Scribd
Upload
2 views

Optimal Control Notes

The document outlines exam questions for the International Master in Systems, Control and Information Technologies program from 2014 to 2019, focusing on topics such as Dynamic Programming, Linear Quadratic Regulators, and the Calculus of Variations. Each section includes specific problems requiring performance measures, Hamilton-Jacobi equations, and optimal control strategies. The exams emphasize thorough motivation for each solution step and allow the use of a double-sided page of notes.

Uploaded by

Eva Adhikary
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
2 views

Optimal Control Notes

The document outlines exam questions for the International Master in Systems, Control and Information Technologies program from 2014 to 2019, focusing on topics such as Dynamic Programming, Linear Quadratic Regulators, and the Calculus of Variations. Each section includes specific problems requiring performance measures, Hamilton-Jacobi equations, and optimal control strategies. The exams emphasize thorough motivation for each solution step and allow the use of a double-sided page of notes.

Uploaded by

Eva Adhikary
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 59

Optimization and Optimal Control

Exam 2014-2015 (S1), 1h30


E. WITRANT, UJF/UFR PhyTEM, December 11, 2015.
International Master in Systems, Control and Information Technologies (MiSCIT)

• Authorized documents: one double-sided page, to be handed back with your exam.
• Observations:
– each step of the solution has to be thoroughly motivated;
– write your name on each sheet and number them, including the total number.
– 7 points per question

I. Dynamic Programming
Consider the problem of bringing to rest at the origin in minimal time the state of the forced harmonic
oscillator ẍ + x = u using controls satisfying u ∈ [−1, 1].
For this problem:
1) define the appropriate performance measure;
2) give the corresponding Hamilton-Jacobi equation;
3) solve this equation to find u∗ (t).

II. Linear Quadratic Regulators


Consider the second order system:
ẍ + a1 ẋ + a2 x = bu
regulated with a PID controller:
Z t
u(t) = k p x(t) + ki x(η)dη + kd ẋ(t)
0

1) Extend the state to obtain the integral and check for controllability;
2) Design an LQR to find the optimal PID gains.

III. Calculus of Variations


Find the equation of the curve that is an extremal for the functional:
Z tf
1
J(x) = t ẋ(t) + ẋ(t) dt
0 2
for the boundary conditions:
1) t f = 1, x(0) = 1 and x(1) = 2?
2) t f = 2, x(0) = 1 and x(2) is free?
3) t f is free, x(0) = 1 and x(t f ) = 5?
Optimization and Optimal Control
Exam 2016-2017 (S1), 1h30
E. WITRANT, UJF/UFR PhyTEM, December 14, 2016.
International Master in Systems, Control and Information Technologies (MiSCIT)

• Authorized documents: one double-sided page, to be handed back with your exam.
• Observations:
– each step of the solution has to be thoroughly motivated;
– write your name on each sheet and number them, including the total number.
– 7 points per question

I. Dynamic Programming
Consider the problem of bringing to the desired final state in minimal time the state of the system
ẋ = ax2 + u using controls satisfying u ∈ [−1, 1].
For this problem:
1) define the appropriate performance measure;
2) give the corresponding Hamilton-Jacobi equation;
3) solve this equation to find u∗ (t).

II. The Variational Approach to Optimal Control Problems


Consider the second order system:
ẍ = xu
with x(0) = x0 and ẋ(0) = 0.
1) Define the performance measure to track the reference r(t) with limited control effort, fixed final
time and free final state.
2) Find the necessary conditions for optimality.
3) Find the optimal input such that |u(t)| ≤ 1.

III. LQ Controller
Consider a controlled object with the transfer function of the form
b1 s + b0
G(s) =
s2 + a1 s + a 0
where b0 = 2, b1 = 0.1, a0 = 4, a1 = 3.
1) Write this system in the state-space form.
2) Is this system controllable and observable?
3) Design a feedback LQ controller with integrator based on state observation (choose Q = I and R = 1
in the cost function).
4) What happens if b0 = 0?
Optimization and Optimal Control
Exam 2017-2018 (S1), 1h30
E. WITRANT, UJF/UFR PhyTEM, December 21, 2017.
International Master in Systems, Control and Information Technologies (MiSCIT)

• Authorized documents: one double-sided page, to be handed back with your exam.
• Observations:
– each step of the solution has to be thoroughly motivated;
– write your name on each sheet and number them, including the total number.

I. The Performance Measure (5 points)

Fig. 1. Edward diwheel from the University of Adelaide, Australia. Left: photograph of the diwheel driving in inversion mode. Right:
Schematic showing the coordinate systems, mass distributions and states. Exerpt from [1].

Consider the diwheel robot depicted on Fig. 1 where:


• θ is the rotation of the inner frame assembly about the z-axis;
• φ is the rotation of the wheels about the z-axis;
• x is the displacement of the diwheel (centre) about an earth-centred frame;
• τ is the torque applied to the wheels by a controlled motor.
For this system:
1) determine the set of physical contraints that should be considered;
2) propose performance measures to:
a) reach a target destination in minimum time in the inversion mode;
b) minimize the input usage and its rate of change;
c) track a reference velocity trajectory.

II. The Variational Approach to Optimal Control Problems (10 points)


The stability of the diwheel robot depicted on Fig. 1 can be controlled by controlling θ, the rotation of
the inner frame assembly about the z-axis. Using the small angle approximation, the dynamics of θ can
be simplified as:
θ̈ = a1 θ + a2 θ̇ + a3 θθ̇2 + bu(t) + cv(t)
2

where {a1,2,3 , b, c} are constant parameters, u(t) is the controlled input and v(t) is a known exogeneous
input (not controlled, related to φ̇). Derive an optimal controller for this system by considering that the
system is initially at rest, ends at rest at a free final time, and that the state and input trajectories minimize
the performance measure:
1 T 2
Z
J(u) = θ (t) + r u2 (t)dt
2 0
More precisely:
1) write the state-space description and the Hamiltonian;
2) obtain the necessary conditions;
3) check that you obtain a consistent number of equations to solve your problem.
Then propose a simplified controller by linearizing the system, considering a constant v(t) and an infinite
time-horizon.

III. State estimation and Kalman filter (5 points)


Consider the system described by the dynamics:
" # " # " #
3 1 1 1
ẋ(t) = x+ u + ζ x (t), x(0) =
2 0 0 0
h i
y(t) = 0 1 x(t) + ζ(t),
where ζ x (t) and ζ(t) are gaussian noises, uncorrelated with time or with the initial state. The variance of
these noises is supposed to be known.
Compute the Kalman filter that provides an estimate of x1 (t).

Reference
1 B. Cazzolato, J. Harvey, C. Dyer, K. Fulton, E. Schumann, T. Zhu, Z. Prime B. Davis, S. Hart, E.
Pearce and J. Atterton, ”Modeling, simulation and control of an electric diwheel”, Proceedings of
Australasian Conference on Robotics and Automation, Monash University, Melbourne, Australia, 7-9
Dec 2011.
Optimization and Optimal Control
Exam 2018-2019 (S1), 1h30
E. WITRANT, UJF/UFR PhyTEM, December 20, 2018.
International Master in Systems, Control and Information Technologies (MiSCIT)

• Authorized documents: one double-sided page.


• Observations:
– each step of the solution has to be thoroughly motivated;
– write your name on each sheet and number them, including the total number.

I. Dynamic Programming (5 points)


2

II. The Variational Approach to Optimal Control Problems (10 points)

III. Optimization Methods (5 points)


Explain the Levenberg-Marquardt algorithm.

You might also like