0% found this document useful (0 votes)
3 views

Multi-Variables Control (1)

Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
3 views

Multi-Variables Control (1)

Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 78

Kendali Variabel Jamak

(Multi-variables Control)

AE3200 (II-2020/2021) 1
Multi-variable Control : Pole Placement
State Feedback

Controlled system
D

UC + u +
.
x x
+ y
B C
_ +
+

𝑥̄̇ 𝑡 = 𝐴𝑥̄ 𝑡 + 𝐵𝑢̄ 𝑡


Controller
𝑦̄ 𝑡 = 𝐶𝑥̄ 𝑡 + 𝐷𝑢̄ 𝑡
ec
K 𝑒̄ 𝑡 = 𝐾𝑥̄ 𝑡

AE3200 (II-2020/2021) 2
State Feedback Control signal

𝑥̄̇ 𝑡 = 𝐴𝑥̄ 𝑡 + 𝐵𝑢̄ 𝑡


𝑦̄ 𝑡 = 𝐶𝑥̄ 𝑡 + 𝐷𝑢̄ 𝑡

𝑒̄ 𝑡 = 𝐾𝑥̄ 𝑡 Closed loop system

K is a (m x n) matrix
m is the size of input vector
(number of column of B)
n is the size of state variable
vector

Open loop system characteristic:

Closed loop system characteristic:


det 𝑠𝐼 − 𝐴 − 𝐵𝐾 = 𝑠 + 𝑏 𝑠 + ⋯+𝑏 = 𝑠 −𝑝 𝑠−𝑝 ⋯ 𝑠−𝑝
AE3200 (II-2020/2021) 3
Pole Placement

Closed loop characteristic can be altered by choosing appropriate matrix K

Consider a system with 1 input (B has 1 column) and n state variables.


Assuming that all the states can be used as feedback signal.
A gain matrix K:

can be inserted in the feedback loop, giving the following closed loop system

𝑥̄̇ 𝑡 = 𝐴 − 𝐵𝐾 𝑥̄ 𝑡 + 𝐵𝑢̄ 𝑡

𝑎 𝑎 … 𝑎 𝑏 𝑎 −𝑏 𝑘 𝑎 −𝑏 𝑘 ⋯ 𝑎 −𝑏 𝑘
𝑎 ⋱ ⋮ 𝑏 𝑎 −𝑏 𝑘 ⋱ ⋮
𝐴 = 𝐴 − 𝐵𝐾 = ⋮ − 𝑘 𝑘 ⋯ 𝑘 =
⋮ ⋮
𝑎 … 𝑎 𝑏 𝑎 −𝑏 𝑘 ⋯ 𝑎 −𝑏 𝑘
Closed loop system characteristic:
𝑎 −𝑏 𝑘 𝑎 −𝑏 𝑘 ⋯ 𝑎 −𝑏 𝑘
𝑎 −𝑏 𝑘 ⋱ ⋮
det 𝑠𝐼 − 𝐴 = det 𝑠𝐼 − =𝑠 +𝑏 𝑠 + ⋯+ 𝑏 = 𝑠 − 𝑝 𝑠−𝑝 ⋯ 𝑠−𝑝 =0

𝑎 −𝑏 𝑘 ⋯ 𝑎 −𝑏 𝑘

AE3200 (II-2020/2021) 4
Suppose we want the closed loop system to have a particular dynamics
characteristic/behaviour, which is reflected by a pre-defined poles location:

Which can be translated into the desired characteristic polynomial

𝑃 𝑠 = 𝑠−𝑝 𝑠−𝑝 … 𝑠−𝑝 =𝑠 +𝑑 𝑠 +𝑑 𝑠 +⋯+𝑑

Then the problem is to find (k1, k2,…,kn) such that

which can be obtained by equating the coefficients:

𝑏 =𝑑
𝑏 =𝑑

𝑏 =𝑑
where bn-1, bn-2, …, b0 are functions of k1,k2,…,kn
AE3200 (II-2020/2021) 5
Example :
Poles :
2 0 −5 0
𝑥̄̇ 𝑡 = 3 −1 0 𝑥̄ 𝑡 + −1 𝑢̄ 𝑡
-2.4107
0 1 0 0
1.7054 + 1.8204i
1.7054 - 1.8204i
Fully controllable
We want to move the poles to new locations : -2.5 ; -0.7+0.4i ; -0.7-0.4i
by implementing a feedback loop with state feedback gain K=[k1 k2 k3]
The closed loop system is represented as :
2 0 −5 0 2 0 −5
𝐴 = 𝐴 − 𝐵𝐾 = 3 −1 0 − −1 𝑘 𝑘 𝑘 = 3+𝑘 −1 + 𝑘 𝑘
0 1 0 0 0 1 0

𝑠−2 0 5
det 𝑠𝐼 − 𝐴 = det − 3 + 𝑘 𝑠 − −1 + 𝑘 −𝑘
0 −1 𝑠

= 𝑠 − 2 𝑠 − −1 + 𝑘 𝑠 − 𝑘 +5 3+𝑘

= 𝑠 + −1 − 𝑘 𝑠 + −2 + 2𝑘 − 𝑘 𝑠 + 5𝑘 + 2𝑘 + 15

AE3200 (II-2020/2021) 6
Based on the desired poles locations, the following characteristic polynomial
can be obtained :

Equating the coefficients of Pd(s) to those of det(sI-ACL) gives the following


relations:
−1 − 𝑘 = 3.9 ⇒ 𝑘 = −4.9

−2 + 2𝑘 − 𝑘 = 4.15 ⇒ 𝑘 = −15.95

5𝑘 + 2𝑘 + 15 = 1.625 ⇒ 𝑘 = 3.7

Hence, the state feedback gain becomes:

Implementing this gain matrix in the feedback loop gives the following
closed loop system matrix:

The closed loop poles are located at :


AE3200 (II-2020/2021) -0.7037 + 0.3830i ; -0.7037 - 0.3830i ; -2.4926 (as desired) 7
Controllable Canonical Form

The characteristic polynomial of the above form is

Consider a (3x3) problem, implementing a feedback loop with gain matrix


K=[k1 k2 k3] gives:

0 1 0 0 0 1 0
𝐴 = 𝐴 − 𝐵𝐾 = 0 0 1 − 0 𝑘 𝑘 𝑘 = 0 0 1
−𝑎 −𝑎 −𝑎 1 −𝑎 − 𝑘 −𝑎 − 𝑘 −𝑎 − 𝑘

AE3200 (II-2020/2021) 8
The characteristic polynomial is :

𝑠 −1 0
det 𝑠𝐼 − 𝐴 = det 0 𝑠 −1
𝑎 +𝑘 𝑎 +𝑘 𝑠+ 𝑎 +𝑘

=𝑠 + 𝑎 +𝑘 𝑠 + 𝑎 +𝑘 𝑠+ 𝑎 +𝑘

Suppose we want the closed loop poles to be at :


s = pd1 ; s = pd2 ; s = pd3
which can be represented in the following polynomial:

𝑃 𝑠 = 𝑠−𝑝 𝑠−𝑝 𝑠−𝑝 =𝑠 +𝑑 𝑠 +𝑑 𝑠+𝑑

Hence, by equating the polynomials

The controller gain can be obtained :

AE3200 (II-2020/2021) 9
So, if the system under consideration is represented in a controllable
canonical form, the feedback gain matrix can be computed easily by
evaluating the coefficients of the system characteristic polynomial and
those of the desired polynomial

For (nxn) system :

Note that the state feedback algorithm already explained is valid for a
single input system (SIMO).
The problem is how to transform a system into a canonical form.

AE3200 (II-2020/2021) 10
Controllable Canonical Form

Consider the following system:

Suppose there exists an (n x n) non-singular matrix T, such that:


0 1 0 0 0
0 0 1 0 0
𝐴 =𝑇 𝐴𝑇 = 0 0 , 𝐵 =𝑇 𝐵= ⋮
0 0 1 0
−𝑎 −𝑎 −𝑎 1

and,

Which means that the eigenvalues (characteristic polynomial and its roots) of A’
is equal to those of A  the coefficients of the characteristic polynomial of A is
equal to those of A’
Suppose a gain matrix K is applied in the feedback loop, such that the closed
loop system becomes:

AE3200 (II-2020/2021) 11
Incorporating the matrix T :
𝐴 =𝑇 𝐴 − 𝐵𝐾 𝑇 = 𝑇 𝐴𝑇 − 𝑇 𝐵𝐾𝑇 = 𝐴 − 𝐵 𝐾

Similar to the open loop system,


det 𝑠𝐼 − 𝐴 = det 𝑇 𝑠𝐼 − 𝐴 + 𝐵𝐾 𝑇

= det 𝑠𝐼 − 𝐴 − 𝐵 𝐾

=𝑠 +𝑏 𝑠 + ⋯+ 𝑏

Suppose the desired closed loop poles are represented by the following
polynomial:
𝑃 𝑠 = 𝑠−𝑝 𝑠−𝑝 … 𝑠−𝑝 =𝑠 +𝑑 𝑠 + ⋯+ 𝑑 𝑠 + 𝑑

Hence, by equating the polynomials:

The feedback gain matrix K’ can be obtained as:


𝐾 = 𝐾𝑇 = 𝑘 𝑘 𝑘 ⋯ 𝑘 = 𝑑 −𝑎 𝑑 −𝑎 ⋯ 𝑑 −𝑎 𝑑 −𝑎

So that,

AE3200 (II-2020/2021) 12
Computing T

Evaluating :

𝐴 =𝑇 𝐴𝑇 ⇒ 𝑇𝐴 = 𝐴𝑇 (∗) , 𝐵 =𝑇 𝐵 ⇒ 𝐵 = 𝑇𝐵 (∗∗)

The column of T can be determined recursively.


From (**)

From (*)

0 1 0 0
0 0 1 0
𝐴𝑇 = 𝑇𝐴 ⇒ 𝐴 𝑡 𝑡 ⋯ 𝑡 = 𝑡 𝑡 ⋯ 𝑡 0 0
0 0 1
−𝑎 −𝑎 −𝑎

AE3200 (II-2020/2021) 13
Since tn is already determined, the last (nth)column of the equation above can be
used for computing tn-1

0 1 0 0
0 0 1 0
𝐴𝑡 𝑡 ⋯ 𝑡 = 𝑡 𝑡 ⋯ 𝑡 0 0 ⇒ 𝐴𝑡
0 0 1
−𝑎 −𝑎 −𝑎
0
Hence, 0
= 𝑡 𝑡 ⋯ 𝑡 ⋮
1
−𝑎
Evaluating the (n-1)th column, we get

0 1 0 0 0
0 0 1 0 ⋮
𝐴𝑡 𝑡 ⋯ 𝑡 = 𝑡 𝑡 ⋯ 𝑡 0 0 ⇒ 𝐴𝑡 = 𝑡 𝑡 ⋯ 𝑡 1
0 0 1 0
−𝑎 −𝑎 −𝑎 −𝑎

Hence,

AE3200 (II-2020/2021) 14
Repeating the procedure until the 1st column of AT, we get the recursive
calculation for obtaining T :

AE3200 (II-2020/2021) 15
Example: Poles:
−0,254 0 −1,76 0,322 0 -8.4328
−9,08 −8,4 2,19 0 29
𝑥̄̇ = 𝑥̄ + 𝑢̄
2,55 −0,35 −0,76 0 −0.222 -0.4934 + 2.3352i
0 1 0.0875 0 0
-0.4934 - 2.3352i
0.0056
𝑃(𝑠) = s + 9.41s + 13.97s + 47.96s−0.27
a3 = 9.41 ; a2 = 13.97 ; a1= 47.96 ; a0 = -0.27

Computing T = [t1 t2 t3 t4] :


𝑡 =𝑡 =𝐵 −0.254 0 −1.76 0.322 0 0 0.39
0 −9.08 −8.4 2.19 0 29 29 28.92
𝑡 = 𝑡 = 𝐴𝑡 + 𝑎 𝑡 = + 9.41. =
29 2.55 −0.35 −0.76 0 −0.222 −0.222 −12.05
=
−0.222 0 1 0.0875 0 0 0 28.98
0

−0.254 0 −1.76 0.322 0.39 0 30.45


−9.08 −8.4 2.19 0 28.92 29 132.11
𝑡 = 𝑡 = 𝐴𝑡 + 𝑎 𝑡 = + 13.97. =
2.55 −0.35 −0.76 0 −12.05 −0.222 −3.05
0 1 0.0875 0 28.98 0 27.9

−0.254 0 −1.76 0.322 30.45 0 6.6


−9.08 −8.4 2.19 0 132.11 29 −2.03
𝑡 = 𝑡 = 𝐴𝑡 + 𝑎 𝑡 = + 47.96. =
2.55 −0.35 −0.76 0 −3.05 −0.222 23.17
0 1 0.0875 0 27.9 0 131.85
AE3200 (II-2020/2021) 16
The matrix becomes
6.6 30.45 0.39 0
−2.03 132.11 28.92 29
𝑇= 𝑡 𝑡 𝑡 𝑡 =
23.17 −3.05 −12.05 −0.222
131.85 27.87 28.9 0

And we get

0 1 0 0
0 0 1 0
𝐴 =𝑇 𝐴𝑇 =
0 0 0 1
0.27 −47.96 −13.97 −9.41

0
0
𝐵 =𝑇 𝐵=
0
1

AE3200 (II-2020/2021) 17
Suppose, by adding a feedback, we want the closed loop system to have poles in
the following locations :
s= -10
s= -1.5 + 2i
s= -1.5 - 2i
s= -1.2
Then we have the desired characteristic polynomial :
𝑃 (𝑠) = s + 14.2s + 51.85s + 106s + 75

d3 = 14.2 ; d2 = 51.85 ; d1= 106 ; d0 = 75

Since the system already transformed into a canonical form, hence we can get
the feedback gain as :
𝐾 = 𝐾𝑇 = 𝑘 𝑘 𝑘 𝑘

= 𝑑 −𝑎 𝑑 −𝑎 𝑑 −𝑎 𝑑 −𝑎

= 75 + 0.27 106 − 47.96 51.85 − 13.97 14.2 − 9.41

= 75.27 58.04 37.88 4.77


And the controller gain matrix is

AE3200 (II-2020/2021) 18
Applying the feedback gain matrix on the original system, we get:

Which has the following characteristic roots:


s= -10.0000
s= -1.5000 + 2.0000i
s= -1.5000 - 2.0000i
s= -1.2000
as required.

AE3200 (II-2020/2021) 19
Ackermann’s Formula

Ackermann formula can be used for obtaining the state feedback gain matrix for
system with single input, provided that the system is controllable.

Consider a system below,

The controllability matrix of the system above is:

Suppose a feedback gain controller K is needed to make the closed loop poles
reside at s=pd1, s=pd2, …, s=pdn . This poles configuration is represented by a
polynomial

AE3200 (II-2020/2021) 20
The required gain matrix K can be computed using Ackermann’s formula:

where Co is the controllability matrix, and

This formula can also be used for obtaining the estimator gain.

AE3200 (II-2020/2021) 21
Example:

Poles:
−0,254 0 −1,76 0,322 0
−9,08 −8,4 2,19 0 29 -8.4328
𝑥̄̇ = 𝑥̄ + 𝑢̄
2,55 −0,35 −0,76 0 −0.222
0 1 0.0875 0 0 -0.4934 + 2.3352i
-0.4934 - 2.3352i
0.0056
The controllability matrix:
0 0.39 26.8 −251
29 −244.09 2025 −17050
𝐶𝑜 = 𝐵 𝐴𝐵 𝐴 𝐵 𝐴 𝐵 =
−0.222 −9.98 94 −712
0 28.98 −245 2033

Suppose, by adding a feedback, we want the closed loop system to have poles in
the following locations :
s= -10
s= -1.5 + 2i
s= -1.5 - 2i
s= -1.2
Then we have the desired characteristic polynomial :
𝑃 (𝑠) = s + 14.2s + 51.85s + 106s + 75
AE3200 (II-2020/2021) 22
d3 = 14.2 ; d2 = 51.85 ; d1= 106 ; d0 = 75
𝑞(𝐴) = 𝐴 + d A + d A + d A + d I
and,
= A + 14.2A + 51.85A + 106A + 75I

−120.32 −5.60 6.99 8.77


−523.61 −581.17 8.76 18.94
=
−26.68 −15.98 −100.06 32.02
67.51 74.02 64.04 61.62

Then we can compute

This matrix K gives the following closed loop system matrix

−0.25 0 −1.76 0.32


−22.35 −12.96 31.25 −21.07
𝐴 = 𝐴 − 𝐵𝐾 =
2.65 −0.32 −0.98 0.16
0 1 0.0875 0

whose poles are located at : s= -10 ; s= -1.5 + 2i ; s= -1.5 - 2i ; s= -1.2 (as required)

AE3200 (II-2020/2021) 23
Multi-variable Control : Linear Quadratic Control
For a dynamical system :

a state feedback controller K (regulator case, r = 0), such that

can be determined so that it minimizes a cost function

Where Q is a positive semidefinite Hermitian (symmetric) matrix and R is a


positive definite Hermitian (symmetric) matrix.
Q and R determine the relative importance of the system state deviation and
the control energy.

AE3200 (II-2020/2021) 24
Optimization formulation
• The control u = -Kx can be obtained by solving the
following optimization problem :

Min (cost function)

Subject to (constraint function)

• The problem can be solved by solving a


minimization problem on a function that combines
the cost and constraint functions via Lagrange
multiplier
• Solution K can also be obtained by solving Riccati
equation
AE3200 (II-2020/2021) 25
Implementing the state feedback controller K, a closed loop system is obtained:

𝑥̄̇ = 𝐴𝑥̄ + 𝐵 −𝐾𝑥̄

= 𝐴 − 𝐵𝐾 𝑥̄ ∗

Assuming that (A-BK) is stable (all of its poles has negative real part) and inserting
the definition of control u into the cost function gives:

𝐽= 𝑥̄ 𝑄𝑥̄ + −𝐾𝑥̄ 𝑅 −𝐾𝑥̄ 𝑑𝑡

= 𝑥̄ 𝑄𝑥̄ + 𝑥̄ 𝐾 𝑅𝐾𝑥̄ 𝑑𝑡

= 𝑥̄ 𝑄 + 𝐾 𝑅𝐾 𝑥̄ 𝑑𝑡

AE3200 (II-2020/2021) 26
Defining
𝑉 𝑥̄ = 𝑥̄ 𝑃𝑥̄

Hence
𝑑 𝑑
− 𝑉 𝑥̄ = − 𝑥̄ 𝑃𝑥̄ = −𝑥̄̇ 𝑃𝑥̄ − 𝑥̄ 𝑃𝑥̄̇
𝑑𝑡 𝑑𝑡

Equating the equation above to the terms inside the integral operator on the
previous equation:

𝑥̄ 𝑄 + 𝐾 𝑅𝐾 𝑥̄ = −𝑥̄̇ 𝑃𝑥̄ − 𝑥̄ 𝑃𝑥̄̇

Substituting 𝑥̄̇ with (*) gives

𝑥̄ 𝑄 + 𝐾 𝑅𝐾 𝑥̄ = − 𝐴 − 𝐵𝐾 𝑥̄ 𝑃𝑥̄ − 𝑥̄ 𝑃 𝐴 − 𝐵𝐾 𝑥̄
= −𝑥̄ 𝐴 − 𝐵𝐾 𝑃𝑥̄ − 𝑥̄ 𝑃 𝐴 − 𝐵𝐾 𝑥̄
= −𝑥̄ 𝐴 − 𝐵𝐾 𝑃 + 𝑃 𝐴 − 𝐵𝐾 𝑥̄

AE3200 (II-2020/2021) 27
Then it can be seen that

𝐴 − 𝐵𝐾 𝑃 + 𝑃 𝐴 − 𝐵𝐾 = − 𝑄 + 𝐾 𝑅𝐾 ∗∗

 If (A-BK) stable, then a positive definite P exists and satisfies (**)

For a stable system

system matrix A has its Eigenvalues on the Left Half side of complex plane.
And there exists a positive definite Hermitian P such that

Where Q is a positive definite Hermitian matrix.

AE3200 (II-2020/2021) 28
Since (A-BK) is assumed to be stable, then as t inf, x(inf)  0, hence

Then the cost J can be determined based only the initial condition.

AE3200 (II-2020/2021) 29
Consider the following :

𝐴 − 𝐵𝐾 𝑃 + 𝑃 𝐴 − 𝐵𝐾 = − 𝑄 + 𝐾 𝑅𝐾

Which can be rewritten as:

𝐴 − 𝐾 𝐵 𝑃 + 𝑃 𝐴 − 𝐵𝐾 + 𝐾 𝑅𝐾 + 𝑄 = 0 ; 𝑃 =𝑃

Since R is a symmetric positive definite matrix, it can be defined as R=T’T, hence:

𝐴 𝑃 − 𝐾 𝐵 𝑃 + 𝑃𝐴 − 𝑃𝐵𝐾 + 𝐾 𝑇 𝑇𝐾 + 𝑄 = 0 ; 𝑅=𝑇 𝑇

𝐴 𝑃 + 𝑃𝐴 + 𝑄 + 𝐾 𝑇 𝑇𝐾 − 𝐾 𝐵 𝑃 − 𝑃𝐵𝐾 = 0

Consider the terms :

𝐾 𝑇 𝑇𝐾 − 𝐾 𝐵 𝑃 − 𝑃𝐵𝐾

AE3200 (II-2020/2021) 30
𝐾 𝑇 𝑇𝐾 − 𝐾 𝐵 𝑃 − 𝑃𝐵𝐾

which can be reformulated as:

𝐾 𝑇 𝑇𝐾 − 𝐾 𝑇 𝑇 𝐵 𝑃 − 𝑃𝐵𝑇 𝑇𝐾

𝐻 𝐻−𝐻 𝐺−𝐺 𝐻 ; 𝐻 = 𝑇𝐾 ; 𝐺 = 𝑃𝐵𝑇

Recalling the form

𝐻−𝐺 𝐻−𝐺 = 𝐻 −𝐺 𝐻−𝐺

=𝐻 𝐻−𝐺 𝐻−𝐻 𝐺+𝐺 𝐺

Hence,

𝐻 𝐻−𝐻 𝐺−𝐺 𝐻 = 𝐻−𝐺 𝐻−𝐺 −𝐺 𝐺

And we can have 𝐾 𝑇 𝑇𝐾 − 𝐾 𝑇 𝑇 𝐵 𝑃 − 𝑃𝐵𝑇 𝑇𝐾

= 𝑇𝐾 − 𝑇 𝐵 𝑃 𝑇𝐾 − 𝑇 𝐵 𝑃 − 𝑇 𝐵 𝑃 𝑇 𝐵 𝑃

AE3200 (II-2020/2021) 31
𝑇𝐾 − 𝑇 𝐵 𝑃 𝑇𝐾 − 𝑇 𝐵 𝑃 − 𝑃𝐵𝑇 𝑇 𝐵 𝑃

= 𝑇𝐾 − 𝑇 𝐵 𝑃 𝑇𝐾 − 𝑇 𝐵 𝑃 − 𝑃𝐵𝑅 𝐵 𝑃

Hence the equation becomes:

𝐴 𝑃 + 𝑃𝐴 + 𝑄 + 𝑇𝐾 − 𝑇 𝐵 𝑃 𝑇𝐾 − 𝑇 𝐵 𝑃 − 𝑃𝐵𝑅 𝐵 𝑃=0

To minimize J with respect to K, we need to minimize the following


(w.r.t. K)
𝑥 𝑇𝐾 − 𝑇 𝐵 𝑃 𝑇𝐾 − 𝑇 𝐵 𝑃 𝑥

Since the formulation is a quadratic form, i.e. It is nonnegative, hence


the minimum is attained when it is zero, implying:

𝑇𝐾 − 𝑇 𝐵 𝑃=0

Giving the controller form :


𝐾=𝑇 𝑇 𝐵 𝑃=𝑅 𝐵 𝑃
AE3200 (II-2020/2021) 32
The control signal then is

Where P is the one that satisfies the following equation

 Algebraic Riccati Equation

 Solving ARE is complicated, needs specific treatment (numerical,


optimization), especially whn the system is “big” and “complicated”
 Solving ARE will give us a positive definite matrix P
 The controller then can be computed as :

AE3200 (II-2020/2021) 33
AE3200 (II-2020/2021) 34
Consider the cost function :

 To compute K, we need to define Q (symmetric positive semidefiinite matrix),


and R (symmetric positive definite matrix), as the weighting matrices in cost
function J.
 These Q and R matrices are the design parameters that can affect the result
 Q is related to system variables energy, and R is related to control variables
energy
 Q determines how much the controller is intended to suppress the system
energy, i.e. to keep system variables “small”
 R determines how much the controller is allowed to vary its variables value
in order to do the task, i.e. to keep control variables “small”
 There must be a compromise in altering the system and control variables. To
suppress the magnitude of system variables we need more control energy.
In the other side, keeping the control energy restricted at a quite small value,
will reduce the ability to maintain system variables “small enough”
AE3200 (II-2020/2021) 35
Tracking Controller

AE3200 (II-2020/2021) 36
PID Controller

 The controller Gb(s) can be applied either in feedback loop or forward loop
(before the plant G(s)
 The coefficients Kp, Ki, and Kd must be determined such that good closed loop
characteristics/response is obtained
 If Ki = Kd = 0  proportional controller  root locus, Ki = 0  PD controller, Kd
= 0  PI controller
AE3200 (II-2020/2021) 37
PID Control : Ziegler-Nichols Method – Closed Loop

Ziegler-Nichols is one of methods that can be used for


determining the value of Kp, Ki, Kd
The formulations are :

 First, set Kd=Ki =0 (proportional controller), and find Kp


such that the system response becomes oscillatory
Km Denotes the value of Kp when the response becomes
oscillatory (when the corresponding root locus crosses the
imaginary axis)
 ωm denotes the oscillation frequency
AE3200 (II-2020/2021) 38
example: 400
𝐺 𝑠 =
𝑠 𝑠 + 30𝑠 + 200

Km= 14.68

ωm= 14

AE3200 (II-2020/2021) 39
K=13

K=14
K=15

AE3200 (II-2020/2021) 40
𝐾 = 0,6. 𝐾
Kp = 8.8080
𝐾 .𝜔
𝐾 =
𝜋 Ki = 39.2514

𝐾 =
𝐾 .𝜋 Kd = 0.4941
4𝜔

AE3200 (II-2020/2021) 41
• Other PID Coefficients Table

• Kcr is the value of


proportional gain
when the response
becomes oscillatory
(harmonic)
• Pcr is the period of
the oscillation

AE3200 (II-2020/2021) 42
Ziegler-Nichols : Open Loop / Step Response Method

Suppose a step input is given to the plant, and its response is obtained

Examining the step response, it can be determined the following parameters :

 Step response has to be


in a “S-shape”
 Approximated as a first
order lag with delay

 Parameters T and L
AE3200 (II-2020/2021) 43
• Open Loop / Step Response Method
Using the parameters T and L, the PID controller coefficients can be determined as:

AE3200 (II-2020/2021) 44
• Effects of Kp, Ki, Kd on system response

AE3200 (II-2020/2021) 45
Final Value & Initial Value Theorem
Suppose a system G(s) is excited by an input (example:step function), then the
output will be y(s)=G(s)u(s).

The output y when t∞ (stationary/steady state condition) can be


expressed as:

The steady state value of y can also be obtained from its formulation in
Laplace domain using final value theorem:

→ →

Also, the output y when t 0 can also be known using initial value
theorem:

→ →

AE3200 (II-2020/2021) 46
Steady state error (tracking):

→ →

In Laplace domain:

𝐸 𝑠 =𝑅 𝑠 −𝑌 𝑠 =𝑅 𝑠 −𝐺 𝑠 𝐸 𝑠
𝐸 𝑠 +𝐺 𝑠 𝐸 𝑠 =𝑅 𝑠

1
⇒ 𝐸 𝑠 = 𝑅 𝑠
1+𝐺 𝑠

Using final value theorem

→  Depends on the type of the


input R(s)
AE3200 (II-2020/2021) 47
For step input r(t)=R0  R(s)=R0/s

→ →  Position Error
constant

For ramp input r(t)=R0t  R(s)=R0/s2

→ →  Velocity Error
constant

For parabolic input r(t)=R0t2  R(s)=R0/s3


 acceleration Error
→ → constant
AE3200 (II-2020/2021) 48
Tracking Controller

 Reguator problem
 Changing the closed loop
characteristics
 Modifying transient response

Tracking problem :
y(t)  r(t)
To control system variable so
that it can follow a given
reference command r(t)

AE3200 (II-2020/2021) 49
Tracking Control – Multivariable system
A. System with integrator

AE3200 (II-2020/2021) 50
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 51
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 52
Tracking Control – Multivariable system

• Recall Pole placement method (Direct, Canonical form,


Ackermann)
• Other multivariable method : Linear Quadratic (Optimal
Control)

AE3200 (II-2020/2021) 53
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 54
Tracking Control – Multivariable system

• Defining the polynomial coefficients required


for computing the controller
AE3200 (II-2020/2021) 55
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 56
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 57
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 58
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 59
Tracking Control – Multivariable system
B. Tracking with intergral action

AE3200 (II-2020/2021) 60
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 61
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 62
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 63
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 64
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 65
Tracking Control – Multivariable system

AE3200 (II-2020/2021) 66
AE3200 (II-2020/2021) 67
AE3200 (II-2020/2021) 68
AE3200 (II-2020/2021) 69
Open loop poles:
Suppose another set of closed poles is chosen : 0.0000 + 0.0000i
• -10.5000 + 0.0000i -2.5000 + 2.5960i
• -7.0000 + 2.0000i -2.5000 - 2.5960i
• -7.0000 - 2.0000i
• -5.7500 + 0.0000i
Then the following K is obbtained :
• K = [ -13.1915 -1.9573 -1.8281 151.6293 ]
• Giving the following tracking response

AE3200 (II-2020/2021)
Choosing new poles that are farther away from the open loop poles will require 70
more control effort
Tracking Controller : Optimal Control

AE3200 (II-2020/2021) 71
Tracking Controller : Optimal Control

AE3200 (II-2020/2021) 72
Tracking Controller : Optimal Control

AE3200 (II-2020/2021) 73
Tracking Controller : Optimal Control

AE3200 (II-2020/2021) 74
Tracking Controller : Optimal Control

AE3200 (II-2020/2021) 75
Tracking Controller : Optimal Control

• R element value is
bigger, than the system
will limit the value of
control variable to
smaller value
• It will affect the system
variables response
AE3200 (II-2020/2021) 76
Tracking Controller : Optimal Control

The 4th state is the “error” state  the error is suppressed more

AE3200 (II-2020/2021) 77
Tracking Controller : Optimal Control

• The error state is


suppressed more,
hence the tracking
variable (theta) try
harder to follow
reference value r
AE3200 (II-2020/2021) 78

You might also like