parameters P - PID using the nominal model based on nominal perfor-
mance requirements;
6. Robust analysis and design: verify robust stability with respect to uncertainties
associated with ±3𝜎 ranges on all parameters, if RS is not achieved redesign the
controller;
7. Robust performance: verify if the obtained controller satisfies RP condition.