Beruflich Dokumente
Kultur Dokumente
Introducing
Module Process integration
5 – Controllability for Environmental Control in Engineering Curricula
Analysis PIECE
1
NAMP PIECE
Nap tha
Light g as o il
Hea vy g as oil
- m1 y1
Gc1 G11
+ +
+
G12
G21
+
m2 + + y2
Gc2 G22
- Loop 2
This net change is the sum of the main effect of m1 on y1 and the
interactive effect provoked by m1 interacting with the other loop.
y* K11 1 K11 *
K12 K 21
K11K 22
A good measure of how well a system can be controlled (λ) if m1
is used to control y1 is:
y1m y1m
11
y * y1m y1r
Module 5 – Controllability Analysis 7
NAMP PIECE
Case 1 : λ11=1
This case is only possible if y1r is equal to zero. In physical
terms, this means that the main effect of m1 on y1, when all the
loops are opened, and the total effect, measured when the
other loop is closed, are identical.
This will be the case if:
• m1 does not affect y2, and thus, there is no retaliatory control
action from m2, or
• m1 does affect y2, but the retaliatory control action from m2
does not cause any change in y1 because m2 does not affect y1.
Under these circumstances, m1 is the perfect input
variable to control y1 because there will be NO
interaction problems.
Module 5 – Controllability Analysis 8
NAMP PIECE
Case 2 : λ11=0
This condition indicates that m1 has no effect on y1,
therefore y1m will be zero in response to a change in m1.
Note that under these circumstances, m2 is the perfect
input variable for controlling y2, NOT y1. Since m1 does not
affect y1, y1 can be controlled with m2 without any
interaction with y1.
Case 4 : λ11>1
This is the condition where y1r is the opposite sign of y1m, but it is
smaller in absolute value. In this case y1* (y1r +y1m) is less than
the main effect y1m, and therefore a larger controller action m1 is
needed to achieve a given change in y1 in the closed loop than in the
open loop. For a very large and positive λ11 the interaction effect almost
cancels out the main effect and closed-loop control of y1 using m1 will
be very difficult to achieve.
Case 5 : λ11< 0
This is the case when y1r is not only opposite in sign, but also
larger in absolute value to y1m. The pairing of m1 with y1 in this
case is not very desirable because the direction of the effect of m1 on y1
in the open loop is opposite to the direction in the closed loop. The
consequences of using such a pairing could be catastrophic.
Module 5 – Controllability Analysis 11
NAMP PIECE
Quiz#4
y i
m j all loops open-loopgain
ij open ij for loop i under
y i closed-loopgain the control of m
j
n1 n 2 nn
PROPERTIES OF THE
RELATIVE GAIN ARRAY
i 1
ij ij 1
j 1
PROPERTIES OF THE
RELATIVE GAIN ARRAY
PROPERTIES OF THE
RELATIVE GAIN ARRAY
PROPERTIES OF THE
RELATIVE GAIN ARRAY
j all loops
open
ij
y i
m j all loops closed
except for
Module 5 – Controllability Analysis the m j loop 20
NAMP PIECE
Due to the fact that the equations found on the previous slide
represent steady-state, open-loop conditions, the differentiation
for the numerator portion of the relative gain is:
y1
K11
m1 all loops open
The second partial derivative (the denominator) requires Loop 2
to be closed, so that in response to changes in m1 , the second
control variable m2 can be used to restore y2 to its initial value
of 0. To obtain the second partial derivative, we first find from
Eq. 1b the value of the m2 must be to maintain y2=0 in the face
of changes in m1, what effect this will have on y1 is deduced by
substituting this value of m2 into Equation 1a.
Module 5 – Controllability Analysis 21
NAMP PIECE
11
1 K12K 21
where =
1 K11K 22
Module 5 – Controllability Analysis 23
NAMP PIECE
1
12 21
1
and 22 11
1
Module 5 – Controllability Analysis 24
NAMP PIECE
lim G ( s ) K
s 0
R K
1 T
ij K ij rij
Solution:
For this system, the steady-state gain matrix (K) is the following.
K11 K12
K
K 21 K 22
Module 5 – Controllability Analysis 28
NAMP PIECE
1 1 K 22 K12
K K
K 21 K 11
Quiz#5
• What advantages does the Matrix Method have over
the First Principles Method?
1
yields an RGA
0 1 0
G( s) s 1
4
3
0 1
3s 1 4 s 1
Note that since the element g12(s) is zero, the input m2 does not have
an effect on the output y1, however, the input m1 does influence the
output y2 as can be seen due to the fact that the g21 element is
nonzero. Upsets in Loop 1 requiring action by m1 would have to also be
handled by the controller of Loop 2. So, even though the RGA is ideal,
Loop 2 would be at a disadvantage. Thus, in deciding on loop pairing,
one should distinguish between ideal RGAs produced from diagonal or
triangular transfer function matrices.
Module 5 – Controllability Analysis 42
NAMP PIECE
• RULE #1
Pair input and output variables that have positive RGA
elements closest to 1.0.
Consider the following examples to demonstrate this rule.
For a 2x2 system with output variables y1 and y2, to be paired
with m1 and m2
If the RGA is… 0.8 0.2
0.2 0.8
0.3 0.7
0.7 0.3
NIEDERLINSKI INDEX
Niederlinski Index
Pairing Rule #1 is usually sufficient in most
cases, it is often necessary to use this rule in
conjunction with the theorem found on the
next slide developed by Niederlinski and later
modified by Grosdidier et al. This theorem is
especially useful if the system is 3x3 or larger.
NIEDERLINSKI INDEX
y(s)=G(s) u(s)
1.Rational, and
2.Open-loop stable
Module 5 – Controllability Analysis 48
NAMP PIECE
NIEDERLINSKI INDEX
NIEDERLINSKI INDEX
2. For a 2x2 system with a negative relative gain, ζ >1, the Niederlinski
index is always negative; hence 2x2 systems paired with
negative relative gains are ALWAYS structurally unstable.
• RULE #2
Any loop pairing is unacceptable
if it leads to a control system
configuration for which the
Niederlinski Index is negative.
5
3 1 1
= 1 1
K = G(0) 1
3
1 1
1
3
Module 5 – Controllability Analysis 54
NAMP PIECE
First, we need to take the inverse of this matrix, then take the
transpose of this matrix to obtain R, being:
10 4.5 4.5
4.5 1 4.5
4.5 4.5 1
10 4.5 4.5
4.5 4.5 1
4.5 1 4.5
K 0.148
N n
0
K
5/3
ii
i 1
From the RGA seen, there is only one feasible pairing, because all
of the other pairings violate Rule 2. The only feasible pairing is
a 1-1,2-2,3-3 pairing, but you will notice that this pairing
violates Rule 1, as the RGA element 1-1 is negative, but
according to the Niederlinski Theorem this system would NOT
be structurally unstable.
If the first loop is opened (the m1, y1 elements dropped from the
process model) the new steady-state gain matrix relating the 2
remaining input variables with the 2 remaining output variables
is:
~ 2 1
K
3 1
Module 5 – Controllability Analysis 61
NAMP PIECE
Quiz #6
Underdefined Systems
m inputs n outputs
As seen in the system above, there are less inputs m than there
are outputs n, thus is defined as an underdefined system.
m=the number of inputs = 2 m<n
n=the number of outputs = 4
Module 5 – Controllability Analysis 80
NAMP PIECE
Underdefined Systems
The main issue with underdefined systems is that not all
outputs can be controlled, since we do not have
enough input variables.
The loop pairing is easier if we make the following
consideration
By economic considerations, or other such means,
decide which m of the n output variables are the most
important, these m output variables should be paired
with the m input variables; the less important (n-m)
output variables will not be under any control.
Module 5 – Controllability Analysis 81
NAMP PIECE
Overdefined Systems
B4
m inputs n outputs
As seen in the system above, there are less inputs m than there
are outputs n, thus is defined as an underdefined system.
m=the number of inputs = 3
m>n
n=the number of outputs = 2
Module 5 – Controllability Analysis 82
NAMP PIECE
Overdefined Systems
Deciding the loop pairing of overdefined systems presents a real
challenge. In this case, there is an excess of input variables,
therefore we can achieve arbitrary control of the fewer output
variables in more than one way.
The situation we are faced with is as follows: since there are m
input variables to control n output variable (m>n), there are many
more input variables to choose from in pairing the inputs and the
outputs, and therefore, there will be several different square
subsystems from which the pairing is possible. There are m
possible square subsystems. n
m m!
Recall that: n = n! (m-n)!
Module 5 – Controllability Analysis 83
NAMP PIECE
The second approach is more time consuming, and involves too many
upsets to the process; for these reasons it is not desirable in practice.
Therefore, the first approach is preferred.
Module 5 – Controllability Analysis 87
NAMP PIECE
2. The main criticism of the RGA is that the RGA only provides
information about the steady-state interactions within a
process systems, and therefore, dynamic factors are not taken
into account by the RGA analysis.
Quiz#7
We have discussed this first point a great deal in the past slides,
this should signify importance of the choice of loop pairing in
controller design.
Now, we must address the issue of tuning the individual
controllers.
Module 5 – Controllability Analysis 92
NAMP PIECE
1.With the other loops on manual control, tune each control loop
independently until satisfactory closed-loop performance is obtained.
2.Restore all the controllers to joint operation under automatic control
and readjust the tuning parameters until the overall closed-loop
performance is satisfactory in all the loops.
When the interactions between the control loops are not too
significant, the procedure mentioned before can be quite useful.
However, for systems with significant interactions, the
readjustment of the tuning in Step 2 can be difficult and tedious.
One can cut down on the amount of guesswork that goes into
such a procedure by noting that in almost all cases, the controllers
will need to be made more conservative (ie. the controller gains
will have to be reduced and the integral times increased) when all
the loops are closed in comparison to when all of the individual
controllers are operating individually, with all of the other loops
open. The process of this changing of the control parameters is
referred to as “detuning”.
DESIGN OF MULTIVARIABLE
CONTROLLERS-Introduction
Design of Multivariable Controllers
In the next section, we will discuss the design of true
multivariable controllers that utilize all of the available process
output information jointly to determine what the complete input
vector u should be. Thus each control command from the
multivariable controller will be based on all of the output
variables, not just based on one. In principle, it will be possible to
eliminate all of the interactions between the process variables.
The objective of the next section is to present some of the
principles and techniques used for designing multivariable
controllers, as designing multivariable controllers is one of the
more challenging problems faced in industrial process control. We
will start by addressing loop decoupling, the most widely used
multivariable controller technique. We will then address Singular
Value Decomposition (SVD) which is a means of determining
when it is structurally unstable to apply decoupling to a system.
Module 5 – Controllability Analysis 97
NAMP PIECE
yd + - ε1 v1 + u1 +
y1
1
gc1 g11
+ +
gI1
g12
Please consider the following system:
g21
gI2
+
yd
+
ε2 v2 + +
u2 + y2
gc1 g22
2
-
DECOUPLING INTRODUCTION
Decoupling:
In Decoupling, as seen in the Figure on Slide 132, additional
transfer function blocks are introduced between the single-loop
controllers and the process, functioning as links between the
otherwise independent controllers. The actual control action
experienced by the process will now contain information from all
of the controllers. For example, a 2x2 system, whose individual
controller outputs are gc1ε1 and gc2ε2 if the decoupling blocks for
each loop have transfer functions of gI1 and gI2 respectively, then
the control equations will be given by:
u1=gc1ε1+gI1 (gc2ε2)
u2=gc2ε2+gI2 (gc1ε1)
Module 5 – Controllability Analysis 101
NAMP PIECE
Decoupling Introduction
We know from our discussion of input/output pairing that the
pairing of y1-u1, y2-u2,…yn-un couplings are desirable; it is however
the yi-uj cross-couplings, by which yi is influenced by uj (for all i
and all j with i≠j), that are undesirable: they are responsible for
the control loop interactions.
It is clear that any technique that eliminates the undesired cross-
coupling will improve the performance of control systems. It is
however NOT possible to ELIMINATE the cross-couplings; that is
a physical impossibility since it will require altering the physical
nature of the system. Consider an example of this on the
following slide.
Module 5 – Controllability Analysis 102
NAMP PIECE
yd + ε v u y
Gc GI G
- Interaction
Single Loop
Controller Compensation
SIMPLIFIED DECOUPLING
SIMPLIFIED DECOUPLING
Which Yields
y1=(g11+g12gI2)v1+(g11gI1+g12)v2 (Eq.1-D)
y2=(g21+g22gI2)v1+(g22+g12gI1)v2 (Eq.2-D)
Module 5 – Controllability Analysis 109
NAMP PIECE
If we now substitute our expressions for gI1 and gI2 into Equations
1-D and 2-D respectively we will yield:
g12g21 g12g21
y1= g11 - v1 y 2 = g22 - v2
g22 g11
- y1
yd1 + v1 g12g21
gc1 g11 -
g22
yd2 + gc2
v2 g g y2
g22 - 12 21
- g11
Let’s consider that the closed loop system is under steady state. If
the steady state gain for an element gij =Kij, observe how the
system is expressed at steady-state.
K K K 12K 21
y1= K 11 - 12 21 v1 y 2 = K 22 - v2
K 22 K 11
yd1 + - v1 + u1 + y1
gc1 g11 +
+
gI12 + g12 +
gI13 g13
u3 u 2 u1
gI21 g21 +
yd2 + +
u2 y2
v2 + +
gc2 +
g22
+
gI23 g23
gI31 g31
gI32 + g32
yd3 +
v3
+
u3
+ + y3
gc3 + g33
+
-
Module 5 – Controllability Analysis 115
NAMP PIECE
GENERALIZED DECOUPLING
Generalized Decoupling
Please refer to Figure 1-D which we will use this figure to outline
a more generalized procedure for decoupler design.
- ε1 v1 +
u1
yd + +
gc1 g11
1 + +
gI1
g12
gI2
g21
+
yd + ε2 v2 + +
u2 + y2
2 gc1 g22
-
GENERALIZED DECOUPLING
y=Gu
u=GIv
So that:
y=GGIv
GENERALIZED DECOUPLING
GGI=GR(s)
y=GR(s)v
GI=G-1 GR
GR=Diag[G(s)]
Quiz #8
LIMITATIONS OF
DECOUPLING
LIMITATIONS OF
DECOUPLING
To illustrate the idea of stable and casual, please consider the 2x2
compensators we saw in Figure 1-D whose transfer functions are
GI1 and GI2 must be casual (no e+αs terms) and stable.
To satisfy causality for the 2x2 system, any time delays in g11
must be smaller than the time delays in g12 and a similar condition
must hold for g22 and g21.
To satisfy stability, a second condition that g11 and g22 must not
have any right hand plane zeros and also g12 and g21 must not
have any right hand plane poles. This leads to the following
general conditions that must be satisfied in order to implement
simplified dynamic decoupling for N x N systems.
Module 5 – Controllability Analysis 123
NAMP PIECE
LIMITATIONS OF
DECOUPLING
1.Causality: In order to ensure causality in the compensator
transfer functions the time-delay structure in G(s) must be such
that the smallest time-delay in each row occurs on the diagonal.
For simplified decoupling, this is an absolute requirement, but it is
possible to add delays to the inputs u1,u2…un, to satisfy the
requirement if the original process G does not comply. This is
equivalent to defining a modified process as Gm:
Gm=GD
ed11s 0
Where D is a
ed22s
diagonal matrix of D(s)=
time delays dnns
0 e
Module 5 – Controllability Analysis 124
NAMP PIECE
LIMITATIONS OF
DECOUPLING
LIMITATIONS OF
DECOUPLING
PARTIAL DECOUPLING
Partial Decoupling
If some loop interactions are weak or if some of the loops do not
need to achieve high performance, the partial decoupling is a
method one should consider. If this is the case, only a subset of
the control loops where the interactions are important and high
performance is important are focused on.
STEADY-STATE
DECOUPLING
Steady-State Decoupling
The difference between dynamic decoupling and steady-state
decoupling is that dynamic decoupling uses the complete,
dynamic version of each transfer function element to obtain the
decoupler, and steady-state decoupling only uses the steady-state
gain portion of each of the transfer elements.
STEADY-STATE DECOUPLING
FOR A 2X2 SYSTEM
Simplified steady-state decoupling for a 2x2 system
K12 K 21
Here: gI1 = - and gI2 = -
K11 K 22
STEADY-STATE DECOUPLING
FOR A 2X2 SYSTEM
Quiz #9
W is the m x m matrix
W=w1 w2 wm
Whose columns wi, i=1,2,…,m are called the left singular vectors
of A; these are normalized (orthonormal) eigenvectors of AAT.
V is the n x n matrix:
V=v1 v2 vn
Whose n columns vi, i=1,2,…,n are called the right singular
vectors of A; these are normalized (orthonormal) eigenvectors of
ATA.
Module 5 – Controllability Analysis 133
NAMP PIECE
WTW=I=WWT
So that W-1=WT
Also VTV=I=V VT
So that V-1=VT
Module 5 – Controllability Analysis 134
NAMP PIECE
A=W ∑ VT
1 2
A 2 1
2 1
Therefore,
T 9 2
A A=
2 6
Module 5 – Controllability Analysis 137
NAMP PIECE
Therefore:
2 1
5 5
v =
1 2
5 5
2 1
5 5
VT =
1 2
5 5
To find the adjoint of this matrix, we first find the cofactors and
take the transpose of the matrix of cofactors. In this case,
0 0 0
adj(AA T - 1 I) = 0 25 25
0 25 25
Module 5 – Controllability Analysis 143
NAMP PIECE
0 0
1 1
w 2 = 0
1 w3 =
w1 = 2
2
1
0 1
2 2
0 1 0
1 1
W = 0
2 2
1 1
0
2 2
STEADY-STATE DECOUPLING BY
SINGULAR VALUE DECOMPOSITION
STEADY-STATE DECOUPLING BY
SINGULAR VALUE DECOMPOSITION
η= WTy
And
μ =∑ VT u
Module 5 – Controllability Analysis 147
NAMP PIECE
STEADY-STATE DECOUPLING BY
SINGULAR VALUE DECOMPOSITION
η=∑μ
Because ∑ is diagonal, this indicates that the system is completely
decoupled at steady state.
yd ηd + μ u y
WT Gc∑ V G
-
η
WT
Module 5 – Controllability Analysis 148
NAMP PIECE
STEADY-STATE DECOUPLING BY
SINGULAR VALUE DECOMPOSITION
REFERENCES
References: