Control of Multiple-Input, Multiple-

Output (MIMO) Processes

Chapter 18

18.1 Process Interactions and Control Loop Interactions

18.2 Pairing of Controlled and Manipulated Variables
18.3 Singular Value Analysis
18.4 Tuning of Multiloop PID Control Systems
18.5 Decoupling and Multivariable Control Strategies
18.6 Strategies for Reducing Control Loop Interactions

1
 Control of Multivariable Processes

• In practical control problems there typically are a

number of process variables which must be controlled
Chapter 18

and a number which can be manipulated.

Example: product quality and throughput

must usually be controlled.

• Several simple physical examples are shown in Fig.

18.1.

Note the "process interactions" between controlled and

manipulated variables.

2
Chapter 18

SEE FIGURE 18.1

in text.

3
 Chapter 18

4
 • Controlled Variables: xD , xB , P, hD , and hB
• Manipulated Variables: D, B, R, QD , and QB
Note: Possible multiloop control strategies = 5! = 120
Chapter 18

5
 • In this chapter we will be concerned with characterizing process
interactions and selecting an appropriate multiloop control
configuration.

• If process interactions are significant, even the best multiloop

control system may not provide satisfactory control.
Chapter 18

• In these situations there are incentives for considering

multivariable control strategies.

Definitions:

• Multiloop control: Each manipulated variable depends on

only a single controlled variable, i.e., a set of conventional
feedback controllers.

• Multivariable Control: Each manipulated variable can depend

on two or more of the controlled variables.

Examples: decoupling control, model predictive control 6

 Multiloop Control Strategy
• Typical industrial approach
• Consists of using n standard FB controllers (e.g., PID), one for
each controlled variable.

• Control system design

Chapter 18

1. Select controlled and manipulated variables.

2. Select pairing of controlled and manipulated variables.
3. Specify types of FB controllers.
Example: 2 x 2 system

Two possible controller pairings:

U1 with Y1, U2 with Y2 (1-1/2-2 pairing)
or
U1 with Y2, U2 with Y1 (1-2/2-1 pairing)
Note: For n x n system, n! possible pairing configurations. 7
 Transfer Function Model (2 x 2 system)

Two controlled variables and two manipulated variables

(4 transfer functions required)
Chapter 18

Y1( s ) Y1( s )
= GP11( s ), = GP12 ( s )
U1 ( s ) U2( s )
(18 − 1)
Y2 ( s ) Y2 ( s )
= GP 21( s ), = GP 22 ( s )
U1 ( s ) U2( s )
Thus, the input-output relations for the process can be
written as:
Y1( s ) = GP11( s )U1( s ) + GP12 ( s )U 2 ( s ) (18 − 2 )
Y2 ( s ) = GP 21( s )U1( s ) + GP 22 ( s )U 2 ( s ) (18 − 3)
8
 Or in vector-matrix notation as,
Y ( s ) = G p ( s )U ( s ) (18 − 4)
Chapter 18

where Y(s) and U(s) are vectors,

⎡ Y1 ( s ) ⎤ ⎡U1 ( s ) ⎤
Y(s )= ⎢ ⎥ U( s ) = ⎢ ⎥ (18 − 5)
Y
⎣ 2 ( s ) ⎦ U
⎣ 2 ( s ) ⎦
And Gp(s) is the transfer function matrix for the process

⎡ GP11( s ) GP12 ( s ) ⎤
G p( s ) = ⎢
G ( s ) G ( s )
⎥ (18 − 6)
⎢⎣ P 21 P 22 ⎥⎦

9
 Chapter 18

10
 Control-loop Interactions

• Process interactions may induce undesirable

Chapter 18

interactions between two or more control loops.

Example: 2 x 2 system
Control loop interactions are due to the presence
of a third feedback loop.
• Problems arising from control loop interactions
i. Closed-loop system may become destabilized.
ii. Controller tuning becomes more difficult.

11
 Chapter 18

12
 Chapter 18

13
 Block Diagram Analysis

For the multiloop control configuration, the transfer

function between a controlled and a manipulated variable
Chapter 18

depends on whether the other feedback control loops are

open or closed.
Example: 2 x 2 system, 1-1/2 -2 pairing
From block diagram algebra we can show
Y1( s )
= GP11( s ), (second loop open) (18-7)
U1( s )
Y1( s ) GP12GP 21GC 2
= GP11 − (second loop closed) (18-11)
U1( s ) 1 + GC 2GP 22
Note that the last expression contains GC2. 14
 Chapter 18

15
 Chapter 18

16
 Chapter 18

17
 Chapter 18

18
 Relative Gain Array

• Provides two types of useful information:

Chapter 18

1. Measure of process interactions

2. Recommendation about best pairing of
controlled and manipulated variables.

• Requires knowledge of steady-state gains

but not process dynamics.

19
 Example of RGA Analysis: 2 x 2 system

• Steady-state process model,

y1 = K11u1 + K12u2
Chapter 18

y2 = K 21u1 + K 22u2
The RGA, Λ, is defined as:
⎡ λ11 λ12 ⎤
Λ = ⎢ ⎥
λ
⎣ 21 λ22 ⎦

where the relative gain, λij, relates the ith controlled

variable and the jth manipulated variable
( ∂yi / ∂u j )u open-loop gain
λij  = (18 − 24 )
( ∂yi / ∂u j ) y closed-loop gain
20
 Scaling Properties:
i. λij is dimensionless
ii. ∑λ = ∑λ
ij ij = 1.0
Chapter 18

i j

For a 2 x 2 system,
1
λ11 = , λ12 = 1 − λ11 = λ21 (18-34)
K12 K 21
1 −
K11K 22

Recommended Controller Pairing

It corresponds to the λij which have the largest

positive values that are closest to one.
21
 In general:
1. Pairings which correspond to negative pairings should
not be selected.
2. Otherwise, choose the pairing which has λij closest
to one.
Examples:
Chapter 18

Process Gain Relative Gain

Matrix, K : Array, Λ :
⎡ K11 0 ⎤ ⎡1 0 ⎤
⎢ 0
⎣ K 22 ⎥⎦
⇒ ⎢0 1 ⎥
⎣ ⎦
⎡ 0 K12 ⎤
⎡0 1 ⎤
⎢K
⎣ 21 0 ⎥⎦ ⇒ ⎢1 0 ⎥
⎣ ⎦
⎡ K11
⎢ 0
K12 ⎤
K 22 ⎥⎦
⇒ ⎡1 0⎤
⎢0 1 ⎥
⎣
⎣ ⎦
⎡ K11
⇒
0 ⎤
⎢K ⎡1 0⎤
⎣ 21 K 22 ⎥⎦ ⎢0 1 ⎥
⎣ ⎦ 22
 For 2 x 2 systems:
1
y1 = K11u1 + K12u2 λ11 =
K12 K 21
, λ12 = 1 − λ11 = λ21
1−
y2 = K 21u1 + K 22u2 K11K 22

Example 1:
Chapter 18

⎡ K11 K12 ⎤ ⎡ 2 1.5⎤

K =⎢ =
⎣ K 21 K 22 ⎥⎦ ⎢⎣1.5 2 ⎥⎦
∴
⎡ 2.29 −1.29 ⎤ Recommended pairing is Y1
∴ Λ= ⎢ ⎥ and U1, Y2 and U2.
⎣ −1. 29 2 .29 ⎦

Example 2:
⎡ −2 1.5⎤ ⎡ 0.64 0.36 ⎤
K =⎢ ⎥ ⇒ Λ= ⎢ ⎥
⎣1.5 2 ⎦ ⎣ 0 .36 0. 64 ⎦
∴
Recommended pairing is Y1 with U1 and Y2 with U2.
23
Chapter 18 EXAMPLE: Thermal Mixing System

The RGA can be expressed in two equivalent forms:

Wh Wc Wh Wc
⎡ T − Tc Th − T ⎤ ⎡ Wc Wh ⎤
W⎢ W⎢
Th − Tc Th − Tc ⎥ Wc + Wh Wc + Wh ⎥
K= ⎢ ⎥ and Λ= ⎢ ⎥
⎢ Th − T T − Tc ⎥ ⎢ Wh Wc ⎥
T ⎢ T ⎢
⎣ Th − Tc Th − Tc ⎥⎦ ⎣Wc + Wh Wc + Wh ⎥⎦

Note that each relative gain is between 0 and 1. The recommended

controller pairing depends on nominal values of T, Th, and Tc.
24
 RGA for Higher-Order Systems
For and n x n system,
u1 u2 " un
y1 ⎡ λ11 λ12 " λ1n ⎤
y2 ⎢⎢λ21 λ22 " λ2 n ⎥⎥
Λ = (18 − 25)
Chapter 18

# ⎢ # # % # ⎥
⎢ ⎥
yn ⎢⎣ λn1 λn1 " λnn ⎥⎦
Each λij can be calculated from the relation,
λij = Kij H ij (18 − 37 )
where Kij is the (i,j) -element of the steady-state gain K matrix,

y = Ku
Hij is the (i,j) -element of the H = K ( ).
-1 T

Note :
Λ ≠ KH
 Example: Hydrocracker

The RGA for a hydrocracker has been reported as,

Chapter 18

u1 u2 u3 u4
y1 ⎡ 0.931 0.150 0.080 −0.164 ⎤
y2 ⎢ −0.011 −0.429 0.286 1.154 ⎥
Λ = ⎢ ⎥
y3 ⎢ −0.135 3.314 −0.270 −1.910 ⎥
⎢ ⎥
y4 ⎣ 0.215 −2.030 0.900 1.919 ⎦

Recommended controller pairing?

26
 Singular Value Analysis
• Any real m x n matrix can be factored as,
T
K=WΣV
• Matrix Σ is a diagonal matrix of singular values:
Σ = diag (σ1, σ2, …, σr)
Chapter 18

• The singular values are the positive square roots of the

T T
eigenvalues of K K ( r = the rank of K K).
• The columns of matrices W and V are orthonormal. Thus,
T T
WW = I and VV = I
• Can calculate Σ, W, and V using MATLAB command, svd.
• Condition number (CN) is defined to be the ratio of the largest
to the smallest singular value,
σ1
CN 
σr
• A large value of CN indicates that K is ill-conditioned.
27
 Condition Number
• CN is a measure of sensitivity of the matrix properties to
changes in individual elements.
• Consider the RGA for a 2x2 process,
Chapter 18

⎡ 1 0⎤
K =⎢ ⎥ ⇒ Λ= I
⎣10 1 ⎦
• If K12 changes from 0 to 0.1, then K becomes a singular
matrix, which corresponds to a process that is difficult to
control.
• RGA and SVA used together can indicate whether a process
is easy (or difficult) to control.
⎡10.1 0 ⎤
∑ (K ) = ⎢ ⎥ CN = 101
⎣ 0 0.1⎦
• K is poorly conditioned when CN is a large number
(e.g., > 10). Thus small changes in the model for this
28
process can make it very difficult to control.
 Selection of Inputs and Outputs
• Arrange the singular values in order of largest to
smallest and look for any σi/σi-1 > 10; then one or
more inputs (or outputs) can be deleted.
Chapter 18

• Delete one row and one column of K at a time and

evaluate the properties of the reduced gain matrix.
• Example:
⎡ 0 .4 8 0 .9 0 − 0 .0 0 6 ⎤
K = ⎢⎢ 0 .5 2 0 .9 5 0 .0 0 8 ⎥⎥
⎢⎣ 0 .9 0 − 0 .9 5 0 .0 2 0 ⎥⎦

29
 ⎡ 0.5714 0.3766 0.7292 ⎤
W = ⎢⎢ 0.6035 0.4093 −0.6843⎥⎥
⎢⎣ −0.5561 0.8311 0.0066 ⎥⎦

⎡1.618 0 0 ⎤ ⎡ 0.0541 0.9984 0.0151⎤

∑ = ⎢⎢ 0 1.143 0 ⎥⎥ V = ⎢⎢ 0.9985 −0.0540 −0.0068⎥⎥
⎢⎣ 0 0 0.0097 ⎥⎦ ⎢⎣ −0.0060 0.0154 −0.9999 ⎥⎦
18
Chapter18

•
CN = 166.5 (σ1/σ3)
Chapter

The RGA is:

⎡ −2.4376 3.0241 0.4135 ⎤
Λ = ⎢⎢ 1.2211 −0.7617 0.5407 ⎥⎥
⎢⎣ 2.2165 −1.2623 0.0458 ⎥⎦

Preliminary pairing: y1-u2, y2-u3, y3-u1.

CN suggests only two output variables can be controlled. Eliminate one

input and one output (3x3→2x2).
30
Chapter 18

Question:
How sensitive are these results to the scaling of inputs and
outputs?

31
 Alternative Strategies for Dealing with Undesirable
Control Loop Interactions

1. "Detune" one or more FB controllers.

2. Select different manipulated or controlled variables.
e.g., nonlinear functions of original variables
Chapter 18

3. Use a decoupling control scheme.

4. Use some other type of multivariable control scheme.

Decoupling Control Systems

• Basic Idea: Use additional controllers to compensate for
process interactions and thus reduce control loop interactions

• Ideally, decoupling control allows setpoint changes to affect

only the desired controlled variables.

• Typically, decoupling controllers are designed using a simple

process model (e.g., a steady-state model or transfer function
model)
32
 Chapter 18

33
 Decoupler Design Equations

We want cross-controller, T12, to cancel the effect of U2 on Y1.

Thus, we would like,

T12 GP11U 22 + GP12U 22 = 0 (18 − 79 )

Chapter 18

Because U22 ≠ 0 in general, then

GP12
T12 = − (18 − 80 )
GP11
Similarly, we want T12 to cancel the effect of U1 on Y2. Thus, we
require that,
T21GP 22U11 + GP 21U11 = 0 (18 − 76 )
GP 21
∴ T21 = − (18 − 78)
GP 22
Compare with the design equations for feedforward control based on
block diagram analysis 34
 Variations on a Theme

1. Partial Decoupling:
Use only one “cross-controller.”

2. Static Decoupling:
Chapter 18

Design to eliminate SS interactions

Ideal decouplers are merely gains:
K P12
T12 = − (18 − 85)
K P11
K P 21
T21 = − (18 − 86 )
K P 22
3. Nonlinear Decoupling
Appropriate for nonlinear processes.
35
 Wood-Berry Distillation Column Model
(methanol-water separation)
Chapter 18

CT

Feed F
Reflux R Distillate D,
composition (wt. %) XD

Steam S

CT
Bottoms B,
composition (wt. %) XB

36
36
 Wood-Berry Distillation Column Model
⎡ 12.8e− s −18.9e −3s ⎤
⎢ ⎥
16.7 s + 1 21s + 1 ⎥
⎡ y1 ( s ) ⎤ ⎢ ⎡ u1 ( s ) ⎤
⎢ y ( s) ⎥ = ⎢ ⎥ ⎢u ( s ) ⎥ (18 − 12)
Chapter 18

⎣ 2 ⎦ ⎢ −7 s ⎥ ⎣ 2 ⎦
⎢ 6.6e −19.4e −3s ⎥
⎢⎣10.9 s + 1 14.4 s + 1 ⎥⎦

where:
y1 = xD = distillate composition, %MeOH
y2 = xB = bottoms composition, %MeOH
u1 = R = reflux flow rate, lb/min
u1 = S = reflux flow rate, lb/min

3737
Chapter 18