CHAPTER TWO

DESIGN OF REJECTION TO DISTURBANCES FOR SINGLE VARIABLE

CONTROL SYSTEMS

§ 2-1 Introduction

As well-known, one of the key problems in the design of process

control systems is their ability to resist disturbances. Any process

control system should have good ability to resist disturbances in order

to keep their controlled parameters being held at desired values with

some accuracy after certain periods.

If only the resistance to disturbances at steady state is required,

then integral control is enough to do so,

But, if the resistance to disturbances is required not only at

steady state, but also at transient process, then a system with full abi-

lity to resist disturbances is needed. Obviously, this is a kind of

process control systems with very high accuracy and to realize such a

system is important in both theory and practice.

Recently, a general terminology " Rejection to Disturbances" is

widely used for this design and some older terminologies, such as " In-

variance Principles " etc, have been ruled out graduately.

We should discuss:

(I) For single variable process control systems, what meathods

can be used to realize the design of rejection to disturbances?

(2) For multivariable process control systems, if only the c~assical

control theory is applied, can we realize a system with the ability not

only to decouple , but also to reject disturbances ?

These two problems are met with now very frequently in the develop-

ments of modern process control engineering theory and practice and we

will discuss them in detail in this book.

By the opinion of Cecil Smith, there are five kinds of process con-

trol systems practially being considered as " advanced process control

 ?0

s y s t e m s " . They a r e : (25)

(1) Cascade c o n t r o l ,
(2) The c o m b i n a t i o n of f e e d f o r w a r d and f e e d b a c k c o n t r o i ,
(3) Smith p r e d i c t o r ,
(4) Adapted c o n t r o l ,
(5) Multivariable decoupling control.

Certainly, the combination of the above five kinds of systems will

give more advanced control systems.

What we will study in this book is just the combination of these

systems except the fourth. Thus, from the view-point of process control

theory, the systems researched here belong to advanced ones and they

are the emphatically researched objects in process control engineering

now indeed.

In our study, two assumptions are taken:

(i) The systems are with linear constant coefficients,

(2) All disturbances are non-stochastic and measurable.

In fact, the following analysis will show that the first assumption

may be unnecessary in some cases.

In addition, we assume that in the system block diagrams, the dis-

turbances have definite input points. Thus, we can make the assort-

ment of the disturbances according to their locations in the block

diagrams. And we always start our study from the most basic control

systems the feedback control systems.

We discuss the single variable control systems at first.

Our book is to discuss the decoupling design of MPCS and why should

we discuss single variable control systems here?

This is because that the design of rejection to disturbances of

single variable control systems is essentially the primary decoupling

idea. For any single variable control system, certainly there is a

reference and all other inputs are disturbances. Because for a single
 71

variable control system, only one reference input and one output are

considered, so the decoupling idea has no meaning here. If we want to

use this idea here , then it means that the output is only influenced

by the reference input~ in other words, the system is with the ability

to resist disturbances fully,

Therefore, the design of rejection to disturbances for single vari-

able system is the primary idea of decoupling design.

For a single variable system, three kinds of disturbances, i.e. the

reference disturbance, the supply disturbance and the load disturbance,

should be considered and we will discuss how to realize the design of

rejection to each kind of these three disturbances.

The basic ideas of this chapter are not new indeed and most readers

are familiar with them. But, however, we will give a systematic discus-

sion on this problem and many important conclusions will be derived

At the meantime , we should point out here that the general prin-

ciples discussed here are not difficult in theory , so the problem to

realize a system with the ability to r~ject the disturbances fully is

not in theory ,but in practice, for example the measurement of the dis-

turbances , the rcalization of the compensation elements, etc. And some

practical problems are not so easily to be settled , thus sometimes

although the system with the ability to reject the disturbances can be

design theoretically, yet it can only be realized approximately.

This problem is not only with the single variable systems, but al-

so with the multivariable control systems. We will see in the coming

chapters that sometimes we can design a decoupling control ~or a multi-

variable system, yet we can only realize it approximately.

 72

S 2-2 Design of Rejection to Reference Disturbances

We discuss the reference disturbances at first.

This disturbance co-exists with the reference and is imported

at the same point. Obviously, it will give the influence on the output

in the same manner as the reference does.

We do not consider the high frequency disturbances. If the dis-

turbance is with high frequency, then because of the filter character

of the system it will not give remarkable influence on the output. So

the disturbances considered here all are in the band width of the

system and they can give explicit influence on the system.

Obviously, in order to eliminate the influence caused by the

disturbance on the output, it is necessary to take some special measure.

A system with a special function is shown in Fig 2-2-1.

For this system, we get:
U1
c wcwp+ wpwf
- (2-2-l)
U1 1 + WcWpW m
X+
Therefore, the condition -----

for realizing rejection to

reference disturbances is:

Fig 2-2-1
wf = -We (2 - 2 - 2 )

and U 1 will give no influence on the output at any time.

Because W c is a physically realizable element, so Wf is easily

to be realized.

In practical design, Wf does not include the control valve cha-

racter , i.e. only the character of the regulator is included, and

the output of Wf is added on the output of the regulator.

We should point out that this system is different from the well-

known Moore system. The Moore system is shown in Fig 2-2-2.

The difference is: The feedforward channel of the Moore system

 73

is derived from the reference ~ Wcl ' .~

but the £eedforward channel of

Fig 2-2-1 is derived from

the disturbance.

The Moore system is a very useful system

system and it is a combination of Fig 2-2-2

a feedforward system with a feedback system. When Wcl is of different

characters, this system may have various outputs.

From Fig 2-2-2, we get:

E - X - CW m (2-2 -3)
C = EWcW P + XWciWp S
namely:
E = X - ( EWcW P + XWclW p )W m (2-2-4)

E ( 1 + WcWpW m ) -- ( 1 - WcWpWm)X (2-2-5)

( 1 - WclWpWm )x
E = (2-z-6)
1 + WWW
c pm
We can realize a system without any error, if we let:

Wc I = 1 CZ-2-7)
WpW
When W --i,
m 1
Wcl = (2-2-8}
W
P
Therefore , the transfer function of its feedforward channel is

WcIWp=I and we can say that this system utilyzes the feedforward channel

to realize the ideal control model and utilizes the feedback control

to eliminate the influence of different disturbances.

But the system shown in Fig 2-2-1 utilizes the feedback control to

realize its control model and utilizes the feedforward channel to elimi-

nate the influence of the reference disturbance.

For process control systems, because the reference disturbance is not

serious in general, so the design of re~ection to reference disturbances

is not very important either.

 74

§ 2-3 Design of Rejection to Supply Disturbances

In process control systems, the most important disturbances are

the supply disturbances and the load disturbances.

The so-called supply disturbance occurs with the medium entering

the plant. It is also the disturbance of the manipulated variable. So

in the block diagram, it enters into the system before the plant; as

for the load disturbance, it occurs with the controlled parameters, so

it is always added on the output.

Because the supply disturbance is a very popular disturbance, so

we will discuss how to design a system with the ability to reject such

a disturbance in detail.

The most common method is to use the combination control system

to realize such demands. The combination means to combine feedfor-

ward control and feedback control. U

Fig 2-3-1 shows shch a system. F - - - - - - - ~ ~ ~
From i t we can get: X Wp ~-~
C _ Wp + WfWRWm I_
U 1 + WRWpW m
(2 -3-1)
~
I
Wm
U
Ix
l
For r e a l i z i n g the r e j e c t i o n Fig 2-3-1
to U, it is necessary:

Wp + WfWRW m = 0 (2-3-29
name ly :
1
Wf -- WR (2-3-39

It seems that the condition is very simple and the structure of

the system is not complicated either,then is there anything worth dis-

cussing? In fact, some useful conclusions can be obtained here.

(I) As well-known, the basic idea of closed-loop control is to

measure the response values of the output and to compare the results

measured with the reference in order to get the deviation. This devia-
 75

tion may be caused by different disturbances and the system just ad-

justs its manipulated variables according to the deviation to perform

the control function.

When the feedforward channel is introduced, this channel measures

the disturbance itself and gives an inverse influence on the system,

thus, the influence of the disturbance can be eliminated without mea-

suring the deviation. Therefore, a feedback control system combined

with a feedforward channel can eliminate the influence of the supply

disturbance more quickly and better than that without feedforward channel.

(2) From (2-3-1) we know that after introducing the feedforward chan-

nel,the characteristic equation of the system remains unchanged; it

means that the introducing the feedforward channel does not influence

the stability problem of the system. In other words, no matter what ele-

ment the feedforward element is, it gives no influence on the system

stability.

(3) In practical process control systems, the energy level of the

compensation signals should be considered. Certainly, it would be better

that the energy required for the compensation signals were small. We

know that in general the degree of the signal energy levels increases

along the direction of the element connection for the main channel.

Therefore, the required energy will be the least when the feedforward

channel is introduced into the system at the reference import point.

But, on the other hand, in which case will the effect of the feedfor-

ward channel be the most prominent? In fact, the smaller the lag between
the entrance point of feedforward signal and the entrance of the distur-

bance than the lag of the plant, the better the effect to eliminate

the influence of the disturbance, Thus, from this meaning, the feedfor-

ward channel should be arranged near the entrance point of the distur-
bance as close as possible.

Hence, the feedforward channel is always not connected before the

 76

regulator, but at the exit of the regulator.

Fig 2-3-2 shows such a system.

Fig 2-3-2
From this figure, we get:

C _ Wp + WfWvW p
(2-3-4)
U 1 + WRWvWpW m
Thus, in order to realize r e j e c t i o n to U, it is necessary:

1
Wf=- - - (2-3-5)
WV
Notice thay W V may not be only the character of the control valve.

In most cases, the Wf obtained from (2-3-3) and (2-3-5) is phy-

sically irrealizable , n a m e l y the order of the n u m e r a t o r of its trans-

fer function is higher than that of its denominator.

This p r o b l e m will not be very serious in process control enginee-

ring. This is because in process control systems, the disturbances are

always with low frequency, so we can introduce additional poles in the

transfer functions to realize Wf within suitable frequency band.

Now, we discuss a practical example which is a water temperature

control system. As we know, in the chemical process, if we want to in-

crease the temperature of a stream of cold water, a direct m e t h o d is

to mix it with hot water.

The hot water is heated by a gas boiler and its control character

is a first lag. The heated water enters into the mixer to be mixed with

the cold water and the character of the mixer is also a first lag. The

measurement value of the temperature of the mixed water is compared

with the reference and a control system is farmed in this manner.

The block diagram is shown in Fig 2-3-3.

 77

I°

i i:10 1 Issi
Fig 2-3-3

Suppose the regulator being a propotional regulator with the gain

equal to 5 and the gain of the valve ( including the second regulator)

being 2 . The process is performed by adjusting the stroke of the con-

trol valve to change the gas flow entering the burning chamber of the

boiler to control the temperature of the hot water. The variation of

the boiler character is always very slow and suppose it with the constant

of 60 seconds. The constant of the mixer is assumed to be 15 seconds

and 5 seconds for the m e a s u r e m e n t device.

In the figure, C is the final temperature of the mixed water, X is

the desired temperature and U is the disturbance caused by the variation

of the temperature of the cold water.

In order to eliminate the influence of U on the controlled parameter

C, it is necessary to design a system with the ability to reject U. In

fact, it means to determine Wf.

From (2-3-5), we get"

0.5Wf = - 1 _
WV 2x0.5
I+60S
namely:
Wr = 1 + 60S _
I
0.5
Thus, it may be realized by a P+D regulator and its transfer function

is:
wf(s) - l°----L,(1 + STd)
P
Where P is the propotional band , Td i s the differential time, so
 78

it is only necessary to let the propotional band be 50% and T d be 60

seconds.

(4) For a feedback control system without the feedforward channel,

as shown in Fig 2-3-4, its response caused by the disturbance U is:

C Wp = 1 WRWp _ G (2-3-6)
U 1 + WRWpW m WR i+ WRWpW m WR

Where G is the closed-loop transfer function of the system.

U

system, if we want to ~ WR Wp

weaken the influence

of the d i s t u r b a n c e , then Wm I
the gain of the regulator Fig 2-3-4
should be increased and when we want to realize the rejection to the

disturbance, theoretically, the gain of W R should be infinite.

This demand is not only difficult to be realized, but also with the

problem of stability, i.e. it will cause unstability.

We have said before when we use feedforward channel to eliminate

the influence of U, there is no such a problem as s t a b i l i t y . Certainly,

this is ideal case, but, however, even for the restriction of physical

realizable conditions the feedforward channel could not realize perfect

rejection to the disturbance, such a design can weaken the most part of

the influence caused by the disturbance and the remaining part may

be weakened by feedback control; for example , we can introduce inte-

gral control in the system to eliminate the steady influence of the dis-

turbance. Thus, comparing with the pure feedback control system without

feedforward channel, the feedforward-feedback combination control sys-

tem can reduce the gain of the regulator and it will certainly improve

the system stability.

(5) Because the introduction of the feedforward channel eliminates

 79

the most influence of the disturbance, so the frequency band of W R and

Wv can be reduced also and the investment of these elements will also

decrease.

On the other hand, owing to the existence o£ the feedforward chan-

nel the failure of the main regulator will not cause the perfect outage

of the whole system and a feedforward control channel still exists , so

the manual control can be performed.

Therefore, although the design of rejection to supply disturbances

seems very simple, the expounded analysis shows that it contains very

abundant contents.

Theoretically, a system shown in Fig 2-5-5 is also capable to reject

the disturbance U,

From it, we can get: U

C = W£+ Wp Wf

U X
1 + WRWpW
m
(2-3-7)
thus, if we let:

Fig 2-3-5
wf = - C2-3-8)

then the full rejection to the disturbance U is realized.

From the analysis of the block diagram, it seems that this project

is more rational since Wp is certainly a realizable element for the

practical plant, so Wf determined by (2-5-8) is also an easily realizable

element.
Does it really mean that this project is better than that of Fig 2-5-1
or Fig 2-5-2 ?

No. It is not true.

The reason we have mentioned before. Because in any practical process

control system, especially in chemical process control systems, the system

output C always is some parameter of a process with large capacity, such

 80

as temperature, pressure, flow, composition etc~ and all these para-

meters in general may be measured by some small measurement device

( in control systems it is the measurement element located at the

feedback channel), but, however, it is impossible to imagine to in-

fluence such parameters by adding some output of small compensation

elements since there is a problem about the quantity degree of energy

and material. For instance, in the above example, how can the output

of the small compensation element Wf influence the temperature of the

mixer ? Obviously, it is impossible. Therefore, we can not expect to

use it to compensate the vaiation of the temperature in the mixer.

Hence, the project of Pig 2-5-5 seems reasonable in block diagram,

but it is unavailable in practice indeed.

The application of the combination of the feedforward control

with the feedback control now is widely seen in industries and a lot

of experience has proved that this is a very effective measure to re-

ject the supply disturbances.

Besides the combination of the feedforward control with the feed-

back control, another effective way to reduce the influence of the

supply disturbances is the cascade control. Although the cascade con-

trol can not eliminate the influences of the supply disturbances en-
tirely, it c a n reduce them remarkably.

So, if there are several supply disturbances , we can not use the

project introduced here to eliminate all of them because this will re-

Sult in a very complicated system, and in such a case to combine the

cascade control with the principles discussed here is certainly a good

idea to reduce the influences of the supply disturbances.

We will discuss this idea in the next paragraph.

 81

§ 2-4 Design of Rejection to Supply Disturbances and the Cascade Control

In last paragraph, we d i s c u s s e d the design of rejection to the sup-

ply disturbances.

But, we should discuss two problems further:

(i) If there is an auxiliary control loop before the supply distur-

bance, how will the case be?

(2) If there are two suppiy disturbances with different characters~

how should we deal with this case?

Now, we discuss the first problem and the system is shown below:

1- I°
X I, Wp2 =

[ - ~ 2 i-~
Fig 2-4-1

Because only the influence of the disturbance is considered, s o we

can let :

x = 0 (z-4-1)

then from Fig 2-4-1, it yields:

M -- wfwswV~plU- WmlWsWv~lM (z-4-2)

It follows :

WfWsWvWpIU
M -- (2-4-3)
1 + WmlWsWvWpl

where, W s is the auxiliary regulator ; Wml is the measurement element of

the auxiliary loop ; Wpl is the plant character of the auxiliary loop.

On the other hand, we have:

Q = H + U (2-4-4)
If we want to realize the full rejection to U, it is necessary:
 82

Q = o C2-4-s)
namoly:
WfWs--WplUWv + U - 0 (2-4-6)
i ÷ WmlWsWvWpl
It follows:

I ÷ WmlWsWvWpI
Wf = - ( 2-4-7 )
WsWvWp 1

We should point out that the system shown in Fig 2-4-1 is practi-

cally meaningful since it is a combination of a cascade control

system with the design of rejection to the supply disturbance. As we

know that the cascade control is very capable to eliminate the inlu-

ence of the disturbances which enter into the auxiliary loop, but for

the disturbances outside the auxiliary loop, its function to eliminate

the influence of disturbances is not striking. Thus, for the distur-

bances outside the auxiliary loop, we should adopt the design of re-

jection to disturbances to eliminate their influence, i.e. the form

of Fig 2-4-1 should be adopted.

That is to say that the principle of realizing rejection to dis-

turbances can be combined with cascade control and it certainly results

in improving the control effect.

Naturally, there is a problem: When there are two different kinds

of disturbances, and the combination principle mentioned above is used,

which one should be weakened by the cascade control?

This problem is difficult to be answered analytically since we

can not find a definite mathematical form to express it.

Now, we are going to explain it by an example of the level con-

trol of a boiler drum and we can learn some useful experience from it.

The plant is a boiler drum and its input is the fed water. The

flow of the fed water is determined not only by the stroke of the re-

gulating valve, but also by the disturbances. Some disturbances show

 83

explicit influence on this process and they may be c a l l e d as s u b s t a n t i a l

disturbances; some do not express E h c m s e l v e s as e x p l i c i E d i s t u r b a n c e s p

but they are i n d e e d and thus they can be c a l l e d as n o n - s u b s t a n t i a l .

Now, let us analyze the disturbances.

As we k n o w that the demands on the steam flow depend on the boiler

load. W h e n the load increases, then the d e m a n d on the steam will enhance

also. But w h a t does it m e a n to increase the steam flow? It means more

water in the drum b e i n g v a p o r i z e d and from the v i e w - p o i n t of water level

change it corresponds to r e d u c i n g the fed water.

Therefore, the v a r i a t i o n of the steam flow may be c o n s i d e r e d as a

negative d i s t u r b a n c e to the fed water , i.e. a n e g a t i v e supply disturbance.

As a supply d i s t u b a n c e of this process, this d i s t u r b a n c e is not ex-

plicit, n a m e l y we can not detect it by the variation of the fed water since

it does not exert p r a c t i c a l influence on the fed water.

On the contrary, the v a r i a t i o n of the p r e s s u r e exerted on the valve

gives an e x p l i c i t influence on the fed w a t e r , so it is a s u b s t a n t i a l sup-

ply d i s t u r b a n c e .

For a s u b s t a n t i a l supply disturbance, because its influence can be

measured by the fed water, so we can use cascade control to e l i m i n a t e its

influence; but for the v a r i a t i o n of the steam flow, it is not an e x p l i c i t

disturbance and its influence can only be e l i m i n a t e d by the design of re-

jection to supply d i s t u r b a n c e s .

Fig 2-4-2 shows such a system, i U1

Fig 2-4-2
 84

Where: C is the controlled level,

W R is the main regulator,

W s is the auxiliary regulator . Suppose it being P+I form.

W V is the control valve,

W is the plant,
P
Wms is the measurement elemen~ of the auxiliary loop,

Wf is the feedforward element under design.

For the auxiliary loop, its closed-loop character is:

WsW V
G = (2-4-8)
1
1 + WsWvWms

Because W s is a P+I regulater, so its gain at low frequency is

very high and we get:

G1 [S) . ~ 1 (2 - 4 - 9 )

at low frequency.

Then, from (2-3-5), it yields:

Wf(S) : 1 (2-4-10)

By using such a feedforward channel , we can eliminate the influ-

ence of steam flow change on the control of the level.

Therefore, the proje¢t of the combination of feedforward control

with the feedback control can be further combined with the cascade con-

trol , but, however, some practical problems should be carefully treated

because there is no general rule for combination.

We will give more discussion on the design of this system later on.

This example and the discussions below are taken from the British

broadcasting course on automatic control.

 85

§ 2-5 How to Utilize Feedback to Simplify the System Design

We have discussed the example of the level control of a boiler.

There are two supply disturbances in it. One is explicit ( the change

of the chamber pressure of the control valve) and the other is implicit

( the influence of the variation of the steam flow). We have pointed

out that we can use the combination of cascade control and the design

of rejection to sypply disturbances to reduce the influence of both

disturbances.

In such a design, two regulators are needed. One is the main re-

gulator of the level control and the other is the auxiliary regulator

in the inner loop.

Now, let us discuss further. In order to reduce investment, can

we use only one regulator to satisfy the control demands?

We return to Fig 2-4-2.

From that figure, we see that the disturbances U1 and U 2 are sent

to the same point on the main channel but with opposite signs. Although

it is impossible to measure U 1 how to be input in the system practically,

its location in the block diagram is definite.

In addition, in that figure Wf = i, therefore, from the analysis of

block diagram, Fig 2-4-2 is entirely corresponding to Fig 2-5-1.

WR Ws

Fig 2-5-1

In this figure, ff is the practical fed water to the boiler ;

fn is the net fed water obtained from the block diagram with the consi-

deration of U I. In fact fn can not be measured but it exists in the block

diagram and it should be fed back in the inner loop.

 86

Thus, all disturbances are included in the inner loop and if P+I

regulator is adopted in it, then there is no steady error for the step

disturbances in the inner loop.

Now, we consider the main loop. The plant is an integral element

since the drum can not hold the balance of the level itself. This in-
tegral element gives the guarantee that the output response to the step

reference is without steady error for the main loop. Therefore, we

can choose a propotional regulator for WR.

Because the main regulator is with propotional character, then the

system design may be simplified further.

The method to simplify the design is to utilize the character o£ the

feedback elements.

For example, we can let:

Wms(S) = ~ (2-5-1)

where, is between 0 a n d 1.

Then , the main regulator may be o m i t t e d and t h e system structure

becomes :
U2 ] U1

Fig 2-5-2

Why can the main regulator be omitted?

This is because decreasing the feedback gain, i.e. ~ I , can

increase the whole loop gain of the inner loop and this is equivalent

to preset a propotional regulator with suitable gain before the inner

loop. Hence, to adjust ahe value of ~ practically may adjust the gain

of the whole open system.

 87

In fact, if G 1 is the closei-loop transfer function of the inner

loop, then:
WsWV 1
G1 = -- (2,-S-2)
I • WsWv 1
wsw v
Because W s is a P+I regulator and is with quite large gain at low
frequency, therefore at steady state,the closed-loop gain of the inner

loop is i / ~ .

Thus, when ~ < l , then 1 >l and this corresponds to the closed-loop

gain of the inner loop equal to one with a preset propotional regulator

( its gain is 1
-~-) before it.

The choice of ~ s h o u l d consider to guarantee the stability of the

inner loop and the main loop. When ~ i n c r e a s e s , then the gain of the open

loop of the inner loop increases and it is easy to cause the unstability

in the inner loop; on the contrary, the decrease of ~ will increase the

open loop gain of the main loop and will be able to cause the unstability

of the main loop, so a balance should be held here on the value of ~.

But, all the above discussions are carried out in the block diagram

and we have pointed out for several times that some conclusions may be

rational in the analysis of the block diagram but are not available in

practice. How about the conclusion now?

In fact, we know that f can not be measured and we can only measure
n

ff, so Fig 2-5-2 is unavailable in practice.

The practically available form is shown in Fig 2-5-3. This is still

a combination of cascade control with the design of rejection to supply

disturbances. Now , we determine Wf.

The transfer funvtion of the inner loop is:
K(I + I/TiS)K V KKv( 1 + TiS)
Gl(s) ~ = =
1 + K(I + I / T i S ) K v ~ K K v ~ ( I + T i S ) + TiS
KKv( i + TiS ) 1 + ST i
(z-s-3)
KKv~( 1 + TiS(1 + 1 / K K v ~ ) ) £( 1 + STiP<)
 88

U1
Fig 2 - 5 - 5

Where:

d = 1 ÷ (2-5 -4)
KKv~
Therefore:
1 + ST.o<
wf(s) - l = ~ ~ (z-s-s)
GI(S) 1 + ST i

In Wf(S), the term

~ is used to eliminate the steady influence
1 + TiS~
of t h e disturbances and t h e term is used to eliminate the
1 + ST i
dynamic influence of the disturbances.

We notice that there is a ~ on the feedback channel and there is

also a [ on the feedforward channel of U 1 and both Y are input to the

same point of the systemp so we can combine them and one amplifier can

be s a v e d .

The final system is shown in Fig ?-5-4.

'v21 I

I l+STi I
U
Fig 2-5-4 1
 89

The practical structure of this system is shown

Steam Flow
in Fig 2-5-5.

i + ST i

Fig 2-5-5

Up t o now, we h a v e e x p o u n d e d t h e principles of the rejection de-

sign to the supply disturbances. We m u s t say that these principles are

available in practice but the project of the feedforward control with

the feedback control is not the unique project to realize the rejection

design.

Another possible project for realizing rejection design will be

given in § Z-8.
 90

§ 2-6 Design of Rejection to Load Disturbances

In process control systems, another important disturbance is the

load disturbance occuring on the outputs. U

Fig 2-6-1 shows such a system.

From this figure , we have:

_ wu.2 wfwp x
U 1 4 WRW p (2-6-1)

Therefore, if we want to rea-

Fig 2-6-1
lize full rejection to disturbance U,

then it is necessary:

WU
Wf = - (2-6-2)
Wp

And we know:

(i) Wf is related to the character of the plant,

(2) When W U approximates Wp, then Wf will be simple. In fact,

when W U = Wp , then Wf = i.

In addition, when both W U and Wp are with the same lags, then

Wf can be realized by a propotional element.

But, for Wf determined by (2-6-2), there is not only a problem

about physical realization mentioned before, i.e. the problem of the

numerator being with higher degree than the denominator , but also ano-
ther physical realization problem.

This is because in chemical process control systems, the plants

always have dead time delays, i.e. there is a term e -IS in Wp. In such

cases, if we realize Wf according to (2-6-2), then certainly Wf will

contain the factor e ~S , which is impossible to be realized by such

transfer functions.

In such cases, we can expand ~TS into fraction form o£ polyno-

mials by using Pad6 approximation.

 81

For example, the common Pad6 a p p r o x i m a t i o n terms are shown below:

-x
e

1 - x 1 - x ++x 2
1 1

1 - -~x 1 - --F-x + x2
1 + x
l + + x l + + x

1 - --~-x
1 1 - --f-x -
1 2

1 2 2
1 + x +-~--x I + ---~X + T1 2 x i + +x + ~2--x 2
-fS
Expanding e by Pad6 approximation, we can get Wf(S) by using

(2-6-2), b u t now Wf(S) may be a a l l - p a s s network.

(2-6-2) shows the c o n d i t i o n of full rejection to the disturbance U

and we see because the practical character of the plant, sometimes it is

difficult to realize Wf, thus, the full r e j e c t i o n to U is not easy to

be realized. That means that in general we can only realize partial re-

jection ~o the disturbance.

Then, a problem arises: Is there any other m e t h o d being able to do

so? Certainly, there is. For instance, the cascade control is also a good

way to reduce the influence of the load d i s t u r b a n c e s .

Flg 2-6-2 shows a system using cascade control to reduce the influ-

ence of the l o a d d i s t u b a n c e . U ~1 WU ] - - - - - - ]

=l lw l
wH ]

Fig 2 - 6 - 2
 92

When t h e control system is not designed, the relation between the

output and the disturbance is:

C
•, WU (2-6-3)
U

When the above system is adopted but the main channel is broken

off, i,e. W R is broken off,then:

c Wu
(2-6-4)
U i + Am
Where:
Am = WpW H (2-6-5)

When the main channel is closed, we have:

C WU
= (2-6-5)
U ( i + Am) ( I + AM()
Where:
A
AM c : mu (2-6-6)
I+A m

Amu = WRW p (2-6-7)

therefore:

C WU
- A (2-6-8)
U (i
mu ) +A )( 1 +
m 1 +Am
In general,the disturbance is always with low frequency and at low

frequency since W R is a P+I regulator, so Ainu is with quite large value

and:
(2 - 6 - 9 )

then:
A A
I +
mu ~. mu (2-6-10)
i + Am i + Am
It yields :

C _ WU WU
(z-6-11)
U Amu WRW p
Therefore, when the cascade control is adopted, the influence of

the low frequency disturbance is reduced t o the degree of i/Amu of the

 95

original influence.Notice that this is about the reduction of the dynamic

influence, as for the steady influence, since U is approximate to a step

disturbance, so its steady influence will be eliminated entirely.

When W R is a propotional regulator with small gain, i.e. WR is

small, then it may be:

(2-6-13)
But, if :

I'm1>>' (2-6-14)
then : A
mu
i + ""l,-.. (2-6-15)
1 +A m
So :
c wu
- (2-6-14)
A
m

Thus, if the open loop gain of the inner loop is large enough, then

the influence of the disturbance may also be reduced. In general, the

propotional gain of the auxiliary regulator of the inner loop is quite

large, so the c o n d i t i o n I%l >l i s always h e l d .

For Wf determined by (2-6-2), as we said before, when both Wp and

W U are with same lags and the valve character is omitted, then Wf is a

pure propotional element. When the valve is with first lag, the Wf will

be a P+D element. All these can be realized by using conventional regula-

tors. But, when Wp and W U are with different lags, especially when Wp is

with dead time delay, then it is difficult to realize Wf.

Recently, some authors suggest to utlize the on-line control to treat

such control problems. By experience, in most process control systems, no

matter how complicated the characters of the plant and the disturbance
channel will be, using:

K( 1 + UIS )
wf = - (z-6-17)
1 + ~2 S
and setting the parameters on-line, i.e. to determine the parameters of
 94

Wf according to the practical measurement of the influence of U on C,

we can always get satisfactory effect to reject the disturbance. For

example, Shinskey, Nisenfield and others did some research work in this
field.(7)(50)

Furthermore, some authors, Wood and Pucey, pointed out that for

most process control systems, we can let:

1
Wf = -K (2-6-18)
1 +E S
and set the parameters on-line, then practically we can get good re-

sult to reject the disturbance.

All these denote although by (2-6-2) we can get ideal full rejec-

tion to the disturbance, it is difficult to be realized practically.

So , we have to use some simplified methods to realize approximate

rejection to disturbances and among these methods, the on-line control

is a good one.

Up to now, we have introduced the most popular or conventional

principles for realizing the rejection design to both supply distur-

bances and the load disturbances. We see that these principles are not

difficult in theory but sometimes are not convenient to be applied in

practice. The basic reasons are:

(I) The compensation elements may be difficult to be realized,

(2) There may be several disturbances,

(3) The disturbances may be difficult to be measured,

C4) The disturbances may vary.

Because of these reasons, so in practice only approximate rejec-

tion ~esign can be realized.

 95

§ 2-7 Disturbance Analysis and Rejection Design

In the previous several paragraphs, we discussed how to realize the

design of rejection to the reference disturbances, supply disturbances

and load disturbances. We have learnt that it is possible to take some

available measures to eliminate the influence of each disturbance.

Further research will propose such a problem: When several kinds

of disturbances co-exist in a process control system, how should we do

the rejection design?

In § 2-4, we have said something about it. There, an explicit supply

disturbance was reduced by using the cascade control , while an implicit

disturbance was t r e a t e d by the d e s i g n of rejection. But, however, it was

a special case indeed. If two d i s t u r b a n c e s both are explicit, how s h o u l d

we treat them?

In fact, two p r o b l e m s s h o u l d be d i s c u s s e d here:

(1) In such a case, whether it is absolutely necessary to adopt dif-

ferent ways introduced before for design? If it is, certainly the final

system is very complicated.

(2) When different disturbances co-exist, is it necessary to do full

rejection design to each one?

In fact, the essentiality of these two problems is to discuss the

characteristics of disturbances and the functions of rejection design.

Now~ we discuss the first problem. The key of this problem is:

Can some rejection design have several rejection abilities at a time?

It is not impossible.

Let us consider a resolution control system. Such a system is shown

in Fig 2-7-1.

Where, f is the resolved medium and U is the resolvent.We hope that

in the mixer, the solution can have invariant density. In order to reach

this demand~ the density control system must be with the ability to

reject disturbances.
 96
f

The f l o w of t h e r e s o l v e d m e d i ~ can I

be e a s i l y held invariable,
b u t the f l o w o f t h e r e s o l v e n t U

is remarkably influenced
by the variation of pressure

and will change frequently.

So we should design a density

control system with the abi-

lity to reject the influence

of the resolvent flow variation.

We suppose that the control valves are with high speed response,

i.e. their characters are not considered in the design, then a change of

the flow of ~he resolved m e d i ~ will require a corresponding ch~ge of

resolvent with the same sign in order to keep the density in the mixer

being invariant.

So, the flow of the resolved medium is the manipulated variable

and R 1 and R 2 are two regulators.

There are two plants for this system. They are:

C(S) (2-7-1)
WPl= fCS)

c(s) (z-7-z)
WPz~ u(s)
The block diagram is shown in Fig 2-7-2.

~ [ WP2

Fig 2-7-2
 97

Obviously, the disturbance enters into the system at two points. One

is at the reference setting point and the other, is at the outlet of the

system. By our above analysis~ it means that a reference disturbance and

a load disturbance co-exist. In such a case, is it necessary to set two

devices to eliminate the influence of U ?

It is not necessary to do so indeed. Obviously, Fig 2-7-2 can be trans-

ferred into the form of Fig 2-7-5.

"1 Wp 2

Fig 2-7-5

From Fig 2-7-5 we know that from m to C is a normal single variable

control system and we can design it by conventional method, i.e. because

Wpl is known, so it is easy to design R 2.

In addition, from Fig 2-7-5, we get:

C = R2Wpl RIR2WpI ÷ Wp2 = . RIR2Wpl + Wp2 (2-7-2)

U
i + RZWpl R2WpI i + R2Wpl

Obviously, if we let:

R1 = - - Wp2 (2-7-5)
R2Wpl
then C is not influenced by U. That means that only one element com-

pletes the rejection to the disturbances at two points.

Now, we discuss the second problem. It is: When there are several

disturbances in a system, is it necessary to treat each of them with full

rejection design ?

It is not necessary either. In general, if all the disturbances are

 98

explicit, then we should design a full rejection to the most serious

disturbance and for the others which have smaller influence we can re-

duce them by other methods.

Consider the example of § 2-5 again. The water temperature con-

trol system is shown in Fig 2-7-4.

U1 U2

X O.S 0.2 C
I+6OS I "-" [ I÷15S

1 I_
I+5S I--

Fig 2-7-4

There are two disturbances in this system. One is the variation

of the pressure of the boiler burning gas and the other is the change

of the temperature of the cold water.

From Fig 2-7-4, we get:

C(S) = 0.2 (1+60S) (1+5S) (2-7-4)

U2(S ) (I+60S)(I+15S)(I+5S)+I

C(S) _ 0.1(1+5S) (2-7-5)

U1(S) (l+60S) (I+15S) (l+Ss) +i

Now we discuss which disturbance gives more serious influence on

the system.

At first, from the meaning of the feedback we can know that a sig-

nal is produced along the closed loop to eliminate the influence of

the disturbance due to the existence of the feedback (suppose X=0),

so the smaller the time delay between the output and the entrance

point of the disturbance, the faster the correction effect of the

feedback and consequently the smaller the influence of the disturbance.

For example, the peak value of the disturbance response will be reduced.

By this principle we can see that for the above system there are
 99

5 second time delay between C and U I and 65 second time delay between C

and U 2, so it is certainly that U 2 will have more serious influence then

U 1 does. It is so from the dynamic analysis and from the static analysis

we can also learn if both U 1 and U 2 are step disturbances, then from

(2-7-4) and (2-7-5),the steady deviation due to U 2 is 0.i and that due to

U 1 is 0.05, so in both dynamic and static cases, disturbance U 2 is with

more serious influence. Consequently, U 2 should be treated by full rejec-

tion. Then, how to reduce the influence of U 1 ?

The cascade control may be used now and we can design a system shown

in Fig 2-7-5. U21U

~
~ o . s !~
i iVl

| I l+ss I
[ . ] 1
1+55
'I
|

Fig 2-7-5

From ( 2 - 5 - 5 ) , we c a n g e t IVf:

Wf = - 2 (I+5S)(I+60S)+ 5 (2-7-6)
5 1 + SS

Thus, we combined cascade control with full rejection. In fact,

in the above system, if a P+I regulator is adopted for the main regulator,

then the system is not only with the ability to fully reject to U Z, but

also without steady deviation due to U 1 and this is a very good control

indeed.

So, in general, for supply disturbances, we can use cascade control

to reduce t~heir influence and for the load disturbances, if the realiza-

tion is not very difficult, then it would be better to eliminate their

influence by using full rejection design.

 100

§ Z-8 The Application of State Feedback to Realize the Design of Full

Rejection to Disturbances

The above analysis denotes that for a single variable process

control system, no matter where the disturbance is, we can realize the

full rejection design to disturbances and all the projects used before,

by a special terminology of control, are based on the combined system

by disturbance control.

But, we can propose a question: Is this the unique method for

rejection design?

In most textbooks, there is a common conclusion: It is impossible

to realize the design of full rejection to disturbances by deviation

control. Now , we discuss this conclusion.

As we know that control by deviation is the most popular and basic

control mode.The so-called deviation means the difference between the

reference and the output of the feedback element. From Fig 2-8-1 the

deviation is : U
I
= x - M
Inasingleloopcontrol
c2-8-l)
~- -I'~J--~sr---~...Pl~-'~ I-'&-7 c

, , eo,.m,, ooo,,urooen
element of the system output C.
It has been pointed out Fig 2-8-1

in § 2-3 i f we w a n t t o r e a l i z e the full rejection to disturbances by

this manner, then the gain of the regulator s h o u l d be i n f i n i t e . At f i r s t ,

it is difficult t o do so i n p r a c t i c e and, on t h e o t h e r hand, in general

it will cause unstability in the system. Thus, in order to avoid the

unstability problem, cascade control is widely used, namely quite a part

of the propotional gain of the system regulator is set to the auxiliary

regulator in the inner loop. This measure can improve the system ability

to resist disturbances very much, but it can not realize the full rejec-

tion to disturbances yet.

 101

Then, is it really that it is impossible to realize the full rejec-

tion design to disturbances by deviation control?

This problem is worth being discussed.

In fact, if we do not consider the element Wf simply as the measure-

ment element of output C, then we can reach another conclusion.

We discuss the state variable feedback control system now. At first,

we discuss two simple examples of state variable feedback control sys-

tems. Suppose an open loop character :

W(S) = K (2-8-1)
S(S+I) (S+2)
and the following closed-loop character is expected:

C _ SO = SO (2-8-2)
R (S+2)(S2+6S+25) $3+8S2+37S+50

Because in state variable feedback control systems, the state varia-

bles are expressed by X l , x 2 ...... , so we d e n o t e the reference by R h e r e .

Now, we build a state variable control system as shown in Fig 2-8-2.

From the control theory, this system can be transferred into a single

loop control system as F i g 2-8-3.

R C
~-I•, 1S+l I - ' 1 ls
"J
i
Fig 2-8-2

~ E ~-[
f

[
s(s+l) (s+2)

KI+KzS+K3S(I+S)
K

]~
I L.

Fig 2-8-3
 102

From Fig 2 - 8 - 3 , we g e t :
C K
R (S) = S(S+I}(S+2)÷K(KsSZ+(KS+K'2)S+I)

K
(2-8-3)
S3+(3+KK3}S2+(2+K(K2÷K3))S÷KK 1
Equalizing (2-8-2) with (2-8-5) yields:

K = 50 1
KK1 = 50
(2-S-4)
3 + KK2= 8
2 + K(K2+K3~__ = 37
and we g e t :
KI = I, K2 = 0.6, K3 = 0.I, K = 50
Therefore, the character o£ the feedback element is:

Wm(S) = K3s.cI+S)+K2S+K1 = 0 . I ( S + 2 } ( S + 5 } (z-s-s)

and:
c (s) = K
R s(s+1)Cs+2) + 0 . 1 ( s + z ) ( s + s )
K =
50 (2-8-6)
(S+2)(S(S+I)+0.1K(S+5)) (S+2)($2+6S+25)
So, the control demand is reached.
The second example is with the following plant :

WpCS] = 1
(2-8-7)
S[S+l](S+10]
The r e g u l a t o r character is:
Wc(S ) = K(s+2~ (2-s-s)
s+o~
The following closed-loop character is expected:
72(S+2)
(s) =
R ($2+1.414S+1)(S+9)(S+10)

= 72(s+2) (2-s-9)
$4+26.4S3+180.4S2+229S+144

Now, we build a state variable control system as shown in Fig

2-8-4 and it can be also turned into a single loop control system

as shown in Fig 2-8-5.

 103

s+2J
S+I0

Fig 2-8-4
......

R tXh E K(S+2) C
S(S+,~) (S+I) (S+10)
4-T %---
l K1 +K2S+K3S(S+10) +K4S (S+l) (S+10)
/
Fig 2-8-5
The c o e f f i c i e n t s c~ , K,K1,K2,K 3 and K4 s h o u l d be d e t e r m i n e d .
From Fig 2 - 8 - 5 , we g e t :

,,,C (S) = ,. K(S+2)

R(S) ~ (KK4*I)S4+[K(K3+I3K4)+(II+~4)IS3+[K(K2+IZK3+32K4)+

(2-8-10)
÷ ( 1 0 + 1 1 ~ ) ] $2+ ~K(K1+ 2K2+30K3+BOK4*10~]S*2KK1~
Equalizing ( 2 - 8 - 9 ) w i t h (2-8-10) yields:
K = 72
KK4 + 1 = 1
K(KB+I3K4)+(11+~) = 26.4 (2-8-11)
K(K2+I2K3+52K4)+(10+II~ ) = 180.4
K(KI+2K2+30K3+20K4)+I0~= 229
2KK 1 = 144
From i t we g e t ;
K = 72, KI= 1, K2= 0.0154,
K3= 0.0014, K4= 0, ~ = 15.3
and the character of the regulator is:
 ~04

WcCS
_ ) : 72(S + 2) (2-8-12)
S + 15.3

The above examples tell us that we can utilize the state variable

feedback to realize control systems with certain demanded merits and

these systems are with deviations indeed. So, they should belong to the
U2 UI

R X ~ X3 4 ~ ~ ~ L _ . . X.
'6
(

Fig 2-8-6

control by deviation but their feedback element characteristics are

with special forms.

Now, let us discuss further whether the state variable feedback

can be used to realize the full rejection to disturbances.

Fig 2-8-6 is such a system and we want to realize the full rejec-

tion to U 1 and U 2.

It is easily to get the state equations and they can be expressed

in the following

0
-W 1

1
matrix

-W 2
form:

0
xl]
X2 =
UIW 1
U2W 2
(2-8-13)
0 0 1 -W 3 x3 ] 0
!
WRH 1 WRH 2 WRH 3 I+WRH 4 ~ X4 J • RWR j

For instance, if the rejection to U 1 is expected, i.e. XI= 0,

then suppose R=0 and U2=0 and we get the conditions:

 105

-WIX 2 = UIW 1
X z - WzX 3 - 0 (2-8-14)
X3 - X4W 3 = 0

WRH2X2+WRH3X3+(I+WRH4)X4 = 0

From this set of equations, we g e t :

When H3 = H 4 = 0,

1
Hz - (2-8-18)
WRW2W 5
When H 2 = H 4 = 0,

H3 1. (2-s-163
WRW s
When H 2 = H 3 = 0,

1
H4= - - - (2-8-17)
WR
Obviously, if any o n e o£ (2-8-15),(2-8-16) and (2-8-17) is held,

then this system is with full rejection to U 1 and it is r e a l i z e d by

deviation control.

H 1 is easily to be determined. For example, when Hz=Hs=0 , the system

transfer function is:

X1 WRWIW2W 3
C2-8-18)
R 1 + WRH 4 + W R W I W 2 W 3 H 1

But :
I
H4 =
WR
SO:

X1 1
- (2-8-19)
R H1
This is just ideal control, namely this is the best control the

system can reach.

Therefore, i£ the control demands a r e given, then it is not diffi-

cult to determine H1.

If HI= i, then:

X1 = R (2-8-20)
 106

and there is no deviation in this system.

Thus, by using s t a t e variable feedback control system, we can rea-

lize full rejection design to disturbances . We need not to measure

the disturbances and , if necessary, we can realize the ideal control

without the deviation at the meantime.

Obviously, in the above analysis, if any one of (2-8-16) and

(2-8-17) is held, then the system is with full rejection to both U 1

and U 2 by deviation control.

In control systems, some state variables (not all~ may be measured

more conveniently than disturbances and, by the above analysis, if some

state variable is difficult to be measured) then we can measure another

one , for instance there are three different forms can he adopted for

realizing a system with full rejection to UI, so the design is quite

flexible . Furthermore) either of (2-8-16) and {2-8-17~ gives the full

rejection to both U 1 and U 2 , so it is not necessary to consider two

disturbances separately.

Besides, no special stability problem will arise then, so the

realization of full rejection to disturbances by state variable feed-

back is an available method.

But, however, this system is physically irrealizablein general,

i.e. the degree of the numerator of H2, H 5 or H 4 is alway higher than

that of the denominator. This means that differential feedback chan-

nels are introduced and the high frequency disturbances will be able

to be fed back. So, when this project is applied, this problem should

be considered carefully.
 107

§ 2-9 The R e j e c t i o n nesign to Disturbances for Smith Predictor

As well-known) in practical process control systems) the plants
are often with dead time delays. It means that the transfer functions
of plants of Zen contain the term e
-TS

Because of the existence of the dead time delay, the system will
meet troubles in stability. In fact, when ~ i s quite large, the system

is difficult to be stable.
In order to overcome the influence of dead lag, a very effective
control model is the Smith predictor.
A typical Smith predictor is shown in Fig 2-9-1.
X L,,q C
,C .I e- z Sl

e
-'CS

Fig 2-9-1
For this system, the closed-loop transfer function is:

WpWR e - ~ S /(i + WpWR(I- e - r S ) ) =

c(s) =
-rS
xcs) 1 + WpWRe

-ZS)
1 + WpWR(I- e
-~S -TS
WpWRe = WpWRe
(2-9-1)
I+WpWR(I_e- z-S)+WpWRe-Y~S i+ WpW R

Obviously, for this system, the influence on the stability by the

dead delay has been eliminated . This is because the compensation element
(Wp-Wpe S) is adopted.

Because the Smith predictor has the ability to eliminate the influ-
ence of the dead delay on system stability, so it is widely applied in
practice and has been considered as an advanced control system.

Now, we discuss the case when the Smith predictor is with distur-

bances and research whether we can also get full rejection design to

disturbances for it.

The following analysis will show that the prominent merit of the

Smith predictor is: It not only can eliminate the influence of the dead

lag on system stability) but also can reach full rejection design to

disturbances without any measurement to them. In fact, if we observe

Fig 2-9-1 carefully, we can know that the Smith predictor is a state va-

riable control system indeed and we have pointed out that by using

the state variable feedback control we can realize the full rejection

design to disturbances. So) we will see that the Smith predictor realizes

the full rejection design to d i s t u r b ~ c e s by using state variable feed-

back control essentially. U2 U1

Fig 2-9-2 shows such a d e s i g n . [ /

Fig 2-9-2
U2 U1

e-TS

L [ l_

Fig 2=9-3
 109

For convenience, Fig 2-9-2 is transferred into Fig 2-9-3.

From Fig 2-9-3) we get:

WR -TS
I + WpWf + WRWp(l-e-r S) Wpe
c(s) =
x(s) -~ S
I + WRWpe

i+ WpWf+ WRWp(l- e -~S)

= WRWp e-IS

1 ÷ WpWf÷ W R W ( I - e-IS)+ WRWpe-ZS

-IS
WRWpe
(2-9-2)
1 + WpWf+ WR{ Wpe -IS + Wp(l- e-TS))

and:
c(s) = I + WfWp - WRW p( e -Ts - I )
{2-9-3)
u1(s)
1 + WpWf + W R ( Wp e-rS + Wp(l-e-rS))

So t h e o u t p u t is:
WRWpe-~ SX(S) + ( I+ WpWf- WRWp( e -Zs - I) )U 1 {S)
c(s) = (z-9-4)
i * WpWf* WR( Wpe - ~ s - Wp( e -'CS
. - i ))
Thus, if this system is with full rejection to U I, then it is
necessary and sufficient:
-~S
i + WpWf - WRWp( e i ) - 0 (2-9-5)

and it yields:
WRW p( e - Z S - i) - i
Wf (2-9-6)
Wp
Furthermore, the output of the system then is:

WRWpe - Y S
COS) -- X(S) - X(S) (2-9-7)
-'cS
WRWpe
So, this system is not only with full rejection to U I, but also
 110

without deviations.
This is a very ideal result and it is an important merit of the
Smith predictor indeed.
When the disturbance U 2 is considered, we have:

C(S) = WRWpe-rSx(s) + Wpe-~S(l + WfWp- W R W p ( e - ~ S - l)U2(S)

1 + ~wf+ wR( ~ e - ~ s - Wp( e - ~ S _ I ) )

(Z-9-8)
and the condition for full rejection to U 2 is:
-YS
Wf = WRWp( e - 1 )
(z-9-9)
Wp

Obviously, both (2-9-6) and (2-9-9) are identical.

That means that the adoption of Wf given by (Z-9-6) or (2-9-9)
results in full rejection to both U 1 and U 2 and this result has nothing
to do with the forms of disturbances and no measurement of disturbances
is needed either. Besides, it reachs the ideal control , i.e. without

deviations, at the meantime.

When the measurement element is with the character Wm, then the

Smith predictor has the following form:

Fig 2-9-4
The closed-loop transfer function of this system is:

-qS
c(s) w_w e
= _K p (2-9-10)
X(S) 1 + WRWpW m
 111

And the Smith predictor with the ability to fully reject the dis-
turbances is shown below: U2
U1
C

Fig 2-9-5
The system response is: ( only U 1 is considered)

WRWpe-[Sx(s ) + (I+ WfWmWp+ WRWmWp(l-e-TS)) UI(S)

C(S) = (2-9-11)
I + WfWmWp+ WRWmWp(I-e-[S ) + WRWmWpe - r S

If the full rejection to U I is expected, then it is necessary:

- Z S)
I + WfWmW p + WRWmWp(I - e = 0 (2-9-12)
name ly :
Wf = WRWmWp(e--£ S_ i) - 1
(z-9-13)
W mW p

Then, the closed-loop transfer function is:

-'(S
C fS) = WRWpe
(2-9-14)
X(S) WRWmWpe - "fS W
m

When the supply disturbance U 2 is considered, the system output

is:

C(S) = WRWpe-[ SX(S)+ Wpe-~S(I+WfWmWp + WRWmWp(I_e--rS))u2(s )

I+ WfWmWp+WRWmWp (l-e- -g S)+ WRWmWp e- 15S

(2-9-1s)
The condition for full rejection to U 2 is:
 112

WRWmWp(e- ~ S _ i )
Wf= (2-9-16)
WmW p

It is the same as (2-9-13).

So the adoption of a Wf given before can give the full rejection to

both U 1 and U 2.
The system is also without deviation then.

When ~ = O, t h e n :

Wf = 1 (2-9-17)
WW
mp
the channel 1 - e - T S i s b r o k e n o f f now. T h a t m e a n s , for the conven-

tional plant ( -6 = 0 ) , the adoption of an inner loop can result in

full rejection to disturbances a s shown i n F i g 2-9-6.

But t h i s is practically corresponding to infinite open-loop gain.

•U 2 U1

wR Wp ~

WW
mp

WmW p

Wm I_

Fig 2-9-6
 113

§ 2-10 The Independence Principle of Rejection Design to Disturbances

Although we did not point out especially in the above discussions,

the systems under consideration are assumed to be linear with constant

coefficients.
Then, a problem arises: Can the discussed principles of rejection

design to disturbances be also available to the non-linear systems or

linear systems with time varying coefficients ?

Now, we discuss this problem.

In process control systems, the non-linear ( or time raring coef-

ficients> cases in general are: either the control system contains non-

linear elements or the disturbances are imported into the system through

non-linear devices. U
Fig 2-10-1 shows such a system.

F3

Fig 2-10-1

In this figure, each of Wp2, W n and W R may be non-linear or time

varying. If they are non-linear elements, then they should be expressed

bY describing functions.
From this figure, we get:

C(S) (WfWpI+ Wu)WnWp2U(S ) WoW .N ~X

= + K p± pz (2-10-1>

1 + WRWp1Wp2 1 + WRWplWp2
In order to reach full rejection to the disturbance, it is necessary:

WfWpl + W U = 0 (2-i0-2>
name Iy :
 114

WU
wf (2-io-3)
w
pl

Thus, the condition for full rejection to the disturbance is inde-

pendent of the forms of Wp2, W R and W n.

From the analysis of concepts, in order to check the influence of

the disturbance we can let R=0 and for a system with full rejection to

~he disturbance, we get:

q = E = 0 (2-10-4)
Therefore, it is not important what forms the regulator W R and the

plant Wp2 are, i.e. they may be non-linear elements or linear time

varying elements.
On the other hand, the output of the system is:

( WfWpl + WU )WnWp2U(S)
c(s) = (2-1o-s)
1 + WRWplWp2

Now that the condition (2-10-2) gives the guarantee that C(S)=0,

so in the above equation W may be any form. It means that W may also
n n
be a non-linear element or a linear time varying element.

Thus~ we can say that the methods discussed in this chapter for

the design of full rejection to disturbance~ are universal. They can

be applied not only to linear systems, but also to non-linear systems.

In other words, it means that the conditions of full rejection to dis-

turbances are independent of the forms of the elements which are not

included in these conditions.

 115

§ 2-11 The Disturbance-Rejection Design for a Two-Variable System

with Only One Controlled Output

Suppose there is a two-variable system but only one output in it

is controlled, then this is also a single variable system. For instance,

in the example of § I-Ii, if only the level or the flow is controlled,

then the system is just one we will discuss now.

Without doubt, in this case the other manipulated variable will

become a disturbance for the controlled channel through the coupling

channel. Now , we discuss to do disturbance-rejection design for this

system.

We discuss two cases.

The first is: If open loop control is carried out for this system,

then the disturbance-rejection design is with the following form:

M1 -~ I C1
I I

x C~
Fig 2-11-1

In order to consider the influence of the disturbance, we may let

X = 0, so the output is:

-FR
C 2 = MIP21 + MIPII P22 (2-11-1)
i + FRPI2

Therefore:

C2 _ P21 + RF(PI2P21- PIIP22 )

(2-11-2)
M1 1 + FRPI2

Obviously, if M 1 is expected to have no influence on the controlled

 116

output C2, t h e n it is necessary :

F = P21 (2-11-s)
R( PIIP22 - PI2P21 )

Obviously, when P21=0, then M 1 is not a disturbance for C 2 and we

have consequently F= 0.

When P12 = O, we have;

P2
F = 1 (2-11-4)
RPIIP22

The second case is: The closed-loop control is carried out to C 2

and the system structure is shown in Fig 2-11-2.

ci

IFi

Fig 2-11-2
M

I
X=0
T C
2

Fig 2-11-5
 117

In order to get the relation between C 2 and MI, we let X= 0.

The influence of M 1 on C 2 is exerted through two channels. The first

channel is through P21 and it is shown in Fig 2-11-3.

From this figure, we get:

C2 P21 ( 1 + FRP12)
=
(2-zz-s)
M1 1 + FRP12 + RHP22

The second channel is through PII and the system is shown below:

J " I"
M1 ] C2
R I -[ 221

Fig 2-11-4

Fig 2-11-4 can be transferred

into Fig 2-11-5.

M1
_C 2

PI2 I
Fig 2-11-5
Fig 2-11-5 can be transferred into

Fig 2-11-6 further.

C2

I +RHP22

Fig 2-11-6
 118

From Fig 2-11-6 we can get the relation between C 2 and MI:

C2 -Pll FR(I+ RHP22) P22

M1 1 + RHP22 + FRPI2 1 + RHP22

-FRPIIP22
(2-11-6)
1 + RHP22 + FRPI2
The final relation between C 2 and M 1 is the sum of (2-11-5)

and (2-11-6).

Thus:

C2 P21 + RF( PI2P21 - PIIPz2 )

- (2-11-7)
M1 1 * RHP22 + RFPI2

If the full rejection to M 1 is expected, then it is necessary:

F = P21 (2-11-8)
R(PIIP22-PI2P21)

It is identical with (2-11-3). It means that no matter the con-

trol loop is open or closed, the condition for rejection to M 1 is the

same.

We must point out that the conclusions obtained here are only va-

lid when only one variable in this two-variable system is controlled

and the other output is free. In such a case , the m a n i p u l a t e d variable

of the free output m a y be considered as a disturbance of the controlled

system. If, however, both variables are under control, then although

the plant is still the same, the system would be a two-variable system

and M 1 is no longer able to be considered as a disturbance of the C 2

control loop .

In such a case, the d e c o u p l i n g d e s i g n principles should be con-

sidered for this two-variable control system.

 119

§ 2-12 The Rejection Design to the Disturbance with Ambiguous Input

Channels(58)

In our above discussion, no matter what kind of disturbances is

considered, if we want to realize a full rejection design to it, we must

know the exact character of the disturbance import channel.

But, howevers sometimes the input channel of the disturbance

may be very complicated and it can not be expressed by a simple transfer

function.

Fig 2-12-1 shows such a system. In this system, U is the system

disturbance and we expect that the output C would not be influenced by U.

KT I~,

' ~ 411
.LJ
4 S+ 83.3 ] ~-

21.0

] s2.8 I~

U
i
~1 S+171

Fig 2-12-1

This is a single variable system but the input channel of the distur-

bance is not easily to be determined.

 120

In such a case, how to realize the d i s t u r b a n c e - r e j e c t i o n design

for this system?

Now, we express this system in d i f f e r e n t i a l equations:

dC = -136.5C + 31.7T + 1 0 . 6 W
dt

dT = -83.3T + 41.IC + 21.0m (2-12-1)

dt

dm
- -171m + 82.8T - 67.6U
dt
In order to discuss the influence of the disturbance, let X = 0

and a system with full r e j e c t i o n to U gives:

dC
- 0 I
dt k
(2-12-2)
C = 0

Thus, the following three equations are obtained:

0 = 31.7T + 10.6W (2-12-3)

dT
= -83.3T + Z1.0m (2-12-4)
dt

dm
= -171m + 82.8T - 6 7 . 6 W (2-12-s)
dt
From (2-12-3), we get:

W
= -2.99 - KT (2-12-6)
m
Subsituting this result into (2-12-4) yields:

W 62.8
= Km (2-12-7)
m
S + 83.3

Now, we substitute (2-12-6) and (2-12-7) into (2-12-5) and get:

W 424S
= (2-12-s)
U S 2 + 254.35 + 12506
That means that the s a t i s f a c t i o n of anyone of t h e above three

conditions (2-12-6), (2-12-7) and (2-12-8) can result in full rejec-

tion to U.
 121

Therefore, it seems t h a t this system is very complicated, but the

condition for rejection to the disturbance is easily to be determined.

This is an available method and certainly it can be used in the si-
milar c a s e s .
But there is another method being able to solve this problem.
Take Laplace transform to (2-12-1) and write it into the matrix form:

-41.I S+ 8 3 . 3 -21 0 = (2-12-9)

0 -82.8 S+ 171 O -67.6U

Now, we introduce the control equation:

W = KcC + KTT + Kmm + KuU (2-12-10)

Then, the system equation becomes:

0
-41.1 S+ 8 3 . 3 -21 0 0
(2-12-11)
0 -82.8 S+ 171 0 -67.6U

-Kc (S) -KTCS) -Km(S) i KuU

Solving for C, we get:

0 -31.7 0 -10.6

c(s) = 0 S+ 8 3 . 3 -21 0 (2-12-12)

-67.6U -82.8 $+171 0

K (S)U -KT(S) -K (S) i

U m

where:
S+ 136.3 -31.7 0 -10.6

Zl =

Expanding the numerator

I -41
o

I -Kc (S)
S+83.3

of
-82.8

-KT(S )

(2-12-12),
-21
S+171

-Kin(S)

we g e t :
0
0

i
(2-12-13)
 122

U(S){-(31.7) (21) (67.6)-(10.6) (21) (67.6)KT(S)-

-(I0.6)(67.6)(S+83.3)Km(S ) + (i0.6) [(S+83.3)(S+171)-
-(21)(s2. s)]%
c(s) :

A (s)
(2-12-141
When t h e rejection to the disturbance is expected, it is neces-

sary that the numerator of (2-12-14) is equal to zero.

There are a lot o f m e t h o d s w h i c h c a n make t h e n u m e r a t o r be z e r o .

In fact, there are infinite methods indeed. But, however, the fol-

lowing three m e t h o d s may be t h e simplest:

Ku(S) = Km(S) = 0, KT(S)= -2,99 (2-12-15)

KuCS) = KmCS) = 0, KIn(S)- 62.8 C2-12-16)

S+83.3

KT(S) = KIn(S) = 0

Ku(S ) = 4245
S2 + 254.3S + 12506
} (2-12-17)

Substituting (2-12-15),(2-12-16) and (2-12-17) into A(S), res-

pectively , we f i n d A(s) p o.

Thus, each of the above three conditions can give the full rejec-

tion to the disturbance U.

In fact, they are identical with (2-12-6), (2-12-7) and (2-12-8).

The l a t t e r method sometimes is called as t h e application of the

Principle of Invariance.