1234567898

8.1

a) For step response,

M
input is u ′(t ) = M , U ′( s ) =
s
 τ s +1   τD s +1 
Ya′ ( s ) = K c  D U ′( s ) = K c M  
 ατ D s + 1  s (ατ D s + 1) 

K c Mτ D KcM
Ya′ (s ) = +
ατ D s + 1 s (ατ D s + 1)

Taking inverse Laplace transform

K c M −t /( ατ D )
y ′a (t ) = e + K c M (1 − e −t /( ατ D ) )
α

As α →0
∞
e − t /( ατ D )
y a′ (t ) = K c Mδ(t ) ∫ dt + K c M
t =0
α
y a′ (t ) = K c Mδ(t )τ D + KcM

Ideal response,

 τ s + 1 KM
Yi′( s ) = Gi ( s )U ′( s ) = K c M  D  = KcMτD + c
 s  s

yi′ (t ) = K c Mτ D δ(t ) + K c M

Hence y a′ (t ) → y i′ (t ) as α → 0

For ramp response,

M
input is u ′(t ) = Mt , U ′( s ) =
s2

Solution Manual for Process Dynamics and Control, 2nd edition,

Copyright © 2004 by Dale E. Seborg, Thomas F. Edgar and Duncan A. Mellichamp.

8-1
  τ s +1   τDs +1 
Ya′ ( s ) = K c  D U ′( s ) = K c M  2 
 ατ D s + 1  s (ατ D s + 1) 

K c Mτ D K M
Ya′ (s ) = + 2 c
s (ατ D s + 1) s (ατ D s + 1)

1 ατ D   − ατ D 1 (ατ D ) 2 
= K c Mτ D  −  + K M  + + 
 s ατ D s + 1 s 2 ατ D s + 1
c
 s

Taking inverse Laplace transform

[ ] [
y a′ (t ) = K c Mτ D 1 − e − t /( ατ D ) + K c M t + ατ D (e − t /( ατ D ) − 1) ]
As α →0

y a′ (t ) = K c Mτ D + K c Mt

Ideal response,

 τ s + 1 K c Mτ D K c M
Yi′(s ) = K c M  D 2  = + 2
 s  s s

yi′ (t ) = K c Mτ D + K c Mt

Hence y a′ (t ) → y i′ (t ) as α → 0

b) It may be difficult to obtain an accurate estimate of the derivative for use

in the ideal transfer function.

c) Yes. The ideal transfer function amplifies the noise in the measurement by
taking its derivative. The approximate transfer function reduces this
amplification by filtering the measurement.

8.2

 K 2 τ1 
P ′( s ) K1 K + K 2 τ1 s + K 2  K + K s + 1
a) = + K2 = 1 = ( K1 + K 2 )  1 2

E ( s ) τ1 s + 1 τ1 s + 1  τ1 s + 1 
 
 

8-2
b) Kc = K1 + K2 → K2 = Kc − K1

τ1 = ατ D

K 2 τ1 K ατ
τD = = 2 D
K1 + K 2 K1 + K 2

K 2α
or 1=
K1 + K 2

K1 + K 2 = K 2 α

K 1 = K 2 α − K 2 = K 2 (α − 1)

Substituting,

K 1 = ( K c − K 1 )(α − 1) = (α − 1) K c − (α − 1) K 1

Then,
 α −1
K1 =  K c
 α 

c) If Kc = 3 , τD = 2 , α = 0.1 then,

− 0 .9
K1 = × 3 = −27
0 .1

K 2 = 3 − (−27) = 30

τ1 = 0.1 × 2 = 0.2

Hence

K1 + K2 = -27 + 30 = 3

K 2 τ1 30 × 0.2
= =2
K1 + K 2 3

 2s + 1 
Gc ( s ) = 3 
 0 .2 s + 1 

8-3
8.3

a) From Eq. 8-14, the parallel form of the PID controller is :

 1 
Gi ( s ) = K c′ 1 + + τ′D s 
 τ′I s 

From Eq. 8-15, for α →0, the series form of the PID controller is:

 1 
Ga ( s ) = K c 1 + [τ D s + 1]
 τI s 

 τ 1 
= K c 1 + D + + τ D s
 τI τI s 

 
 
 τ D  1 τDs 

= K c 1 + 
 1+ +
 τI    τ   τ 
 1 + D τ I s 1 + D  
  τI   τI  

Comparing Ga(s) with Gi(s)

 τ 
K c′ = K c 1 + D 
 τI 
 τ 
τ′I = τ I 1 + D 
 τI 
τD
τ′D =
τ
1+ D
τI

 τ 
b) Since 1 + D  ≥ 1 for all τD, τI, therefore
 τI 
K c ≤ K c′ , τ I ≤ τ′I and τ D ≥ τ′D

c) For Kc = 4, τI=10 min , τD =2 min

K c′ = 4.8 , τ′I = 12 min , τ′D = 1.67 min

d) Considering only first-order effects, a non-zero α will dampen all
responses, making them slower.

8-4
8.4

Note that parts a), d), and e) require material from Chapter 9 to work.

a) System I (air-to-open valve) : Kv is positive.

System II (air-to-close valve) : Kv is negative.

b) System I : Flowrate too high → need to close valve → decrease controller

output → reverse acting

System II: Flow rate too high → need to close valve → increase
controller output → direct acting.

c) System I : Kc is positive
System II : Kc is negative

d) Kc Kv Kp Km
System I : + + + +
System II : − − + +

Kc and Kv must have same signs

e) Any negative gain must have a counterpart that "cancels" its effect. Thus,
the rule:

# of negative gains to have negative feedback = 0 , 2 or 4.

# of negative gains to have positive feedback = 1 or 3.

8.5

a) From Eqs. 8-1 and 8-2,

[
p(t ) = p + K c y sp (t ) − y m (t ) ] (1)

The liquid-level transmitter characteristic is

ym(t) = KT h(t) (2)

where h is the liquid level

KT > 0 is the gain of the direct acting transmitter.

8-5
 The control-valve characteristic is

q(t) = Kvp(t) (3)

where q is the manipulated flow rate

Kv is the gain of the control valve.

From Eqs. 1, 2, and 3

[ ] [
q(t ) − q = K v p (t ) − p = K V K c y sp (t ) − K T h(t ) ]
q (t ) − q
KV K c =
y sp − K T h(t )
For inflow manipulation configuration, q(t)> q when ysp(t)>KTh(t). Hence
KvKc > 0

then for "air-to-open" valve (Kv>0), Kc>0 : reverse acting controller

and for "air-to-close" valve (Kv<0), Kc<0 : direct acting controller

For outflow manipulation configuration, KvKc <0

then for "air-to-open" valve, Kc<0 : direct acting controller

and for "air-to-close" valve, Kc>0 : reverse acting controller

b) See part(a) above

8.6

For PI control

 1
t

p (t ) = p + K c  e(t ) + ∫ e(t*)dt * 

 τI 0 

 1
t

p ′(t ) = K c  e(t ) + ∫ e(t*)dt * 
 τI 0 

Since
e(t) = ysp – ym and ym= 2

8-6
 Then

e(t)= -2

 1
t
  2 
p ′(t ) = K c  − 2 + ∫ (−2)dt *  = K c  − 2 − t 
 τI 0   τI 

Initial response = − 2 Kc

2K c
Slope of early response = −
τI

− 2 Kc = 6 → Kc = -3

2K c
− = 1.2 min-1 → τI = 5 min
τI

8.7

a) To include a process noise filter within a PI controller, it would be placed

in the feedback path
b)

 1 
K c 1 + 
 τI s 

1
1 f s 23

c) The TF between controller output P ′(s ) and feedback signal Ym(s) would
be

8-7
 P ′( s ) − K c (τ I s + 1)
= Negative sign comes from comparator
Ym ( s ) τ I s (τ f s + 1)

M
For Ym ( s ) =
s

− KcM  τI s +1  − KcM A B C 
P ′( s ) =  2 =  2 + + 
τI  s (τ f s + 1)  τI  s s τ f s + 1

C
The term gives rise to an exponential.
τ f s +1

To see the details of the response, we need to obtain B (= τI - τf) and A(=1)
by partial fraction expansion.

The response, shown for a negative change in Ym, would be

Slope = KcM/τI

"Ideal" PI
y

Filtered PI
-KcM

-KcM(1-τf/τI)

time

τf
Note that as → 0 , the two responses become the same.
τI
d) If the measured level signal is quite noisy, then these changes might still
be large enough to cause the controller output to jump around even after
filtering.

One way to make the digital filter more effective is to filter the process
output at a higher sampling rate (e.g., 0.1 sec) while implementing the
controller algorithm at the slower rate (e.g., 1 sec).

A well-designed digital computer system will do this, thus eliminating the

need for analog (continuous) filtering.

8-8
8.8

τD
a) From inspection of Eq. 8-26, the derivative kick = K c ∆r
∆t
b) Proportional kick = K c ∆r

c) e1 = e2 = e3 = …. = ek-2 = ek-1 = 0

ek = ek+1 = ek+2 = …= ∆r

p k −1 = p
 ∆t τ 
p k = p + K c ∆r + ∆r + D ∆r 
 τI ∆t 
 ∆t 
p k +i = p + K c ∆r + (1 + i ) ∆r  , i = 1, 2, …
 τI 

τD
Kc ∆r
pk ∆t

∆t
K c ∆r Kc ∆r
τI
p

k-1 k k+1 k+2 k+3

c) To eliminate derivative kick, replace (ek – ek-1) in Eq. 8-26 by (yk-yk-1).

8-9
8.9

a) The digital velocity P algorithm is obtained by setting 1/τI = τD = 0 in Eq.

8-28 as

∆pk = Kc(ek – ek-1)

[ ]
= K c ( y sp − y k ) − ( y sp − y k −1 )

= K c [ y k −1 − y k ]

The digital velocity PD algorithm is obtained by setting 1/τI = 0 in Eq. 8-

28 as
τ
∆pk = Kc [(ek – ek-1) + D (ek – 2ek-1 + ek-2)]
∆t
τ
= Kc [ (-yk + yk-1) + D (-yk – 2yk-1 + yk-2) ]
∆t
In both cases, ∆pk does not depend on y sp .

b) For both these algorithms ∆pk = 0 if yk-2 = yk-1 = yk. Hence steady state is
reached with a value of y that is independent of the value of y sp . Use of
these algorithms is inadvisable if offset is a concern.

∆t
c) If the integral mode is present, then ∆pk contains the term Kc ( y sp − y k ) .
τI
Thus, at steady state when ∆pk = 0 and yk-2 = yk-1 = yk , yk = y sp and the
offset problem is eliminated.

8.10

P ′( s )  1 τDs 
a) = K c 1 + + 
E (s)  τ I s ατ D s + 1 

= Kc
( τ I s(ατ D s + 1) + ατD s + 1 + τD sτ I s )
τ I s (ατ D s + 1)
1 + (τ I + ατ D ) s + (1 + α)τ I τ D s 2 
= Kc  
 τ I s (ατ D s + 1) 

8-10
 Cross- multiplying

( )
(ατ I τ D s 2 + τ I s ) P ′( s ) = K c 1 + (τ I + ατ D ) s + (1 + α )τ I τ D s 2 E ( s )

d 2 p ′(t ) dp ′(t )  de(t ) d 2 e(t ) 

ατ I τ D + τ = K  e (t ) + ( τ + ατ ) + (1 + α ) τ τ 
dt 2 
I c I D I D
dt 2 dt  dt

P ′( s )  τ s + 1  τ D s 
b) = K c  I  
E (s)  τ I s  ατ D s + 1 

Cross-multiplying

τ I s 2 (ατ D s + 1) P ′( s ) = K c ((τ I s + 1)(τ D s + 1) ) E ( s )

d 2 p ′(t ) dp ′(t )  de(t ) d 2 e(t ) 

ατ I τ D + τ = K  e (t ) + ( τ + τ ) + τ τ 
dt 2 
I c I D I D
dt 2 dt  dt

c) We need to choose parameters in order to simulate:

e.g., Kc = 2 , τI = 3 , τ D = 0.5 , α = 0 .1 , M=1

By using Simulink-MATLAB

Step Response
22
Parallel PID with a derivative filter
20 Series PID with a derivative filter

18

16

14

p'(t) 12

10

2
0 2 4 6 8 10
Time
Figure S8.10. Step responses for both parallel and series PID controllers
with derivative filter.

8-11
8.11

P ′( s )  τ s +1
a) = K c  I (τ D s + 1)
E (s)  τI s 

τ I s P ′( s ) = K c ((τ I s + 1)(τ D s + 1) ) E ( s )

dp′(t ) K c  de(t ) d 2 e(t ) 

=  e(t ) + (τ I + τ D ) + τI τD 
dt τI  dt dt 2 

b) With the derivative mode active, an impulse response will occur at t = 0.

Afterwards, for a unit step change in e(t), the response will be a ramp
with slope = K c (τ I + τ D ) / τ I and intercept = K c / τ I for t > 0 .

Impulse at t=0

p'
K c (τ I + τ D )
slope =
τI
Kc
τI

8-12