AN

EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Control of
Time-delay Systems

Anjan Rakshit and Amitava Chatterjee

x(t)

Time Response

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Time-Delay Element
Time-Delay
Element

Tl

x(t - Tl)

Tl = Time-Delay

x(t)

unit step

y(t)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Time-Delay Systems
Example 1

Temperature Measurement of Heat-Exchanger Output

b(t)

a(t)
Sensor

Heat

cold in

hot
out

Heat

Sensor

cold in

v = velocity
d = distance

b(t) = a(t Tl )

d
Tl =
v

hot
out

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Time-Delay Systems
Example 2

Thickness Measurement of Rolled Steel Plate

Thickness
gauge

Steel Plate

Time delay between

change in thickness
and measurement:

d
Tl =
v

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Difference Between Time-Lag and Time-Delay

(a)

(b)

(c)

e sT
In case (c), transfer function of the block = G (s ) =
1 + sT
T = time lag, Tl = delay
l

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Problem due to Presence of Time-Delay

in a Process Control Loop
Due to the presence of time-delay, any corrective action from
the controller cannot be immediately applied to the process.
Thus during that time, process may buildup a deviation,
sometimes quite large, due to any load disturbance or change
in set-point.

x(t)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Transfer Function of Time-Delay Element

Time-Delay
Element

y(t) = x(t - Tl)

Y (s )
sT
=e
X (s )

Taking Laplace Transform:

Transfer Function of a Time-Delay Element

sTl )
(
1
1
2 sTl
1 2 sTl +

e
2!
Now, =
=
e sTl
1 sT
2
1
2 l
e
(
2 sTl )
1

1 + 2 sTl +
2!
1
2

Using First-Order Pade Approximation:

1
2 sTl
-sT

e l
1 + 12 sTl

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Effect of Time-Delay on Process Loop Response

G (s ) = G(s )e

T.F. of a Process with Time-Delay:

sTl

where G/(s) is the transfer function of the delay free part of the
process.

Case Study: Proportional Control of a First-Order Process

with Time-Delay
controller

R(s)

Kp

process

e sTl
1 + sT

Kp = Proportional Gain of the Controller

T = Process Time Constant

C(s)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Case Study: Proportional Control of a First-Order Process

with Time-Delay (contd )
process
controller

R(s)

e sTl
1 + sT

Kp

1+

The Characteristic Eqn. of the C.L. System:

K p e sTl
1 + sT

C(s)

0
=

Using First-Order Pade Approximation:

1
1

T
T
K
T

l
p l
K p + 1
2
2
2
+
=0
s + s
1
1

TT
TTl

2 l
2

1
K p 1 sTl
2 =0
1+
(1 + sT )1 + 1 sTl
2

The standard form:

s + 2n s + =
0,
2

2
n

n =

2 ( K p + 1)
TTl

The Natural Freq.

1
1
T + Tl K PTl
2
2
=
2(K P + 1)TTl

The Damping Ratio

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Case Study: Proportional Control of a First-Order Process

with Time-Delay (contd )
process
controller

R(s)

Kp

e sTl
1 + sT

The damping ratio becomes zero when:

2T + Tl

K p =
Tl

The system becomes unstable when:

2T + Tl

K p >
Tl

Conclusion: Time-delay may cause instability even in

first-order system for large values of Kp.

C(s)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Case Study: Proportional Control of a First-Order Process

with Time-Delay (contd )
controller

R(s)
+

+
_

Root Locus Plot

process

e sTl
1 + sT

Kp

K p e sT
Loop T.F. of the C. L. System:
1 + sT

C(s)

Using First-Order Pade Approximation:

Case I:

T>

Tl
2

Kp =Kc

Kp = 0

2
Tl

K p
2
sT
s

K p 1 l

T Tl
2

=
sTl
(1 + sT ) 1 + s + 1 s + 2
2 T Tl

Kp = 0

1
T

Kp = Kc

Kc: Critical Gain

s-plane

2
Tl

2T + l
2
Kc =
Tl

2T + Tl
=

T
l

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Case Study: Proportional Control of a First-Order Process

with Time-Delay (contd )
controller

R(s)
+

+
_

Root Locus Plot

process

e sTl
1 + sT

Kp

K p e sT
Loop T.F. of the C. L. System:
1 + sT

C(s)

Using First-Order Pade Approximation:

Case II:

Tl
T<
2

1
T

K p
2
sTl
s

K p 1

T
T
2
l

sT
(1 + sT ) 1 + l s + 1 s + 2
2 T Tl

Kp =Kc

Kp = 0

Kp = 0

2
Tl

Kp = Kc

Kc: Critical Gain

s-plane

2
Tl

2T + l
2
Kc =
Tl

2T + Tl
=

T
l

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Case Study: Proportional Control of a First-Order Process

with Time-Delay (contd )
process

controller

R(s)
+

+
_

e sTl
1 + sT

Kp

C(s)

Critical gain Kc

1.67

2T

3T

Tl

The Critical gain falls with increasing Tl.

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Bode Plot

Case I: First Order System without Time-Delay

Kp
Open Loop Transfer Function: G (s ) =
1 + sT

Kp
G ( j ) =
1 + j T

|G (j)|

1
T

20 log10Kp
odb

phase
0

- 90o

Conclusion: The phase plot never crosses 180 line, hence the system is always stable.

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Bode Plot

Case II: First Order System with Time-Delay

K p e sT K p sT
Open Loop Transfer Function: G (s ) =
=
e = G1 (s )G2 (s )
1 + sT 1 + sT
l

Frequency Response of Time-Delay Unit:

Phase

G2 ( j) = 1 Tl

Slope = - T l

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Bode Plot

Case II: First Order System with Time-Delay

K p e sT K p sT
Open Loop Transfer Function: G (s ) =
=
e = G1 (s )G2 (s )
1 + sT 1 + sT
l

gain |G (j)|

increasing Kp

GM=0, System becomes oscillatory

20 log 10Kp
odb

Gain crossover freq. (GCF)

For a particular gain Kp, for stability,

GM (gain margin)
with Tl = 0

Phase
0

increasing Tl

- 90

Tl = 0

(GCF ) T

(as Tl is the phase of time-delay)

PM
or Tl <
GCF

PM ( phase margin ) with Tl = 0

- 180o

< PM

PM with Tl = 0

PM = 0, oscillatory system

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Smiths Principle for Control of

Time-Delay Systems

Process with
time-delay:
y(t)

G ( s ) = G ( s ) e sTl

G(s) is the T.F. of the delay free part

of the process and Tl is the time-delay.

y(t) represents the unit step response of G(s) in

a unity feedback closed loop system.

y(t-Tl)

Smiths Principle:

The unit step response desired

from the closed loop system
with the delay introduced in
series with G(s) is y (t Tl).

Tl

If response y(t) satisfies the design criteria

for the delay-free case, then the response to
be designed for the system with time-delay is
the same response but delayed by Tl.

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Smiths Method

Let the controller in the delay free case be Gc(s).

R(s)

+ _

The closed loop transfer function:

controller

process without
time-delay

G'c(s)

G'(s)

Gc/ (s )G / (s )

1 + G (s )G (s )
/
c

C(s)

with time response y(t).

The time response from a system with closed loop transfer function :

Gc/ (s )G / (s )

1 + G (s )G (s )
/
c

e sTl

will be y (t Tl).

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Smiths Method

Let the same response (i.e. y (t Tl)) be available from a controller Gc(s) and
the process with time-delay G(s) in a unity feedback closed loop system.

R(s)

+ _

The closed loop transfer function:

Gc(s)

G(s)

C(s)

Gc (s )G (s )
with time response y (t Tl).
1 + Gc (s )G (s )

Gc (s )G (s )
Gc/ (s )G / (s )e sTl
=
1 + Gc (s )G (s ) 1 + Gc/ (s )G / (s )

According to Smiths Method

Substituting G ( s ) = G ( s ) e

controller

process with
time-delay

sTl

Gc/ (s )
Gc (s ) =
1 + Gc/ (s )G / (s ) 1 e sTl

Gc (s ) : T.F. of the required controller for the process with time-delay

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Smiths Method

The Realization of the Controller Gc(s)

Gc(s)

E(s)
Error

G'c(s)

M(s)
Controller
output

G'(s)

1- e-sT l

For the above realization of Gc(s), knowledge of G(s) and Tl is required.

In practice, model estimates of G(s) and Tl are used.

Model Estimates:

G ( s ) , model of G (s)

Tl

, model of Tl

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Smiths Method

The Realization of the Controller Gc(s)

Gc(s)

E(s)

G'C(s)

M(s)

G ( s )

1 e sTl

Gc ( s ) : The Controller for Delay-Free Process G(s)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Realisation of the Closed Loop System with Gc(s)

Scheme I:

Controller
(designed for G'(s)) Process with
time-delay

R(s)

+
_

G'C(s)

C(s)

G(s)

1 e

G ( s )

sTl

Models

GC(s)

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Realisation of the Closed Loop System with Gc(s)

Scheme II:

Controller (Process with time delay)

R(s)

G'C(s)

C(s)

G(s)

Model of process
with time-delay

sTl
G ( s ) e

Model of delay free

process
G ( s )

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Realisation of the Closed Loop System with Gc(s)

Scheme III:
R(s)

Controller

Process

G'C(s)

G(s)

C(s)

e sTl

Model delay

G ( s )

Model of delay
free process

Smiths method assumes that models are exact, i.e. G ( s ) = G ( s ) and Tl = Tl .

In practice, there is always some mismatch between the model and the actual
parameters, therefore, the closed-loop response may differ depending upon the
degree of mismatch.

AN
EL JA
JA E N
D CT R
AV R AK
PU IC SH
R AL IT
U EN A
N
IV GI ND
ER NE A
SI ER MIT
TY IN A
, K G VA
O DE C
LK P H A
AT AR T
A, T TE
IN ME RJ
D N EE
IA T ,
.

Realisation of the Closed Loop System with Gc(s)

Scheme III:
R(s)

Controller

Process

G'C(s)

G(s)

C(s)

e sTl

Model delay

G ( s )

Model of delay
free process

The model of delay free process provides an estimate of output from delay free
process for the controller. The estimate is, in fact, the prediction of the process
output. Thus the term Smiths Predictor is used for such controllers.