The Concept of Multi-channel

System in Chemical Process

Marthen Luther Doko

Department of Chemical Engineering

Institut Teknologi Nasional Bandung

Abstract

The real Processes in chemical Engineering application, especially for non static model, i.e., we as
chemical engineers are concerned with the problem of dynamics of multi input multi ouput process system.
If such systems are modeled mathematically and give systems of differential equations with initial
conditions.

The basic model for a two-channel plant is the following input-output model in terms of its transfer matrix
Tv:
y1 u T T 12 u1
[ ] [][
y2
=T p 1 = 11
u2 T 21 T 22 u 2
,
][ ] (1)

where T11 Ppxm, T12 Pp×n, T21 Pq×m, and T22 Pq×n We assume that
T11 is strictly proper, (2)
i.e., each entry of T11 is strictly proper. This is a standard simplifying assumption made to avoid
complications concerning the well-defined nature of the feedback loop when a feedback is applied around
the first channel.
This model is widely used for various control problems where it is either convenient or necessary to distinguish
between two types of outputs and inputs: The outputs that can be used as inputs to the controller and those
that are to be modified, the inputs which can be used for control purposes and those with unwanted influences
on the plant. Naturally, some outputs may be included in both channels if they are measurable, i.e., can be
used to drive the controller, while at the same time its behavior needs to be changed. Similarly, a particular
input may have unwanted influences on the plant and it may be suitable for control purposes, in which case it
may be included in both channels. Motivated by application, the output vector 2/1 is called the measured output
and y2the controlled output, the input vector Mi is called the control input and tt2the disturbance input. Thus, the
first channel of the plant is the control channel around which the feedback is applied. The need to use a two-
channel system model can also arise due to geographical separation of various subplants of the original plant
as in the case of large-scale plants. (For such applications, the connotations above need to be modified as we
shall do so in other paper.
The plant transfer matrix can be represented in matrix fractions over S as
T T 12 P U U 12
[
T p= 11
T 21 T 22][ ]
= 1 Q−1
P2
11 =[ R1
[
R2 ] +
]
U 21 U 22
,
(3)
where P1 Spxr,P2 Sqxr,Q11 Srxr, R1 Srxm, R2Srxn, W  Spxm, W12SpXn, W21\ Sqxm, W22 Sqxn with Q11
nonsingular. We assume that this representation is bicoprime, i.e.,
(P1, P2, Q11) is right coprime, {Q11, R1,R2}
is left coprime.
Although the overall representation in 3.13 is bicoprime, the representation of the control input-to-
measured output subplant
−1
T11= P1 Q11 R1 +U (4)
is usually not bicoprime. In fact, the cancellations that occur in the right-hand side of equation (4) are the
unstable input and output decoupling zeros of (P1,Q11,R1,U) and are of primary importance in many control
problems. Two natural bicoprime fractional representations for T11can be obtained by following the procedure
outlined in Section 3.3 through 3.4-3.8. To obtain a left and right coprime fractional representation of T11,
let
−1
Q11 RtQ-1R =Pr Qr
−1
PQ-1=
for some left coprime matrices (Q1, Rl) and right coprime matrices (Pr,Qr) over S.Thus, by Lemma (3.3), we
have
K −L = Q Nl
=I 0 ,
[ R l Q1 −P ][ ][ ]
Ml 0 I (3.15).
and
Q R = M −Pr = I
[ −Lr Kr N Qr ][ 0 ] [ 0I ] , (3.16)
for some matrices K, L, M, N, M t, Nt, K r, L rover S. Further,
−1
Z11 = (PPr + WQr) r = Q−1 (Q1W + R1R).
Q −1
(3.17)
It is easy to see using right coprimeness of (P r, Qr) and (P, Q) that the first representation in 3.17 is
right coprime. Similarly, by left coprimeness of (Q l ,R l ) and (Q, R), it follows that the second
representation in 3.17 is left coprime.
The system matrices associated with various subplants of 3.13 are denoted by
Q R3
Π ij := 11
−Pi W ij [ , i, j =1,2,
] (3.18)
where W 11 := W. Let us further define the nonsingular matrices
C 0
Γ := C 0 , Γ 1 := 1
0 1 [ ] [ ]
0 1
,
(3.19)
where I is the identity matrix of size m, and
D 0
Δ:= D 0 , Δ 1 := 1
0 I [ ] [
0 1
,
] (3.20)
where I is the identity matrix of size p and C, C 1 , D, D 1 are as defined in Section 3.3. Then, system
matrices associated with various reduced fractions are denoted by
D1 Q0 R22 DQ R2
Π C12 :=
[ −P0 W 12
, Π C121 :=
]
−P W 12
,
[ ] (3.21)
and by
QC 1 R Q C R0
Π C12 :=
[ −P2 W 21
D1
, Π 21
]
:= 0
−P2 W 21
,
[ ] (3.22)
Note that the system matrices we just defined are interrelated by
D C C D D1
Π 21=Π 12 Γ =Π 12 Γ 1 , Γ 21=ΔΠ21=Δ 1 Π 21 (3.23)
by 3.18-3.22.
As a consequence of the fact that (P 1, P2 , Q11 ) is right coprime and (Q 11 , R1, R 2) is left coprime,
the pair (P 2 ,C 1 ) is right coprime and the pair (D,R 2 ) as well as (D 2 ,R 2 ) is left coprime. Let
−1 −1
P2 C1 =C̄ 1 T
for some left coprime pair ( C̄,T ) and
−1
-1
D R2 = S D̄ …..
for some right coprime pair (S, D̄ ). The system matrices
……………

have the same nonunit invariant factors as the system matrices

Π D12 , Π C121 , Π 22 , of 3.21, 3.22, 3.18,
respectively, by an application of Fact (3.2). Further note that,
……… (3.25)
and
………. (3.26)
where
12 := KS – LW12 D̄
12 := TM –
C̄1 W21N

12 := RlS –Q1W D̄

12

21 := TPr +
C̄1 W Q
21 r

It follows by the unimodularity of the first and the third matrices in both 3.25 and 3.26 that the nonunit
invariant factors of
Π̄
12 are associates of those of 12 and that the nonunit invariant factors of Π̄ 21
are associates of those of 12.
The matrices in 3.24 and the matrices in 3.21 and 3.22 are interrelated by
………..
where
………..

Figure 3.1 The closed-loop system Zf

3.5 TWO-CHANNEL PLANTS WITH FEEDBACK

The compensation scheme that is used to change the behavior of the plant is a nic output feedback which
processes the measured outputs and feeds the >1 input. There may be an external input to the resulting
closed-loop system control channel. This external input may serve as a new control input for the ing two-
channel system should a further need for control arise. The feedback lamic since the present as well as the
past measured outputs are processed in edback loop. This provides greater flexibility in control than a static
output ick at the cost of (sometimes a great deal of) complexity in implementation.
Let Zc PmxP and consider the feedback law
u1 = -Zcyi +ue1
resulting in the closed-loop plant
−1
y 1 Z 11 −Z 11 Y C Z 11 Z 12−Z 11 Y C Z12 ue1
[ ][ =
y 2 Z 21−Z 21 Y C Z11 Z 22−Z 21 Y C Z 12 ] [ ] +
u2
(3.30)
where the matrix
YC = (I + Zc Z11)-1 Zc
Is in pmxpby thej assumption 3.12. Various control specifications that will be considered in the subsequent
chapters concern the closed-loop transfer matrix
Zdc := Z22- Z21YcZ12
between the disturbance input and the controlled output. The resulting closed-loop system is illustrated in
Figure 3.1. If the transfer matrix Zc of the compensator is written in matrix fractions as
−1
Zc = P c c c
Q R (3.31)
Where Pc Smxs, Qc Ssxs, Rc Ssxp, then we can write a natural matrix fractional representation for the
closed-loop transfer matrix Zf of 3.30
−1
P1 −WPc Q11 R1 Pc R2 R2
zf =
[ =
][
P 2 −W 21 P c −R c P1 Qc +R c WPc ] [ +
Rc W
This expression can be verified by writing the inverse of the denominator matrix
RC W 12 ]
Q11 R 1 Pc
Φ11 =

explicitly as
[
−R c P 1 Qc −R C WP c ]
 Q−1−Q−1−R1 P C T −1 R C P1 Q−1 −Q−1 R 1 Pc T −1
Φ−1
11 =
[ 11 11 C

T −1−R c P1 Q −1
c 11
11 11
c

T −1
CC
…….
] (3.33)
where Tc:=Qc+ RcZ11Pc, carrying out the multiplication, and simplifying the expressions. The resulting
fractional representation from 3.32 for Zdc is clearly
−1
Q11 R c P1 R2
Z dc =[ P2 −W 21 P c ]
[
−R c P1 Q c +R c WPC ][ ] Rc W 12
+W 22
………. (3.34,
We remark that the representation in 3.31 of the compensator is more general than what is really needed. In
fact, either a left.or a right fractional representation for the compensator will do for most of the problems
considered in the book. However, these more special representations and the corresponding closed-loop
expressions can easily be obtained on letting Pc = I or Rc — I in 3.31 and in subsequent equalities.