Sol6 PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

EECS 16B Designing Information Devices and Systems II

Fall 2017 Miki Lustig and Michel Maharbiz Homework 6


This homework is due October 17th, 2017, at noon.

1. Otto the Pilot


Otto has devised a control algorithm so that his plane climbs to the desired altitude by itself. However he
is having oscillatory transients as shown in the figure. Prof. Arcak told him that if his system has complex
eigenvalues
λ1,2 = v ∓ jω
then his altitude would indeed oscillate with frequency ω about the steady state value, 1 km, and that the
time trace of his altitude would be tangent to the curves 1 + evt and 1 − evt , near its maxima and minima
respectively.
Altitude (km)

1 + evt
1.4843

1
0.7654

1 − evt
Time (min)
5 10 15 20

(a) Find the real part v and the imaginary part ω from the altitude plot.
Solution: Solving 1 + e5v = 1.4843 gives us v = −0.1450 min−1 . Then, comparing the maxima that
are separated by an interval of 10 minutes gives ω = 2π
10 = 0.62832 rad/min.
If you solved in units of s and rad/s, then v = −0.0024 s−1 and ω = 0.0105 rad/s.
−1

(b) Let the dynamical model for the altitude be


    
d y(t) 0 1 y(t)
=
dt ẏ(t) a1 a2 ẏ(t)

where y(t) is the deviation of the altitude from the steady state value, ẏ(t) is the time derivative of y(t),
and a1 and a2 are constants. Using your answer to part (a), find out what a1 and a2 are.
 
0 1
Solution: The eigenvalues of A = are given by 0 = λ 2 − a2 λ − a1 , or equivalently
a1 a2
q
a2 ∓ a22 + 4a1
λ= = v ∓ jω.
2

EECS 16B, Fall 2017, Homework 6 1


Solving for a1 and a2 (using the min−1 and rad/min values of v and ω) we get
a22
a2 = 2v = −0.2900 and a1 = −ω 2 − = −0.4158.
4
If you solved using the s−1 and rad/s values of v and ω, then
a22
a2 = 2v = −0.0048 and a1 = −ω 2 − = −1.16 ∗ 10−4 .
4

(c) Otto can change a2 by turning a knob. Tell him what value he should pick so that he has a “critically
damped" ascent with two negative, real eigenvalues at the same location.
Solution: To get two real identical eigenvalues Otto should choose a2 to make a22 + 4a1 = 0. This

means a2 = ±2 −a1 . Since a2 must be negative for the system to be stable, we only look at the
negative root.
Solving with the a1 derived from the min−1 and rad/min values of v and ω, he should tune his knob to
√ √
a2 = −2 −a1 = −2 0.4158 = −1.2897.

If you solved using a1 derived from the s−1 and rad/s values of v and ω, then you get
√ √
a2 = −2 −a1 = −2 1.16 ∗ 10−4 = −0.0215.

2. LED Strip
I have an LED strip with 5 red LEDs whose brightnesses I want to set. These LEDs are addressed as a
queue: at each time step, I can push a new brightness command between 0 and 255 to the left-most LED.
Each of the following LEDs will then take on the brightness previously displayed by the LED immediately
to its left.

(a) What should we use for our state vector? What does it mean that this is a state vector? What is our
input?
Solution: We can use the brightnesses of each LED as our state vector. We can use these values as
our state vector since together with the input, they describe everything about our system that we need
to know in order to predict what our system will do in the future. Our input is the command to the
left-most LED.
(b) Is our system linear? If it is linear, write out the state equations in matrix form. Please choose a
reasonable order for the state variables in the state vector.
Solution: The system is linear because it can be written in the form ~x(t + 1) = A~x(t) + Bu. Ordering
the LED brightnesses in the state vector from left to right, we get:
   
0 0 0 0 0 1
1 0 0 0 0 0
   
~x(t + 1) = 
0 1 0 0 0~x(t) + 0 u(t)
  
0 0 1 0 0 0
0 0 0 1 0 0
If you chose to put the left-most LED’s brightness last in the state vector (so that the LEDs are ordered
right to left and the state vector gets flipped upside down), the A matrix gets transposed and the B
matrix is flipped upside down.

EECS 16B, Fall 2017, Homework 6 2


(c) Is this system controllable? Explain intuitively what this system’s controllability means in terms of
LED brightnesses.
Solution: Testing for controllability we see:
 
0 0 0 0 1
0 0 0 1 0
 4   
A B A3 B A2 B AB B =  0 0 1 0 0

0 1 0 0 0
1 0 0 0 0

which has full rank. This means that the system is controllable. A system is called controllable if from
any initial state, we can reach any final state that we desire at some time in the future.
For our LED strip, controllability means that we can display any set of brighnesses that we desire, but
it may take a few time steps to get there.
(d) Is this system stable?
Solution: The eigenvalues of this discrete-time system are all 0, which is inside the unit circle.
Therefore, the system is stable.
(e) Starting from the pattern of brightnesses (from left to right) [0, 127, 0, 255, 0], can we maintain this
pattern for all future time steps? Can we display any fixed pattern of brightnesses for all time?
Solution: We cannot display [0, 127, 0, 255, 0] for all time. Immediately after we display this set of
brightnesses, we will display [u(1), 0, 127, 0, 255].
If we want to display a fixed and unchanging set of brightnesses, every element in our state vector must
be the same.
Controllability tells us only that we can reach any desired state (sometimes only temporarily). It does
not mean we can keep our system at any desired state for all time.

3. Controllability and discretization


In this problem, we will use the car model

d
p(t) = v(t)
dt
d
v(t) = u(t)
dt
that was discussed in class.

(a) Assuming that the input u(t) can be varied continuously, is this system controllable?
Solution: Introducing states x1 = p and x2 = v, we rewrite this system in state space form
   
0 1 0
ẋ(t) = Ax(t) + Bu(t) = x(t) + u(t).
0 0 1

The controllability matrix  


0 1
C = B AB =
 
1 0
has rank 2. Therefore the continuous-time system is controllable.

EECS 16B, Fall 2017, Homework 6 3


(b) Now assume that we can only change our control input every T seconds. Derive a discrete-time state
space model for the state updates, assuming that the input is held constant between times t and t + T .
Solution: By integrating both sides of the second equation from t to t + T and keeping in mind that
u(t) is constant in this interval, that is u(t + τ) = u(t) for τ ∈ [0, T ):
Z T
v(t + T ) − v(t) = u(t + τ)dτ = Tu(t).
0

Now integrating the first equation and using the fact that v(t + τ) = v(t) + τu(t) we get
Z T
1
p(t + T ) − p(t) = (v(t) + τu(t))dτ = T v(t) + T 2 u(t).
0 2
Introducing states x1 (k) = p(kT ) and x2 (k) = v(kT ) we get the state space model
   1 2
1 T T
x(k + 1) = Ax(k) + Bu(k) = x(k) + 2 u(k).
0 1 T

(c) Is the discrete-time system controllable?


Solution: The controllability matrix
1 2 3 2

C = B AB = 2T 2T
 
T T

has rank 2. So the discrete-time system is controllable.

4. Controllability in circuits
Consider the circuit in Figure 1, where Vs is an input we can control:
C1 C2

+ − + −
V1 V2
+ V
− s R

Figure 1: Controllability in circuits

(a) Write the state space model for this circuit.


Solution:

Vs −V1 −V2 dV1 dV2


I= = C1 = C2
R dt dt
  " 1 #  " #
d V1 − RC1 − RC1 1 V1 1
= 1 1 + RC1 1 Vs
dt V2 − RC2 − RC2 V2 RC2

EECS 16B, Fall 2017, Homework 6 4


(b) Show that this system is not controllable.
Solution: If we calculate AB, we find that it is a linear combination of B:
" #
− RC1 1 ( RC1 1 + RC1 2 ) 1 1
AB = = −( + )B
− RC1 2 ( RC1 1 + RC1 2 ) RC1 RC2

This means that the controllability matrix


 
B AB

Must have rank of 1. Therefore, this system is not controllable.


(c) Explain, in terms of circuit currents and voltages, why this system isn’t controllable. (Hint: think about
what currents/voltages of the circuit we are controlling with Vs )
Solution: We can only control Vs , which in turn controls the amount of current flowing through
the circuit. Since this current is equal through both capacitors and current directly affects the voltage
across a capacitor, there is no way to individually control the voltages across the capacitors.
(d) Draw an equivalent circuit of this system that is controllable. What quantity can you control in this
system?
Solution:
C3

+ −
V3
+ V
− s R

We can control V3 in this circuit.

5. Controllability in 2D
Consider the control of some two-dimensional linear discrete-time system

~x(k + 1) = A~x(k) + Bu(k)

where A is a 2 × 2 real matrix and B is a 2 × 1 real vector.


   
a 0 f
(a) Let A = with a, c, d 6= 0, and B = . Find a B such that the system is controllable no matter
c d g
what nonzero values a, c, d take on, and a B for which it is not controllable no matter what nonzero
values are given for a, c, d. You can use the controllability rank test, but please explain your intuition
as well.    
1 1 a
Solution: With B = , the system is controllable for all nonzeros a, c, d, because [B, AB] = ,
0 0 c
   
0 0 0
which has full rank. With B = the system is not controllable because [B, AB] = , which
1 1 d

EECS 16B, Fall 2017, Homework 6 5


only has rank=1. The intuition is that, due to the zero entry in A, the state x1 evolves autonomously, i.e.,
d
dt x1 (t) = ax1 (t) , hence it needs to be controlled by some input f . On the other hand we can control x2
via controlling x1 , as dtd x2 (t) = cx1 (t) + dx2 (t), which implies that x2 can be “tuned” by manipulating
x1 .
   
a 0 f
(b) Let A = with a, d 6= 0. and B = with f , g 6= 0. Is this system always controllable? If not,
0 d g
find configurations of nonzero a, d, f , g that make the system uncontrollable.
Solution: No. uncontrollable when a = d. In this case the matrix is just a constant a times the identity.
So when you check with the controllability test, AB is just a scalar multiple of B and hence linearly
dependent. The intuition is that the two states are inherently “coupled” as two eigenvalues are the
same. Any control input can only move the states along a line hence the states cannot reach arbitrary
points in R2 .
(c) We want to see if controllability is preserved under changes of coordinates. To begin with, let~z(k) =
V −1~x(k), please write out the system equation with respect to~z.
Solution: ~x(k) = V~z(k), hence we have
V~z(k + 1) = AV~z(k) + Bu(k)
~z(k + 1) = V −1 AV~z(k) +V −1 Bu(k)
.
(d) Now show that controllability is preserved under change of coordinates. (Hint: use the fact that
rank(MA) = rank(A) for any invertible matrix M.)
Solution: The matrix whose rank needs to be tested after the coordinate change is [V −1 B,V −1 AVV −1 B] =
[V −1 B,V −1 AB] = V −1 [B, AB] which has the same rank as [B, AB], since V by assumption is full rank.

6. Understanding the SIXT33N car control model


As we continue along the process of making the SIXT33N cars be awesome, we’d like to better understand
the car model which we will be using to develop a control scheme. As a wheel on the car turns, there is an
encoder disc (see below) which also turns as the wheel turns. The encoder shines a light though the encoder
disc, and as the wheel turns, the light is continually blocked and unblocked, allowing the encoder to detect
rotations of the wheel per ’tick’ of the encoder disc.

The following model applies separately to each motor of the car:

v(t) = d(t + 1) − d(t) = θ u(t) − β

Meet the variables at play in this model:

EECS 16B, Fall 2017, Homework 6 6


• t - The current timestep of the model. Since this is a discrete system, this will advance by 1 on every
new sample in the system.
• d(t) - The current number of ticks advanced by this wheel.
• v(t) - The discrete-time velocity (in units of ticks/timestep) of the wheel, measured by finding the
difference between two adjacent tick counts (d(t + 1) − d(t)).
• u(t) - The input to the system, in terms of "PWMs". As we will observe in lab, the circuit driving
the model controls the amount of power delivered to the wheel in units of PWM intensity. This is
a number between 0 and 255, where 0 means no power is delivered to the motor and 255 means
maximum capable power is delivered to the motor.
• θ - In units of "ticks/PWM", this models how much more the motor turns for every increase in PWM.
This is empirically measured from the car.
• β - In units of ticks/timestep, this models the effect of friction on the car (if no power is applied to the
motors, we’d expect the velocities of the motors to decrease). This is empirically measured from the
car.

In this problem, we will assume that the motor conforms perfectly to this model to get an intuition of how
the model works.

(a) If we wanted to make the motor drive at a certain target velocity v∗ , with what PWM u(t) should we
feed the motor?
Solution:

v∗ = θ u(t) − β
v∗ + β = θ u(t)
v∗ + β
u(t) =
θ

(b) What signs should θ and β have? Should they be positive or negative? Note that applying PWM to the
motor driver circuit can only ever deliver power in a way so as to cause the motor to move forwards
and never backwards, and there are no braking mechanisms on the motor.
Solution: θ > 0 since applying PWM should only ever increase the velocity. If friction is a large
factor in the linear fit for our motor model, we expect that β should be greater than zero since fric-
tion should decrease the speed of the car in the absence of PWM input. However, nonlinearities and
imperfections in our motors may outweight the effect of friction so that β may experimentally end up
positive.
(c) Even if the motor conforms perfectly to the model, our inputs still limit the range of velocities of the
motor. Given that 0 ≤ u(t) ≤ 2551 , determine the maximum and minimum velocities possible with the
motor. What does this tell us about the braking of the car?
Solution: The maximum is θ 255 − β and the minimum is 0 − β = −β .
Since there are no brakes on the motor, we slow down by reducing the PWM.
(d) Our intuition tells us that a motor on a car should eventually stop turning if we stop applying any power
to it. Find v(t) as t → ∞, assuming v(0) = v0 (say v0 > 0) and u(t) = 0. Does our model obey our
intutition? What does that tell us about our model?
1 See https://www.arduino.cc/en/uploads/Tutorial/pwm1.gif for an example of how PWM works and why
this is the case.

EECS 16B, Fall 2017, Homework 6 7


Solution:
v(∞) = −β

However our intuition says that the motor should have stopped: v(∞) = 0. In lab we empirically find
the value of β over a range of PWM values, but our fit does not work very well everywhere and our
model does not match the real behavior near u = 0.

Stay tuned for closed-loop control of the SIXT33N motors!

Contributors:

• Murat Arcak.

• Justin Yim.

• John Maidens.

• Brian Kilberg.

• Yuxun Zhou.

• Edward Wang.

EECS 16B, Fall 2017, Homework 6 8

You might also like