Vous êtes sur la page 1sur 5

Possible answers to open questions

MOOC-TP, v. 2015.3

1.1) Is the linear range similar when negative voltages are applied?

With this kind of electric drive, the system should behave independently of the sign of the applied
command voltage.

We can assume that our system will give the same response but with opposite rotational direction if we
apply a negative voltage. This is due to the fact that the static friction and the behavior of the motor have
symmetrical behaviors.

On the other hand the software saturation (+/- 5[v]) is implemented for security reasons, could have
different min and max values, for ex. +5[v], -4[v]

1.2) Does the step amplitude influence identified parameters?

Identified parameters will remain the same if the command (voltage) is within the identified linear
interval. The linear interval is typically between U0min and Umax (with U0min = 0.5 - 0.8[v] and Umax
= 5[v])

If we apply a step from U0min with an amplitude exceeding Umax (5[v]), the identified static gain K
will be lower. This can be explained by the fact that the command voltage is greater than Umax but the
steady state output is the same as if we had applied Umax (saturated). Thus, the ratio between the steady
state output value and the applied command decrease (K = Yss/Uc).

Note: The step amplitude should be as large as possible within the linear range, if you apply a too small
step (ex. from 1.1[v] to 1.2 [v]) the sensor noise will hide/bias your measurement.

2.1) In closed-loop with the PI controller in speed mode. In experimentation 1, why not applying a
step that uses the full linear range?

In module 1 the identified static gain of the system is ~0.8 and the saturation is at 5[v]. Thus the
maximum speed at which the motor can turn is 5 * 0.8 = ~4 [v]. In other words, a reference of 5[v] is not
reachable, whatever the controller.

1
2.2) In experimentation 2, can the PI controller remove a steady-state error when applying a ramp?

The PI controller is not able to remove the steady state error when applying a ramp.

The error E(s) is expressed as: 1


E(s) = Y (s)
1+ K (s)Gω (s) c

The steady state error is defined as:


ess = lim e(t)
t →∞

= lim sE (s)
s →0

Given the Laplace transform of the ramp input:

A where A is the slope of the ramp (constant).


Yc (s) =
s2

Given also the transfer functions of the controller K(s) and of the physical system G(s):
τ ⎛ 1 ⎞
K (s) = 1+
γ ωτ m ⎜⎝ τ .s ⎟⎠
1
= K p (1 + )
Ti .s
γω
Gω (s) =
1 + τ .s

The steady state error can be computed as following:


A
s
ess = lim s2
s →0 1 + K (s) G(s)

A
= lim
s →0 1
s (1 + )
τ m .s
A
= lim
s →0 s +1/τm
ess = A.τ m ≠ 0

The theory shows that for the given plant model, a PI speed control cannot compensate the steady-state
error when the input is a ramp.

2.3) In experimentation 2, when decreasing Kp to 0.5, did you observe a change in the steady-state
error? Why?

When reducing Kp to 0.5 (divided by ~2), we can observe an increase of the steady-state error when a
ramp is applied. Intuitively we see that the controller reacts less aggressively to the error.

In order to show the increase of the steady-state error we just have to recalculate ess the same way as in
τ τ
question 2.2 but with K p = instead of K p = previously.
2γ ωτ m γ ωτ m
Thus ess =2 Aτ m , the error is twice as big.

2
3.1) In experimentation 1, did you observe a difference between the computed U0 and the one found
experimentally? Why?

You should observe a difference since the formula used to compute ū assumes that there is no friction
while the value obtained experimentally includes the friction.

3.2) In experimentation 2, suggest a way to compute the feed forward command U0 such that it works
for both values proposed (2.5[v] and 0.5[v])?

In the experiment 1 & 2 we computed a feedforward command for a given reference value Yc. If this
reference value changes, the feed forward value should change accordingly. This solution induces a
slight change in your controller architecture where we add a block to compute U0 as a function of the
reference Yc (only valid for this configuration). Note that the feed forward command does not consider
potential perturbations, the controller should compensate for perturbations or model miss match.

3.2) In experimentation 4, assuming that the command can take a very large (infinite) value, would
the ARW measures improve the system response?

The Anti-Reset Windup measures are only active when the command is saturated, so if there is no
saturation, the ARW measure will never be active and neither improve nor worsen the system response.

4.1) In experimentation 2, why a PID controller can follow a ramp signal without steady state error
(trainée) if in position mode?

A PID controller can follow a ramp signal without steady state error (trainée) if in position mode since
we added an integrator by measuring potion instead of speed as in question 2.2. The steady state error
can be computed the same way as in Q. 2.2. Note that G(s) will have an additional integrator:

γθ
Gθ (s) =
s(1 + τ .s)

3
3.2) In experimentation 3, when turning the controller I & D terms off, is the resulting P controller
able to follow the applied square signal without steady state error? Why?

A proportional controller is sufficient to track a square signal without steady state error. As in Q. 3.1
there is an integrator within the system since we measure the position.

Note: in practice we often add an integral term to remove the potential offset that may be induced by
friction.

4.3 In experimentation 5 (position mode), when turning the controller I term off, would the controller
still be able to reject the constant perturbation if Kp is increased?

The constant perturbation is added to the controller value. Thus the error is defined as:

Yc (s) + U 0Gθ (s)


E (s) =
1 + K (s) Gθ (s)

ess = lim sE (s)


s →0

U 0 (s) Gθ (s)
= lim s (If we consider Yc (s)=0 )
s →0 1 + K '(s) Gθ (s)
U0 γ θ
= lim s s s + τ .s 2
s →0 γ
1+ Kp θ
(1 + Td s )
s + τ .s 2
U 0γ θ
= lim
s →0 s + τ .s 2 + γ K (1 + T s )
θ p d

U0
=
Kp
≠0 if U 0 ≠ 0

Increasing Kp would reduce the error but not remove it. Increasing Kp to a large value may also cause
stability issues for a slight change in the reference.

4
5.1) In experimentation 2, did you obtain the same values as in Module 1?

You should get the same identified parameters (to the precision of the tool) since we are measuring the
same values but with different tools.

5.2) In experimentation 4, what is special with a -3[dB] attenuation?

In the above Bode plots, the -3dB attenuation is reached when the frequency of the reference is
ω
f c = c [Hz] ωc = 3.5 [rad/s]. This frequency is called the cut-off frequency and corresponds to the

frequency at which the output power is half the input one.

If we define the power as the square of the amplitude, the output amplitude at low frequencies is equal to
the static gain (~0.8) times the input amplitude, at the cut-off frequency, the output amplitude is
~0.8/sqrt(2) times the input amplitude.

Vous aimerez peut-être aussi