Professional Documents
Culture Documents
```````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````
`````````````````````````````````````````````````
==
n f the more i study the more i dont know… me too
keep at it mate
Lost cause
life aint shit but a fa
Sorry for deleting stuff. goddamnit i forgot to download and edit.
2011,2012
(your name below)
| Zhiyuan Li | Many Past students! | Winston Ng has no |
| |
Please Use a different color for any of your text answers
Try to keep this document clean, preferably make use of the comment feature, please don’t
delete any work without a good consensus.
++++++++++++++++++++++++++
Also, if it’s easier just drag and drop images of handworking, etc
….A collaborative Effort by Control Systems 2014 and 2013 Facebook group
I may tidy up this more in the future, splitting the document into 2 probably doesn't help
much as the comments in this document should be preserved. Anyway the links should
help make it look better. - Zhiyuan
Keep these solutions up, i’ll be needin them next year for
when i repeat - YES!
IS that facebook bloke just referencing the wrong shit or is classical loop shaping not
just pole placement like the stuff on the past 2 exams? From what i can tell the integral
on that page is basically just showing a result; that if you have too many poles around
cross over (high roll off) you risk fucking with the phase margin (you probably dont need to
understand the maths, because if i remember corrently he just basically skimmed over it. if we do then its
“bode gain-
not really testing anything control related so might onoly be worth a couple of marks at most ).
phase tradeoff between slope at crossover and phase margin limits proximity of LOW
and HIGH frequency range, if too close specs may not be acheivable”the dot point down
the bottom is saying as you pass over the cross over you dont want anything beyond a -
20dB/dec. So to answer your original question, pole placement (and maybe constraints
on other parameters such as PM and w_c) would be likely asked of us. YES.
essment
(Appear to be 100% Identical):
1.i. Most likely correct, equations
100
𝑣0 (𝑡)(𝑠𝑡𝑒𝑎𝑑𝑦 𝑠𝑡𝑎𝑡𝑒) 1+100𝐹 100
= =
𝑣𝑖 (𝑡) 1 1+100𝐹
F = 1 @ Vo = 100/101
F = 9/100 @ Vo = 10
Solution:
9/100 < F < 1 I got this i got this too Commented [1]: i got this
For F = .1, the plot is translated down by the difference of the DC gains (20 dB).
DC gain = 40 @ F = 1
DC gain = 20 @ F = .1
Solution:
@ F = 1, wc = 100 (99.5 Matlab), PM= 12.5 (11.5 Matlab)
@ F = .1, wc = 30 (30 Matlab), PM =37.5 (36.9 Matlab)
I understand we need to use F somehow but how does the Transfer function look?
1.vi. May be errors, no way to check
Find the canonical equation to find damping and cs4natural frequency, use these to find settling
time, overshoot and steady state value/error.
You can figure it out through MATLAB step response: step(tf([...], [....])). Numbers of overshoot
step response match that with step response and settling time in MATLAB.
Solution @ F = .1
Damping = .301, wn = 33.167, approx. 37% overshoot, ts = 4/(.301*33.167) = .4 seconds, steady
state value = 10000/1100, steady state error of 1/11
Do we take in to consideration the negative C(s)? Why is there a negative ? <to make our life
easier so instead of subtracting the dotted line from the solid line we just have to add them. Not
sure if its actually easier tho.(negative means -180 of the original phase, mag doesnt change
with negative sign, so we just think the original one start from zero, and add them up. it is
easier~)
How do u tell which pair corresponds to which line? Less K = slower rise time?
Not sure either
- If you expand C(s) out you get a zero=-1/tz and pole = -1/tp the pole/zero combinations
at the lowest frequencies correspond to an earlier ‘rise’ in the frequency domain, thats
how you tell the systems apart. (there may be a quicker way to determine this)
Solution:
The third pair, (ammamatlab)
As taup is bigger than tauz, 1/tauz will be greater than 1/taup, therefore it is possible for lines
approaching infinity on the root locus.
.03 <<< what is this 0.03?
at some point.
2(ii) CAN ANYONE DO THIS>!?SDAPOI!Y#$%^&*&^%&**^%&*( Commented [4]: How do the poles at s=0 affect the
As per fb comment: stability?
sketch a rough bode plot it helps. the two zeros start at 0 give a phase rise of 180, and the two poles start at 0 give a Commented [5]: Can we just say a vector from the
shift of -180. The repeat ed poles at the origin contributed a constant -180. Since the zeros occur before the origin to the intersect of a circle of radius 1, and nyquist
poles (wn^2 = k vs wn^2 = (k + mg/L) ), the overall phase is -180 for small omegae, a small rise due to zero's and thn plot, produces an angle of greater than 45deg? see
section 3 slide 33.
the poles give a drop back to -180 for high frequency oma.eg
Commented [6]: an angle greater than 45 deg
between it and the imaginary axis
Therefore, for increasing w, it starts at -180, and then goes up sligthly, (lets say -178 for arguments sake). since this
is between 0 and -180, we have moved from left to right along the BOTTOM part of the nyquist curve.
Hope that makes sense, Michael explained it for me in a consultation today Commented [7]: So to sum at this part, we have a
nyquist plot that goes in the clockwise direction, but it
does not encircle the -1 point. However, if we increase
gain it will end up enclosing the -1 point. Right?
Commented [8]: no cause the plot loops back around
itself i think (i.e. one of the dotted lines on the outside).
it goes from the w=-0 at the top left and rotates
anticlockwise to w=+0 at the bottom left in a D shape
curve. Let me know what you think
Commented [9]: Hmmm, nah i got the opposite. I've
uploaded the picture here below. To account for the
pole at the origin you get a half-clockwise rotation (two
in this case because two poles at origin). And, my
statement above is wrong. I think no matter how much
you increase gain the system'll still be stable. Let me
know what you think?
Commented [10]: I agree with you on the not
becoming stable bit :) not sure on your direction
though. I used what was written on the sheet, which
says that since the zeros kick in before the non-origin
poles, there will be an increase in phase (leading to a
rotation clockwise from -180 degrees) and when you go
around the origin poles, wont they each contribute a
half-COUNTERclockwise rotation since they are poles?
or am i missunderstanding...
Commented [11]: lol scratch that! we have the same
answers hahaha. i think we are just talking about it
slightly differently
Commented [12]: Alright, nice! (i've got each pole at
the centre gives a half-clockwise rotation btw, but if
your answer is the same as my figure below then we
have the same thing haha)
This is not exponentially stable as an pole zero cancellatio has occurred at s = 0. When a
pole/zero combination is cancelled out, it can automatically be factorized into the closed loop
polynomial. This leaves us with a closed loop pole at s = 0 , and thus, the system is only
marginally stable.
I am not sure if u can use gamma^2 < 4mk here, coz in that case gamma will be a complex
number
3.i. Output should be delta theta, messes everything up → isnt it already delta_theta??
anyone got the right block diagram/the eqns necessary to se
t one up????
-180+30 = phase(p0(.2*5*sqrt(3)j+1)/((l1*5*sqrt(3)j+1)((5*j*sqrt(3))^2-25)))
L1 = 1/15
p0 = 100/√3 = 57.74 Commented [18]: yeh I have no idea how I get exactly
half.. I don't know how to include the u2 caus that's
prob where im going wrong...
Just about to say I got the same results -Winston
Steady state value = 0.0305 → Definitely the right value!
got same
I am getting steady state value 0.01527 with the inclusion of 𝜇2= 0.5
what is your tf? → im using theta/D = u2*G/(1+GC), probably the same as what you have
yep :) sweet
Input Disturbance Sensitivity = Sio(s) =
G(s) / [ 1 + G(s)C(s) ] = B(s)L(s) / [ A(s)L(s) + B(s)P(s) ]
Using the Final Value Theorem
Steady State Value =
lim s * Sio(s) * D(s) , where D(s) = Disturbance Step Signal = 𝜇2/𝑠
(S->0+) .
Thus, Steady State Value = 𝜇2 ∗ 𝑆𝑖𝑜(0)
3.iii.
Dominance of second order dynamics for higher number of poles happens when all other poles
are fast compared to the s+
We know that the open loop transfer function has poles at s=±5and s= -15, but what we need is
the closed loop poles in order to determine which component dominates the response
(exponential decay or 2nd order dynamics). Proble m is how do we solve a cubic equation by
hand?
The characteristic polynomial is : f
(s^2 - 25) ( L1s + 1 ) + p0 ( 0.2s + 1 ), with L1 = 1/15 and p0 = 57.735
Solving for the roots of the equation will give the closed loop poles.
Expanding the (s^2 - 25) as (s - 5)*(s + 5) and factoring out a (s + 5) from (0.2s + 1) yields :
(s + 5) * [ (s - 5)*(L1s + 1) + p0 * 0.2 ] ,
Substituting values in and solving gets poles at :
s = -5, s = -5 + 8.556i, s = -5 - 8.556i.
And so we can infer that as the real pole has a real part equal to the real part of the
complex conjugate pair, they decay at the same rate.
And as they decay at the same rate, the second order complex pair remains and thus the
response has second order dynamics. :)
F
3.iv.a.
𝐴𝑐𝑙 = 5𝑙1 𝑠 3 + 𝑠 2 (5𝑙0 − 25𝑙1 ) + 𝑠(−25𝑙0 + 𝑝1 ) + 𝑝0 this is the denominator of the closed loop
tf i.e. the characteristic polynomial (you must cancel the (0.2s+1) with (s^-
25)=5(0.2s+1)(s-5)., therell be three poles left; thats what is meant by three free cl poles
Matching coefficients gives:
𝑙1 = 0.2, 𝑙0 = 13, 𝑝1 = 925, 𝑝0 = 5000 sorry guys, forgot a 5 in my answer, therefore, i got the
same result as below (updated 14/6 3:02pm)
I got 𝑙1 = 0.2, 𝑙0 = 13, 𝑝1 = 925, 𝑝0 = 5000, confirmed by matlab to give same roots as given
characteristic equation.
If you do it using the matrix and using the correct one on lms then this is how to do it.. :
Commented [19]: its way quicker to the ol' fashioned
way
3.iv.b.
We must ensure the input generating polynomial is contained in the denominator of the
pole-placement controller.
For a step:
𝑥 ̇ = 0; 𝑠𝑋(𝑠) − 𝑥(0) = 0; 𝛤 = 𝑠
Since the controller contains s as a pole, the step input disturbance is cancelled.
4.i. w
Root locus B = Step response C due to undershoot from RHP zero Commented [20]: overshoot?
Root locus C = Step response A due to oscillations?...ya Commented [21]: undershoot caus the settle is in
Root locus A = Step Response B opposite direction to shoot.
Is this not…?
pole/zero plot C = step response A (oscillations in step response due to dominant complex
poles)
pole/zero plot B = step response B (RHP zero
therefore undershoot)
pole/zero plot A = step response C (dominant LHP zero causes overshoot)
4.ii.
Write out the characteristic polynomial and let s = w - 2. All constants have to be positive,
leading to conditions K>6 and K>0. Since degree three, also need to do routh array which leads
to additional condition K>7.5. The final result is simply K>7.5. K>7.6666
4.iii.
Angle of Open Loop TF at s=3j = 180 - 45 = 135 degrees
Magnitude of Open Loop TF at s=3j = 1.
I’m pretty sure this is a starting point but not sure where to go from here?
*At 3 rad/s we can afford to lose 15 deg or 15*pi/180 rads of phase.
[15*pi/180 rads] / [3 rad/s] = 5*pi/180 seconds delay (maximum)
5*pi/180 = 2x
x = 5*pi/360 (metres)
Got the same result with a similar approach: -Winston
One anticlockwise encirclement, therefore stability for 2<K<(1/0.136) Commented [23]: how does this work? is it just
reading off from the plot ?
Commented [24]: _Marked as resolved_
Commented [25]: _Re-opened_
Commented [26]: If you expand the Nyquist plot out
by a factor greater than 1/0.136, you'll get zero net
encirclement due to the double loop.
2011 Exam: It is same as JAN 2012 below... Not quite
1.i.
Vo = A*(Vi-B*Vo)
Vo + A*B*Vo = A*Vi
Vo/Vi = A/(1+A*B)
1.ii.
B = Zm/(Zm+Zf)
Zf = 0
B = Zm/Zm
B=1
Poles at s = - 100 and s = - 1,000,000 ← I think pole should be s=-1,000,000,000 (45 degree
PM, as top bold line in mag. plot crosses over at 109 and phase is -135 degrees)
From looking at the changes in decline, it is the top magnitude and bottom phase plot.
Phase margin = 45 degree s
sub in jw and find magnitude and phase. To find the bode graph.
cross over frequency when mag=0. go down to 180 degrees. Find difference between plot and
180.
1.iii.
SS = lim t=>infinity which is S=>0 (sVo) when Vi=1/s
Vo=K/[(t1s+1)(t2s+1)+k] (ignored Vi as it cancels with sVo using FVT)
so error 1-(K/1+K) which is what is also below...
E(s) =Vi-B*Vo how did you get the error signal - i had Vi/(1+A(s)) gg- Vi goes to 1?
B=1
E(s) = (1-Vo/Vi)Vi
E(s)=(1-A/(1+A))Vi from 1.i
E(s)=(1/(1+A))(1/s) The Vi isnt 1 but 1/s i would think, that it vanished from final value theo
E(t)=(1/(1+A))(s/s) as s-->0
E(t) = 1/(1+14142)
Error = 1/14143
Error = 7.07x10^-5....
I got steady state error of 7.07 *10^-8 (i m with u ⇐ me 2<= perhaps numbers are different
between jan 2012 a nd 2011
Percentage overshoot = approx 23 % - can someone elaborate on this answer please? use your Commented [27]: I believe it comes from previous
pm go to page 8 and look @ curve.pm = solid. look down @ dashed and bam its 23 degrees phase margin using the overshoot graphs
Vo/Vi= ______K_______ so from expanding wn=sqrt(K+1/t1*t2) where i divided t1*t2
[(t1s^2+1)(t2s+1)+k] thus 2wn damp = (t1+t2)/t1*t2 rearange and damp = 0.42
1.iv.a.
The case of B = 1 is the previous case, so use the bode plot from 1.ii. As the poles approach -
180 exponentially, they never really cross it, therefore gain margin is infinite.
there is three poles in the transfer function, therefore it should cross -180 i believe. can
someone help???? Commented [28]: not in the nominal case
1.iv.b.I
C=10^-7 is the plot that stays at 0 magnitude and phase for the longest before dropping down.
To find the gain margin of AB, find the magnitude at the frequency where the sum of the phases
of A and B are -180. This is roughly w = 3*10^6, where <A = approx -162 and <B = approx 18.
Gain margin approximately 17 dB. -isnt this calculating the open loop gain margin? zwhen we
want only the feedback loop?
gain margin is a property of the open-loop only. (remember Nyquist) Commented [29]: I think in ECD we learn that gain
margin is a property of LOOP GAIN. In this question, B
is part of the LOOP GAIN so we need to count for the
Im not sure if above is correct. AB = 1/(t1s + 1)(t2s + 1)(Cs +1), contribution of B.
therefore, |AB(w = 10^4)| = 42.82 dB. So, the graph is very similar to graph in (ii). Since it has 3
poles, it should cross -180 degrees at some point. Not sure how to calculate that, though.
Can someone please clarify this?
Commented [30]: what does this mean setting
nominal dc value times root 10? Multiply A by 10^0.5?
1.iv.c or ß?
you can use FVT and find the Steady state to be 3.16 for a step input, so error is 2.16 units...?
Commented [31]: _Marked as resolved_
thus the error difference is massive? so it is behaving like an amp....,
Commented [32]: _Re-opened_
iv.d
unstable
if i got an improper root locus am i wrong?
2.i.
Setting derivatives to zero and subbing equations into each other:
2.ii.
As all coefficients are positive, all zeros/poles are LHP, except the two zeros at (0,0). Since no
RHP poles, we need 0 encirclements of -1 to be stable. response for -w values, never actually
going at 180 or -180 degrees to encircle the critical point. There is probably a better way to
explain this.
*i always thought of it as if a plot reaches -180 before crossover then u have instability
(negative pm). therefore since this reaches -180 at w->infinity then it will always be stable.
Visually it will look something like this, where no amount of6 gain will cause an encirclement.
2.iii.
Just create an equation for the transfer function of F to delta1 and then rearrange.
Q1(ii)
- If Zm is an open circuit, the only feedback is from Zf
- If Zf is a short circuit, it has a gain of 1 (no effect, direct feedback)
Q1(iii)
𝑉𝐸 = 𝑉𝑖 − 𝑉𝑚 = 𝑉𝑖 − 𝛽𝑉𝑜
𝐴𝛽
𝑉𝐸 = (1 − )𝑉
1+𝐴𝛽 𝑖
𝑉𝑖 = 1/𝑠
𝐴(0)𝛽(0)
𝑙𝑖𝑚 𝑠𝑉𝐸 = 1 − 1+𝐴(0)𝛽(0)
𝑠→0
𝛽=1
𝐴(0)
=1−
1 + 𝐴(0)
2
𝐾
𝐴(0) = =𝐾
(0+1)(0+1)
Q1(iv.a)
A(s) never touches -180 degrees
therefore, Gain margin = infinite
Q1(v.a)
L=10^4 bode plot are the two right mhed lines Commented [33]: how do you know which line is
which?
Commented [34]: the break frequency, so one is at
frequency where phase is -180 = 3*10^9 10^6/10^-4= 10^10 the other is 10^6/(7*10^-3)=
(1/7)*10^-9 so at break there should be -45 deg phase
( we can also have phase = -180 at 2.5 * 10^8 and gain margin is around 12.5 dB ) so the L=10^-4 one has -45 and 10^10 which is the
right one
Gain at this frequency = -17dB
Therefore, Gain Margin = 17dB
For √10𝐾, DC gain is: you could just qualitatively say "the sqrt(10) increases
20𝑙𝑜𝑔(√10𝐾) = 20𝑙𝑜𝑔(100.5 ) + 20𝑙𝑜𝑔(𝐾) = 10𝑑𝐵 + 𝐷𝐶 𝑔𝑎𝑖𝑛 𝑓𝑟𝑜𝑚 𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑠𝑦𝑠𝑡𝑒𝑚 the gain, which moves the crossover frequency to the
left.
𝐴𝛽
𝑉𝐸 = (1 − )𝑉
1 + 𝐴𝛽 𝑖
𝐴𝛽 1
𝑠𝑉𝐸 = 𝑠(1 − )
1 + 𝐴𝛽 𝑠
𝐴(0)𝛽(0)
𝑙𝑖𝑚 𝑠𝑉𝐸 = 1 −
𝑠→0 1 + 𝐴(0)𝛽(0)
For configuration in (ii), steady state error = 7.07 x 10^-8, larger error but still very small.
Dividing by sqrt(10) gives approx above error. Overshoot given in part ii.
Q1.iv.d.
Q1.iv.e9.
Find the characteristic polynomial, divide by everything that is not multiplied by L, should get
something like this: would like some enlightenment on this part here^^^ <dont worry I got it
Model is not strictly proper
Asymptote = -1
Q2.i.
As per problem set 1, 2014, at equilibrium all signals are constant. Therefore, all derivatives are
0. Using this we can find x1q and x2q by themselves. (x1 = x1q, x2 = x2q)
CW encirclements = 0
Open-loop RHP poles = 0
Unstable CLosed-loop poles = 0
Hence, CLosed loop is stable
if 𝑏 2 − 4𝑚𝑘 < 0, you get complex conjugates with negative real part
𝑚2 𝑠 2
𝑚1 𝑠 2 + 𝑏1 𝑠 + 𝑘1
2 zeros at s=0, 2 poles at some frequency
so subtracting the pole contribution from the ‘initial’ zero contribution, phase t w=inifinity
therefore
0 < ∠𝑚2 𝑠 2 . . . . < 180
therefore:
−180 + 0 < ∠𝛬(𝑗𝜔) < 0 + 180
−180 < ∠𝛬(𝑗𝜔) < 180
Therefore, it never crosses the real axis, and can nev
er make an encirclement for any mass spring damper values
I think it will cross the real axis(when ∠𝛬(𝑗𝜔)=0 which is possible). the point is it will
never cross the negative real axis, as such it will never hit (-1+0j) and the tf will always be
stable considering there are no RHP poles.
(NOTE: it could get very very close to 180 (ie very small phase margin), but technically its still
stable)
2iv
using final value theorem on q 2iii
ss value = 1/k1 ← seems right
possible if L1 = (g(m1+m2) + 1)/k1 then it would return to equilibrium value ← how did u
get this condition? ***note: i forgot the k1 in denominator (updated 13/06 715pm)
I got ess=1/k2, is there anyone who got the same answer? < < make all s terms 0 and you
get left with k2/k1k2 ----> 1/k1
can someone check this? not really sure. i think you want SS = 0 for this case, which makes
sense if you apply a force to a spring and dont remove it the spring wont return to equilibrium..
anybody else agree? therefore not possible? → yeah this does make sense, a step force input
means the force will always be there (at least for t>0 or something), so can’t really expect the
spring to return to its equilibrium point can we?
𝑥𝑞 = 𝑙1 − (𝑚1 + 𝑚2 )𝑔/𝑘1
1/𝑘1 = 𝑙1 − (𝑚1 + 𝑚2 )𝑔/𝑘1 etc
Isn’t that wrong as for xq calculated earlier, it said assume F=0? And in this question F is a step
input so its not 0...
Q2(v)
After a week and emailing Cantoni, I have the answer -Winston
𝑘2
for ≠ 𝜔2 , there are poles on the imaginary axis, due to the 𝑠 2 + 𝜔2term
𝑚2
Therefore, Final Value Theorem is not valid, the steady state response to the sinusoid input is a
sinusoid, which does not have a constant final value
𝑘2
for = 𝜔2
𝑚2
𝑚2 (𝑠 2 +𝜔2 ) 𝑚2 𝐴𝜔
𝛥1 (𝑠) = =
𝑚1 𝑚2 𝑠 4 ........𝑒𝑡𝑐 𝑚1 𝑚2 𝑠 4 ........𝑒𝑡𝑐
𝑚2 𝐴𝜔
𝑙𝑖𝑚 = 𝑙𝑖𝑚 𝑠𝛥(𝑠) = 𝑙𝑖𝑚 𝑠 =0 <-in order to make the first two limits
𝑡→𝑖𝑛𝑓𝑖𝑛𝑖𝑡𝑦 𝑠→0 𝑠→0 𝑚1 𝑚2𝑠4 ........𝑒𝑡𝑐
equal do we have to state that we are making the assumption that xq = 0? Otherwise im not
sure how we’ve come to the conlclusion that: lim(xi(t)) as t-> infty = lim s Delta(s). Cheers
𝑘2
This agrees with the Internal Model Principle, since when = 𝜔2 , 𝛤(𝑠) = 𝑠 2 + 𝜔2 is cancelled
𝑚2
out, presumably by 𝐶(𝑠)
I just thought looking at the function from F(s) to 𝛥1 (𝑠) if b2=0 the zero is 𝑚2 𝑠 2 + 𝑘2 so will
𝐴𝜔 𝑘
cancel out 𝐹(𝑠) = when 𝑚2 = 𝜔2 … is that too short an explanation? I award you no
𝑠2 +𝜔 2 2
marks and fail you This looks good to me, and is consistent with the internal model principle. But hey,
what do I know :D
Q2)vi) -?
In describing a qualitative benefit that adding a small non zero b2 might add it is worthy
identifying what is actually wrong with the bode plot in figure 7 to begin with. At the
cross over frequencies, the phase drops to -180, rise sharply, then drops to -180 again.
This is bad as it cause the phase margin to be almost zero and hence makes the system
vulnerable to disturbances (not entirely sure of the vulnerable to disturbances bit, but
thats my understanding anyhow).
In short: Either way a possible benefit might be to increase the phase margin
at these points.
yeah basically by creating a small b2 youre going to remove 2 zeros from the imaginary
axis and “shove” them into the LHP. This in turn creates the better CL stability.
you’re also right. our answers in addition would most definitely bring full marks (Like)
Note that by examining the open loop transfer function, by letting b2 be positive, we are
in fact taking both poles on the imaginary axis and shifting it to LHP, thus making the
closed loop transfer function now more stable relatively speaking.
My explanation is: when b2=0, the peak of loop gain is infinite at w=1. According to Bode’s gain-
phase relation, steep slope makes the phase margin small. So by making a small value of b2,
the loop gain looks less steep at the cross-over frequency, which means more stable.
Another point of view utilizes root-locus plot. If you rearrange the loop gain in terms of b2, and
plot the root-locus plot by matlab, as b2 increases, the poles moves from imaginary axis
towards LHP, BUT, if b2 is too large, the poles will move back to the zeros on the imaginary
axis. During exam, we can yield a polynomial in term of b2, but it takes a bit of time and it’s hard
to get poles or zeros from a high-order polynomial even with CASIO fx-991ES. So although
rlocus is straightforward, it might not be the answer to this question under exam condition.
I think with b2 an appropriate small value, it will add b2*s+10 in the new bode plot, thus the
phase of the system in high frequency will increase by 90 degree, and the Mp will increase. Also
at the frequency of w=1, the gain will not be infinite.
The figure is b2=1
Q3
can someone do 3i, it looks like a massive bitch. theres no helpful hints like he gave in
the july exam for all the sin and cos terms
and seriously, who the fuk knows the derivative of theta dot sin(theta)cos(theta) in an
exam < bitta primary school maths. 2sinx.cosx=sin2x or use prdocut rule
Dont need to worry about the the sin(theta)cos(theta) bit really because it’s mulitplied by
theta dot, so when you are doing the partial derivative of theta, you end up with it being
multiplied by theta dot which is zero. ye strangely enough i still cant seem to get my
head around this linearisation shit 100% and really cbf doing it the longggggg way
The linearisation is not as hard as it looks like since the equilibrium values are all 0, the
derivative of those sin^2(x)are all 0, quite simple.
Q3 ii if you look at figure 9 it says what C(s) is equal to and relates Vr and theta, so if you solve
.the block you get the equation.. guess its right?? where K is the back emf not the gain which is
next question...
Commented [40]: this is not correct,
V_m is the motor voltage (read error)
D is a disturbance
But this looks weird, the D is a disturbance i would assume when reading the question... like its
all switched around... then it makes the other bits confusing. --I think that this is wrong.
I also think this is wrong
This is what I came up with….. Im so lost on this one, thoughts feelings and emotions are
welcome. how did you get the plant model in this one??
Q3 iii)
cant be stabilized with a gain feedback as the poles are at +10 and -10 with no other poles or
zeros, thus break is between them at 90deg which will coincide with the imaginary, resulting in
marginally stable result. however when a disturbance comes in from the movement will tip the
result to be unstable...
Q3iv) Can’t guarantee these numbers, but l1 = 0.01716, po = 236.4? (Po is wrong) ⇐ if the Commented [41]: How come your G(s) is 10/(s^2 -
numbers are bodeplotted there is a phase margin of 22.7 at 110 rads... that is jsut G*C where 100) and not 25/(s^2-100)?
G=10/(s^2 -100) and C is given with the values...
When i found these numbers I had a G(s) = 25/(s^2-100)? The 0.4 will make it 10, I just don’t
quite get where it goes/why you multiply it by G(s). i thought it would be 0.4*C(s)? This will give
the same as below (l0 = 0.01716 and po = 23.64) o
neither can i but i got 𝐿1 = 1.524 ∗ 10−3 and Po=18 ywhich when plotted gives the
values...
to solve is soooo not worth 3 marks and took 2 pages of workings.... i think this exam is
going to be a joke!!
but to solve properly i would think quick routh to check for stability....
find that 0<L1<1/10 and Po>1 for stability....
then you sub s=jw and solve arg(A(jw)) where A is the open loop which is GC=
10(𝐿1 𝑠+1)
closed loop will be so steady state will be 10/(-100+po) but if it
(𝑠2 −100)(𝐿1 𝑠+1)+𝑃𝑜 (0.1𝑠+1)
was the open loop the steady state will be jsut -Po.... which is frecken huge!!! like -18 or
-236 in the other persons results!!! but the closed is more acceptable... but still kkind
a not really as closed loop for Po=18 steady state is -0.12 from a step of 1...
steady state error will be final value theorem but of the closed loop??? or the open
loop?i think its closed loop. its the signal that goes into C(s), which is Vr-Θ
^^^^^ not confident with ANYTHING I HAVE DONE!!! please comment if i fucked it up!!
Update: using L1=0.0172 & p0=23.64 seems to satisfy the requirements. I got these
results as well using the formulas on the phase-lead control slide. ← can you say a bit
more on this? i used the formulas on the slides and they still don’t seem to match up
with the answer....
worked out necessary phase (phi) pluged that into the arcsin(xxxxx) formula to get l1.
then used the formula for K to get K (in both these cases using wc = 20 rad/s)
i think u mean 15rad/s but its ok. however, we cannot just use wc = 1/sqrt(tz tp) to get the
00tp?the Jan exam has 20 rad/s the 2011 has 15 rad i believe.oh yes my bad.
but can we still use the formula?
Im not sure :S
i will try again thanks for discussion :)
try looking at worksheet 5, problem 1, part 3. very similar. thank you :D
im getting l1 = 1/60 and po about 23.something, cant remember exactly. this is for 20rad/s
^^ i got this answer (well the first part anyway) - I unsure about the plant TF (𝐺(𝑠) =
25×0.4
)... I just solved these by hand (𝜔𝑐 = 20rad/s):
𝑠 2 −100
|𝛬(𝑗𝜔𝑐 )| = 1 and 𝑀𝑝 = 𝜋 + ∠𝛬(𝑗𝜔𝑐 )
ye i had 2.36 to begin with but then i realised i forgot to take remove the 10 on the
numerator since it cancels out with the (s+10) together with the (0.1s+1)
=> 10(0.1s+1) = (s+10) <- denominator term from G(s)
^ ^ numerator of C(s)
numberator of G(s)
𝛩(𝑠) 𝐾𝐺𝐶
steady state value = 𝑙𝑖𝑚 𝜃(𝑡) − 𝜃𝑞 = 𝑙𝑖𝑚𝑠𝛩(𝑠) = 𝑙𝑖𝑚𝑠 𝑉𝑟(𝑠) =( ) |𝑠 =
𝑡→𝑖𝑛𝑓 𝑠→0 𝑠→0 𝑉𝑟(𝑠) 1+𝐾𝐶𝐺
0 =1.733
steady state error for step input = 1 - 1.733 = -0.733 I think the steady state error should
be negative since its 1-1.733. This suggests that the output is larger than the reference
which is the case you you plot in matlab. Anyone e lse get -0.733? yep ur right <<
THIS IS RIGHT!! <Another Vote
where 𝜃𝑞 =0, KG=10/(s^2-100), C=23.64(0.1s+1)/(0.0172s+1)
**verified using matlab step function
i got -0.0733
𝜃 = 𝐸. 𝐶. 𝐺 + 𝐷. 𝐺
𝐸 = −𝜃 → i thought error is E=Vr-Vtheta, no?It is if you’re analysing the SS in response to a
step from the input. V_r=0 because we’re analysing it from the input disturbance
𝐸 = −𝐸. 𝐶. 𝐺 − 𝐷. 𝐺
𝐸 = −𝐺/(1 + 𝐶𝐺) ∗ 𝐷
𝐸 = −𝐺(0)/. . . 𝑒𝑡𝑐use FinalValueTheorem
𝐸 = −0.0733seems a bit stupid for 1 mark 1
Since the question ask about the steady state error in response to the reference signal, shouldnt
we set the disturbance to zero? The questions says “what dominates the transients to a step
disturbance input”. Unless im interpretting that completely wrong, we’re analysing the SS to a
input disturbance. → ***I think you’re reading question 3(vi), we’re doing question 3(iv) here lol
how dyslexic
How to calculate p0 without using bode plot? setup eqns relating phase margin>
|\lambda(jw_c)|=1 and arg(lambda(jw_c))= -180+PM. its a fuckin pain but the numbers flop out
Commented [42]: unsure if I have the right approach...
has anyone used the matrix way?? caus i found P1=3820... even tho i agree its 3920.. Commented [44]: Is P1 3820 or 3920?
Commented [45]: P1 is 3920. I think you used the
wrong matrix inversion formula. You have to use the
one in the seperate pdf which gives 3920
Commented [46]: is it just me, or is the matrix
unecessary
i find it easier to equate co-effs
Commented [47]: yeh so much faster and easier... but
when ya have shit loads more variables gets bit hairy...
I got 3830 as well and omg cant believe I used the wrong matrix after he
went to such lengths to give us the right one :P *face
i hope you realise thats the wrong inverse matrix^^ ur missing a 100 under the 10 in column 1 of palm
the inverse
<===The correct one.
Q3vi
From part (v), the slowest closed loop pole is at s=-10 hence this dominates the response, i.e.
the response is dominated by exponential decay term rather than 2nd order dynamics
oscillation.Besides s=-10, there are two complex poles. Together with these dominate the
transient response.
WHY IS S = -10?? Isnt it just the three roots: s = -100, -11+16.7j , -11-16.7j
The steady state error is zero since there is an integrator in the loop (pole at s=0 in the
controller transfer function).
what about CL zero’s. They also have an effect in the time domain.a
wouldnt the steady state error be when the input disturbance sensitivity? ie S_io(0) as this maps
1D(s) to theta.... then the error results in sG(0)=-25/100 as D=1/s and C has a pole at origin....
The former statement is correct. The latter is wrong.
An integrator gives infinite gain at DC, not 0 steady state error. Since we are looking at the
response to an output disturbance it would make sense to use the output disturbance sensitivity
function (S_0(s)). Applying the step input 1/s and using the final value theorem (which is for
signals, not for transfer functions) leads to the result that the final value at the output is S_0(0).
Since the error signal is E(s) = R(s)-THETA(S) = 0 - S_0(s) we can see that the steady state
error is just -S_0(0).
Q4.
(i) The goal of the modified Nyquist test is to ensure that the closed-loop poles reside in the left-
half of some Real{s} = a in the complex plane. For these requirements, the Nyquist path is Commented [48]: Question still lacks some
changed to a contour that contains the vertical line crossing real axis at ‘a’ and the left-hand justification, but seems to work (I did a matlab code to
check it).
side of this vertical line.
DO NOT KNOW THE JUSTIFICATION: s moves around the contour in the COUNTER
CLOCKWISE direction
.
By Cauchy’s principle of argument: Z = N +P where
Z = number of zeros to the left of Real{s}=a
N = number of counter clockwise encirclements made by lambda(jw+a) as s traverses the
contour in the counter clockwise direction.
P = number of poles of open-loop poles to the left of Real{s}=a
The modified Nyquist plot: magnitude and phase of lambda(s) = G(s) evaluated at s=j*w+a
By adjusting K value to be negative the condition can be met. However, K is not allowed to be
less than 0. Hence, the system cannot achieve closed loop poles that lie on the left of Alpha.
2.By modifing Nyquist plot: magnitude and phase of lambda(s) = KG(s) evaluated at s=j*w-a, we
can test whether the poles lies in the left side of a, without changing the definition of N , P
and Z. i.e.we still conider :
N=#clockwise encirclement of -1 made by lamda(s-a);
P=#RHP poles of *lamda(s-a).
The poles all lies to the left of a if N=-P.s
Actually, if we go on doing the question (ii), I think the test method is consistent with my second
opinion. Q(iii) is consistent with my 1st opinion.(Did I make any mistake? :P) Cheers!!!
***To stick more to the standard nyquist test, can we also use a contour that covers the right
side of Re(s) = alpha (traversed clockwise)? In this case we would want no closed loop poles
inside that contour i.e. Z=0 and the usual condition of N = -P applies (N is clockwise
encirclement of -1+j0, as usual). In your method you propose that you want Z=total number of
closed loop poles, which works fine as well. However, Nyquist plot suggestst that encirclement
is required for stability€
I also think the right part of Re(s)=alpha should be considered as D-contour. Then the modified
Nyquist stability test is that the Nyquist plot of lambda(jw+alpha)=KG(jw+alpha) encircles (-1,j0)
P times in counter-clockwise direction. P is the number of poles of KG(s) having larger real
value of alpha.
(ii)
G(s) = (s+5)/[(s+2)(s+9)]
s=-9 is a OL pole to the left of Real{s}=a => P=1
Closed loop has degree 2 for proportional controller => 2 poles
It’s desired that both poles are to the left of Real{s}=a => Z =2 Commented [49]: Why does both poles to the left
By Cauchy’s principle argument: N = Z – P = 1 counter clockwise encirclement correspond to Z = 2?
This happens for 1/0.1 > K => 10 > K
The number of open-loop pole with bigger real value of -4 : 1 (for s=-2).
Thus the modified Nyquist plot should encircle (-1, j0) in counter-clockwise direction once.
So K>(-1/-0.1) => K>10