MULTIVARIABLE

CONTROL
.

to this point, the fundamentals of process dynamics and control have been il-
lustrated by single-input single-output (SISO) systems. The processes encountered
in the real world are usually multiple-input multiple-output systems (MIMO). To
explore these concepts, consider the interacting, two-tank liquid-level system in
Rig.” 30.1 where there is one input, the flow to tank 1 and one output, the
in tank 2 In this figure, is related to m by a second-order transfer
function. From the point of view of a SISO system, the relation between and
may be represented by the block diagram in Fig. One may place a
feedback control system around the open-loop system of Fig. 30. lb to maintain
control of
Now consider the same process of Fig. 30.1 in which there are two inputs
and t wo outputs and This system is shown in Fig.
change in m alone will affect both outputs and A change in
 alone will also change both outputs. (Remember that this is an interacting process
which the level in tank. 1 is affected by the level in tank 2.) The interaction
between inputs and outputs can be seen more clearly by the block diagram of Fig.
In this diagram, the transfer functions show how the change in one of the
inputs affects both of the outputs. For example, if a change occurs in only
responses of and are .

: .
..

transfer functions in Fig. will be worked out for a specific set of pro-
\‘:: parameters in Example 30.1. (If the tanks were noninteracting, G = 0, with
 STATE-SPACE METHODS

FIGURE 30-1
Single-input single-output system (SISO): (a) two-tank interacting level system, block diagram
for SISO system.

Ml
A

FIGURE 30-2
Multiple-input multiple-output system (a) level process, (b) block diagram.
 CONTROL 455

the result that a change in flow to tank 2 would not affect H If both H and
are to be controlled, a single control loop will not be sufficient; in this case two
control loops are needed. The addition of control loops to the interacting system
will be considered in the next section.

CONTROL OF INTERACTING SYSTEMS

The problem of controlling the outputs of an MIMO system will be discussed
by means of a 2 X 2 system in Fig. 30.3. The problem can be extended
to the case of more than two pairs of inputs and outputs by the same procedure
described here. The control objective is to control C and independently, in
spite of changes in and or other load variables not shown. control
loops are added to the diagram of Fig. 30.3 as shown in Fig. 30.4. Each loop
has a block for the controller, the valve, and the measuring element. In principle,
the multiloop control system of Fig. 30.4 will maintain control of and
However, because of the interaction present in the system, a change in will also
cause to vary because a disturbance enters the lower loop through the transfer
function Because-of interaction, both outputs and will change if a
change is made in either input alone. If and provide weak interaction, the
two-controller scheme of Fig. 30.4 will give satisfactory control. In the extreme,
if = = 0, we have no interaction and the two control loops are isolated
from each other.
To completely eliminate the interaction between outputs and set points, two
more controllers (cross-controllers) are added to the diagram of Fig. 30.4 to give
the diagram shown in Fig. 30.5. In principle, these cross-controllers can be de-
signed to eliminate interaction. The following analysis, which is expressed in
matrix form, will lead to the method of design for cross-controllers that will
eliminate interaction.

Ml

FIGURE303
MIMO system for two pairs of inputs and outputs.
456 STATE-SPACE METHODS

FIGURE 30-4
Multiloop control system with two controllers.

Response of Multiloop Control System

From Fig. 30.5, we may write by direct observation the following relationships
in the form of the matrix expression
C =

We also may write from Fig. 30.5

= + (30.2)
= + (30.3)

FIGURE. 30-5
Multiloop control system with two primary controllers and two cross-controllers,
 CONTROL 457

where and are the transfer functions for the valves. Equations (30.2) and
(30.3) may be written in matrix form as
M = (30.4)

where =
1 (valve matrix)

=
c21 1 
(controller matrix)

E =
I

From Fig. 30.5, we write directly

= (30.5)
= (30.6)
where and are the error signals from the comparators. Equations (30.5) and
(30.6) can be written in the matrix form
E = (30.7)

where (measuring element matrix)

From Eqs. (30.1) and we obtain

C = (30.8)
If we let = Eq. (30.8) becomes
C = (30.9)
Combining Eqs. (30.7) and (30.9) gives
C = (30.10)
We may now solve Eq. (30.10) for C to obtain
C = [I + (30.11)

Notice that the closed-loop behavior expressed by this matrix equation is

analogous to the closed-loop response of a SISO system, which may be written

(30.12)
458 STATE-SPACE METHODS

FIGURE
Block diagram for MIMO control sys-
tem in terms of matrix blocks.


The matrix term is equivalent to the scalar term

A block diagram equivalent to the diagram for the MIMO control system in
Fig. 30.5 is shown in Fig. 30.6. In this diagram, the blocks are filled with the
matrices in Eqs. and (30.7). The double line indicates that more
than one variable is being transmitted. Each block contains a matrix of transfer
functions that relates an output vector to an input vector. The diagram can be
simplified by multiplying the three matrices in the forward loop together and
calling the result as was done to obtain Eq. (30.9). The simplified diagram
is shown in Fig. 30.7.

In order for no interaction to occur between C and in Fig. 30.5 (i.e., R affects
only C and affects only the off-diagonal elements of
in Eq. (30.11) must be zero. Since and are diagonal, [I + will
be diagonal if is diagonal. Multiplication of the matrices in the expression for
is now shown:
=

=
The result of multiplying these matrices gives
+
+
Setting the off-diagonal elements to zero and solving for
+
+
and
1 (30.13)

give

Gc l 2 = (30.14)

G c21 = (30.15)

Reduced block diagram for control system where

 459

The following example will give some experience with the computations
involved in applying the theory developed so far in this chapter.

Example 30.1. For the two-tank, interacting liquid-level system shown in Fig. 30.8,
develop the block diagram for an MIMO system corresponding to Fig. 30.3.
Material tank 1 and tank 2 give following differential equations:

= + (30.17)

Introducing the parameters given in Fig. 30.8 into Eqs. (30.16) and (30.17) gives
= + (30.18)
= + (30.19)
These equations may be written in matrix form as

where

We use (29.15) to obtain

C(S) = (30.20)
Writing (30.20) in the form of (30.1) gives
C =
where =
After several steps involving the inversion of A) and multiplying the result of
inversion by B, one gets

= 4
(s +
4
+ 3)
+ 7)
1 (30.21)

FIGURE
Process for Example 30.1: =
1, = = = 2,
R3 = 1.
460 STATE-SPACE METHODS

r
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2

0 1.5 t 3

FIGURE 30-9
Open-loop response for Example 30.1. (a) = l/s, = 0, (b) = l/s, =

The block diagram can now be drawn as shown in Fig. 30.3 with
4
= (s + + 7) = (s + + 7)
4 + 3)
= + + 7) = (s + 7)
Notice that the diagonal elements of are of the form
+
++ 7)
These elements, which relate cl to m and to will produce a second-order
response to a step change in input that has a finite slope at the origin because of
the numerator term + . In contrast, the off-diagonal elements have second-order
transfer functions without numerator dynamics, for which case the step response
will be second-order with zero slope at the origin. The responses of and for
unit-step changes in and taken separately are shown in Fig. 30.9.

Example 30.2. For the two-tank liquid-level system of Example 30.1, determine the
controller transfer function matrix needed to eliminate interaction. The primary
controllers are to be proportional, i.e., = = The diagram of
the control system is shown in Fig. 30.10. The block labeled controller contains the

FIGURE 30.10
Process for Example 30.2. =
A2 = = = 2,
= = =
 461

four transfer functions that the elements of G,. In this problem, is a unit
diagonal matrix i.e., = = 1.
From (30.14) and (30.15) we obtain
4 (s + + 7)
Gc l 2 = - - =
(s + +
or

Gc l2 = (30.22)

4 (s + + 7)
G c21 = - - =
(s + + + 3)
or

= (30.23)

Having found the transfer functions for the cross-controllers, we can now determine
the nature of the uncoupled response of c to a change in r and of to a change
in
Inserting = = 1 and the expressions for and from
(30.14) and (30.15) into (30.13) gives for

(30.24)

Inserting the appropriate elements of the matrix and the

matrix in (30.24) gives after considerable simplification

= + (decoupled system) (30.25)

0

The block diagram for this decoupled MIMO system is shown in Fig. 30.11.
Assuming that the measurement matrix is a unit diagonal matrix, the diagram

o 30-11
Block diagram for decoupled system in Example
30.2.
462 STATE-SPACE METHODS

FIGURE 30-12
Simplified block diagram for Example 30.2.

in Fig. 30.11 can be simplified to the unity feedback diagram of Fig. 30.12. From
Fig. 30.12, we may write directly
C=
E = R - C

therefore C =
or

1
From this expression, we may write

Solving for gives

Inserting from (30.25) gives

= + (30.26)

In a similar way, one can show that

The result shows that the cross-controllers of Eqs. (30.22) and (30.23) give two
separate noninteracting control loops as shown in Fig. 30.13.
The response of the control system of Fig. 30.10 is shown in Fig. 30.14
for a unit-step change in In Fig. no cross-controllers present in
the matrix G,. In Fig. cross-controllers having the transfer functions given
by Eq. (30.22) and (30.23) present. As expected, for the case of no
controllers, one sees from Fig. that a request for a unit-step change in r
 CONTROL 463

causes and to change. For case where cross-controllers are present,

one sees from Fig. a change in does not affect as demanded by
a decoupled system.

To avoid the offset associated with proportional control, we can use PI

controllers for the primary controllers for the decoupled system. To study the
effect of PI controllers for the decoupled system, let

and

For this case, the cross-controller transfer functions may be obtained from,
(30.14) and (30.15); the results are

+ + 1)
Gcl2 = =
+ 5 ) + 3)
A simulation using these four controller transfer functions with = = 4 is
shown in Fig. 30.15. From the transient response, we see that moves toward
the set point of 1.0 and that does not change, as is expected for a decoupled
system.

1.0
r

0 1.5 '0 1.5

FIGURE 30-14
Response for control system in Example 30.2 for = = 0, = = 4, =
= 4. (a) no cross-controllers, (b) cross-controllers present.
464 STATE-SPACE METHODS

Cougnanowr
1.0,

Response of decoupled control system in Example 30.2

for PI primary controllers: = = +
= = 0.

STABILITY OF MULTIVARIABLE SYSTEMS

Determining the stability for a multivariable control system, such as the one in
Fig. 30.4 or Fig. 30.5, can be much more complicated than for an SISO system.
The transfer function for the closed-loop response of an MIMO system is given
by Eq. (30.11):
C = [I +
To invert this expression, we write
+ (30.28)

The numerator of this expression is an matrix; the denominator is a

nth order polynomial. To simplify the following argument, let the matrix in Eq.
(30.28) be 2 X 2. Let the elements of the numerator, after expansion, be written
as follows:

Let the elements of

+ =
be written as follows:
1 (30.29)

(30.30)

Expansion of the determinant in Eq. using Eq. is shown below

= +

or
+ = + + (30.31)
Equation (30.31) is a polynomial expression, for which the order will depend on
the order of the transfer functions in and G,. Equation (30.28) can now be
written in terms of the expansions shown in Eqs. (30.29) and (30.31) as follows:
 CONTROL

Since each term contains the polynomial + in the denominator, the

stability of the multivariable system will depend on the roots of the polynomial
equation
= o (characteristic equation) (30.32)
Equation (30.32) is the characteristic equation of the multivariable system.
Although Eq. (30.32) has been derived here for the case where is a 2
2 matrix, one can show that Eq. (30.32) applies to the general MIMO system
of Fig. 30.7 in which is a matrix of any size X n). If the roots of
the characteristic equation are in the left half of the complex plane, we know
that the system is stable. One method to be used for examining the stability of a
multivariable system is to apply the Routh test to the characteristic equation of Eq.
(30.32). In practice, the characteristic equation can be of high order for a simple
2 X 2 multivariable control system. Example 30.3 illustrates the determination of
stability for a multivariable control system.

Example 30.3. For the control system of Example 30.2, which is shown in Fig.
30.10, determine stability for the case where = = and there
are no cross-controllers present (i.e., = = 0) also let and be
unit matrices. From Example 30.1, we have for the elements of
4
= (s + 7) = (s + + 7)

4 + 3)
= (s + + 7) = (s + + 7)

Since = I, = Since = I, the characteristic equation of Eq.

(30.32) can now be written as
= 0 (30.33)

Introducing the elements of the matrices and into Eq. (30.33) gives, after
expansion of the determinant

[(s + + + + 7) + + = 0
For given values of and this expression can be expanded into a fourth
order polynomial equation of the form
+ + + + A = 0 (30.34)
where , y, and A will include the gains K and
466 STATE-SPACE METHODS

The Routh test can be applied to Eq. (30.34) to determine whether or not the system
is stable. From this simple example, the reader can appreciate the algebraic tedium
that may be needed to determine the stability of a multivariable system.
One way to express the stability of this system is to plot the stability boundaries
on a graph of versus The region within the boundaries gives the combinations
of values of and for which the system is stable. Since the details of stability
boundaries is beyond the scope of this chapter, the reader may consult Seborg,
Edgar, and Mellichamp (1989) for examples of stability boundaries for multivariable
systems.

SUMMARY
Most of the systems encountered are multiple-input multiple-output (MIMO) sys-
tems. Such systems have several inputs and several outputs that are often inter-
acting, meaning that a disturbance at any input causes a response in some or all
of the outputs. This interaction in an MIMO system makes control and stability
analysis of the system very complicated compared to a single-input single-output
(SISO) system. A convenient way to describe an MIMO system is by means of
a block diagram in which each block contains a matrix of transfer functions that
relates an input vector to an output vector.
It is often desirable to have a control system decoupled so that certain outputs
can be controlled independently of other outputs. A systematic procedure was
described for decoupling a control system by including cross-controllers along
with the principal controllers. This approach to decoupling requires an accurate
model of the system; the number of controllers (principal controllers and
controllers) increases rapidly with the number of inputs and outputs. A system
represented by two inputs and two outputs requires as many as four controllers; a
system of three inputs and three outputs requires as many as nine controllers, and
so on.
The characteristic equation for a multivariable control system, from which
one can determine stability by examining its roots, can be of high order for
a relatively simple system. Expressing stability boundaries in terms of controller
parameters becomes complex because of the large number of controller parameters
that can be adjusted.

PROBLEMS
30.1. For the liquid-level system shown in Fig. determine the cross-controller trans-
fer functions that will decouple the system. Fill in each block of the diagram shown
in Fig. 30.5 with a tmnsfer function obtained from an analysis of the control system.
The transfer function for each feedback measuring element is unity. The following
data apply:
= 1, = 0.5, = 0.5, = = =

The resistance on the outlet of a tank has been denoted by Res to avoid confusion
with the symbol for set point (R).
 R es FIGURE

30.2. (a) For the interacting liquid-level system shown in Fig. draw very neatly
a block diagram that corresponds to Fig. 30.4. Each block should contain a
transfer function obtained from an analysis of the liquid-level system. There
are no cross-controllers in this system. The transfer function for each feedback
element is unity. The following data apply:
= = Rest = = 2, = 1
(b) Obtain the characteristic equation of this system in the form
+ + +...

Obtain expressions for CY, etc. in terms of = 1)

(c) How would you determine stability limits for this interacting control system?

Pr op ortio nal con tro lle r Pr op ortio nal con tro lle r

FIGURE