1234567898

7.1

In the absence of more accurate data, use a first-order transfer function as

T '( s ) Ke −θs
=
Qi '( s ) τs + 1
T (∞) − T (0) (124.7 − 120) o
F
K= = = 0.118
∆qi 540 − 500 gal/min
θ = 3:09 am – 3:05 am = 4 min

Assuming that the operator logs a 99% complete system response as “no
change after 3:34 am”, 5 time constants elapse between 3:09 and 3:34 am.

5τ = 3:34 min − 3:09 min = 25 min

τ = 25/5 min = 5 min

Therefore,

T '( s ) 0.188e−4 s
=
Qi '( s ) 5s + 1

To obtain a better estimate of the transfer function, the operator should log
more data between the first change in T and the new steady state.

7.2

h(5.0) − h(0) (6.52 − 5.50) min

Process gain, K = = = 0.336 2
∆qi 30.4 × 0.1 ft
a) Output at 63.2% of the total change

= 5.50 + 0.632(6.52-5.50) = 6.145 ft

Interpolating between h = 6.07 ft and h = 6.18 ft

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

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

7-1
 (0.8 − 0.6)
τ = 0.6 + (6.145 − 6.07) min = 0.74 min
(6.18 − 6.07)
b)

dh h(0.2) − h(0) 5.75 − 5.50 ft ft

≈ = = 1.25
dt t = 0 0.2 − 0 0.2 min min

Using Eq. 7-15,

KM 0.347 × (30.4 × 0.1)

τ= = = 0.84 min
 dh  1.25
 
 dt t =0 
 h(t i ) − h(0) 
c) The slope of the linear fit between ti and z i ≡ ln 1 −  gives an
 h(∞) − h(0) 
approximation of (-1/τ) according to Eq. 7-13.

Using h(∞) = h(5.0) = 6 .52, the values of zi are

ti zi ti zi
0.0 0.00 1.4 -1.92
0.2 -0.28 1.6 -2.14
0.4 -0.55 1.8 -2.43
0.6 -0.82 2.0 -2.68
0.8 -1.10 3.0 -3.93
1.0 -1.37 4.0 -4.62
1.2 -1.63 5.0 -∞

Then the slope of the best-fit line, using Eq. 7-6 is

 1  13Stz − St S z
slope =  −  = (1)
 τ  13Stt − ( St )
2

where the datum at ti = 5.0 has been ignored.

Using definitions,

St = 18.0 Stt = 40.4

S z = −23.5 Stz = −51.1

Substituting in (1),

 1
 −  = −1.213 τ = 0.82 min
 τ

7-2
 d)

6.8

6.6

6.4

6.2

Experimental data
Model a)
5.8 Model b)
Model c)

5.6

5.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure S7.2. Comparison between models a), b) and c) for step response.

7.3

a)
T1′( s ) K1 T2′( s ) K2
= =
Q′( s ) τ1s + 1 T1′( s ) τ2 s + 1
T2′( s ) K1 K 2 K1 K 2 e −τ2 s
= ≈ (1)
Q′( s ) (τ1s + 1)(τ2 s + 1) (τ1s + 1)

where the approximation follows from Eq. 6-58 and the fact that τ1>τ2 as
revealed by an inspection of the data.

T1 (50) − T1 (0) 18.0 − 10.0

K1 = = = 2.667
∆q 85 − 82
T (50) − T2 (0) 26.0 − 20.0
K2 = 2 = = 0.75
T1 (50) − T1 (0) 18.0 − 10.0

Let z1, z2 be the natural log of the fraction incomplete response for T1,T2,
respectively. Then,

7-3
  T (50) − T1 (t )  18 − T1 (t ) 
z1 (t ) = ln  1  = ln  
 T1 (50) − T1 (0)   8
 T (50) − T2 (t )   26 − T2 (t ) 
z2 (t ) = ln  2  = ln  
 T2 (50) − T2 (0)   6
A graph of z1 and z2 versus t is shown below. The slope of z1 versus t line
is –0.333 ; hence (1/-τ1)=-0.333 and τ1=3.0

From the best-fit line for z2 versus t, the projection intersects z2 = 0 at

t≈1.15. Hence τ2 =1.15.

T1 ' ( s ) 2.667
= (2)
Q ' ( s ) 3s + 1
T2 ' ( s ) 0.75
= (3)
T1 ' ( s ) 1.15s + 1

0.0
0 5 10 15 20
-1.0

-2.0

-3.0
z 1,z 2

-4.0

-5.0

-6.0
-7.0

-8.0
time,t

Figure S7.3a. z1 and z2 versus t

b) By means of Simulink-MATLAB, the following simulations are obtained

28

26

24

22

20
T1 , T2

18

16

14 T1
T2
12 T1 (experimental)
T2 (experimental)
10
0 2 4 6 8 10 12 14 16 18 20 22
time

Figure S7.3b. Comparison of experimental data and models for step change

7-4
7.4

2 1.5
Y (s) = G( s) X ( s) = ×
(5s + 1)(3s + 1)( s + 1) s

Taking the inverse Laplace transform

y (t ) = -75/8*exp(-1/5*t)+27/4*exp(-1/3*t)-3/8*exp(-t)+3 (1)

a) Fraction incomplete response

 y (t ) 
z (t ) = ln 1 −
 3 

0.0
-1.0 0 10 20 30 40 50

-2.0
-3.0
-4.0
z(t)

-5.0
-6.0
-7.0
-8.0 z(t) = -0.1791 t + 0.5734

-9.0
time,t

Figure S7.4a. Fraction incomplete response; linear regression

From the graph, slope = -0.179 and intercept ≈ 3.2

Hence,

-1/τ = -0.179 and τ = 5.6

θ = 3.2
2e −3.2 s
G (s) =
5.6s + 1

b) In order to use Smith’s method, find t20 and t60

y(t20)= 0.2 × 3 =0.6

y(t60)= 0.6 × 3 =1.8

Using either Eq. 1 or the plot of this equation, t20 = 4.2 , t60 = 9.0
Using Fig. 7.7 for t20/ t60 = 0.47
ζ= 0.65 , t60/τ= 1.75, and τ = 5.14

7-5
 2
G (s) ≈
26.4 s + 6.68s + 1
2

The models are compared in the following graph:

2.5

1.5

Third-order model
y(t)

First order model

Second order model
1

0.5

0
0 5 10 15 20 25 30 35 40

time,t

Figure S7.4b. Comparison of three models for step input

7.5

The integrator plus time delay model is

K
G(s) e −θs
s
In the time domain,
y(t) = 0 t<0
y(t)= K (t-θ) t≥0

Thus a straight line tangent to the point of inflection will approximate the
step response. Two parameters must be found: K and θ (See Fig. S7.5 a)

1.- The process gain K is found by calculating the slope of the straight
line.
1
K= = 0.074
13.5

2.- The time delay is evaluated from the intersection of the straight line
and the time axis (where y = 0).
θ = 1.5

7-6
 0.074 −1.5 s
Therefore the model is G(s) = e
s

y(t)

Slope = KM

Figure S7.5a. Integrator plus time delay model; parameter evaluation

From Fig. E7.5, we can read these values (approximate):

Time Data Model

0 0 -0.111
2 0.1 0.037
4 0.2 0.185
5 0.3 0.259
7 0.4 0.407
8 0.5 0.481
9 0.6 0.555
11 0.7 0.703
14 0.8 0.925
16.5 0.9 1.184
30 1 2.109

Table.- Output values from Fig. E7.5 and predicted values by model

A graphical comparison is shown in Fig. S7.5 b

1
0.9
0.8
0.7
0.6
Output

0.5
0.4 Experimental data
Integrator plus time delay model
0.3
0.2
0.1
0
0 5 10 15 20 25 30
Time

Figure S7.5b. Comparison between experimental data and integrator plus time
delay model.

7-7
7.6

a) Drawing a tangent at the inflection point which is roughly at t ≈5, the

intersection with y(t)=0 line is at t ≈1 and with the y(t)=1 line at t ≈14.
Hence θ =1 , τ = 14−1=13
e−s
G1 ( s ) ≈
13s + 1

b) Smith’s method

From the graph, t20 = 3.9 , t60 = 9.6 ; using Fig 7.7 for t20/ t60 = 0.41

ζ = 1.0 , t60/τ= 2.0 , hence τ = 4.8 and τ1 = τ2 = τ = 4.8

1
G (s) ≈
(4.8s + 1) 2

Nonlinear regression

From Figure E7.5, we can read these values (approximated):

Time Output
0.0 0.0
2.0 0.1
4.0 0.2
5.0 0.3
7.0 0.4
8.0 0.5
9.0 0.6
11.0 0.7
14.0 0.8
17.5 0.9
30.0 1.0

Table.- Output values from Figure E7.5

In accounting for Eq. 5-48, the time constants were selected to minimize
the sum of the squares of the errors between data and model predictions.
Use Excel Solver for this Optimization problem:

τ1 =6.76 and τ2 = 6.95

1
G (s) ≈
(6.95s + 1)(6.76s + 1))

The models are compared in the following graph:

7-8
 1

0.9

0.8

0.7

0.6

Output 0.5

0.4

0.3

0.2

0.1 Non linear regression model

First-order plus time delay model
Second order model (Smith's method)
0
0 5 10 15 20 25 30 35 40 45 50
time

Figure S7.6. Comparison of three models for unit step input

7.7

a) From the graph, time delay θ = 4.0 min

Using Smith’s method,

from the graph, t20 + θ ≈ 5.6 , t60 + θ ≈ 9.1

t20 = 1.6 , t60 = 5.1 , t20 / t60 = 1.6 / 5.1 = 0.314

From Fig.7.7 , ζ = 1.63 , t60 / τ = 3.10 , τ = 1.645

Using Eqs. 5-45, 5-46, τ1 = 4.81 , τ2 = 0.56

b) Overall transfer function

10e −4 s
G (s) = , τ1 > τ2
(τ1s + 1)(τ2 s + 1)

Assuming plug-flow in the pipe with constant-velocity,

7-9
 −θ p s 3 1
G pipe ( s ) = e , θp = × = 0.1min
0.5 60

Assuming that the thermocouple has unit gain and no time delay

1
GTC ( s ) = since τ2 << τ1
(τ2 s + 1)

Then

10e −3 s
GHE ( s ) = , so that
(τ1s + 1)
 10e −3 s  −0.1s  1 
G ( s ) = GHE ( s )G pipe ( s )GTC ( s ) =   (e )  
 τ1s + 1   τ2 s + 1 

7.8

a) To find the form of the process response, we can see that

K K M K M
Y (s) = U ( s) = =
s (τs + 1) s (τs + 1) s (τs + 1) s 2

Hence the response of this system is similar to a first-order system with a

ramp input: the ramp input yields a ramp output that will ultimately cause
some process component to saturate.

b) By applying partial fraction expansion technique, the domain response for

this system is

A B C
Y(s) = + 2+ hence y(t) = -KMτ + KMt − KMτe-t/τ
s s τs + 1
In order to evaluate the parameters K and τ, important properties of the
above expression are noted:

1.- For large values of time (t>>τ) , y(t) ≈ y′(t ) = KM (t-τ)

2.- For t = 0, y′(0) = −KMτ

These equations imply that after an initial transient period, the ramp input
yields a ramp output with slope equal to KM. That way, the gain K is

7-10
 obtained. Moreover, the time constant τ is obtained from the intercept in
Fig. S7.8

y(t)

Slope = KM

−ΚΜτ

Figure S7.8. Time domain response and parameter evaluation

7.9

For underdamped responses,

   1 − ζ2  ζ  1 − ζ2  
y (t ) = KM 1 − e − ζt / τ cos t + sin  t   (5-51)
   τ 
 1 − ζ2  τ


 

a) At the response peaks,

ζ   1 − ζ2  ζ  1− ζ2 
dy 
= KM  e −ζ t / τ cos  t+ sin  t 
dt  τ   τ 
 1 − ζ2  τ




 1− ζ2  1− ζ2  ζ  1 − ζ2   
−e −ζt / τ  − sin  t  + cos  t   = 0
 τ  τ  τ  τ  
    

Since KM ≠ 0 and e − ζt / τ ≠ 0

 ζ ζ   1− ζ   ζ2 1 − ζ2   1 − ζ2 
2
0 =  −  cos t +  +  sin  t
 τ τ   τ   τ 1 − ζ2
  τ   τ
 



7-11
  1− ζ2  πτ
0 = sin t  = sin nπ , t= n
 τ  1 − ζ2
 
where n is the number of peak.

πτ
Time to the first peak, tp =
1 − ζ2

b) Graphical approach:

Process gain,

wD (∞) − wD (0) 9890 − 9650 lb

K= = = 80
∆Ps 95 − 92 hr
psig

a 9970 − 9890
Overshoot = = = 0.333
b 9890 − 9650

From Fig. 5.11, ζ ≈ 0.33

tp can be calculated by interpolating Fig. 5.8

For ζ ≈ 0.33 , tp ≈ 3.25 τ

Since tp is known to be 1.75 hr , τ = 0.54

K 80
G (s) = =
τ s + 2ζτs + 1 0.29s + 0.36s + 1
2 2 2

Analytical approach

lb
The gain K doesn’t change: K = 80
hr
psig
To obtain the ζ and τ values, Eqs. 5-52 and 5-53 are used:

a 9970 − 9890
Overshoot = = = 0.333 = exp(-ζπ/(1-ζ2)1/2)
b 9890 − 9650

Resolving, ζ = 0.33

7-12
 πτ
tp = = 1.754 hence τ = 0.527 hr
1− ζ2
K 80
G (s) = =
τ s + 2ζτs + 1 0.278s + 0.35s + 1
2 2 2

c) Graphical approach

From Fig. 5.8, ts/τ = 13 so ts = 2 hr (very crude estimation)

Analytical approach

From settling time definition,

y = ± 5% KM so 9395.5 < y < 10384.5

(KM ± 5% KM) = KM[ 1-e(-0.633)[cos(1.793ts)+0.353sin(1.793ts)]]

1 ± 0.05 = 1 – e(0.633 ts) cos(1.793 ts) + 0.353e (-0.633 ts) sin(1.7973 ts)

Solve by trial and error…………………… ts ≈ 6.9 hrs

7.10

T '( s ) K
a) = 2 2
W '( s ) τ s + 2ζτ + 1

T (∞) − T (0) 156 − 140 o

C
K= = = 0.2
∆w 80 Kg/min

From Eqs. 5-53 and 5-55,

a 161.5 − 156
Overshoot = = = 0.344 = exp(-ζπ (1-ζ2)1/2
b 156 − 140
By either solving the previous equation or from Figure 5.11, ζ= 0.322
(dimensionless)

7-13
 There are two alternatives to find the time constant τ :

1.- From the time of the first peak, tp ≈ 33 min.

One could find an expression for tp by differentiating Eq. 5-51 and

solving for t at the first zero. However, a method that should work
(within required engineering accuracy) is to interpolate a value of
ζ=0.35 in Figure 5.8 and note that tp/τ ≈ 3

33
Hence τ ≈ ≈ 9.5 − 10 min
3 .5

2.- From the plot of the output,

2πτ
Period = P = = 67 min and hence τ =10 min
1− ζ2

Therefore the transfer function is

T ' (s) 0.2

G (s) = =
W ' ( s ) 100 s + 6.44 s + 1
2

b) After an initial period of oscillation, the ramp input yields a ramp output
with slope equal to KB. The MATLAB simulation is shown below:

160

158

156

154

152
Output

150

148

146

144

142

140
0 10 20 30 40 50 60 70 80 90 100
time

Figure S7.10. Process output for a ramp input

7-14
 We know the response will come from product of G(s) and Xramp = B/s2
KB
Then Y ( s ) = 2 2 2
s (τ s + 2ζτs + 1)
From the ramp response of a first-order system we know that the response
will asymptotically approach a straight line with slope = KB. Need to find
the intercept. By using partial fraction expansion:

KB α α α s + α4
Y (s) = = 1 + 22 + 2 2 3
s (τ s + 2ζτs + 1) s s
2 2 2
τ s + 2ζτs + 1

Again by analogy to the first-order system, we need to find only α1 and

α2. Multiply both sides by s2 and let s→ 0, α2 = KB (as expected)

Can’t use Heaviside for α1, so equate coefficients

KB = α1s (τ2 s 2 + 2ζτs + 1) + α 2 (τ2 s 2 + 2ζτs + 1) + α 3 s 3 + α 4 s 2

We can get an expression for α1 in terms of α2 by looking at terms

containing s.

s: 0 = α1+α22ζτ → α1 = -KB2ζτ

and we see that the intercept with the time axis is at t = 2ζτ. Finally,
presuming that there must be some oscillatory behavior in the response,
we sketch the probable response (See Fig. S7.10)

7.11

a) Replacing τ by 5, and K by 6 in Eq. 7-34

y (k ) = e−∆t / 5 y (k − 1) + [1 − e−∆t / 5 ]6u (k − 1)

b) Replacing τ by 5, and K by 6 in Eq. 7-32

∆t ∆t
y (k ) = (1 − ) y (k − 1) + 6u (k − 1)
5 5

In the integrated results tabulated below, the values for ∆t = 0.1 are shown
only at integer values of t, for comparison.

7-15
 y(k) y(k) y(k)
t
(exact) (1t=1) (1t=0.1)
0 3 3 3
1 2.456 2.400 2.451
2 5.274 5.520 5.296
3 6.493 6.816 6.522
4 6.404 6.653 6.427
5 5.243 5.322 5.251
6 4.293 4.258 4.290
7 3.514 3.408 3.505
8 2.877 2.725 2.864
9 2.356 2.180 2.340
10 1.929 1.744 1.912

Table S7.11. Integrated results for the first order differential equation

Thus ∆t = 0.1 does improve the finite difference model bringing it closer
to the exact model.

7.12

To find a1′ and b1 , use the given first order model to minimize

10
J = ∑ ( y (k ) −a1′ y (k − 1) − b1 x(k − 1)) 2
n =1

∂J 10
= ∑ 2( y (k ) −a1′ y (k − 1) − b1 x(k − 1))(− y (k − 1) = 0
∂a1′ n =1
∂J 10
= ∑ 2( y (k ) −a1′ y (k − 1) − b1 x(k − 1))(− x(k − 1)) = 0
∂b1 n =1
Solving simultaneously for a1′ and b1 gives

10 10

∑ y (k )y (k − 1) − b1 ∑ y (k − 1)x(k − 1)
a1′ = n =1
10
n =1

∑ y (k − 1)
n =1
2

10 10 10 10

∑ x(k − 1) y(k )∑ y(k − 1)2 − ∑ y (k − 1)x(k − 1)∑ y(k − 1) y(k )

b1 = n =1 n =1 n =1 n =1
2
10 10
 10

∑
n =1
x(k − 1) ∑ y (k − 1) −  ∑ y (k − 1)x(k − 1) 
n =1
2

 n =1
2

7-16
Using the given data,

10 10

∑ x(k − 1) y(k ) = 35.212 ,

n =1
∑ y(k − 1) y(k ) = 188.749
n =1

10 10

∑ x(k − 1) 2 = 14
n =1
, ∑ y(k − 1)
n =1
2
= 198.112

10

∑ y (k − 1) x(k − 1) = 24.409
n =1

Substituting into expressions for a1′ and b1 gives

a1′ = 0.8187 , b1 = 1.0876

Fitted model is y (k + 1) = 0.8187 y (k ) + 1.0876 x(k )

or y (k ) = 0.8187 y (k − 1) + 1.0876 x(k − 1) (1)

Let the first-order continuous transfer function be

Y ( s) K
=
X ( s ) τs + 1

From Eq. 7-34, the discrete model should be

y (k ) = e −∆t / τ y (k − 1) + [1 − e −∆t / τ ]Kx(k − 1) (2)

Comparing Eqs. 1 and 2, for ∆t=1, gives

τ=5 and K = 6

Hence the continuous transfer function is 6/(5s+1)

7-17
 8
actual data
fitted model

5
y(t)

0 1 2 3 4 5 6 7 8 9 10
time,t

Figure S7.12. Response of the fitted model and the actual data

7.13

To fit a first-order discrete model

y (k ) = a1′ y (k − 1) + b1 x(k − 1)

Using the expressions for a1′ and b1 from the solutions to Exercise 7.12,
with the data in Table E7.12 gives

a1′ = 0.918 , b1 = 0.133

Using the graphical (tangent) method of Fig.7.5 .

K = 1 , θ = 0.68 , and τ = 6.8

The response to unit step change for the first-order model given by

e −0.68 s
is y (t ) = 1 − e −( t −0.68) / 6.8
6.8s + 1

7-18
 1
0,9
0,8
0,7
y(t) 0,6
0,5
0,4 actual data
0,3 fitted model
0,2 graphical method

0,1
0
0 2 4 time,t 6 8 10

Figure S7.13- Response of the fitted model, actual data and graphical method

7-19