Chapter 2 Degrees of freedom for the simplified model: 2.

Parameters : ρ, V
a) Mass Balances:
Variables : w1, w2, T1, T2, T3
NE = 1 dh1
2.1 ρA1 = w1 − w2 − w3 (1)
NV = 5 dt

Thus, NF = 5 – 1 = 4 dh
a) Overall mass balance: ρA2 2 = w2 (2)
dt
Because w1, w2, T1 and T2 are determined by upstream units, we assume
d (ρV )
= w1 + w2 − w3 (1) they are known functions of time: Flow relations:
dt
w1 = w1(t) Let P1 be the pressure at the bottom of tank 1.
Energy balance: w2 = w2 (t)
T1 = T1(t) Let P2 be the pressure at the bottom of tank 2.
d ρV (T3 − Tref )  T2 = T2(t)
C = w1C (T1 − Tref ) + w2C (T2 − Tref )
dt (2) Let Pa be the ambient pressure.
− w3C (T3 − Tref )
Thus, NF is reduced to 0.
P1 − P2 ρg
Then w2 = = (h1 − h2 ) (3)
Because ρ = constant and V = V = constant, Eq. 1 becomes: R2 g c R2

w3 = w1 + w2 (3) 2.2
P1 − Pa ρg
w3 = = h1 (4)
R3 g c R3
b) From Eq. 2, substituting Eq. 3 Energy balance:
d (T3 − Tref ) b) Seven parameters: ρ, A1, A2, g, gc, R2, R3
dT d ρV (T − Tref ) 
ρCV = ρCV 3= w1C (T1 − Tref ) + w2C (T2 − Tref )
dt dt (4) Cp = wC p (Ti − Tref ) − wC p (T − Tref ) − UAs (T − Ta ) + Q Five variables : h1, h2, w1, w2, w3
dt
− ( w1 + w2 ) C (T3 − Tref )
Simplifying Four equations
Constants C and Tref can be cancelled:
dT Thus NF = 5 – 4 = 1
ρVC p = wC p Ti − wC p T − UAs (T − Ta ) + Q
dT3 dt
ρV = w1T1 + w2T2 − ( w1 + w2 )T3 (5) 1 input = w1 (specified function of time)
dt dT
ρVC p = wC p (Ti − T ) − UAs (T − Ta ) + Q 4 outputs = h1, h2, w2, w3
dt
The simplified model now consists only of Eq. 5.

b) T increases if Ti increases and vice versa.

T decreases if w increases and vice versa if (Ti – T) < 0. In other words, if

Q > UAs(T-Ta), the contents are heated, and T >Ti.

Solution Manual for Process Dynamics and Control, 4th edition

Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III

2-1 2-2 2-3

1/ γ
2.4 2.5  V2 γ 
1 dP2
  P2
(1− γ ) / γ
= wb − wc
γ  C  dt
 
Assume constant liquid density, ρ . The mass balance for the tank is a) For linear valve flow characteristics,
as the new dynamic model. If the ideal gas law were not valid, one would
d (ρAh + m g ) Pd − P1 P − P2 P2 − Pf use an appropriate equation of state instead of Eq. 3.
= ρ(qi − q ) wa = , wb = 1 , wc = (1)
dt Ra Rb Rc
Mass balances for the surge tanks
Because ρ, A, and mg are constant, this equation becomes
dm1 dm2
dh = wa − wb , = wb − wc (2)
A = qi − q (1) dt dt
dt 2.6
where m1 and m2 are the masses of gas in surge tanks 1 and 2,
The square-root relationship for flow through the control valve is respectively.
a) Assumptions:
1/ 2
 ρgh  If the ideal gas law holds, then
q = C v  Pg + − Pa  (2) 1. Each compartment is perfectly mixed.
 gc 
P1V1 =
m1
RT1 , P2V2 =
m2
RT2 (3) 2. ρ and C are constant.
M M 3. No heat losses to ambient.
From the ideal gas law,
where M is the molecular weight of the gas Compartment 1:
(m g / M ) RT T1 and T2 are the temperatures in the surge tanks.
Pg = (3) Overall balance (No accumulation of mass):
A( H − h)
Substituting for m1 and m2 from Eq. 3 into Eq. 2, and noticing that V1, T1, 0 = ρq − ρq1 thus q1 = q (1)
where T is the absolute temperature of the gas. V2, and T2 are constant,
Energy balance (No change in volume):
Equation 1 gives the unsteady-state model upon substitution of q from Eq. 2 and V1M dP1 V2 M dP2
= wa − wb and = wb − wc (4)
of Pg from Eq. 3: RT1 dt RT2 dt dT1
V1ρC = ρqC (Ti − T1 ) − UA(T1 − T2 ) (2)
dt
1/ 2
dh  (mg / M ) RT ρ gh  The dynamic model consists of Eqs. 1 and 4.
A = qi − Cv  + − Pa  (4) Compartment 2:
dt  A( H − h) gc  b) For adiabatic operation, Eq. 3 is replaced by
Overall balance:
Because the model contains Pa, operation of the system is not independent of Pa. γ γ
V  V 
For an open system Pg = Pa and Eq. 2 shows that the system is independent of Pa. P1  1  = P2  2  = C , a constant (5) 0 = ρq1 − ρq2 thus q2 = q1= q (3)
 m1   m2 
1/ γ 1/ γ
 PV γ   PV γ  Energy balance:
or m1 =  1 1  and m2 =  2 2  (6)
 C   C  dT2
V2ρC = ρqC (T1 − T2 ) + UA(T1 − T2 ) − U c Ac (T2 − Tc ) (4)
dt
Substituting Eq. 6 into Eq. 2 gives,

1/ γ
b) Eight parameters: ρ, V1, V2, C, U, A, Uc, Ac
1  V1 γ  dP1 Five variables: Ti, T1, T2, q, Tc
  P1
(1− γ ) / γ
= wa − wb
γ  C  dt
  Two equations: (2) and (4)

2-4 2-5 2-6

 2 manipulated variables: Q, wi
Thus NF = 5 – 2 = 3 1 disturbance variable: Ti Eqs.1 and 2 comprise the dynamic model for the system.

2 outputs = T1, T2
(b) In this part, two controllers have been added to the system. Each controller
3 inputs = Ti, Tc, q (specify as functions of t) provides an additional equation. Also, the flow out of the tank is now a
manipulated variable being adjusted by the controller. So, we have 2.9
c) Three new variables: ci, c1, c2 (concentration of species A). 4 parameters: C , ρ , Tsp , Vsp
Two new equations: Component material balances on each compartment.
6 variables: V , T , wi , Ti , Q, w Assume that the feed contains only A and B, and no C. Component balances for
c1 and c2 are new outputs. ci must be a known function of time.
4 equations A, B, C over the reactor give.

The degrees of freedom are 6 − 4 =2 . To specify the two degrees of freedom, we V

dc A
= qi c Ai − qc A − Vk1e − E1 / RT c A (1)
2.7 set the variables as follows: dt

2 output variables: T , V dcB

V = qi cBi − qcB + V (k1e − E1 / RT c A − k2e − E2 / RT cB ) (2)
As in Section 2.4.2, there are two equations for this system: 2 manipulated variables (determined by controller equations): Q, w dt
2 disturbance variables: Ti , wi
dV 1 dc
= ( wi − w) V C =−qcC + Vk2e − E2 / RT cB (3)
dt ρ dt
dT wi Q 2.8
= (Ti − T ) + An overall mass balance over the jacket indicates that qc = qci because the volume
dt V ρ ρVC of coolant in jacket and the density of coolant are constant.
Additional assumptions:
Results: Energy balance for the reactor:
(i) Density of the liquid, ρ, and density of the coolant, ρJ, are constant.
d (Vc A M A S A + VcB M B S B + VcC M C SC ) T 
(a) Since w is determined by hydrostatic forces, we can substitute for this
(ii) Specific heat of the liquid, C, and of the coolant, CJ, are constant.
variable in terms of the tank volume as in Section 2.4.5 case 3. = ( qi c Ai M A S A + qi cBi M B S B ) (Ti − T )
dt
Because V is constant, the mass balance for the tank is:
dV 1  V  −UA(T − Tc ) + (−∆H1 )Vk1e− E1 / RT c A + (−∆H 2 )Vk2e− E2 / RT cB (4)
=  wi − Cv 
dt ρ  A  dV
ρ = q F − q = 0 ; thus q = qF where MA, MB, MC are molecular weights of A, B, and C, respectively
dT wi Q dt
= (i )
T − T + SA, SB, SC are specific heats of A, B, and C.
dt ρV ρVC U is the overall heat transfer coefficient
Energy balance for tank:
A is the surface area of heat transfer
This leaves us with the following:
dT
ρVC = q F ρC (TF − T ) − Kq J A(T − TJ )
0.8
(1) Energy balance for the jacket:
5 variables: V , T , wi , Ti , Q dt
4 parameters: C , ρ , Cv , A dTc
Energy balance for the jacket: ρ j S jV=
j ρ j S j qci (Tci − Tc ) + UA(T − Tc ) (5)
2 equations dt
where:
dT
The degrees of freedom are 5 − 2 =3 . To make sure the system is specified, we ρ J VJ C J J = q J ρ J C J (Ti − TJ ) + Kq J
0.8
A(T − TJ ) (2) ρj, Sj are density and specific heat of the coolant.
dt Vj is the volume of coolant in the jacket.
have:
where A is the heat transfer area (in ft2) between the process liquid and the Eqs. 1 - 5 represent the dynamic model for the system.
2 output variables: T , V
coolant.

2-7 2-8 2-9

2.10 2.11

The plots should look as shown below: a)

Note that the only conservation equation required to find h is an overall Rearrange Eq. 5 to get the feedforward (FF) controller relation,
mass balance:

dm d (ρAh) dh w2 = C v hR − w1 where hR = 2.25 m

= =
ρA =+w1 w2 − w (1)
dt dt dt
w2 = (2.13)(1.5) − w1 = 3.2 − w1 (6)

Note that Eq. 6, for a value of w1 = 2.0, gives

ρg
Valve equation: w = C v′ h = Cv h (2)
gc w2 = 3.2 –1.2 = 2.0 kg/s which is the desired value.
ρg
where C v = C v′ (3) If the actual FF controller follows the relation, w2 = 3.2 − 1.1w1 (flow
gc
transmitter 10% higher), w2 will change as soon as the FF controller is
Substituting the valve equation into the mass balance, turned on,

dh 1 w2 = 3.2 –1.1 (2.0) = 3.2 – 2.2 = 1.0 kg/s

= ( w1 + w2 − C v h ) (4)
dt ρA (instead of the correct value, 1.2 kg/s)
Steady-state model:
Then C v h = 2.13 h = 2.0 + 1.0
0 = w1 + w2 − C v h (5)
3
or h= = 1.408 and h = 1.983 m (instead of 2.25 m)
2.13
w1 + w2 2.0 + 1.2 3.2 kg/s
b) =
Cv = = = 2.13 1/2
h 2.25 1.5 m
2.25 − 1.983
Notice that the functions are only good for t = 0 to t = 18, at which point the tank Error in desired level = × 100% =
11.9%
is completely drained. The concentration function blows up because the volume 2.25
function is negative.
c) Feedforward control

2-10 2-11 2-12

 The sensitivity does not look too bad in the sense that a 10% error in flow 2.13 1
measurement gives ~12% error in desired level. Before making this = [(72 + 50) − 220 x3 )]
(800)(1.75)π
conclusion, however, one should check how well the operating FF
controller works for a change in w1 (e.g., ∆w1 = 0.4 kg/s). Model of process
= 0.027738 − 0.050020x3
Overall material balance:
Integrating,
dh x3 f t
2.12 ρAT = w1 + w2 − w3 = w1 + w2 − C v h (1) dx3
dt ∫ 0.027738 − 0.050020 x = ∫ dt
x3 o 3 0

a) Model of tank (normal operation): Component:

where x3o=0.5 and x3f =0.555 – (0.555)(0.01) = 0.549
dh d (hx3 )
ρA = w1 + w2 − w3 (Below the leak point) ρAT = w1 x1 + w2 x 2 − w3 x3 Solving,
dt dt
π (2) 2 t = 47.42 min
A= = π= 3.14 m 2 dx3 dh
4 ρAT h + ρAT x3 = w1 x1 + w2 x 2 − w3 x3
dt dt c) If w1 is changed to 100 kg/min without changing any other input variables,
dh then x3 will not change and Eq. 1 can be solved to find the time to achieve
(800)(3.14) = 120 + 100 − 200 = 20 Substituting for dh/dt (Eq. 1) 99% of the h response. From the material balance, the final value of the
dt
tank level is h =1.446 m.
dh 20 dx3
= = 0.007962 m/min ρAT h + x3 ( w1 + w2 − w3 ) = w1 x1 + w2 x 2 − w3 x3 dh
dt (800)(3.14) dt 800π = 100 + 100 − Cv h
dt
Time to reach leak point (h = 1 m) = 125.6 min. dx3
ρAT h = w1 ( x1 − x3 ) + w2 ( x 2 − x3 ) (2) dh 1 
dt = 200 − 166.3 h 
dt 800π  
Model of tank with leak and w1 , w2 , w3 constant:
b)
or
dx3
=
1
[w1 ( x1 − x3 ) + w2 ( x2 − x3 )] (3) = 0.079577 − 0.066169 h
dt ρAT h
dh
ρA =20 − d q4 =20 − ρ(0.025) h − 1 = 20 − 20 h − 1 , h ≥ 1 where ho=1.75 and hf =1.446 + (1.446)(0.01) = 1.460
dt a) At initial steady state ,
By using the MATLAB command ode45 ,
To check for overflow, one can simply find the level hm at which dh/dt = w3 = w1 + w2 = 120 + 100 = 220 Kg/min
0. That is the maximum value of level when no overflow occurs.
220 t = 122.79 min
Cv = = 166.3
0 = 20 − 20 hm − 1 or hm = 2 m 1.75
Numerical solution of the ode is shown in Fig. S2.13
b) If x1 is suddenly changed from 0.5 to 0.6 without changing flowrates, then
Thus, overflow does not occur for a leak occurring because hm < 2.25 m. level remains constant and Eq.3 can be solved analytically or numerically
to find the time to achieve 99% of the x3 response. From the material
balance, the final value of x3 = 0.555. Then,

dx3 1
= [120(0.6 − x3 ) + 100(0.5 − x3 )]
dt (800)(1.75)π

2-13 2-14 2-15

dX Substituting these two equations into Eq. 10,

1.8 =0 ∴ rg = DX
dt
then,  DK S 
=
DX YX / S  S f − D (11)
1.7 µ X = DX ∴ D= µ (4)  mmax − D 
h(m) A simple feedback strategy can be implemented where the growth rate For Yx/s = 0.5, Sf = 10, Ks = 1, X = 2.75, μmax = 0.2, the following plot can be
1.6 is controlled by manipulating the mass flow rate, F, so that F/V stays generated based on Eq. 11.
constant.

1.5 c) Washout occurs if dX/dt is negative for an extended period of time;

that is,

1.4 rg − DX < 0 or D>µ

0 50 100 150 200 250 300
time (min)
Figure S2.13. Numerical solution of the ode for part c) Thus, if D > µ the cells will be washed out.

d) In this case, both h and x3 will be changing functions of time. Therefore, d) At steady state, the dynamic model given by Eqs. 1, 2 and 3 becomes:
both Eqs. 1 and 3 will have to be solved simultaneously. Since
concentration does not appear in Eq. 1, we would anticipate no effect on 0 = rg - DX DX = rg (5)
the h response.
0 = rp - DP DP = rp (6)
1
0 = 𝐷𝐷�𝑆𝑆𝑓𝑓 − 𝑆𝑆� − 𝑌𝑌 𝑟𝑟𝑔𝑔 (7)
2.14 𝑋𝑋/𝑆𝑆

a) The dynamic model for the chemostat is given by:

From Eq. 5,
F Figure S2.14. Steady-state cell production rate DX as a function of dilution rate D.
Cells: V
dX
= Vrg − FX or
dX
= rg −   X (1) DX = rg (8)
dt dt V  From Eq. 7 From Figure S2.14, washout occurs at D = 0.18 h-1 while the maximum
production occurs at D = 0.14 h-1. Notice that maximum and washout points
dP dP F are dangerously close to each other, so special care must be taken when
Product: V = Vrp − FP or = rp −   P (2) rg = Y X / S ( S f − S ) D (9)
dt dt V  increasing cell productivity by increasing the dilution rate.

dS 1
Substrate: V = F (S f − S ) − Vrg Substituting Eq. 9 into Eq. 8,
dt YX / S

or DX = Y X / S ( S f − S ) D (10)
dS  F  1 1
=  ( S f − S ) − rg − rP (3)
dt  V  YX / S YP / S
From Eq. 4
b) At steady state, DK S
S=
mmax − D

2-16 2-17 2-18

2.15 2.16 2.17
For V = constant and F = 0, the simplified dynamic model is:
a) We can assume that ρ and h are approximately constant. The dynamic
dX S
model is given by: = rg = µ µax X (a) Using a simple volume balance, for the system when the drain is closed
dt Ks + S (q = 0)
dM
rd = − = kAc s (1) dP S 𝑑𝑑ℎ
dt = rp = YP / X µ µax X 𝐴𝐴 = 𝑞𝑞1 (1)
Notice that: dt Ks + S 𝑑𝑑𝑑𝑑
dM dV
M = ρV ∴ =ρ (2) dS 1 1 Solving this ODE with the given initial condition gives a height that is increasing
dt dt =− rg − rP at a rate of 0.25 ft/min.
dt YX / S YP / X
dV dr dr So the height in this time range will look like:
V = πr h 2
∴ = (2πrh) =A (3) Substituting numerical values:
dt dt dt
Substituting (3) into (2) and then into (1), dX SX
= 0.2
dt 1+ S
dr dr
− rA = kAc s ∴ −r = kc s
dt dt dP SX
= (0.2)(0.2)
Integrating, dt 1+ S

r kcs t kc s dS SX  1 0.2 
∫ r dr =
o
−
ρ ∫0
dt ∴ r (t ) = ro −
r
t (4)
dt
= 0.2 − −
1 + S  0.5 0.1 

By using MATLAB, this system of differential equations can be solved. The time
Finally,
to achieve a 90% conversion of S is t = 22.15 h.
M = rV = rπhr 2

then (b) the drain is opened for 15 mins; assume a time constant in a linear
2
 kc  transfer function of 3 mins, so a steady state is essentially reached. (3 < t < 18).
M (t ) = ρπh ro − s t  Assume that the process will return to its previous steady state in an exponential
 ρ 
manner, reaching 63.2% of the response in three minutes.
b) The time required for the pill radius r to be reduced by 90% is given by
Eq. 4:

kc s 0.9ρo ρ (0.9)(0.4)(1.2)
0.1ro = ro − t ∴ t= = = 54 min
r kc s (0.016)(0.5)

Therefore, t = 54 min .

Figure S2.16. Fed-batch bioreactor dynamic behavior.

2-19 2-20 2-21

2.18
The graph should show an exponential decrease to the previous steady state of 0.5
ft. The initial value should coincide with the final value from part (c).

Parameters (fixed by design process): m, C, me, Ce, he, Ae.

CVs: T and Te.

Input variables (disturbance): w, Ti. Input variables (manipulated): Q.

(c) the inflow rate is doubled for 6 minutes (18 < t < 24) Degrees of freedom = (11-6) (number of variables) – 2 (number of equations) = 3

The height should rise exponentially towards a new steady state value double that
of the steady state value in part b), but it should be apparent that the height does
not reach this new steady state value at t = 24 min.. The new steady state would The three input variables (w, Ti, Q) are assigned and the resulting system has zero
be 1 ft. degrees of freedom.

Putting all the graphs together would look like this:

(d) the inflow rate is returned to its original value for 16 minutes (24 < t < 40)

2-22 2-23 2-24

 2.20
2.19

(a) First we simulate a step change in the Fuel Gas Purity (FG_pur) from 1 to
(a) First we simulate a step change in the vapor flow rate from 0.033 to 0.045 0.95.
m3/s. The resulting plots of Oxygen Exit Concentration (C_O2) and Hydrocarbon
The resulting plots of xD and xB are shown below. Outlet Temperature (T_HC) are shown below.

Figure: Plot of xD, xB, and zF versus time for a step change in zF from 0.5 to 0.55

By examining the resulting data, we can find the steady-state values of xD and xB
before and after the step change in zF.

Start End Change

Figure: Plot of xD, xB, and V versus time for a step change in V from 0.033 to 0.045 Figure: Plot of C_O2, T_HC, and FG_pur versus time for a step change in FG_pur from 1
m3/s. xD 0.73 0.80 +0.066 to 0.95.

By examining the resulting data, we can find the steady-state values of xD and xB xB 0.0050 0.0068 +0.0018 By examining the resulting data, we can find the steady-state values of C_O2 and
before and after the step change in V. T_HC before and after the step change in FG_pur.
(c) Increasing V causes xD and xB to decrease, while increasing zF causes both
Start End Change xD and xB to increase. The magnitude of the effect is greater for changing V Start End Change
than for changing zF. When changing V, xB changes more quickly than xD.
xD 0.85 0.73 -0.12 C_O2 0.92 1.06 0.14
xB 0.15 0.0050 -0.145 T_HC 609 595 -14

(b) Next we simulate a step change in the feed composition (zF) from 0.5 to 0.55. (b) Next we simulate a step change in the Hydrocarbon Flow Rate (F_HC_sp)
Note that the vapor flow rate, V, is still set at 0.045 m3/s. from 0.035 to 0.0385. Note that the Fuel Gas Purity, FG_pur, is still set at
0.95.

2-25 2-26 2-27

( 4 )  =
1
2.21 τ AR
= = 2 min
2

Part a
The key to this problem is solving the mass balance of the tank in each part.
dh
Mass balance: A = qi − qo (Mass Balance)
dt
d
( ρ Ah
= ) ρ qi − ρ qo dh
dt 4 =2 (Separable ODE)
dt
- ρ (density) and A (tank cross-sectional area) are constants, therefore:
1
dh
∫ dh = ∫ 2 dt
A = qi − qo
dt
1
h(t ) = t +C h(0) =
1
- The problem specifies qo is linearly related to the tank height 2

1
qo =
1
h h(t ) = t +1 0≤t <3
R 2

dh 1
A = qi − h
Figure: Plot of C_O2, T_HC, and F_HC_sp versus time for a step change in F_HC_sp dt R
from 0.035 to 0.0385.

By examining the resulting data, we can find the steady-state values of C_O2 and - Next, we can obtain R (valve constant) from the steady state information in the
T_HC before and after the step change in F_HC_sp. problem

dh
Start End Change = 0 at steady state
dt
C_O2 1.06 1.06 0
1
T_HC 595 572 -23 0= qi − h
R

(c) Decreasing FG_pur causes C_O2 to increase, while T_HC decreases. 1

0= 2 − (1)
Increasing F_HC_sp causes T_HC to decrease while C_O2 stays the same. R
The change in T_HC occurs more quickly when changing F_HC_sp versus
changing FG_pur. 1 ft 2
∴ = 2 R= 0.5
R min

- In addition, we can find that

2-28 2-29 2-30

Part b Part d 2.22

dh 1 Same as part b with h (33) = 2

A = qi − h (Mass Balance) To solve the problem, we start by writing the mass balance for each tank 1-4.
dt R
h(t ) = 1 + e − (t −33) / 2 33 ≤ t ≤ 50 To write the mass balance for each tank, we start with the most general form,
dh where the change in mass in the tank over time is equal to the mass flowing into
4 = 2 − 2h
dt the tank minus the mass flowing out of the tank. The general form of the
equations are shown below, where i represents the tank number (1, 2, 3, 4). The
dh 1 1
+ h= (Solution by integrating factor = et / 2 ) mass can be written as the density multiplied by the tank volume, and the mass
dt 2 2
flow rates can be written as the density multiplied by the volumetric flow rate.
1 t/2
∫ d (e h) = ∫
t/ 2
e dt d ( ρVi )
2 = ρ qin ,i − ρ qout ,i
dt
he t/2
=
1e + c t/2
h(3) =
2.5
With density assumed constant over time, it can be pulled out of the derivative.
h = 1 + ce − t / 2 Also, we write the volume of the tank as the height of liquid in the tank, hi,
multiplied by the cross-sectional tank area, Ai.
2.5 = 1 + ce −3/ 2
ρ Ai d (hi )
c = 1.5e3/ 2 = ρ qin ,i − ρ qout ,i
dt
Ai d (hi )
h(t ) = 1 + (1.5)e − ( t −3) / 2
3 ≤ t < 18 = qin ,i − qout ,i
dt
Part c
The flow exiting each tank through the bottom can be written as:
dh
4 = 4 − 2h (Mass balance)
dt qexit ,i = Ci hi

dh 1
+ h=
1 (Solution by integrating factor) Where Ci is the proportionality constant for each tank.
dt 2

∫ d (e h )
= ∫ 1e dt h(18) 1
=
t/2 t/2

Results:
- Method is same as part b. a) The final equations for the height of liquid in each tank are shown below.

h(t ) = 2 − e − (t −18) / 2 18 ≤ t < 33

2-31 2-32 2-33

γ
dh1
dt
=
C C
− 1 h1 + 3 h3 + 1 F1
A1 A1 A1
(1) Chapter 3 © 3.2

dh2 C C γ
=
− 2 h2 + 4 h4 + 2 F2 (2) Break the pulse into three step functions. First, a step up to 10 at t=0. Then, a step
dt A2 A2 A2 down by 8 at t=1. Finally, a step down by 2 at t=3:
dh3 C (1 − γ 2 ) 3.1
=
− 3 h3 + F2 (3) f(t) = 10 S(t) – 8 S(t-1) – 2 S(t- 3)
dt A3 A3
dh4 C ( 1 − γ1 ) (a)
dt
=
− 4 h4 +
A4 A4
F1 (4)
f (t )  5  e 3t
 te 4t F (s )  =
1
s
 10 - 8e-s - 2 e-3s 
Transform each term using rules 2, 5, and 7 from Table 3.1, respectively.
b) Now we can substitute γ= γ= 0.5 5 1 1
1 2
F ( s)   
s s  3 ( s  4)2 3.3
dh1 C C 0.5
=
− 1 h1 + 3 h3 + F1
dt A1 A1 A1 (b)
5 a) Pulse width is obtained when x(t) = 0. Since x(t) = h – at
dh2 C C 0.5 f (t )  sin(4t )  (t  3) S (t  3)  e (t 3) S (t  3) 
=
− 2 h2 + 4 h4 + F2 t
dt A2 A2 A2 t : h  at = 0 or t = h/a
 To transform sin(4t), use rule 14 from Table 3.1 b)
dh3 C 0.5  To transform (t-3)S(t-3) use rules 3 and 26 together. To use rule 26,
=
− 3 h3 + F2
dt A3 A3 set f(t) = t and t0 =3.
dh4 C 0.5  To transform e-(t-3)S(t-3) use rules 5 and 26 together. To use rule h

=
− 4 h4 + F1 26, set f(t) = e-t and t0 =3. slope = -a
x(t)
slope = a
dt A4 A4 x(t)
 Note that there is no Laplace transform for 1/t.
The differential equations for the tank heights are coupled, so the heights 4 e 3 s e 3 s 1
F (s)  2   5  
cannot be solved for or controlled independently. F1 and F2 can be used to s  16 s 2 s  1 t 
slope = -a

control h3 and h4 independently, but h1 and h2 will be affected in an

uncontrolled manner. (c) x(t) = hS(t) – atS(t) + a(t -t) S(t-t)
t h a ae  st h e  st  1
c) In the extreme case where γ= γ= 0 , we get: f (t )  e  t cos(4t )  c) X ( s)   2  2  
1 2 5 s s s s s2
 To transform the first term, use rule 18 from Table 3.1
dh1
=
C C
− 1 h1 + 3 h3  To transform the second term, use rule 3 from Table 3.1 d) Area under pulse = h t/2
dt A1 A1 s 1 1
F ( s)  
dh2 C C ( s  1)2  16 5s 2
=
− 2 h2 + 4 h4 3.4
dt A2 A2
(d)
dh3 C F
=
− 3 h3 + 2 f (t )  S (t  1) cos(4(t  1))  t 2
dt A3 A3 (a) Laplace transform on the ODE gives:
 To transform the first term, use rules 15 and 26 together. To use rule
dh4 C F 26, set f(t) = cos(4t) and t0 = 1.
=
− 4 h4 + 1 d2y dy
dt A4 A4  To transform the second term, use rule 4 of Table 3.1. ( )  6 ( )  8 ( y)  3b (e2t )
s 2 dt 2 dt
F ( s)  e s 2  1
These equations make sense with the process diagram because now F1 and s  16 s 3 s 2Y ( s)  sy (0)  y '(0)  6sY ( s)  6 y(0)  8Y ( s)  3b
F2 only affect tanks h3 and h4 directly (they no longer flow into tanks 1 s2
and 2 at all). However, F1 and F2 indirectly affect tanks 1 and 2 through h3
and h4.

2-34 [Type here] 3-1 [Type here] 3-2

 s 2Y ( s)  6sY ( s)  8Y ( s ) 
3b s( s  1) 1 1
3  6 A= 
s2 ( s  2)( s  3) (4)(3) 12
s  4

Thus: Multiply by (s+4) and let s 4

3b 3b a1 a2 a3
Y ( s)      1 6 6
( s  2)( s 2  6s  8) ( s  2) 2 ( s  4) ( s  2) 2 ( s  2) ( s  4) X ( s)    and x(t )  e2t  6e3t  6e4t 4  1 3 3
s2 s3 s4 B=  
(4)(4  3) (4)(1) 4
Regardless of the numerical values of a1, a2 and a3, the inverse Laplace transform
indicates that y(t) includes e2t , te2t , and e 4t . s2  2 Multiply by (s+3) and let s-3
b) X ( s)   1  (1)
( s  1)2 s  1 ( s  1)2
(b) When u = ct, Laplace transform gives: 3  1 2 2
C=  
2  (s  2) s 1  1 (3)(3  4) (3)(1) 3
3c
s Y ( s )  6sY ( s )  8Y ( s )  2
2

s
In Eq. 1, substitute any s-1 to determine 1. Arbitrarily using s=0, Eq. 1 Then
a3
gives
3c a a a4 1 12 3 / 4 2 3
Y ( s)   1 2  Xˆ ( s )   
s ( s  2)( s  4) s s 2 ( s  2) ( s  4)
2
2 1 1 s s4 s3
  or 1  1
12 1 12
Regardless of the numerical values of a1, a2, a3, and a4, the inverse Laplace 1 3 4t 2 3t
transform indicates y(t) includes a1 , t , e2t , and e4t . xˆ(t )   e  e
1 1 12 4 3
X ( s)   and x( t )  et  tet
s  1 ( s  1)2
1 1 1
3.5 c) X ( s)  2   Using the Real Translation Theorem,
s  s  1  1  2 3 s  b 2   2
s   
 2 4 1 3 4(t 0.5) 2 3(t 0.5)
55 55 x(t )  xˆ (t  0.5)   e  e
T(t) = 20 S(t) + t S(t) – (t-30) S(t-30) 1 3 12 4 3
30 30 where b  and 
2 2
T ( s) 
20 55 1

s 30 s 2

55 1 30s 20 55 1
30 s 2
e  
s 30 s 2
1  e 30s   t
for t  0.5
1 bt 2 2 3
x(t )  e sin  t  e sin t
 3 2
3.6 3.7
s 1
d) X(s) = e0.5 s
s( s  1)    s( s  4)( s  3) 6( s  1) 6 1  2
a) X ( s)   1  2  3 Y ( s)     2
( s  2)( s  3)( s  4) s  2 s  3 s  4 a) s 2 ( s  1) s 2 s s
To invert, first ignore the time delay term. Using the Heaviside expansion
s( s  1) with the partial fraction expansion, 6
1  1 2  s2 6 1  0
( s  3)( s  4) s2 s 0
s  2 s 1 A B C
Xˆ ( s)     6
s( s  4)( s  3) s s  4 s  3 Y (s)  2
s( s  1) s
2   6
( s  2)( s  4) s  3 Multiply by s and let s  0

3-3 3-4 3-5

12( s  2)   s  3 1 s   2 3s  4  2
b) Y ( s )   1  22 =   5 or (s2 + 4s + 3) X(s) =
s ( s 2  9) s s 9 s 2  2 s  2 ( s 2  2 s  2) 2 s  2 s( s  1)
2
X(s) =
Multiplying both sides by s(s2+9) Multiplying both sides by (s 2  2s  2) 2 (s  2) gives s( s  1)2 ( s  3)

12( s  2)  1 ( s  9)  ( 2 s   3 )( s)
2
or 1 = 1s4 + 41s3 + 61s2 +41s + 2s3 +42s2 +62s +42 + 3s2 +23s + Performing partial fraction expansion and taking the inverse Laplace
12s  24  (1   2 ) s 2   3 s  91 4s + 24 + 5s4 + 45s3 + 85s2 + 85s + 45 transform (either manually or using a symbolic software program), we get:

Equating coefficients of like power of s, 2 e3t et

Equating coefficients of like powers of s, x(t )   tet  
3 6 2
s2: 1 + 2 = 0 s : 1 + 5 = 0
4

s1: 3 = 12 b) Applying the Final Value Theorem (note that the theorem is applicable here)
s0: 91 = 24 s3 : 41 + 2 + 45 = 0
2 2
s2 : 61 + 42 + 3 + 85 = 0 lim x(t )  lim sX ( s)  lim 
Solving simultaneously, t  s 0 s 0 ( s  1)2 ( s  3) 3

8 8 s1 : 41 + 62 + 23 + 4 + 85 = 0

1  , 2  ,  3  12
3 3 3.9
s0 : 42 + 24 + 45 = 1
 8 
  s  12  2 2 A B C
8 1  3  i) Y(s) =    
Y ( s)   Solving simultaneously:
s( s 2  4s) s 2 ( s  4) s 2 s s  4
3s s2  9
1 = -1/4 2 = 0 3=-1/2 4=0 5 = ¼
( s  2)( s  3)     y(t) will contain terms of form: constant, t, e-4t
c) Y (s)   1  2  3
( s  4)( s  5)( s  6) s  4 s  5 s  6  1 / 4s  1 / 2s 1/ 4
Y ( s)    2 2 A B C
s  2s  2 ( s 2  2s  2) 2 s  2
2
ii) Y(s) =    
( s  2)( s  3) s( s 2  4s  3) s( s  1)( s  3) s s  1 s  3
1  1
( s  5)( s  6) s  4
3.8  y(t) will contain terms of form: constant, e-t, e-3t

( s  2)( s  3) 2 2 A B C
2   6 iii) Y ( s)     
( s  4)( s  6) s  5
a) From Eq. 3-66 s( s 2  4s  4) s( s  2) 2 s ( s  2) 2 s  2
t  1
L   f (t * )dt *   F ( s)
( s  2)( s  3) 0  s  y(t) will contain terms of form: constant, e-2t , te-2t
3  6
( s  4)( s  5) s  6 t  1 1
we know that L   e   d = L e t   iv) Y (s) 
2
0  s s( s  1) s( s 2  4s  8)
1 6 6
Y (s)   
s4 s5 s6
 Laplace transforming yields s 2  4s  8  (s 2  4s  4)  (8  4)  (s  2) 2  2 2
1 1 2
Y ( s)   Y ( s) 
(s  1) 
d) 2
2 2
 1 ( s  2) ( s  2s  2) 2 ( s  2)
2 2
s X(s) + 4X(s) + 3X(s) = s[( s  2) 2  2 2 ]
s( s  1)
 y(t) will contain terms of form: constant, e-2t sin2t, e-2tcos2t

3-6 3-7 3-8

 2( s  1) 2( s  1)
700
A Bs C The denominator contains no complex factors; thus x(t) is not oscillatory.
v) Y ( s)      The denominator vanishes at s=1 0; x(t) is divergent. See Fig. S3.10b.
s( s 2  4) s( s 2  2 2 ) s s 2  2 2 s 2  2 2 600

2( s  1) 1
A = lim 2  1
s 3 X (s)  X (s) 
500
s 0 ( s  4) 2 c)
s 1
2
400
1 1
2
2(s+1) = A(s +4) + Bs(s) + Cs X (s)  2 

x(t)
( s  1)( s 3  1) 1 3 1 3
( s  j )( s  j )( s  1)( s   j )( s  
300
j)
2s+2 = As2 + 4A + Bs2 + Cs 2 2 2 2
200

Equating coefficients on like powers of s The denominator contains complex factors; x(t) is oscillatory. The 100

1 denominator vanishes at real s = 0 and s= ½; thus x(t) is not convergent. See

s2 : 0=A+B  B = A =  Fig. S3.10c.
2 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

s1 :  4 time
2= C C=2 d) s 2 X ( s)  sX ( s) 
1 s Figure S3.10b. Simulation of X(s) for case b)
s: 0
2 = 4A  A= 4 4
2 X (s)  
s ( s 2  s ) s 2 ( s  1)
80

1 2  (1 2) s 2
 Y(s)   2 
s s  22 s 2  22 60
The denominator of [sX(s)] contains no complex factors; x(t) is not
1 1 2 oscillatory. The denominator of [sX(s)] vanishes at s = 0; x(t) is not
y(t) =  cos 2t  sin 2t convergent. See Fig. S3.10d.
40

2 2 2

x(t)
3.5 20

1
y(t) = (1  cos 2t )  sin 2t 3
2 0

2.5

-20
2
3.10

x(t)
1.5 -40
0 1 2 3 4 5 6 7 8 9 10
time

a) Laplace transform of the equation gives 1

Figure S3.10c. Simulation of X(s) for case c)

0.5

3
s 3 X ( s )  2s 2 X ( s )  2sX ( s )  X ( s ) 
s 0

3 3
X (s)   -0.5
0 1 2 3 4 5 6 7 8 9 10
s ( s 3  2s 2  2s  1) 1 3 1 3 time

s ( s  1)( s   j )( s   j)
2 2 2 2 Figure S3.10a. Simulation of X(s) for case a)
The denominator of [sX(s)] contains complex factors so that x(t) is
oscillatory, and the denominator vanishes at real values of s= 1 and -½
which are all <0; thus x(t) is converges. See Fig. S3.10a.
2
b) s 2 X ( s)  X ( s) 
s 1
2 2
X ( s)  
( s  1)( s 2  1) ( s  1) 2 (s  1)

3-9 3-10 3-11

b 1 1
18 b) If u(t) = be-t/ then U(s) = 1  
s  1 ( s 3  4) 5
16 s 1

1 1
14 Kb 2  
X b ( s)    1   2 ( s  1)( s  0.79  1.37 j )( s  0.79  1.37 j ) 19.6
12 (s  1)( 1 s  1)(  2 s  1) s  1.59

1
10  3  j 3   0.74  0.59 j
xb(t) = fb(e-t/ , e-t/1, e– t/2) ( s  1)( s  1.59)( s  0.79  1.37 j )
x(t)

s 0.791.37 j
8

c
If u(t) =ce-t/ where  = 1 , then U(s) =
6
c) 1 1
1 s  1 
 0.074  0.059 j  0.074  0.059 j
X(s)  5  19.6  
4

Kc s  1 s  1.59 s  0.79  1.37 j s  0.79  1.37 j

2
X c ( s) 
(1 s  1) 2 ( 2 s  1)
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1 1 1.59t
x(t )  e t  e  2e 0.79t (0.074 cos1.37t  0.059 sin 1.37t )
xc(t) = fc(e– t/1, t e– t/1, e– t/2)
time
5 19.6
Figure S3.10d. Simulation of X(s) for case d)
d dx
d) If u(t) = d sin t then U(s) = b)  12 x  sin 3t with x(0)  0
3.11 s 2  2 dt

Kd 3
X d ( s)  sX (s)  12 X(s) 
Since the time function in the solution is not a function of initial conditions, take ( s 2   2 )( 1 s  1)(  2 s  1) s2  9
the Laplace transform with:
xd(t) = fd(e– t/1, e– t/2, sin t, cos t)
time
3 3
dx(0) X ( s)  
x(0)  0 ( s 2  9)( s  12) ( s  3 j )( s  3 j )( s  12)
dt
3.12 1  j1 1  j1 3
12s2X(s) + (1+2)sX(s) + X(s) = KU(s)   
s 3j s 3j s  12
K dx3 d 2 x(0) dx(0)
X ( s)  U (s) a)  4 x  et with   x(0)  0 3 3 1 4
1  2 s 2  (1   2 ) s  1 dt 3 dt 2 dt 1  j1     j
( s  3 j )( s  12) s  3 j
 18  72 j 102 102
Factoring the denominator Take the Laplace transform of the equation:
3 1
K 1 3  
X (s)  U ( s) s 3 X(s)  4 X(s)  ( s 2  9 ) s 12 51
(1 s  1)(  2 s  1) s 1
1 4 1 4 1
a 1 1   j   j
a) If u(t) = a S(t) then U(s)= X ( s)   X (s)  102 102  102 102  51
s ( s  1)( s 3  4) ( s  1)( s  1.59)( s  0.79  1.37 j )( s  0.79  1.37 j ) s3j s 3j s  12

Ka 1 2  3  j 3  3  j 3
X a (s)  1   2     x(t )  
1 1
(cos 3t  4 sin 3t )  e12 t
s(1 s  1)(  2 s  1) s  1 s  1.59 s  0.79  1.37 j s  0.79  1.37 j 51 51

xa(t) = fa( S(t), e-t/1, e– t/2)

3-12 3-13 3-14

 d 2x dx dx(0) K  lim G ( s)  1 (from (1)) 7
c)  6  25 x  et with  x(0)  0 s 0 Y ( s 2  5s  6)  s  5 
dt 2 dt dt s
c) For a step change in x s 2  5s  7
Y ( s  5s  6) 
2
1 1  (1  4s)e 2 s 1.5 s
s 2 X( s )  6sX( s )  25X( s )  X( s )  or X( s )  X(s) =
1 .5
and Y(s) = 2
s 1 ( s  1 )( s 2  6s  25 ) s ( s  3s  1) s s 2  5s  7
Y
1    2 j  2  2 j s ( s  2)( s  3)
X (s)   1  2 
( s  1)( s  3  4 j )( s  3  4 j ) s  1 s  3  4 j s  3  4 j Therefore, yˆ (t ) = 1.5 + 1.5e-1.5t cosh(1.11t) + 7.38e-1.5t sinh(1.11t)
1 1 Using MATLAB-Simulink, y(t)= yˆ (t  2) is shown in Fig. S3.13 Now perform partial fraction expansion.
1  
( s  6 s  25)
2
20 1.5
s  1
s 2  5s  7 A B C
  
1
s ( s  2)( s  3) s s  2 s  3
1 1 1
 2  j 2    j 7 1 1
( s  1)( s  3  4 j ) s  3 4 j
40 80 0.5
A  , B   ,C 
6 2 3
0
7 1 1
1 1 1 1 1 X ( s)   
  j   j -0.5
6 s 2( s  2) 3( s  3)
X ( s )  20  40 80  40 80 7 1 2t 1 3t
s 1 s  3  4 j s 3 4 j -1 x(t )   e  e
6 2 3
1 t 1 1 -1.5

x(t )  e  e 3t ( cos 4t  sin 4t )

0 5 10 15 20 25 30

20 20 40
Figure S3.13. Output variable for a step change in x of magnitude 1.5 3.15

3.13
3.14 f (t )  hS (t )  hS (t  1 / h)
 4 x  hS (t )  S (t  1 / h)
dx
d 2 y(t ) dy(t ) d ( x  2) , x(0)=0
3  y(t )  4  x(t  2) dt
dt 2 dt dt First, take the Laplace transform of each term in the equation
Take the Laplace transform assuming zero initial conditions: Take the Laplace transform,
 d2y  2  1 e s / h 
 2   s Y ( s )  sy (0)  y '(0)  s Y ( s )  s
2
sX ( s )  4 X ( s )  h  
s2Y(s) + 3sY(s) + Y(s) = 4 e-2ssX(s) e-2sX(s)
 dt  s s 
Rearranging,  dy  1   
 5   5( sY ( s )  y (0))  5sY ( s )  5 X ( s)  h(1  e  s / h )  h(1  e  s / h )  1  2 
 dt  s( s  4)  s s  4
Y (s)  (1  4 s )e 2 s  6 y   6Y ( s) 1 1 1 1
 G (s)  2 (1) 1   , 2  
X (s) s  3s  1 s  4 s 0 4 s s  4 4
7
7 
s
a) The standard form of the denominator is : 2s2 + 2s + 1 h 1 1 
X ( s)  (1  e  s / h )   
The final transformed equation is: 4  s s  4
From (1) ,  = 1 ,  = 1.5
h 1 e s / h 1 e s / h 
Thus the system will exhibit overdamped and non-oscillatory responses.      
4 s s s  4 s  4
b) Steady-state gain

3-15 3-16 3-17

s 3  8s 2  2 s  8
0 t <0 Y(s) =
( s  3) 2 ( s 2  1) 1 / 8  1 / 8( s  2) 3/ 4
Y ( s)   
x(t ) 
h
(1  e  4t ) 0 < t < 1/h s ( s  2) 2  2 2 ( s  2) 2  2 2
4 To find y(t) we have to expand Y(s) into its partial fractions

4

h  4 ( t 1 / h )
e  e 4t  t > 1/h 1 1  3
y(t) =      e-2t cos 2t +   e-2t sin 2t
Y ( s) 
A

B

Cs

D 8 8 8
( s  3) 2 s  3 s 2  1 s 2  1

1
h=1
y(t) = Ate-3t + Be-3t + C cost + D sin t 3.17
h=10
0.9 h=100

s 1
0.8
b) Y(s) = Laplace transform of the system of ODEs gives:
0.7
s( s 2  4s  8)
 dy1 
   y2    e 
0.6
t

2
4
Since  8 , there are complex factors.  dt 
x(t)

0.5

4
 dy2 
  3  y2   2  y1 
0.4


0.3
 complete the square in denominator  dt 
0.2

s2 + 4s + 8 = s2 + 4s + 4 + 84 1
0.1
sY1  Y2  (1)
0
s 1
= s2 + 4s + 4 + 4 = (s+2)2 + (2)2 { b = 2 , =2}
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time sY2  3Y2  2Y1 (2)
Figure S3.15. Solution for values h= 1, 10 and 100
 Partial fraction expansion gives Next solve Equation 2 for Y2 in terms of Y1

3.16 A B( s  2) C s 1 Y2 ( s  3)  2Y1
Y ( s)    
s s 2  4s  8 s 2  4s  8 s( s 2  4s  8) 2Y
Y2  1 (3)
a) Take the Laplace transform: s3
Multiply by s and let s0 Substitute equation 3 into equation 1 and solve for Y1

s Y (s)  sy(0)  y(0) 6sY (s)  y(0)  9Y (s)  s s 1

2
A=1/8 2Y1 1
sY1  
2

Multiply by s(s2+4s+8) s  3 s 1
s  2  1
(s + 6s + 9)Y(s)  s(1)  2 –(6)(1)= 2
2
Y1  s  
s 1 A(s2+4s+8) + B(s+2)s + Cs = s + 1  s  3  s 1
1
s As2 + 4As + 8A + Bs2 + 2Bs + Cs = s + 1 Y1 
(s2 + 6s + 9)Y(s) = 2 2
+s+8 ( s  1)( s  )
s 1 s3
1
s2 : A+B=0  B = A =  Expand using partial fractions:
s  s 3  s  8s 2  8 8
(s2 + 6s + 9)Y(s) = 1 1 3
s2 1 1
 C = 1 + 2   4  = s3 1 1 2
s: 4A + 2B + C = 1 Y1    
8 8 4 ( s  1)2 ( s  2) s  2 s  1 ( s  1)2
1
s0 : 8A = 1 A= (This checks with above result)
8

3-18 3-19 3-20

 Now go back and substitute into equation 3 to get Y2 and expand using partial   t 
q
1 
5 21

5 21 21 5  21 5
c(t) = c i 1  e V 
t t t t
fractions: g[t_]:=  21e 2  11 21e 2  21e 2
 11 21e
 42  2

  41  

2Y1 2 2 2 2 Plot[g[t],{t,0,100},AxesLabel{time,Y},PlotRange{{0,100},{0 ,2}}]
Y2      Using MATLAB, the concentration response is shown in Fig. S3.18. (Consider
s  3 ( s  1)2 ( s  2) s  2 s  1 ( s  1) 2 V = 2 m3, ci=50 Kg/m3 and q = 0.4 m3/min) y
4
Finally, get both time-domain solutions using the inverse Laplace transform:
50

y1 (t )  e 2t  e  t  2te  t 45

y2 (t )  2(e 2t  e  t  te  t ) 40
3

35

30
3.18 2

c(t)
25

dc
V  qc  qci 20

dt
15 1

Since V and q are constant, taking Laplace transforms give 10

5
sVC(s) + qC(s) = q Ci(s)
0
0 t
0 5 10 15 20 25 30 0 2 4 6 8 10
Note that c(t = 0) = 0 Time
Figure S3.19a: Tank level response to a unit step change in flow rate.
Figure S3.18. Concentration response of the reactor effluent stream.
Also, ci(t) = 0 , t 0 (b) Define the time when y(t) reaches its maximum as tmax. This time occurs
ci(t) = c i , t>0 when y′(t)=0. Solve for this time using Mathematica and find that tmax=0.877
3.19 and y(tmax)=1.558. Therefore, the tank will not overflow.
Taking Laplace transform of the input function, a constant, gives
(c) Now find the general solution for any input step size, M (the solution is
ci (a) Take the Laplace transform of each term, taking into account that all initial denoted in this case as YM(s) and yM(t) for clarity). The input U(s) is M/s.
Ci ( s )  conditions are zero:
s
M
so that U (s) 
ci qci s 2Y  sy (0)  y '(0)  5sY  5 y (0)  Y  8sU  8u (0)  U s
sVC(s) + qC(s) = q or C(s) =
s 2Y  5sY  Y  U (8s  1) M (8s  1)
s ( sV  q)s YM ( s  5s  1) 
2

1 s
U ( s)  M (8s  1)
Dividing numerator and denominator by q s YM   MY
8s  1 s ( s 2  5s  1)
ci Y ( s 2  5s  1) 
C(s) = s
V  8s  1
YM is the previous Y, multiplied by the size of the step, M. Since M is a constant,
 q s 1 s Y taking the inverse Laplace transform gives:
  s ( s 2  5s  1)
yM (t )  My(t )
Use Transform pair #3 in Table 3.1 to invert ( =V/q) Now use symbolic mathematical software (ex. Mathematica) to solve for y(t).
Now solve the equation:
InverseLaplaceTransform[(8*s+1)/(s*(s^2+5*s+1)),s,t]

3-21 3-22 3-23

where ci(t) is a pulse of magnitude A and width tw. A pulse can be described by the
yM (tmax )  2.5  My (tmax )  M (1.558) b) Taking Laplace transform of above equations and eliminating sum of two step functions. The first will be a step function of magnitude A at time
2.5 C1 (s) and C2 (s) gives 0. The second will be a step function of –A at t=tw.
M  1.605
1.558  q  q  q 
    ci (t )  AS (t )  AS (t  t w )
C3 ( s )   V1   V2   V3 
The maximum step change in the flow rate into the tank that will not overflow the Ci ( s ) A A  tw s  1  e  tw s 
tank is 1.605.  s  q / V1  s  q / V2  s  q / V3  Ci ( s ) 
 e  A  (2)
s s  s 
Now substitute Equation (2) into Equation (1). For simplicity, define a new variable
y
Since ci (t )   (t ), Ci (s)  1
4 f=q/V.

V1  V2  V3  V
A  f  (1  etw s )
3
1.
( q / V )3 1 2 3 C3 ( s) 
ss  f 
3
3 C3 ( s )    
( s  q / V )3 ( s  q / V ) ( s  q / V ) 2 ( s  q / V )3
Now use a symbolic mathematics software to find the inverse Laplace transform,
giving c3(t). The solution is formulated as a function of t, f, A, and tw. Then as an
c3 (t )  1e(q /V )t  2te(q /V )t  3t 2e(q /V )t example, we plot c3(t) for f=1/20, A=10, and tw=1.
2

2.
V1  V2  V3  V1 In Mathematica, take the inverse Laplace transform:

( q /V1 )t ( q /V2 )t ( q /V3 )t

InverseLaplaceTransform[𝐴 ∗ 𝑓^3
c3 (t )   4e   5e   6e
1
∗ (1 − Exp[−tw ∗ 𝑠])⁄(𝑠 ∗ (𝑠 + 𝑓)^3) , 𝑠, 𝑡]

(c) Yes, amount of tracer can be calculated by measuring c3 (t ) , The solution:

0 t 
0 2 4 6 8 10
amount of tracer =  qc3 (t )dt , which can be evaluated numerically c3(t) =
Figure S3.19b: Tank level response to a 1.605 step change in flow rate. 0 ½ 𝐴(ⅇ −𝑓𝑡 (−2 + 2ⅇ 𝑓𝑡 − 2𝑓𝑡 − 𝑓 2 𝑡 2 )
− ⅇ 𝑓(−𝑡+tw) (−2 + 2ⅇ 𝑓(𝑡−tw) − 2𝑓(𝑡 − tw)
− 𝑓 2 (𝑡 − tw)2 )HeavisideTheta[𝑡 − tw])
3.20 3.21
Define the function in terms of the parameters:
Start with the Laplace version of the equations from Exercise 3.20:
a) Given constant volumes, overall balances on the three tanks indicate that 𝑔[t_, f_, A_, tw_]
the flow rate out of each tank is equal to q 1
 q  q  q  ≔ 𝐴(ⅇ −𝑓𝑡 (−2 + 2ⅇ 𝑓𝑡 − 2𝑓𝑡 − 𝑓 2 𝑡 2 ) − ⅇ 𝑓(−𝑡+tw) (−2
    2
Component balance for tracer over each tank,  V1   V2   V3  + 2ⅇ 𝑓(𝑡−tw) − 2𝑓(𝑡 − tw) − 𝑓 2 (𝑡 − tw)2 )HeavisideTheta[𝑡
C3 ( s )  C ( s)
 s  q / V1  s  q / V2  s  q / V3  i − tw])
dc1
V1  q (ci  c1 )
dt Then plot the concentration over time, assuming f=1/20, A=10, and tw=1.
Since V1=V2=V3, this equation reduces to:

dc2 3 Plot[𝑔[𝑡, 1⁄20 , 10,1], {𝑡, 0,200}, AxesLabel → {time, C3}]

V2  q (c1  c 2 ) q
dt  
C3 ( s )    3 Ci ( s )
V
(1)
dc3 s  q /V 
V3  q (c2  c3 )
dt

3-24 3-25 3-26

 Chapter 4 ©
C3
7.12/12=0.593.
0.12
e) Again use the final value theorem.
3e  s (1  e  s )
0.10 4.1 Y (s) 
(10 s  1) s
0.08 Y (s) d 3e  s (1  e  s )
 sY ( s ) 
U (s) bs  c (10 s  1)
0.06
a) Gain K can be obtained by setting s = 0 3e  s (1  e  s ) 3(1  1)
d lim  0
0.04 K s 0 (10 s  1) 1
c
0.02
Alternatively, the transfer function can be placed in the standard gain/time From the Final Value Theorem, y(t) = 0 when t
constant form by dividing the numerator and denominator by c:
Y ( s) K d b
time
 , where K  and   . f)
50 100 150 200 U ( s)  s  1 c c 3e  s
Figure S3.21: Plot of c3 over time in response to a pulse in ci of amplitude 10 and b) In order to determine the boundedness of the output response, consider a step Y (s)  1
width 1, with f=1/20. (10 s  1)
input of magnitude M. Then use U(s) = M/s and
3se  s
Y ( s) 
K M sY ( s ) 
 s 1 s (10 s  1)
3.22 From Table 3.1, the step response is: 3se  s
y(t )  KM (1  et / ) lim 0
s 0 (10 s  1)
Solve this problem using a symbolic software program such as Mathematica. The By inspection, this response will be bounded only if τ > 0, or equivalently,
following script will solve the problem (note that only 4 of the 5 possible initial only if b/c > 0.
From the Final Value Theorem, y(t)= 0 when t
conditions on y and its derivatives are included, otherwise the problem is over-
specified). 3e  s 10 30e  s
4.2 g) Y (s)   then
(10 s  1) ( s 2  4) (10 s  1)( s 2  4)
DSolve[{𝑦''''[𝑥] + 16 ∗ 𝑦'''[𝑥] + 86 ∗ 𝑦''[𝑥] + 176 ∗ 𝑦′[𝑥] + 105 ∗ 𝑦[𝑥] =
= 1, 𝑦[0] == 0, 𝑦′[0] == 0, 𝑦''[0] == 0, 𝑦'''[0] == 0}, 𝑦[𝑥], 𝑥]
a) K=3  10  ( t 1)
1 
b) =10 y (t )  30 S (t  1)  e 10
 (sin(2(t  1))  20 cos(2(t  1))) 
Running this script will give the result:  401 802 
c) We use the Final Value Theorem to find the value of y(t) when t.
ⅇ −7𝑥 (−1 + ⅇ 𝑥 )4 (5 + 20ⅇ 𝑥 + 29ⅇ 2𝑥 + 16ⅇ 3𝑥 ) 12e  s
{{𝑦[𝑥] → }} Y (s)  The sinusoidal input produces a sinusoidal output and y(t) does not have a
1680 s (10 s  1) limit when t.
12e  s
Use the Expand[ ] command to expand this solution into its individual terms. sY ( s )  These solutions can be verified by using mathematical software such as
(10 s  1)
Mathematica or Simulink.
1 ⅇ −7𝑥 ⅇ −5𝑥 ⅇ −3𝑥 ⅇ −𝑥 12e  s
{{𝑦[𝑥] → + − + − }} lim  12
105 336 80 48 48 s 0 (10 s  1)

If desired, the fractions can be approximated as decimals: From the Final Value Theorem, y(t) = 12 when t

y(t )  0.003e 7t  0.0125e 5t  0.021e 3t  0.021e t  0.01

d) y(t) = 12(1e(t-1)/10) , then y(10) = 12(1-e-9/10) = 7.12

[Type here] 4-1 4-2

3-27 [Type here]

y
y
12 4.4 4.5
10 0.25
dy1
8 0.20
From Exercise 4.2, a) 2 = -2y1 – 3y2 + 2u1 (1)
6 0.15
dt
s dy 2
4 0.10 Y (s) 3e = 4y1 – 6y2 + 2u1 + 4u2 (2)
 dt
2 0.05 U ( s ) 10s  1
time time Rearrange, Taking Laplace transform of the above equations and rearranging,
10 20 30 40 50 10 20 30 40 50
Fig. S4.2a. Output for parts c) and d). Fig. S4.2b. Output for part e).
Y ( s )[10s  1]  3e  sU ( s ) (2) (2s+2)Y1(s) + 3Y2(s) = 2U1(s) (3)
y
0.30 y
Take L-1 of (2), -4 Y1(s) + (s+6)Y2(s)=2U1(s) + 4U2(s) (4)
0.25
2 dy
0.20 10  y  3 u (t  1) (3) Solving Eqs. (3) and (4) simultaneously for Y1(s) and Y2(s),
0.15 dt
1
(2s  6)U1 ( s)  12U 2 ( s) 2( s  3)U1 ( s)  12U 2 ( s)
0.10 Take L of (3) for y (0)=4, Y1(s) = 
time 2s 2  14s  24 2( s  3)( s  4)
0.05 10 20 30 40 50
10[ sY ( s )  4]  Y ( s )  3e  sU ( s )
time 1
10 20 30 40 50 (4s  12)U1 ( s)  (8s  8)U 2 ( s) 4( s  3)U1 ( s)  8( s  1)U 2 ( s)
Substitute U ( s)  2 / s and rearrange to give, Y2(s) = 
2s 2  14s  24 2( s  3)( s  4)
Fig. S4.2c. Output for part f). Fig. S4.2d. Output for part g).
6e  s
10sY-40+Y= Therefore,
s
4.3
6e  s Y1 ( s ) 1 Y1 ( s ) 6
Y (10 s  1)   40  , 
s U1 ( s ) s  4 U 2 ( s ) ( s  3)( s  4)
The transfer function for the pressure transmitter is given by, Partial fraction expansion:
Pm ( s) 1 Y2 ( s ) 2 Y2 ( s ) 4( s  1)
 (1)  , 
P(s) 10s  1 Y (s)  e s
6

40 U1 ( s ) s  4 U 2 ( s ) ( s  3)( s  4)
s (10 s  1) (10 s  1)
and P( s)  15 / s for the step change from 35 to 50 psi. Substituting (1) and
6 a a2
rearranging gives:  1 4.6
1 15 s (10 s  1) s 10 s  1
Pm ( s) 
10s  1 s Find 1: Multiply by s and set s  0  1  6 a)
Taking the L-1 gives,
From item #13 in Table 3.1, the step response is given by: Find  2 : Multiply by 10s  1 and set s  0.1   2  60
Pm (t )  15 (1  et /10 ) (2) x (t )  0.09e t /10 and x(t )  x  x (t )  0.3  0.09e  t /10
6 6  4
Y (s)  e s   
 s s  0.1  ( s  0.1) The intial values are x (0)  0.09 and
Let ta be the time that the alarm sounds. Then,
Pm (ta )  45  35  10 psi (3) x(0)  x (0)  x  0.09  0.3  0.39.
Take L-1 ,

Substituting (3) and t=ta into (2) and solving gives ta = 11s. Thus, the alarm will y (t )  6 S (t  1)(1  e  ( t 1)/10 )  4e  t /10 The plot of the concentration response is shown in Fig. S4.6.
sound 11 seconds after 1:30PM. Check: At t =0, y (0)=4.

4-3 4-4 4-5

 a) Let the right-hand side of Eq. 1 be denoted as f(L, x0, x1, V, y1, y2)
4.8
dx1  f   f   f 
H  f ( L, x0 , x1 , V , y1 , y 2 )    L    x0    x1
dt  L  s  x0  s  x1  s The material balance is,
x
 f   f   f 
  V     y1   y  y 2 d (Ah)
 V  s  y1  s  2 s  wi  Rh 1.5
dx1 dx1 dt
Substituting for the partial derivatives and noting that  : or
t dt dt dh 1 R 1.5
dx1  wi  h
H  ( x0  x1 ) L  L x0  L x1  ( y 2  y1 )V   V y 2  V y1 (3) dt A A
dt
Use a Taylor series expansion to linearize
Fig. S4.6. Transient response. Similarly,
 g  dh  1 R 1.5  1 1.5Rh 0.5
y1  g ( x1 )    x1  (a1  2a 2 x1  3a3 x1 2 ) x1 (4)  wi  h  ( wi  wi )  (h  h )
The transfer function is given by:
 x1  s dt  A A  A A
X ( s) 0.6

X i( s) 10s  1 b) For constant liquid and vapor flow rates, L  V   0 Since the bracketed term is identically zero at steady state,
For the impulse input, xi(t )  1.5 (t ) , and from Table 3.1, X i(s)=1.5 . Thus,
Taking Laplace transforms of Eqs. 3 and 4, dh  1 1.5 Rh 0.5
X ( s ) 
0.9  wi  h
10 s  1 dt A A
HsX 1 ( s)  L X 0 ( s)  L X 1 ( s)  V Y2( s)  V Y1( s) (5)
b) Initial Value Theorem: Rearranging
0.9 Y1(s)  (a1  2a2 x1  3a3 x1 ) X 1(s)
2
(6) A dh 1
x(0)  lim sX ( s)  0.09  h  wi
s  10 1.5Rh 0.5 dt 1.5Rh 0.5
Thus, x(0)  x(0)  x  0.09  0.3  0.39 From Eqs. 5 and 6, the desired transfer functions are: Thus,
H ( s ) K
c) For the steady-state condition, 
L V Wi( s) s  1
 
x(0)  x  0.3 X 1 ( s ) X 1 ( s ) where,
 H ,  H
X 0 ( s ) s  1 Y2 ( s ) s  1 1 h h  height 
K   
1.5Rh 0.5 1.5Rh 1.5 1.5w  flowrate 
d) As indicated in the plot, the impulse response is discontinuous at t=0. The L
(a1  2a 2 x1  3a3 x1 ) 
2
results for parts (a) and (b) give the values of x(0) for t=0+ while the result for (c) Y1( s )
gives the value for t=0-.  H

A

Ah

V

mass  time
X 0 ( s ) s  1
1.5Rh 0.5 1.5Rh 1.5 1.5w mass / time
V
(a1  2a 2 x1  3a3 x1 ) 
2
4.7 Y1( s ) H

Y2 ( s ) s  1
The simplified stage concentration model becomes
where:
dx1
H  L( x 0  x1 )  V ( y 2  y1 ) (1) 
H
dt
L  V (a1  2a 2 x1  3a3 x1 )
2
y1 = a0 + a1x1 + a2x12 +a3x13 (2)

4-6 4-7 4-8

4.9 4.10 4.11

a) The model for the system is given by Additional assumptions a) Mass balances on the surge tanks:

dT 1. Perfect mixing in the tank dm1

mC  wC (Ti  T )  h p A p (Tw  T ) (2-51)  w1  w2 (1)
dt 2. Constant density  and specific heat C. dt
dT 3. Ti is constant. dm 2
mw C w w  hs As (Ts  Tw )  h p A p (Tw  T ) (2-52)  w2  w3 (2)
dt dt
Assume that m, mw, C, Cw, hp, hs, Ap, As, and w are constant. Rewriting the Energy balance for the tank,
above equations in terms of deviation variables, and noting that Ideal gas law:
dT m1
VC  wC (Ti  T )  Q  (U  bv 2 ) A(T  Ta ) P1V1  RT (3)
dt M
dT dT  dTw dTw
 ,  m
dt dt dt dt Let the right-hand side be denoted by f(T,v), P2V2  2 RT (4)
dT  M
mC  wC (Ti  T )  h p A p (Tw  T ) E Driving Force
dt dT  f    f   Flows (Ohm's law: I   )
dT 
VC  f (T , v)    T   v (1) R Resistance
m w C w w  hs As (0  Tw )  h p A p (Tw  T ) dt  T  s  v  s 1
dt  f  w1  ( Pc  P1 ) (5)
   wC  (U  bv ) A R1
2

Taking Laplace transforms and rearranging,  T  s 1
 f  w2  ( P1  P2 ) (6)
   2vbA(T  Ta ) R2
(mCs  wC  h p Ap )T ( s)  wCTi( s)  h p Ap Tw ( s) (1)  v  s 1
w3  ( P2  Ph ) (7)
(mw C w s  hs As  h p Ap )Tw ( s)  h p Ap T ( s) (2) R3
dT dT 
Substituting for the partial derivatives in Eq. 1 and noting that =
Substituting in Eq. 1 for Tw (s) from Eq. 2,
dt dt Degrees of freedom:
dT   number of parameters : 8 (V1, V2, M, R, T, R1, R2, R3)
h p Ap VC    wC  (U  bv 2 ) A T   2vbA(T  Ta )v
(mCs  wC  h p Ap )T ( s)  wCTi( s)  h p Ap T ( s ) dt  number of variables : 9 (m1, m2, w1, w2, w3, P1, P2, Pc, Ph)
(mw C w s  hs As  h p A p ) dT 
VC   wC  (U  bv 2 ) A T   2vbA(T  Ta )v  number of equations : 7
Therefore, dt   number of degrees of freedom that must be eliminated = 9  7 = 2
Taking the Laplace transform and rearranging
T ( s ) wC (m w C w s  hs As  h p A p ) Because Pc and Ph are known functions of time (i.e., inputs), NF = 0.

Ti( s ) (mCs  wC  h p A p )( m w C w s  hs As  h p A p )  (h p A p ) 2 VCsT    wC  (U  bv 2 ) A T   2vbA(T  Ta )V 
b) Model Development
 VCs   wC  (U  bv 2 ) A  T   2vbA(T  Ta )V 
 T ( s)  wC (hs As  h p A p )    Substitute (3) into (1) :
MV1 dP1
 w1  w2 (8)
b) The gain is    2vbA(T  Ta ) RT dt
 Ti( s)  s 0 wC (hs As  h p A p )  hs As h p Ap T  V MV2 dP2
 VCs   wC  (U  bv 2 ) A  Substitute (4) into (2) :  w2  w3 (9)
   RT dt
c) No, the gain would be expected to be one only if the tank were insulated so 2vbA(T  Ta )
that hpAp= 0. For the heated tank, the gain is not one because heat input T ( s) wC  (U  bv 2 ) A

changes as T changes. V ( s )  VC 
  s 1
 wC  (U  bv ) A 
2

4-9 4-10 4-11

 Substitute (5) and (6) into (8): m dT ' hA
 (T ' T 'i )  e e (T 'e  T ') (3) 4.13
w dt wC
MV1 dP1 1 1
 ( Pc  P1 )  ( P1  P2 ) meCe dT 'e

Q'
 (T 'e  T ') (4)
RT dt R1 R2 he Ae dt he Ae A mass balance yields:
MV1 dP1 1 1 1 1 dm
 Pc (t )  (  ) P1  P2 (10) Eliminate one of the output variables, T'(s) or T'e ( s ), by solving   qi   q (1)
RT dt R1 R1 R2 R2 dt
(4) for it, and substituting into (3). Because T'e ( s ) is the intermediate
Substitute (6) and (7) into (9): variable, remove it. Then rearranging gives: The mass accumulation term can be written, noting that dV=Adh=wtLdh, as
 mmeCe 2  meCe meCe m   dm dV dh
s     s  1 T '( s )    wt L (2)
 dt dt dt
MV2 dP2 1 1  whe Ae  he Ae wC w  
 ( P1  P2 )  ( P2  Ph )
RT dt R2 R3 mC  1 where wtL represents the changing surface are of the liquid. Substituting (2) into
  e e s  1 T 'i ( s )  Q '( s ) (1) and simplifying gives:
MV2 dP2 1 1 1 1  he Ae  wC
 P1  (  ) P2  Ph (t ) (11)
RT dt R2 R2 R3 R3 Because both inputs influence the dynamic behavior of T', it is necessary to dh
dP1 develop two transfer functions for the model. The effect of Q' on T' can be wt L  qi  q (3)
Note that  f 1 ( P1 , P2 ) from Eq. 10 dt
dt derived by assuming that Ti is constant at its nominal steady-state value, Ti .
dP2 The geometric construction indicates that wt/2 is the length of one side of a right
 f 2 ( P1 , P2 ) from Eq. 11 Thus, T'i = 0 and the previous equation can be rearranged as:
dt triangle whose hypotenuse is R. Thus, wt/2 is related to the level h by
T '( s ) 1/ wC
This system has the following characteristics:   G1 ( s ) (T 'i ( s )  0) A mass balance yields:
nd
(i) 2 -order denominator (2 differential equations) Q '( s ) b2 s 2  b1s  1 wt
(ii) Zero-order numerator (See Example 4.7 in text)  R 2  ( R  h) 2
Similarly, the effect of T'i on T' is obtained by assuming that Q=Q 2
W ( s ) (that is, Q'=0): After rearrangement,
(iii) The gain of 3 is not equal to unity. (It cannot be because the
Pc( s ) meCe wt  2 ( D  h)h (4)
units for the two variables are different). s 1
T '( s ) hA with D = 2R (diameter of the tank). Substituting (4) into (3) yields a nonlinear
 e e  G2 ( s ) (Q '( s )  0) dynamic model for the tank with qi and q as inputs:
T 'i ( s ) b2 s 2  b1s  1
dh 1
where  (qi  q )
4.12 dt 2 L ( D  h)h
meCe meCe m
b1 is defined to be   To linearize this equation about the operating point (h  h ) , let
he Ae wC w
(a) First write the steady-state equations: qi  q
mmeCe f 
b2 is defined to be 2 L ( D  h) h
0  wC (Ti  T )  he Ae (Te  T ) whe Ae
Then
0  Q  he Ae (Te  T ) By the superposition principle, the effect of simultaneous changes in both  f  1
inputs is given by   
Now subtract the steady-state equations from the dynamic equations  qi  s 2 L ( D  h )h
dT T '( s )  G1 ( s )Q '( s )  G2 ( s )T 'i ( s )
mC  wC (Ti  Ti )  (T  T )   he Ae  (Te  Te )  (T  T )  (1)  f  1
dt (b)   
The limiting behavior of m eCe going to zero has b 2  0  q  s 2 L ( D  h )h
dTe
meCe  (Q  Q )  he Ae (Te  Te )  (T  T )  (2)  
dt and b1  m / w and simplifies the last equation to  f  1 
Note that dT / dt  dT '/ dt and dTe / dt  dTe '/ dt. Substitute     qi  q      0

T '( s ) 
1/ wC
Q '( s ) 
1
T 'i ( s )  h  s  h  2 L ( D  h )h  s
deviation variables; then multiply (1) by 1/wC and (2) by 1/(h e Ae ). m m
s 1 s 1
w w

4-12 4-13 4-14

The last partial derivative is zero, because qi  q from the steady-state relation, and K2   q  3.153  10  4.364  10 
6 7
q 
Taking the Laplace transforms and rearranging,     13.84     
 q  13840 
the derivative term in brackets is finite for all 0<h<D. Consequently, the linearized 0.15766q   1000  T2  T2
model of the process, after substitution of deviation variables, is
Vs  q  Vk (T )C  (s)  qC  (s)  Vc
A Ai A k (T )
20000
T ( s) (5)  8.87  10 3
dh ' 1 T2
 (q 'i  q ')
dt 2 L ( D  h )h  20000 
VCs  qC  (H )VcA k (T ) T 2  T ( s)  (H )Vk (T )C A ( s) (6)
4.15
Recall that the term 2L ( D  h)h in the previous equation represents the variable
Substituting for CA (s) from Eq. 5 into Eq. 6 and rearranging,
surface area of the tank. The linearized model treats this quantity as a constant that
Assumptions:
depends on the nominal (steady-state) operating level. Consequently, operation of
T ( s ) HVk (T )q (7) 1. Constant physical properties
the horizontal cylindrical tank for small variations in level around the stead-state 
 (s)
C Ai  20000  20000 2. Perfect mixing
Vs  q  Vk (T )  VCs  qC  (H )Vc A k (T ) T 2   (H )c AV k (T ) T 2
2 2
value would be much like that of any tank with equivalent but constant liquid
 
surface. For example, a vertical cylindrical tank with diameter D’ has a surface are
c A is obtained from the steady-state version of Eq.1, Dynamic model: Balances on cell mass and substrate concentration
of liquid in the tank equal to  ( D ')2 / 4  2 L ( D  h )h . Note that the coefficient
qcAi dX
1 cA   0.001155 mol/cu.ft.   ( S ) X  DX  f1 ( S , X , D) (1)
L ( D  h )h is infinite for h  0 or for h  D and is a minimum at h  D / 2 . q  Vk (T ) dt
2 dS
Thus, for large variations in level, this equation would not be a good approximation, Substituting the numerical values of T , , C, (  H), q, V, c A into Eq. 7 and    ( S ) X / YX / S  D( S f  S )  f 2 ( S , X , D, S f ) (2)
simplifying gives, dt
because dh/dt is independent of h in the linearized model. In these cases, the
where:
horizontal and vertical tanks would operate very differently.
T ( s) 11.38  S
  ( S ) X is defined as m X ,
C Ai ( s) (0.0722s  1)(50s  1) Ks  S
4.14
 T ( s)  D is defined as
F
b) The gain K of the above transfer function is equal to   ,
 C Ai ( s)  s 0
V
Assumptions
Linearization of (1) about the nominal steady state gives a linearized model of the
1. Perfectly mixed reactor 0.15766 q
2. Constant fluid properties and heat of reaction. K (8) form:
 q c  q  c
  3.153 106 A2    13.84   4.364.107 A2
a) Component balance for A,  1000 T   1000  T dX  f1 f1 f1
 S  X D
dc dt S X D
V A  q (c Ai  c A )  Vk (T )c A (1) It is obtained by setting s=0 in Eq. 7 and substituting numerical values for ,
ss ss ss

dt
Energy balance for the tank, C, (  H), V. Evaluating sensitivities gives,
dX   m ( K s  S )  m S    S 
dT   X  S '  m  D  X ' X D (3)
VC  qC (Ti  T )  (  H )Vk (T )c A (2) dK K K2  q c  dt  ( K s  S )2   Ks  S 
dt   2  0.01384  3153 A2   6.50  10 4
Since a transfer function with respect to cAi is desired, assume the other inputs, dq q 0.15766q  10 6 T 
namely q and Ti, to be constant. Linearization of (2) about the nominal steady state:
dc A dc A dT dT  dK K 2  q   3.153  10 c A  2  2  4.364  10 c A 
6 7
    13.84    dS 
Linearize (1) and (2) and note that , ,   f f 2 f 2 f 2
dt dt dt dt dT 3.153  1000  T3  T3   2 S' X ' D'  S f
dt S ss X ss D ss S f
dc  20000  2.57  10 5 ss

V A  qc Ai  (q  Vk (T ))c A  Vc A k (T ) T (3)

dt T2 dS   1 m ( K s  S )  m S   1 m S 
 X ' ( S f  S ) D '  D S f
dK dK dc A   X  D  S '  
dT   20000   
VC   qC  HVc A k (T ) T  + (H )Vk (T )cA (4) dc Ai dc A dc Ai dt  YX / S ( K s  S )2   YX / S K s  S 
 
2
dt T
(4)

4-15 4-16 4-17

 Substituting the numerical values gives: Chapter 5 1500

dX '
 0.113S '  2.25D ' 5.1
dt
No, the time required for the output Y(s) to reach steady state does not depend on
dS '
  0.326 S '  0.2 X '  9 D '  0.1S f the magnitude of the step input in U, it only depends on the time constant 𝜏𝑖 and 1000
dt delay θ . Since the Laplace transform of a step change is M/s, we have:
Taking Laplace transforms, assuming steady state initially: KM
Y  s   G  s U  s   e s
s  1s  1 2 s  1 3 s  1 4 s  1

T'
sX ' ( s)  0.113S ' ( s)  2.25D' ( s )
The inverse Laplace transform takes the following form:
sS '( s)   0.326 S '( s)  0.2 X ( s)  9 D( s)  0.1 S f ( s)
500
  12  22 
 e  t / 1  e  t / 2  
1   2 1   3 1   4  1   2  2   3  2   4 
In order to derive the transfer function between X and D, assume that Sf is Y  t   KMu  t     
  32  42 
constant at its nominal steady-state value, S f (t )  S f ; thus Sf  0. Rearranging  e  t / 3
 e  t / 4

  1   3  3   2  3   4  1   4  4   2  4   3  
gives, 0
As shown in above equation, the settling time is not related to the magnitude of 0 1 2 3 4 5 6 7 8 9 10
0.113 2.25
X ' ( s)  S ' (s)  D' ( s ) (5) input signal M. Time(/min)
s s
Figure S5.2. Temperature response for a ramp input of magnitude 0.5 Kw/min.
 0.2   9 
S '( s)    X '( s)    D '( s) (6) 5.2 5.3
 s  0.326   s  0.126 
(a) For a step change in input of magnitude M: The contaminant concentration c increases according to this expression:
Substitute (6) into (5) and rearrange gives,
y  t   KM 1  e  t /   y  0  c(t) = 5 + 0.2t
X '( s)  (2.25s  1.7)
 (7)
D '( s) s 2  0.326s  0.0226 We note that KM  y     y  0   500  100  400 C Using deviation variables and Laplace transforming,

400 C 0.2
Rearrange (7) to a standard form: Then K   400 C / Kw c(t )  0.2t or C ( s) 
 2  1 Kw s2
X '( s ) K ( s  1) 400  100 Hence
 2 2 a At time t  4 , y  4   400 C ; thus,  1  e4/ , or   2.89 min 1 0.2
D '( s )  s  2 s  1 500  100 C m ( s)  
T ' s 400 10s  1 s 2
where:    C / Kw


K  77.4 g h /L P '  s  2.89 s  1  and applying Eq. 5-21

 a  0.778 h
(b) For an input ramp change with slope a  0.5Kw / min : cm (t )  2(et /10  1)  0.2t
  6.65 h
Ka  400  0.5  200 C / min
  1.08
This maximum rate of change will occur as soon as the transient has died out, i.e., As soon as cm (t )  2 ppm the alarm sounds. Therefore,
Note that the step response will be overdamped because  >1. after 5  2.89 min  15 min have elapsed.
Solution Manual for Process Dynamics and Control, 4th edition t = 18.4 s (starting from the beginning of the ramp input)
Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III

4-18
5-1 5-2

The time at which the actual concentration exceeds the limit (t = 10 s) is 5.5
subtracted from the previous result to obtain the requested t . 6

a) Energy balance for the thermocouple,

t = 18.4  10.0 = 8.4 s 5

dT
2.5
mC  hA(Ts  T )
4
dt

where m is mass of thermocouple

2
3 C is heat capacity of thermocouple
C'

h is heat transfer coefficient

1.5 2
A is surface area of thermocouple
t is time in sec
c'm

1
1 Substituting numerical values in (1) and noting that

dT dT 
0.5
0
0 2 4 6 8 10 12 14 16 18 20 Ts  T and  ,
time
dt dt
Figure S5.4. Exit concentration response for a rectangular input. dT 
0
0 2 4 6 8 10 12 14 16 18 20
15  Ts  T 
dt
time
b) By inspection of Eq. 1, the time at which this function will reach its
Figure S5.3. Concentration response for a ramp input of magnitude 0.2 Kw/min. T ( s ) 1
maximum value is 2, so maximum value of the output is given by Taking Laplace transform, 
c (2)  8(1  e 1 )  8(1  e 0 / 2 ) S (0) Ts( s) 15s  1
5.4

and since the second term is zero, c(2)  5.057 b) Ts(t) = 23 + (80  23) S(t)
a) Using deviation variables, the rectangular pulse is
c) By inspection, the steady state value of c(t ) will be zero, since this is a Ts  T  23
0 t<0 first-order system with no integrating poles and the input returns to zero.
cF = 2 0t<2 To obtain c() , simplify the function derived in a) for all time greater From t = 0 to t = 20,
0 2t than 2, yielding
57
Ts(t )  57 S(t) , Ts( s ) 
Laplace transforming this input yields
c (t )  8(e  (t 2) / 2  e t / 2 ) s
1 57
T ( s )  Ts( s ) 
CF' ( s) 
2
s
1  e2s  which will obviously converge to zero. 15s  1 s (15s  1)

Substituting c(t )  0.05 in the previous equation and solving for t gives Applying inverse Laplace Transform,
The input is then given by
t = 9.233 T (t )  57(1  e t / 15 )
8 8e2 s
C ' ( s)  
s(2s  1) s(2s  1) Then

and from Table 3.1 the time domain function is T (t )  T (t )  T  23  57(1  e t / 15 )

c ' (t )  8(1  et /2 )  8(1  e (t 2)/2 ) S (t  2) Since T(t) increases monotonically with time, maximum T = T(20).

5-3 5-4 5-5

 lim y(t )  lim(25e t / 5  15e t / 3  10) M  10M Then, lim y (t ) does not converge.
t  t  t 
Maximum T(t) = T(20) = 23 + 57 (1-e-20/15) = 64.97 C
c)
Or, final value theorem from Chapter 3 applies: (d)
10 M 10 1
lim y (t )  lim sY ( s )  lim sG ( s)U ( s )  lim s  10M Y ( s)  G ( s)U ( s )  (1  e t w s )
50 t  s o s o s o (5s  1)(3s  1) s (5s  1)(3s  1) s
45
41.97 º (b) According to part (a), we have:
10 y(t )  (25e t / 5  15e t / 3  10)  S (t  t w )(25e  (t 10) / 5  15e (t 10) / 3  10)
40
Y ( s )  G ( s)U ( s ) 
(5s  1)(3s  1)
35
a1 a Then,
Y (s)   2 lim y(t )  lim(25e t / 5  15e t / 3  10)  (25e (t 10) / 5  15e  (t 10) / 3  10)  0
30 5s  1 3s  1 t  t 

25
Partial fraction expansion:
T'

20
a1 = 25, a2 = -15.
25 15 5.7
Y ( s)  
15
5s  1 3s  1
10
Inverse Laplace: Assume that at steady state the temperature indicated by the sensor Tm is equal to
5 y (t )  5e  t / 5  5e t / 3 the actual temperature at the measurement point T. Then,

0
Tm ( s ) K 1
Then,  
0 5 10 15 20 25 30 35 40 45 50
time
lim y(t )  lim 5e t / 5  5e t / 3  0 T ( s ) s  1 1.5s  1
t  t 
Figure S5.5. Thermocouple output for parts b) and c)
Tm  T  350 C
Or, final value theorem from Chapter 3 applies:
5.6
10
lim y (t )  lim sY ( s )  lim sG ( s )U ( s )  lim s 0 Tm (t )  15sin t
t  s o s o s o (5s  1)(3s  1)
(a)
10 M (c) where  =2  0.1 rad/min = 0.628 rad/min
Y ( s)  G ( s )U ( s )  10 1
(5s  1)(3s  1) s Y ( s )  G ( s )U ( s ) 
(5s  1)(3s  1) s 2  1 At large times when t/ >>1, Eq. 5-26 shows that the amplitude of the sensor
a a a signal is
Y ( s)  ( 1  2  3 ) M a1 a2 a3  b3 j a3  b3 j
5s  1 3s  1 s Y (s)     A
5s  1 3s  1 s j s j Am 
2 2  1
Partial fraction expansion:
a1 = 125, a2 = -45, a3 = 10. Partial fraction expansion: where A is the amplitude of the actual temperature at the measurement
125 45 10 a1 = 625/26, a2 = -27/2, a3 = -2/13, b3 = 7/26.
Y ( s)  (   )M point.
5s  1 3s  1 s 625 / 26 27 / 2  2 / 13  7 / 26 j  2 / 13  7 / 26 j
Y (s)    
5s  1 3s  1 s j s j A  15 (0.628) 2 (1.5) 2  1 = 20.6C
Inverse Laplace: Therefore
y (t )  (25e t / 5  15e t / 3  10) M Inverse Laplace:
Maximum T  T  A =350 + 20.6 = 370.6
y (t )  125 / 26e t / 5  9 / 2e t / 3  7 / 13sin t  4 / 13 cos t
Then,
Maximum Tcenter = 3 (max T) – 2 Twall

5-6 5-7 5-8

= (3  370.6)(2  200) = 711.8C

2.5
11

Therefore, the catalyst will not sinter instantaneously, but will sinter if operated 2
10

for several hours.

9
1.5

8
h'(t)

h(t)/ft
1
5.8 7

0.5 6
a) Assume that q is constant. Material balance over the tank,
5
dh
 q1  q 2  q
0
A 0 5 10 15 20 25 30 35 40 45 50

dt time
4
0 5 10 15 20 25
Figure S5.8a. Liquid level response for part b) Time/min
Writing in deviation variables and taking Laplace transform
Figure S5.8b. Liquid level response for part d)
As H ( s )  Q1 ( s )  Q2 ( s ) c) h  6.122 ft at the new steady state t  12

H ( s ) 5.9

1 d) q1(t )  10S  t   5S  t  12 
Q1 ( s ) As
a) Material balance over tank 1.
b) q1 (t )  5 S(t) – 5S(t-12) 10 5 12 s
Q1( s)   e
s s dh
A  C (qi  8.33h)
5 5 12 s 10 / A 5 / A dt
Q1 ( s)   e H ( s)  2  2 e12 s
s s s s
1 5/ A 5/ A  10 where A =   (4)2/4 = 12.6 ft2
H ( s)  Q1( s)  2  2 e12 s 4  t  4  0.354t 0  t  12
As s s h t    A
 6.122  0.177t t  12 ft 3 /min
C = 0.1337
5 5 USGPM
h(t )  t S(t)  (t  12) S(t-12) The liquid level will keep increasing and there will be no steady-state
A A value of liquid level h .
AsH ( s)  CQi( s)  (C  8.33) H ( s)

4+
5
t  4  0.177t 0  t  12 H ( s ) 0.12

A Qi( s ) 11.28s  1
h(t) =
5 
4 +   12   6.122 12 < t For tank 2,
A 
dh
A  C (qi  q )
dt

5-9 5-10 5-11

 H ( s ) 0.011 5.11
As H ( s )  CQi( s ) , 
Qi ( s ) s 5.10
KM K K2
b) Qi( s )  20 / s Y(s) =  1
s 2 (s  1) s 2 s(s  1)
a) The dynamic behavior of the liquid level is given by
2.4 2.4 27.1
For tank 1, H ( s)    K1s + K1 + K2s = KM
s (11.28s  1) s 11.28s  1 d 2 h dh
A  Bh  C p (t )
-t/11.28
dt dt K1 = KM
h(t) = 6 + 2.4(1 – e ) K2 = K1 =  KM
where
For tank 2, H ( s )  0.22 / s 2 6 3g 3 Hence
A= B= and C =
R 2 2L 4L KM KM
h(t) = 6 + 0.22t Y(s) = 
s2 s (s  1)
Taking the Laplace Transform and assuming initial values = 0 or
c) For tank 1, h() = 6 +2.4 – 0 = 8.4 ft
y(t) = KMt  KM (1-e-t/)
s 2 H ( s)  As H ( s)  BH ( s)  C P ( s)
For tank 2, h() = 6 + (0.22  ) =  ft After a long enough time, we can simplify to
C/B
d) For tank 1, 8 = 6 + 2.4(1 – e-t/11.28) or H ( s)  P ( s) y(t)  KMt - KM (linear)
1 2 A
s  s 1
h = 8 ft at t = 20.1 min B B
For tank 2, 8 = 6 + 0.22t slope = KM
h = 8 ft at t = 9.4 min We want the previous equation to have the form intercept =  KM

Tank 2 overflows first, at 9.4 min. K That way we can get K and 
e) The red line (h’(t)=2 ft, or h(t)= 8 ft) shows that tank 2 overflows first at H ( s )  P ( s)
 2 s 2  2s  1
9.4 min.
y(t)
1
Hence K = C/B =
9
2g Slope = KM
1/ 2
8 1  2L 
2  then   1 / B   
7 B  3g 
1/ 2
A 3  2 L  
6
2  then    
B R 2   3g 
5
h'(t)

Figure S5.11. Time domain response and parameter evaluation

4 b) The manometer response oscillates as long as 0 <  < 1 or
1/ 2
3  2 L 
3
0   1
R 2   3g 
2
If  is larger , then  is smaller and the response would be more oscillatory.
1

If  is larger, then  is larger and the response would be less oscillatory.

Tank 1
Tank2
0
0 5 10 15 20 25 30 35 40
time

Figure S5.9. Transient response in tanks 1 and 2 for a step input.

5-12 5-13 5-14

5.12  t  t
5.13 Let E = ln1  s   s  3
  
y  Ky  4 y  x t
a) and find value of s that makes E  0 by trial-and-error.
a) The solution of a critically-damped second-order process to a step change 
Assuming y(0) = y (0)  0 of magnitude M is given by Eq. 5-50 in text.
ts/ E
4 0.6094
Y (s) 1 0.25   t 
  y(t) = KM 1  1  e t /   5 -0.2082
X ( s) s 2  Ks  4 0.25s 2  0.25Ks  1     4.5 0.2047
4.75 -0.0008
b) Characteristic equation is Rearranging
 a value of t = 4.75 is ts, the settling time.
s2 + Ks + 4 = 0 y  t
 1  1  e  t /  Ka a a a a4
KM   b) Y(s) =  1  22  3 
 K  K 2  16 s 2 (s  1) 2 s s s  1 (s  1) 2
The roots are s =
2
 t  t /  y We know that the a3 and a4 terms are exponentials that go to zero for large
1  e  1
-10  K < -4 Roots : positive real, distinct   KM values of time, leaving a linear response.
Response : A + B e t / 1 + C et /  2
Ka
When y/KM = 0.95, the response is 0.05 KM below the steady-state value. a2 = lim  Ka
K = -4 Roots : positive real, repeated s 0 (s  1) 2
Response : A + Bet/ + C et/
KM Ka
Define Q(s) =
-4 < K < 0 Roots: complex with positive real part. (s  1) 2
t t 0.95KM
Response: A + et/ (B cos 1   2 + C sin 1   2 )
  dQ  2 Ka
y 
ds (s  1) 3
K=0 Roots: imaginary, zero real part.
Response: A + B cos t/ + C sin t/
1   2 Ka 
Then a1 = lim 
1! s 0  (s  1) 3 
0<K<4 Roots: complex with negative real part.
0 ts
Response: A + e-t/ (B cos 1   2
t t
+ C sin 1   2 )
time a1 =  2 Ka
 
 the long-time response (after transients have died out) is
K=4 Roots: negative real, repeated.
 t s  t / 
Response: A + Be-t/ + C t e-t/ 1  e  1  0.95  0.05 y  (t )  Kat  2 Ka  Ka (t  2)
 
 a(t  2) for K = 1
4 < K  10 Roots: negative real, distinct  t  t
ln1  s   s  ln(0.05)  3.00 and we see that the output lags the input by a time equal to 2.
Response: A + B et / 1 + C et /  2   

Response will converge in region 0 < K  10, and will not converge in
region –10  K  0

5-15 5-16 5-17

   R
R  R   R R  R-8 P  P-15
4
u(t)
3.5 y(t)   0.167 R  R  0.2 P  5
0.127 R
2
3   1.31 R  7.88 R  1.57 P  39.5
R
y x=at yl =a(t-2) 2.5

5.15
2

Y
1.5
P ( s ) 3
  C / kW 
1 T ( s ) (3) 2 s 2  2(0.7)(3) s  1
0 actual response time
Note that the input change p(t )  26  20  6 kw
0.5

(c) . 0
0 0.5 1 1.5 2
Time
2.5 3 3.5 4
Since K is 3 C/kW, the output change in going to the new steady state
1
Output
will be
0.9
Figure S5.13b Computer simulation results on part (b)
95% threshold
T   (3 C / kW )6 kW   18 C
0.8 t 
0.7 5.14
a) Therefore the expression for T(t) is Eq. 5-51
0.6

11.2mm  8mm  0.7 t   1  (0.7) 2   1  (0.7) 2  

 0.20mm / psi
0.5
 
Y

T (t )  70  18 1  e 3  cos t  

a) Gain = 
t  sin 
0 .7
31psi  15psi  
0.4
3  1  (0.7) 2   

     
 
12.7mm  11.2mm
0.3
Overshoot =  0.47
0.2 11.2mm  8mm 25

0.1    
Overshoot = exp    0.47 ,  = 0.234
0  1 2 
0 1 2 3 4 5 6   20

Time
 2 
Figure S5.13a Computer simulation results on part (a) Period =    2.3 sec
 1 2  15

 

T'(t)
1  0.2342
 = 2.3 sec   0.356 sec 10
2
R ( s ) 0.2
 (1)
P ( s ) 0.127s 2  0.167s  1 5

b) From Eq. 1, taking the inverse Laplace transform, 0

0 5 10 15 20 25 30 35 40 45 50

time
  0.167 R   R  0.2 P
0.127 R Figure S5.15. Process temperature response for a step input

5-18 5-19 5-20

b) The overshoot can obtained from Eq. 5-53 or Fig. 5.11. From Figure 5.11    20 
Time to the first peak, tp  ts  
we see that OS  0.05 for =0.7. This means that maximum temperature is 1  2
Therefore, ln
  1   2 

Tmax  70 + (18)(1.05) = 70 + 18.9 = 88.9 y (t p )  KM
b) Overshoot, OS =
KM
From Fig S5.15 we obtain a more accurate value.   t    
5.17
The time at which this maximum occurs can be calculated by taking OS =  exp  cos()  sin( )
derivative of Eq. 5-51 or by inspection of Fig. 5.8. From the figure we see     1  2 
that t /  = 3.8 at the point where an (interpolated) =0.7 line would be.        
 exp    exp   a) Assume underdamped second-order model (exhibits overshoot)
  1     1   
2 2

Δoutput 15  10 ft ft
 tmax  3.8 (3 min) = 11.4 minutes K   1/ 6
y (t 3 p )  KM Δinput 210  180 gal/min gal/min
c) Decay ratio, DR =
y (t p )  KM
5.16 16.5  15 1.5
3 Fraction overshoot =   0.3
where y (t 3 p )  is the time to the third peak. 15  10 5
1 2
For underdamped responses, From Fig 5.11, this corresponds (approx) to  = 0.35
KM e
 t 3 p / 
      2 

   1  2    1  2  
 DR =  exp  (t 3 p  t p )  exp    From Fig. 5.8 ,  = 0.35 , we note that tp/  3.5
y (t )  KM 1  e  t /  cos t  sin t   (5-51) KM e
 t p / 
      1   2 

    
 1  2  


 

 
Since tp = 4 minutes (from problem statement, assuming first peak),
 2 
 exp    (OS) 2 t p 1 ζ2
  1.19 min
 1   
a) At the response peaks, 2


dy 
   1  2    1   2  d) Consider the trigonometric identity 1/ 6 0.17
 KM  e t /  cos  t sin  t   G p (s)  
      
  1  2 (1.19) 2 s 2  2(0.35)(1.19)s  1 1.42s 2  0.83s  1
dt   
 sin (A+B) = sin A cos B + cos A sin B
b) 4 minutes might not be the first peak (as shown in Figure 5.8); thus, the
 1  2  1 2    1 2  
  1 2  solution may be not unique.
et /    sin  t   cos  t    0 Let B =  t  , sin A = 1   2 , cos A = 
            
      

Since KM  0 and et /   0 

y (t )  KM 1  e

 t /  1
1 2
 
1   cos B   sin B 
2


 5.18

     1      1     1      (a)
2 2 2 2
0     cos t   sin t  KM 1 
e  t / 
sin( A  B) τ=1,ζ=0.5 .
         1  2
      
  1  2 
Roots of denominator are: s 2  s  1  0 , s = -0.50+0.87j and -0.50-0.87j.
Imaginary roots suggest oscillation.
 1 2   Hence for t  t s , the settling time,
0  sin  t   sin n , t  n
2 2 4
lim y (t )  lim sY ( s)  lim s 2  lim 4.
   1  2 t  s 0 s 0 s  s  1 s s 0 s 2  s  1
 
where n is the number of the peak.
e  t / 
1  2
 0.05 , or 
t   ln 0.05 1   2  Time to first peak: πτ/ 1-ζ 2 =3.6
Overshoot: 2  2  exp(-πζ/ 1-ζ 2 )=0.652

5-21 5-22 5-23

 2 ft 3 /min H ( s ) C CR 0.0281
Period:  7.25 C = 0.1337   
1-ζ 2 gpm Qi( s ) 1 ARs  1 2.64s  1
As 
h 2.5 ft R
R1 = R2 = 1   0.187 3
Figure can be sketched using Figure S5.8 in Chapter 5 for   1,ζ  0.5 . Cqi 0.1337  100 ft /min   0.0281M
hmax
2.81 ft
Using deviation variables and taking Laplace transforms, Mmax =  100 gpm
0.0281 ft/gpm
H 1 ( s) C CR1 0.025
   Hence, both systems can handle the same maximum step disturbance in qi.
Qi( s) 1 A1 R1 s  1 1.32s  1
A1 s 
R1
M
H 2 ( s) 1 / R1 R2 / R1 1 b) For step change of magnitude M, Qi( s) 
   s
H 1 ( s ) 1 A2 R2 s  1 1.32s  1
A2 s 
R2 For original system,
H 2 ( s) 0.025

Qi( s) (1.32s  1) 2 Q2 ( s) 
1
H 2 ( s) 
1 0.025 M
R2 0.187 (1.32s  1) 2 s
For step change in qi of magnitude M,
1 1.32 1.32 
h1max  0.025M  0.134M    2 
 s (1.32s  1) (1.32s  1) 
h2 max  0.025M since the second-order transfer function
  t  t / 1.32 
q2 (t )  0.134M 1  1  e 
0.025   1.32  
is critically damped (=1), not underdamped
Figure S5.18 Step response of G. (1.32s  1) 2 For modified system,
2.5 ft
Hence Mmax =  100 gpm
(b) 0.025 ft/gpm 1 1 0.0281 M 1 2.64 
Q ( s )  H ( s)   0.134M  
Decay Ratio: exp(2 ζ/ 1-ζ 2 )  0.106 R 0.21 (2.64s  1) s  s 2.64s  1
For the modified system,

dh h

q (t )  0.134M 1  e t / 2.64 
5.19 A  Cqi 
dt R
Original system provides better damping since q2 (t ) < q(t ) for t < 3.4.
c) Computer simulation result
a) For the original system, A  (4) 2 / 4  12.6 ft 2
V = V1 + V2 = 2  7.07ft 2  5ft = 70.7ft3
dh h hmax = V/A = 5.62 ft
A1 1  Cqi  1
dt R1
dh h h h 0.5  5.62 ft
A2 2  1  2 R=   0.21 3
dt R1 R2 Cqi 0.1337  100 ft /min

where A1 = A2 = (3)2/4 = 7.07 ft2

5-24 5-25 5-26

3 dc A
0.14 b) C1 ( s)  V  q(c A i  c A )  Vk (T )c A (1)
Original s dt
Modified 1 .5
0.12 C m ( s ) 
s (0.1s  1)(49s  1) Energy balance for the tank,
0.1  1 
cm (t )  1.51  (0.1e t / 0.1  49e t / 49 ) dT
 (49  0.1)  VC  qC (Ti  T )  (H R )Vk (T )c A (2)
0.08
dt
q/M

0.5 3 1.5 Since a transfer function with respect to cAi is desired, assume the other
0.06 c) C m ( s )  
(49s  1) s s (49s  1) inputs, namely q and Ti, are constant. Linearize (1) and (2) and note that

 
0.04
cm (t )  1.5 1  e t / 49 dc A dc A dT dT 
 ,  ,
0.02 dt dt dt dt

d) The responses in b) and c) are nearly the same. Hence the dynamics of the dcA 20000
0
0 1 2 3 4 5 6 7 8 9 10 V  qcA i  (q  Vk (T ))cA  Vc A k (T ) T (3)
Time/min
conductivity cell are negligible. dt T2
Figure S5.19 Computer simulation results on part (b) 1.5
dT   20000 
VC   qC  H RVc A k (T ) T   H RVk (T )c A (4)
dt  T2 
5.20
Taking Laplace transforms and rearranging
1
a) Caustic balance for the tank,
Vs  q  Vk (T )C  (s)  qC  (s)  Vc
A Ai A k (T )
20000
T ( s) (5)
T2
Cm'(t)

dC  20000 
V  w1c1  w2 c 2  wc VCs  qC  (  H )Vc k ( T ) T ( s )  ( H )Vk (T )C  ( s ) (6)
dt  T 2 
R A R A

0.5
Since V is constant, w = w1 + w2 = 10 lb/min
Substituting C A (s ) from Eq. 5 into Eq. 6 and rearranging,
For constant flows,
T ( s) (H R )Vk (T )q
Part b) 
Vs C ( s )  w1C1 ( s )  w2 C 2 ( s )  wC ( s ) Part c) C Ai ( s)  20000  20000
Vs  q  Vk (T )  VCs  qC  (H R )VcA k (T )  (H R )V 2c Ak 2 (T )
T 2 
0
0 20 40 60 80 100
time
120 140 160 180 200
 T2
C ( s) w1 5 0.5
   Figure S5.20 Step responses for parts b) and c) (7)
C1 ( s) Vs  w (70)(7) s  10 49s  1
c A is obtained from Eq. 1 at steady state,
C m ( s) K
 , K = (3-0)/3 = 1 ,   6 sec = 0.1 min qc Ai
C ( s) s  1 5.21 cA  = 0.01159 lb mol/cu.ft.
(from the graph) q  Vk (T )

C m ( s) 1 0.5 0.5 Assumptions: 1) Perfectly mixed reactor Substituting the numerical values of T , , C, –HR, q, V, c A into Eq. 7
 
C1 ( s) (0.1s  1) (49s  1) (0.1s  1)(49s  1) 2) Constant fluid properties and heat of reaction and simplifying,

a) Component balance for A,

5-27 5-28 5-29

 T ( s) 12.69 5.22

C A i ( s) (0.082s  1)(5s  1)
(a)
For step response, C Ai ( s )  1 / s Step response of a first-order process is:
K1 M
Y ( s)  G ( s )U ( s) 
T ( s ) 
12.69 (s  1) s
(0.082s  1)(5s  1) s Inverse Laplace gives:
y (t )  K1 M (1  e t /  ) (1)
 1 
T (t )  12.691  (0.082e t / 0.082  5e t / 5 ) Taylor series expansion at t = 0: e-t/τ = 1- 1/τ × t. Substitute into Eq. (1):
 (50  0.082) 
K
y(t )  K1 M (1  (1  t /  ))  1 Mt (2)
A first-order approximation of the transfer function is 
Ko
Inverse Laplace on integrator Go ( s )  :
T ( s) 12.69 s Figure S5.22 Step response data to find delay and approximated integrator process gain.
 y (t )  K 0 Mt (3)
C A i ( s) 5s  1

For step response, T ( s ) 

12.69
s (5s  1)

or T (t )  12.69 1  e
t / 5
 Compared Eqs. 2 and 3, we conclude when t is close to zero, or t << τ, first order
5.23
system can be approximated by integrator with:
K
The two step responses are very close to each other hence the K0  1 (4)
 (a)
approximation is valid. The ODE calculation indicates a slightly different 5 (1  e s )
gain due to linearization. Y ( s)  G ( s)U ( s) 
(b) (3s  1) s
K1 Final value theorem:
From part (a), K 0  .
  5 (1  e s )  5(1  e s )
lim y (t )  lim sY ( s)  lim  s   lim 0
(c)
t  s 0 s 0
 (3s  1) s  s 0 (3s  1)
Eq. 3 shows the integrator step response in time domain. With the step test data,
plot the data and approximate the slope of the line. Set the slope equal to K0M and (b)
find K0. The time delay would be estimated to be the time where the line K1 1
Y ( s )  G ( s )U ( s ) 
intersects the x-axis. (3s  1) s 2
Final value theorem:
 5 1 5
lim y(t )  lim sY ( s)  lim  s 2 
 lim  Undefined.
t  s 0 s 0
 (3s  1) s  s 0 (3s  1) s

(c)
For part (a), the heating rate returns to steady state after time 1, the tank
temperature will gradually return to the steady state value once the hotter fluid is
passed out of the stirred tank heater.
Figure S5.21 Step responses for the ODE system, 2nd order t.f and 1st order For part (b), the heating rate rises linearly with time and so does the outlet
approx. temperature. Physical limitations include element burnout, boiling of the liquid,
and constraints on the amount of electrical power available. However, there

5-30 5-31 5-32

should not be any short term physical limitations to the ramp, but it is an unsafe b) T (t )  t  T0 (1)
situation.
Define deviation variables: Tm’ = Tm – Tss = Tm – T0 ; T’ = T – Tss = T – T0,
Step Response
1.4 substitute these into Eq. 1 :

5.24 1.2
1
T ' (t )  t and LT: T ' ( s)  (2)
1 s2

0.8 As known, thermometer can be modeled by a first order system with time
Amplitude

a)
From block algebra, constant 0.1 and gain 1:
0.6

Tm ' ( s ) G 1
𝑌(𝑠) = 𝐺1 (𝑠)𝑈(𝑠) + 𝐺2 (𝑠)𝑈(𝑠) + 𝐺3 (𝑠)𝑈(𝑠) 0.4
  (3) Eq. 5-19 in the book
T ' ( s ) s  1 0.1s  1
0.2
or 𝑌(𝑠) = [𝐺1 (𝑠) + 𝐺2 (𝑠) + 𝐺3 (𝑠)]𝑈(𝑠) Combine Eqs. 2 and 3 :
0
0 2 4 6 8 10 12
After some simple operations, and by account that U(s) = 1, then 1 1 1
Time (seconds)
Tm ' ( s)  T ' (s)   (4)
0.1s  1 0.1s  1 s 2
1 4 −3 4𝑠+1 Apply PFE to Eq. 4:
𝑌(𝑠) = [𝑠 + 2𝑠+1 + 𝑠+1] 𝑈(𝑠) = 𝑠(2𝑠+1)(𝑠+1) Figure S5.24 Step response for part (a)
5.25 a1 a a
1 1 1
or
4𝑠+1 1 4𝑠+1
𝑌(𝑠) = (2𝑠+1)(𝑠+1) × 𝑠 = 2𝑠2 +3𝑠+1 × 𝑠
1 Tm ' ( s )  T ' (s)     2  3
0.1s  1 0.1s  1 s 2 0.1s  1 s s 2
For such an integrating process at steady state, any positive/negative step 1
Notice that this system is equivalent to a step input response of an overdamped a1  2  0.01
s s  10
change in inlet flow will cause the tank level to increase/decrease with time.
( = 1.06) second-order transfer function with numerator dynamics (see 1
a3  1
Example 6.2 in your textbook). Thus, no new steady state will be attained, unless the tank overflows or 0.1s  1 s 0
1 1 a1 a a
For this example, a > 1 (e.g., 4 > 2), so the response will exhibit some empties. set s  10,   0.005;  2  3  0.015  0.1a 2
0.1s  1 s 2 0.1s  1 s s 2
overshoot.
Integrating processes do not have a steady-state gain in the usual sense. Note a 2  0.1
The system poles (-0.5, -1) lie in the LHP, so y(t) will be bounded.
that G(0) is undefined because of dividing by zero. As a result:
Finally,
K 0.01 0.1 1
K  lim G ( s)  lim   Undefined. Tm ' ( s)    2 (5)
s 0 s 0 s 0.1s  1 s s
𝑦(0) = 𝑙𝑖𝑚 𝑠𝑌(𝑠) = 0 𝑦(∞) = 𝑙𝑖𝑚 𝑠𝑌(𝑠) = 1
s⟶∞ s⟶0 Use inverse LT to time domain:
5.26
1 t
Tm ' (t )  0.01 exp(  )  0.1  t
0.1 0.1 Eq. 5-21 in the book
(a) Tm ' (t )  0.1(exp( 10t )  1)  t

At time 0, Tm (0) and T(0) are the same which is t0. Then T (bath temperature) At t = 0.1 min and t = 1.0 min after the change in T(t), the difference would be:
follows a ramp:

5-33 5-34 5-35

 Tm (t )  Tm (0)  0.1(exp( 10t )  1)  t Define deviation variables as: T’(t) = T(t) – T(0) and Tm’(t) = Tm’(t) – T(0) where (b)
T (t )  T (0)  t (6) T(0) = 120°F. Transfer T(t) into T’(s) : when t = 0.5 s, Tm (0.5)  20(1  exp(0.5))  120  127.87F
T ' (t )  T (t )  120  20S (t )  40S (t  10)
Tm  Tm (t )  T (t )  0.1(exp( 10t )  1) when t = 15 s, Tm (2)  20(1  exp(2))  120  100.26F
20 40 10 s
T ' (s)   e (2)
s s
Tm (0.1)  0.0632
Tm (0.1)  0.1 According to the problem, the dynamics of the thermometer follow first order:
Tm '( s ) 1
 (3)
(b) T '( s ) s  1 5.28

By looking at Figure 5.5, the maximum difference occurs when t→∞ and the 1 a a a a4 a5
Y s  2
Combine Eqs. 2 and 3:
corresponding difference is: lim{0.1(exp( 10t )  1)}  0.1  1 2 3  
1 1 20 40 10 s s (2 s  1)3 s s 2  s  1 ( s  1) 2 ( s  1)3
t 
Tm ' ( s )  T ' ( s)  (  e )
0.1s  1 s 1 s s We know that the a3 , a4 , a5 terms are exponentials that go to zero for large values
(c)  20 20  40 40 10 s of time, leaving a linear response.
For large time, exp(-10t) approaches zero and: Tm ' ( s )  (  )(  )e
s 1 s s 1 s 1 a a 1
Tm (t )  Tm (0)  0.1(exp( 10t )  1)  t  Tm (0)  (t  0.1)  2  1  22   a1s  a2
s  2s  1  2s  1
3 3
s s
which indicates there is a 0.1 min time delay between measurement and true value Use inverse LT to time domain:
1
after a long time. Tm ' (t )  20(1  exp( t ))  40(1  exp( (t  10)))S (t  10) (4)  a2  lim 1
 2s  1
s 0 3

Add T(0) back: 1

Define Q  s  
Tm (t )  20(1  exp( t ))  40(1  exp( (t  10)))S (t  10)  120 (5) (2 s  1)3
dQ 6
 
when 0 < t < 10s, Tm (t )  20(1  exp( t ))  120 ; ds (2s  1) 4
when t > 10s, Tm (t )  20(1  exp( t ))  40(1  exp( (t  10)))  120 1  6 
1! s 0  (2s  1)4 
Then a1 = lim

(from Eq. 3-62)

a1 =  6
 the long-time response (after transients have died out) is
y (t )  t  6
We see that the output lags the input by a time equal to 6.
Figure S5.26 T(t) and Tm(t)

5.29

5.27 (a) Energy balance:

dT
ρVc p  UA(TA  T )
dt
The temperature of the bath can be described as:
Figure S5.27 Tm vs. time where ρ is density of water
T (t )  120  20S (t )  40S (t  10) (1) V is volume of water

5-36 5-37 5-38

It takes 5 to reach steady state, thus, the time constant

c p is heat capacity of water
U is heat transfer coefficient 4
Chapter 6
  hr  0.8hr  2880s
A is surface area of tank 5
t is time in mins (b) The interval of step changes for the input should be larger, possibly greater 6.1
Substituting numerical values in and noting that than 4 hours. a)
0.7  s 2  2s  2 
dT dT '
TA  T ;  T ' s K dT
dt dt (c)  ; mc p  UA T  Ta  Because the gain is small and the time G s 
Q  s  s 1 dt s 5  5s 4  9s3  11s 2  8s  6
 dT '
1000   0.52  1 4180   120  60    0.5  1 TA'  T ' 
4 dt constant is large, we can see that the mass, density, heat capacity and furnace
By using MATLAB, the poles and zeros are:
Taking Laplace transform: height are all large.
T s
'
1 Zeros: (-1 +j) , (-1-j)

TA'  s  72.57 s  1 Poles: -3
-1
TA  20   15  20  S  t  ; TA  T  20 (0+j), (0-j)
35
TA'  35S  t   TA'  s    (-1+j), (-1-j)
s
35
T ' s  
s  72.57 s  1 zeros
1.5
Applying inverse Laplace Transform to find when the water temperature reaches poles
0 ̊C. axis
1
T (t )  35(1  et /72.57 )  0  20  35  (1  et /72.57 )
 t  61.45 min 0.5

Imaginary Part
(b) Since the second stage involves a phase change with a constant temperature,
thus, the time spent on phase change can be calculated based on the following 0
equation:
ρV   UA T  TA  t -0.5

 -1
1000kg / m3   0.52 1m3  334 103 J / kg  120W / m2 K    0.5 1m2 15 K  t
4
-1.5

t  386.6 min -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1

So the total time it takes to complete freeze the water in the tank is: Real Part
ttotal  61.45  386.6  448 min Figure S6.1. Poles and zeros of G(s) plotted in the complex s plane.

5.30
b) The process output will be bounded because there is no pole in the right
(a) From the results after 15 hr, we can see: half plane, but oscillations will be shown because of pure imagine roots.
It is a first order system, the gain K is:
c) Simulink results:
0   1 K
K  103 K / kW
0   1000  kW Solution Manual for Process Dynamics and Control, 4th edition
Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III

6-1
5-39 5-40
 6.3 a) Extremum  y (t )  0
Step response
0.5
 1    1  t / 1 1   a   2  t /  2 
Y ( s) K ( a s  1) KM 0   a e   e 0
1  1   2   2  1   2 
M
0.45
 , X ( s)   
0.4
X ( s) (1 s  1) s
From Eq. 6-13 1 1 
0.35
1  a / 2 t   
 
 e  1 2 1 since 1>2
0.3          1   a / 1
y(t) = KM 1  1  a  e  t /1   KM 1  a 1 e  t /1 
Amplitude 0.25   1    1 
0.2
   1   a b) Overshoot  y(t ) > KM
a) y (0  )  KM 1  a  KM
0.15
 1  1
   1 t / 1  a   2 t / 2 
0.1
b) Overshoot  y(t) > KM KM 1  a e  e   KM
0.05  1   2 1   2 
   1 t / 1   1 1
KM 1  a e   KM  a  1 t   
 
0
0 20 40 60 80 100 120 140  1   e  2 1   0 , therefore a>1
Time a  2
Figure S6.1c. Response of the output to a unit step input. or when, a  1 > 0 , that is, a > 1
As shown in Fig. S6.1c, the system is stable but oscillations show up c) Inverse response  y (t )  0 at t = 0+
( a   1 )
because of pure imaginary roots. y   KM et /1  0 for KM  0
 2
1
 1    1  t / 1 1   a   2  t / 2 
KM 0   a e   e   0 at t = 0
+
c) Inverse response  y(t) < 0
 1  1   2   2  1   2  
6.2
4  s  2 8  0.5s  1     1    1  1   a   2 
(a) Standard form: G  s   e 5 s  e 5 s KM 1  a 1 et /1   0   a   0
 0.5s  1 2s  1  0.5s  1 2s  1  1  1  1   2   2  1   2 
(b) Apply zero-pole cancellation:  a  1 a
 e  t /1 or  1  e  t /1  0 at t = 0.
8 1 1 1 1
G s  e5 s  a   

2s  1  Therefore, a < 0.  2 1   0
Gain =8; Pole=-0.5; Zeros=None 1  2
(c)
Since 1 > 2, a < 0.
6.4
1  5 / 2s
1/1 Padéapproximation: e5 s  d) If an extremum in y exists, then from (a):
1  5 / 2s Y ( s) K ( a s  1)
 , 1>2, X(s) = M/s
The transfer function becomes X ( s) (1 s  1)( 2 s  1)  1 1 
 t     1  a 2 
 8  (1  5 / 2s) e  1  2 
  
G( s)    From Eq. 6-15  1   a 1 
 2s  1  (1  5 / 2s)
Gain = 8; poles = -0.5, -0.4; zero = 0.4
   1 t / 1  a   2 t / 2  1  2  1  a 2 
y (t )  KM 1  a e  e  t ln 
 1   2 1   2  1   2  1   a 1 

6-2 6-3 6-4

Put in standard K/ form for analysis:

Q (t )  6(1  e t / 20 )
6.5  K  
K 1    2  s  1
b) R ( s )  Q ( s )  Pm ( s )
 K1 
  
1  22.5s Y ( s)
Using 1/1 Padéapproximation: e45 s  ; G(s) 
1  22.5s U (s) s (s  1)
r (t )  q (t )  p m (t )  p m (t )  p m (0)
1.4  13.5s 1.4 1  9.64 s 
Gp  s   e 30 s  e 30 s a) Order of G(s) is 2 (maximum exponent on s in denominator is 2)
 40s  11  22.5s   40s  11  22.5s  r (t )  p m (t )  12  6(1  e t / 20 )
Gain = 1.4; 0   a   1 , so it is an over damped process, b) Gain of G(s) is K1. Gain is negative if K1 < 0.
r (t  ) 18  12  6(1  0)
c) Poles of G(s) are: s1 = 0 and s2 = –1/ K  6
p (t  )  p (t  0) 42
1.4
True Value s1 is on imaginary axis; s2 is in the left hand plane.
Overshoot,
1.2
d) The zero of G(s) is:
r (t  15)  r (t  ) 27  12  6(1  e15/ 20 )  12
OS    0.514
1 1  K1 r (t  ) 12
sa  
 K  K1  K 2
   2 
0.8
 K 1 
   
OS  exp   0.514 ,   0.2
0.6 K1  1 2 
If  0 , the zero is in right half plane.  
K1  K 2
0.4 Period T for r(t ) is equal to the period for pm(t) because e-t/20 decreases
Two possibilities: monotonically.
0.2 1. K1<0 and K1 + K2 >0
Thus, T = 50  15 = 35
0 e) Gain is negative if K1 < 0
0 50 100 150 200 250 300 T
and  1   2  5.46
Then the zero is RHP if K1 + K2 > 0. This is the only possibility. 2

Figure S6.5. Step response of the system.

Constant term and e-t/ term. Pm ( s) K K
f) c)  
P ( s)  2 s 2  2s  1 s  1
g) If input is M/s, the output will contain a t term that is not bounded.


K  s
2 2
 ( K  2 K ) s  ( K  K )
6.6 ( 2 s 2  2s  1)(s  1)

K1 K K K  6.7
Y ( s)  U ( s)  2 U ( s)   1  2 U ( s) Pm ( s ) %
s s  1  s s  1 d) Overall process gain =  K  K  6  3  3
P ( s ) s 0
psi
2
Y ( s) K1 s  K1  K 2 s ( K1   K 2 ) s  K1 a) p(t )  (4  2)S (t ) , P ( s) 
  s
U ( s) s (s  1) s(s  1) 3 3 2
Q( s)  P ( s) 
20s  1 20s  1 s

6-5 6-6 6-7

6.8
d) Simulink simulation co is the inlet concentration to tank 1.

a) Transfer Function for the blending tank: 1.2 V is the volume of each tank.

K bt q is the volumetric flow rate.

Gbt ( s)  1

 bt s  1
C5 ( s)  C ( s)   1 
5 5

  i
3
where K bt 
qin
 1 and  bt 
2m
 2 min
0.8
  ,
C0 ( s) i 1  Ci1 ( s)   6s  1 
q i 1m 3 / min
0.6
Then, by partial fraction expansion,

Output/Kbt
Transfer Function for the transfer line
 
 t 1  t  2 1  t 3 1  t  4 

0.4 c5 (t )  0.60  0.15 1  et / 6 1           
Gtl ( s) 
K tl
  6 2!  6  3!  6  4!  6  
 
tl s  15 0.2
where:

K tl  1 0

Exact model
0.1m 3 Approximate model
 tl   0.02 min -0.2
5  1m 3 / min 0 5 10 15
Time
20 25 30

Thus,
Figure S6.8. Unit step responses for exact and approximate models.
 ( s)
C out K bt

Cin ( s) (2s  1)(0.02s  1) 5
6.9
which is a 6th-order transfer function.
320 1  4s  e3s 80 1  4s  e 3s
Since bt >> tl [ 2 >> 0.02], we can approximate
1
by e-s (a) G  s   
b)
(0.02s  1) 5 24s 2  28s  4  6s  1 s  1
5 Gain= 80; time delay = 3; time constants  1  6, 2  1 ; poles = -1, -1/6; zeros =
where    (0.02)  0.1 0.25
i 1
(b) Since  a  4  0; it will show an inverse response.
 ( s) K bt e 0.1s
Cout
  6.10
Cin ( s) 2s  1

c) Because bt  100 tl, we anticipate that this approximate TF will yield a) The transfer function for each tank is
results very close to those from the original TF (part (a)). This
approximate TF is exactly the same as would have been obtained using a C i( s ) 1
plug flow assumption for the transfer line. Thus we conclude that  ,i = 1, 2, …, 5
Ci1 ( s )  V 
 q  s  1
investing a lot of effort into obtaining an accurate dynamic model for the
transfer line is not worthwhile in this case.  

Note: if bt  tl , this conclusion would not be valid. where i represents the ith tank.

6-8 6-9 6-10

Denote the undelayed response (for =0) by cm (t ). Then, Let ta denote the time that the alarm goes on for the undelayed system; thus, the alarm lights up when
b) Using Simulink, c m (ta )  25 min; i.e., when c m (ta )  25  5  20 min . .Substituting into (7) and solving for ta by trial
cm (t )  cm (t  ) (1) and error gives
0.6
c5
c4 Taking the Laplace transforms give, ta  9.24 min
0.58
c3
c2 Let ta denote the time that the alarm goes off for the system with time delay. It follows from the
0.56
c1
Cm ( s)  e Cm ( s)
s
(2)
definition of a time delay that,
0.54
Concentration

The transfer functions for the delayed and undelayed systems are:
ta  ta    9.24  2.00  11.24 min
0.52
Cm ( s ) e s
 (3)
0.5 C ( s ) s  1
6.12
0.48 Cm ( s ) 1
 (4)
C ( s ) s  1
0.46 a) Using Skogestad’s method

For the ramp input, c(t )  2t; from Table 3.1: 5e (210.2) s 5e3.2 s
0.44 G s  
0 5 10 15 20 25 30 35 40 45 50
(12s  1) 12s  1
time
2
C ( s)  (5)
Figure S6.10. Concentration step responses of the stirred tank. s2
Using Simulink,
The value of the expression for c5(t) verifies the simulation results above:
Substituting (5) into (4) and rearranging gives: 5

  5 5 5  2 3 4
c5 (30)  0.60  0.15 1  e 5 1  5      0.5161
 2! 3! 4!   1  2 
 Cm ( s)  
4
 2  (6)
 s  1   s 
3

The corresponding response to the ramp input is given by Eq. 5-19 with K =1, a =2, and =10:
6.11 2

cm (t )  20 (e  t /10  1)  2t (7)

First, consider then the undelayed response (with =0); then apply the Real 1

Translation Theorem to find the desired delayed response.

0

True
Approximate
-1
0 5 10 15 20 25 30 35 40 45 50

Figure S6.12 Unit step responses for the exact and approximate models.

6-11 6-12 6-13

  Kb Kc  6.14
 
(c) Maximum error =0.265, at t= 9.89s, and the location corresponding to the P2 ( s ) Kb Kc 1 Ka Kc 
  (5) 4e  s 4e  s
maximum error is graphically shown in above figure by black vertical line. Pd ( s ) (1 s  1)( 2 s  1)  K a K c  1  2  2  1   2  Let G ( s)  
  s    s  1  0.4s  1 (2 s 2  3s  1)  0.4s  1 (2 s  1)( s  1)
2 2

1 Ka Kc  1 Ka Kc 
6.13
Substituting for P2 ( s ) from Eq. 5 into 4, We want an approximate model of the form,
From the solution to Exercise 2.5(a) , the dynamic model for isothermal
operation is  Kb  Ke  s
  (2 s  1) Gapprox ( s) 
V1 M dP1 Pd  P1 P1  P2
P1( s )
  1  Ka Kc  (6)
 s 1
  (1) Pd ( s )  12  2  1  2 
RT1 dt Ra Rb  s   s 1
 1  Ka Kc   1  Ka Kc  In order for the approximate model and the original models to have the
V2 M dP2 P1  P2 P2  Pf same steady-state gain, we set K  4.
  (2) To determine whether the system is overdamped or underdamped, consider the
RT2 dt Rb Rc
denominator of the transfer functions in Eqs. 5 and 6.
The largest time constant in G (s) to neglect is 1. Thus,
Taking Laplace transforms, and noting that
1
  2    1  2.5
Pf ( s)  0  1  2  1   2 2
 2    , 2  1  K K
1 Ka Kc  a c
Approximate the smallest time constant by:
since Pf is constant, 1
 e 0.4 s
0.4 s  1
K P  ( s)  K a P2( s) Therefore, Thus,
P1( s)  b d (3)
1 s  1 1
1 (1   2 ) (1  K a K c ) 1  1 2  1     ( ) 2  2 (0.4)    1.8
K P ( s)      2
P2( s)  c 1 (4) 2 (1  K a K c ) 1  2 2  2 1  (1  K a K c )
2 s  1
where
Since x + 1/x  2 for all positive x,
6.15
K a  Ra /( Ra  Rb ) 
1
K b  Rb /( Ra  Rb ) (1  K a K c )
From Eqs. 6-71 and 6-72,
K c  Rc /( Rb  Rc ) Since KaKc  0,
R2 A2  R1 A1  R2 A1 1  R1 A1 R2 A2  1

R2 A1
 1   
V1 M Ra Rb 2 R1 R2 A1 A2 2  R2 A2 R1 A1  2 R1 A2
1 
RT1 ( Ra  Rb ) Hence the system is overdamped.
1
Since x   2 for all positive x and since R1, R2, A1, A2 are positive
V2 M Rb Rc x
2 
RT2 ( Rb  Rc )

1
2  1 R2 A1  1
Substituting for P1 ( s ) from Eq. 3 into 4, 2 2 R1 A2

6-14 6-15 6-16

Q2(s) 1 algebraic equations to obtain the coefficients of the four partial fractions. The

6.16 Q0(s)  2 s  1 second method requires that the numerator and denominator be defined as
coefficients of descending powers of s prior to calling the MATLAB residue
function:
a) Mass balance on Tank 2: b) Mass balances on the two tanks yield (after dividing by , which is constant)

dh2 dh dh MATLAB Commands:
A2  q0  q2 A1 1  q1 A2 2  q0  q1  q2
dt dt dt
>> b = [ 36*0.262 72*0.262]
Dividing by , Valve resistance relations: b=
dh2 9.4320 18.8640
  1  1
A2  q0  q2 q1  (h1  h2 ) q2  h2 >> a = conv([10 1], conv([5 1], [1 0 0.262^2]))
dt R1 R2
b=
For a linear resistance, (cf. Eq. 4-50), c) These equations clearly describe an interacting second-order system; one 50.0000 15.0000 4.4322 1.0297 0.0686
1 or more transfer functions may contain a single zero (cf. Section 6.4). For >> [r,p,k] = residue(b,a)
 q2  h2 the Q2/Q0 transfer function
 we know that the steady-state gain must be r=
R2
equal to one by physical arguments (the steady-state material balance 6.0865  4.9668j
Substitute, around the two tank system is q2  q0 ). 6.0865 + 4.9668j
38.1989
 dh2 1
A2  q0  h2 d) The response for Case (b) will be slower because this interacting system is 50.3718
dt R2
second order, instead of first order. p=
or
dh2 0.0000  0.2620j
A2 R2  R2q0  h2 0.0000 + 0.2620j
dt
 0.2000
6.17
0.1000
Introducing deviation variables and Laplace transforming yields
 k=
The input is Ti(t )  12 sin t where []
H2(s) R2
 2 radians
Q0(s) A2R2 s 1   0.262 hr 1 Note: the residue function re-computes all the poles (listed under p). They
24 hours are, in reverse order: p1 = 0.1( 1  10) , p = 0.2(  2  5) , and the two purely
Because
imaginary poles corresponding to the sine and cosine functions. The
1 The Laplace transform of the input is from Table 3.1,
 Q2(s)  H 2(s)
residues (listed under r) are exactly the coefficients of the corresponding
R2 12 poles; in other words, the coefficients that would have been obtained via a
Ti( s)  manual partial fraction expansion. In this case, we are not interested in the
s 2  2
real poles since both of them yield exponential functions that go to 0 as
we obtain,
 Multiplying the transfer function by the input transform yields t .
Q2(s) 1 R2 1
 
Q0(s) R2 A2R2s 1 A2R2s 1 (72  36s) The two complex poles are interpreted as the sine/cosine terms using
Ti( s) 
(10s  1)(5s  1)( s 2  2 ) Appendix L.

Letting  2  A2 R2 To invert, either (i) make a partial fraction expansion manually, or (ii) use the The coefficients of the periodic terms:

MATLAB residue function. The first method requires solution of a system of

6-17 6-18 6-19

 a2 6.19 F0  RF0  F0 
y (t )  a1ebt cos t  e bt sin t  ... C2'  s   C0'  s 
 V 2 s 2   RF0  RF0  2 F0  Vs   RF0  F0  F0

b= 0, thus the exponential terms = 1. Using (L-13) and (L-15),   0.264 . (a) The mathematical model is derived based on material balance:  1
F02 1  
dc   R C0'  s 
V 1  F0 c0  RF2 c2  F1c1  Vkc1
y(t )  12.136 cos t  9.9336sin t  ... dt V 2 s 2 2 F0 F2
 Vs  2 F0Vs  F02  0
dc R R R
The amplitude of the composite output sinusoidal signal, for large values, V 2  F1c1  1  R  F2c2  Vkc2 F0
dt When R   , C2'  s   C0'  s  , equivalent to a single tank with a
of t is given by
Subtracting the steady-state equation and substituting deviation variables 2Vs  F0
yields: volume = 2V.
A  (12.136) 2  (9.9336) 2  15.7 The gain of above transfer function is 1.
dc '
V 1  F0c0'  RF2c2'  F1c1'  Vkc1'
dt
Thus the amplitude of the output is 15.7 for the specified 12 amplitude
dc ' 6.20
input. V 2  F1c1'  1  R  F2c2'  Vkc2'
dt
The dynamic model for the process is given by Eqs. 2-45 and 2-46,
which can be written as
(b) The transfer function model can be derived based on Laplace transform:
6.18
dh 1
VsC1'  s   F0C0'  s   RF2C2'  s   F1C1'  s   VkC1'  s   ( wi  w) (1)
dt A
150 VsC2'  s   F1C1'  s   1  R  F2C2'  s   VkC2'  s 
dT w Q
148 Solve above equations, we have:  i (Ti  T )  (2)
dt Ah AhC
146 F0 F1
y1 C s  2 2
'
C0'  s  where h is the liquid-level
V s   2Vk   F1  F2  RF2   Vs  F1F2  Vk 1  R  F2  FVk  V 2k 2
2
y2 A is the constant cross-sectional area
144 1

System outputs: h , T
142 (c) When R  0 , we have:
System inputs : w, Q
F0 F1
C2'  s   C0'  s 
y 140
V 2 s 2   2Vk   F1  F2   Vs  F1F2  VkF2  FVk
1  V 2k 2 Assume that wi and Ti are constant. In Eq. 2, note that the nonlinear term
138
F0 F1  dT 
 C' s h  can be linearized as
136 Vs  Vk  F1 Vs  Vk  F2  0  dt 
134 which is equivalent to the transfer function of the two tanks connected in
dT  dT
series. h  h
132 dt dt
130 (c) When k=0, Equation in (b) becomes:
0 1 2 3 4 5 6 7 8 dT  dT
or h since 0
t F0 F1 dt dt
C2'  s   C0'  s 
V 2 s 2   F1  F2  RF2 Vs  F1F2
Figure S6.18 Comparison between y1 and y2 Then the linearized deviation variable form of (1) and (2) is
Since F1  RF2  F2 , F0  F2
dh  1
 w
dt A

6-20 6-21 6-22

dV CAT h
dT   wi 1
Eq. 2 is simplified by substituting for from Eq. 1. Also, replace V by AT h Let  
 T Q dt CqF  UpT h
dt Ah Ah C (where AT is the tank area) and replace A by pT h (where pT is the perimeter of
the tank). Then, Then from Eq. 7
Taking Laplace transforms and rearranging,
dh H ( s) 1 H ( s) 1 H ( s )
H ( s ) K1 H ( s ) T ( s ) T ( s) K2 AT  qF  q (3)  ,  , 0
 , 0 , 0 ,  dt QF ( s) AT s Q( s ) AT s Ts ( s )
W ( s ) s Q ( s ) W ( s ) Q ( s)  2 s  1
dT
CAT h  qF C (TF  T )  UpT h(Ts  T ) (4) And from Eq. 9
1 1 Ah dt
where K 1   ; and K 2  , 2 
A wi C wi  UpT (Ts  T )   C (TF  T ) AT  
Then, Eqs. 3 and 4 are the dynamic model for the system.     s  1
T ( s )  CqF  UpT h   UpT (Ts  T )  
For an unit-step change in Q: h(t )  h T (t )  T  K 2 (1  e t / 2 ) 
, a) Making a Taylor series expansion of nonlinear terms in (4) and introducing QF ( s )  AT s  s  1
deviation variables, Eqs. 3 and 4 become:
For an unit step change in w: h(t )  h  K 1t , T (t )  T
dh  UpT (Ts  T ) 
AT  q F  q  (5)  
dt T ( s )
  CqF  UpT h 
6.21 Q( s )  AT s  s  1
dT 
Additional assumptions: CAT h  C (TF  T )qF  (CqF  UpT h )T   UpT hTs  UpT (Ts  T )h (6)
dt  UpT h 
(i) The density  and specific heat C of the liquid are constant.  
T ( s )  CqF  UpT h 
Taking Laplace transforms, 
Ts( s ) s  1
(ii) The temperature of steam, Ts, is uniform over the entire heat transfer
1 1
area. H ( s)  QF ( s)  Q( s) (7)
AT s AT s
(iii) The feed temperature TF is constant (not needed in the solution).
 CAT h    C (TF  T ) 
Mass balance for the tank is   s  1 T ( s )    QF ( s )
 CqF  UpT h    CqF  UpT h 
dV  UpT h   UpT (Ts  T ) 
 qF  q (1)   Ts( s)    H ( s) (8)
dt  CqF  UpT h   CqF  UpT h 
Energy balance for the tank is Substituting for H ( s) from (7) into (8) and rearranging gives
d [V (T  Tref )]
C  q F C (TF  Tref )  qC (T  Tref )  UA(Ts  T ) (2)  CAT h    C (TF  T ) 
dt  AT s    s  1 T ( s )   AT s  QF ( s )
 CqF  UpT h    CqF  UpT h 
where Tref is a constant reference temperature and A is the heat transfer area
 UpT hAT s   UpT (Ts  T ) 
  Ts( s)    QF ( s)  Q( s) (9)
 CqF  UpT h   CqF  UpT h 

6-23 6-24 6-25

Note: and the gain is positive since T  TF  0 . Furthermore, it has dimensions of
C (TF  T ) AT temperature/volume. 6.22
2  is the time constant in the numerator.
UpT (Ts  T )
UpT h a) The two-tank process is described by the following equations in deviation
(The ratio is dimensionless).
Because TF  T  0 (heating) and Ts  T  0 ,  2 is negative, we can show this CqF variables:
property by using Eq. 2 at steady state:
(b) The h  qF transfer function is an integrator with a positive gain. Liquid level dh1' 1  ' 1 ' '
 w1  (h1  h2  (1)
CqF (TF  T )  UpT h (Ts  T ) accumulates any changes in qF , increasing for positive changes and vice-versa. dt A1  R 

UpT h (Ts  T ) h  q transfer function is an integrator with a negative gain. h accumulates dh2' 1 1 ' ' 
or C (TF  T )   (h1  h2  (2)
qF changes in q, in the opposite direction, decreasing as q increases and vice dt A2  R 
Substituting versa.
hAT Laplace transforming
2   h  Ts transfer function is zero. Liquid level is independent of Ts and
qF
steam pressure Ps . A1RsH1' ( s)  RWi ' ( s)  H1' ( s)  H 2' ( s) (3)
V
Let V  hAT so that 2   =  (initial residence time of tank)
qF T  q transfer function is second-order due to the interaction with liquid A2 RsH ( s)  H ( s)  H ( s)
' ' '
(4)
T ( s )
2 1 2
T ( s ) level; it is the product of an integrator and a first-order process.
For and the “gain” in each transfer function is
QF ( s ) Q( s ) From (4)
T  qF transfer function is second-order due to the interaction with liquid
 Up (T  T )  level; it has numerator dynamics since qF affects T directly as well if (A2 Rs  1) H 2' ( s)  H1' ( s) (5)
K  T s

 AT  CqF  UpT h  
TF  T .
or
T  Ts transfer function is first-order because there is no interaction with H 2' ( s) 1 1
and must have the units of temperature/volume . (The integrator s has units of   (6)
liquid level. H1' ( s) A2 Rs  1 2 s  1
t-1).

To simplify the transfer function gain, we can substitute c) h  qF : h increases continuously at a constant rate. where 2  A2 R

CqF (TF  T ) h  q : h decreases continuously at a constant rate. Returning to (3)

UpT (Ts  T )  
h
h  Ts : h stays constant. (A1Rs  1) H1' ( s)  H 2' ( s)  RWi ' ( s) (7)
from the steady-state relation. Then
T  qF : for TF  T , T decreases initially (inverse response) and then Substituting (6) with 1  A1 R
CqFT (TF  T )
K increases. After long times, T increases like a ramp function.
hAT  CqF  UpT h   1  '
(1s  1)   H1 ( s)  RWi ( s)
'
(8)
T  q : T decreases, eventually at a constant rate.  2 s  1 
T  TF
or K  T  Ts : T increases with a first-order response and attains a new steady
 Up h  or
V 1  T  state.
 CqF 

6-26 6-27 6-28

(12 ) s 2  (12 ) s  H1' ( s )  R(2 s  1)Wi ' ( s ) (9) dh' 1 ' - However, the two-tank TF’s contain a pole (3 s  1) that will “filter
 wi (16)
H1' ( s ) R(2 s  1) dt A out” changes in level caused by changing wi(t).
 (10)
W1' ( s) s  12 s  (1  2 )  - On the other hand, for this special case, we see that the zero in the first
Note that  , which is constant, subtracts out.
tank transfer function ( H i' ( s) / Wi ' ( s)) is larger than the pole:
Dividing numerator and denominator by (1  2 ) to put into standard form
Laplace transforming and rearranging: 2 3   3
H1' ( s ) [ R /(1  2 )](2 s  1) Thus we should make sure that amplification of changes in h1(t)
 (11)
W1' ( s)    H ( s) 1/ A
'
caused by the zero do not more than cancel the beneficial filtering of
s  1 2 s  1  (17)
Wi ' ( s)
 1  2
s the pole so as to cause the first compartment to overflow easily.

Again Now look at more general situations of the two-tank case:
Note that 1
K
A
H1' ( s) K (A2 Rs  1) K (2 s  1)
K
R

R

1

1
(12) H ' ( s) K   (20)
1  2 A1R  A2 R ( A1  A2 ) A  (18) Wi ' ( s)  RA1 A2  s(3 s  1)
Wi ' ( s) s s s  1
 A 
since A  A1  A2 H 2' ( s) K
which is the expected integral relationship with no zero.  (21)
Wi ' ( s) s(3 s  1)
Also, let
For either A1  0 or A2  0 ,
12 2 R 2 A1 A2 RA1 A2 b) For A1  A2  A / 2
s    (13)
1  2 R ( A1  A2 ) A RA1 A2
3  0
2  AR / 2  A
 (19)
so that 3  AR / 4  Thus the beneficial effect of the pole is lost as the process tends to
look more like the first-order process.
H1' ( s) K (2 s  1) Thus 2  23
 (14)
Wi ' ( s) s(3 s  1) c) The optimum filtering can be found by maximizing  3 with respect
We have two sets of transfer functions: to A1 (or A2)
and
One-Tank Process Two-Tank Process RA1 A2 RA1 ( A  A1 )
3  
H 2' ( s) H 2' ( s) H1' (s) 1 K (2 s  1) A A
 
Wi ' ( s) H1' ( s) Wi ' ( s) (2 s  1) s(3 s  1) H ' ( s) K H i' ( s) K (23 s  1)
  3 R

K Wi ' ( s) s Wi ' ( s) s(3 s  1) Find max 3 :  ( A  A1 )  A1 (1)
(15) A1 A
s (3 s  1)
H 2' ( s) K
 Set to 0: A  A1  A1  0
Wi ' ( s) s(3 s  1)
Transfer functions (6), (14) and (15) define the operation of the two-tank
process. Remarks: 2 A1  A

The single-tank process is described by the following equation in - The gain ( K  1/ A) is the same for all TFs. A1  A / 2
deviation variables:
- Each TF contains an integrating element. Thus the maximum filtering action is obtained when A1  A2  A / 2.

6-29 6-30 6-31

 2  2    0.5 1 = 1:
The ratio of 2 / 3 determines the “amplification effect” of the zero on
For the second-order process element with 2 = 2 and this degree of
3.5
h1 (t ). a)
underdamping (  0.5) , the small time constant, critically damped 2nd- 3

2 A2 R A order process element (1 = 0.1) will have little effect.
 
3 RA1 A2 A1
2.5

In fact, since 0.1 << 2 (= 2) we can approximate the critically damped

A
element as e 21 so that 2

Output
As A1 goes to 0, 2   Ke 0.2 s
1.5

3 G ( s) 
4s 2  2s  1 1

Therefore, the influence of changes in wi (t ) on h1 (t ) will be very large,

b) From Fig. 5.10 for   0.5 , OS  0.15 or from Eq. 5-51 0.5

leading to the possibility of overflow in the first tank.

   
Overshoot = exp    0.163 0

Summing up:  2 
Exact model

 1  
Approximate model
-0.5
0 5 10 15 20 25 30

The process designer would like to have A1  A2  A / 2 in order to obtain Time

Hence ymax = 0.163 KM + KM = 0.163 (1) (3) + 3 = 3.5
the maximum filtering of h1 (t ) and h2 (t ). However, the process response Figure S6.23b Step responses for exact and approximate models; 1 = 1.
should be checked for typical changes in wi (t ) to make sure that h1 does
c) From Fig. 5.3, ymax occurs at t/ = 3K or tmax = 6.8 for an underdamped 1 = 5:
not overflow. If it does, area A1 needs to be increased until it is not a 2nd-order process with   0.5 .
problem. 3.5

Adding in the effect of the time delay t = 6.8 + 0.2 = 7.0

Note that 2  3 when A1  A , thus a careful study
3
(simulations)
should be made before designing the partitioned tank. Otherwise, leave d) By using Simulink 2.5

wellenough alone and use the non-partitioned tank.

1 = 0.1:
2

Output
3.5 1.5
6.23
3
1

2.5
The process transfer function is 0.5

Y ( s) K 0
 G( s) 

Output
1.5
(0.1s  1) 2 (4s 2  2s  1)
Exact model
U ( s) Approximate model
-0.5
1 0 5 10 15 20 25 30

Time
where K = K1K2. 0.5

Figure S6.23c Step response for exact and approximate models ; 1 = 5.

The quadratic term describes an underdamped 2nd-order system since 0
Exact model

-0.5
Approximate model
As is apparent from the plots, the smaller 1 is, the better the quality of the
2  4  2 approximation. For large values of 1 (on the order of the underdamped
0 5 10 15 20 25 30
Time

element's time scale), the approximate model fails.

Figure S6.23a Step response for exact and approximate models; 1 = 0.1.

6-32 6-33 6-34

6.26
6.24 6.25

Applying numerical values, equations for the three-stage absorber are:

0
The system equations are:
dx1
-0.2
dh1  0.881y f  1.173x1  0.539x2
1 1
A1  qi  h1 , q1  h1 dt
-0.4
dt R1 R1
 dx2
dh
A2 2 
1
h1 
1
h2 , q 2 
1
h2  0.634x1  1.173x2  0.539x3
-0.6
dt R1 R2 R2 dt
Output

dx3
-0.8
Using a state space representation,  0.634x2  1.173x3  0.539x f
dt
-1
x  Ax  Bu y i  0.72xi
y = Cx + Du
-1.2

 h  Transforming into a state-space representation form:

where x   1 , u  qi and y  q 2
-1.4
0 50 100 150 200 250 300 350 400 h2 
Time
then,  dx1 
 dt 
Fig. S6.24. Unit step response in blood pressure.  
 dh1   1     1.173 0.539 0   x1  0.881
 1   h1   A   dx2    0.634  1.173 0.539  x    0 y
 dt   0
     1  q  dt     2    f
The Simulink- block diagram is shown below     R1 A1 
  1 1      i   
 0 0.634  1.173  x3   0 
    0   dx 
 dh2   R1 A1  h2 
R2 A2     
 dt 
3
   dt 
-1
 h1 
40s+1  1    y1  0.72 0   x1  0 
q 2  0 
0
R2     y    0 0  x   0 y
Step Transfer Fcn Transport

 h2 
Delay1= 30 s
 2   0.72  2    f
 y 3   0 0.72  x3   0 
Scope
-0.4 0
40s+1 Therefore,
Step1 Transfer Fcn1 Transport Therefore, because xf can be neglected in obtaining the desired transfer
Delay = 75 s
 1  functions,
 1   A 
 R A 0   1   1   1.173 0.539 0  0.881
The system appears to respond approximately as a first-order system or A  1 1  , B  , C0  , E 0
overdamped second-order process with time delay.  1

1     R2  A   0.634  1.173 0.539  B   0 
 RA R2 A2   0 
 1 1    0 0.634  1.173  0 

0.72 0 0  0 
C   0 0.72 0  D   0 

 0 0 0.72  0 

6-35 6-36 6-37

 Applying the MATLAB function ss2tf , the transfer functions are:
Chapter 7 b)

Y1( s ) 0.6343s 2  1.4881s  0.6560 dh h(0.2)  h(0) 5.75  5.50 ft ft

    1.25
Y f ( s ) s 3  3.5190s 2  3.443s  0.8123 7.1 dt t 0 0.2  0 0.2 min min

Y2( s) 0.4022s  0.4717 In the absence of more accurate data, use a first-order transfer function: Using Eq. 7-15,

Y f ( s) s 3  3.5190s 2  3.443s  0.8123
T '( s) Kes KM 0.339  (30.1 0.1)
    0.82 min
Qi '( s) s  1  dh  1.25
Y2( s) 0.2550  
  dt t 0 
Y f ( s) s 3  3.5190s 2  3.443s  0.8123 T ()  T (0) (124.7  120) F o
K   0.235  h(t )  h(0) 
qi 520  500 gal/min The slope of the linear relation between ti and z i  ln 1  i
c)  gives
 = 3:08 am – 3:05 am = 3 min  h()  h(0) 
an approximation of (-1/), according to Eq. 7-13.
Assuming that the operator logs a 99% complete system response as “no change
after 3:34 am”, five time constants elapse between 3:08 and 3:34 am. Using h() = h(5.0) = 6 .52, the values of zi are
5 = 3:34 min  3:08 min = 26 min
 = 26/5 min = 5.2 min ti zi ti zi
Therefore, 0.0 0.00 1.4 -1.92
T '( s) 0.235e3s 0.2 -0.28 1.6 -2.14
 0.4 -0.55 1.8 -2.43
Qi '( s) 5.2s  1 0.6 -0.82 2.0 -2.68
To obtain a better estimate of the transfer function, the operator should log more 0.8 -1.10 3.0 -3.93
data between the first change in T and the new steady state. 1.0 -1.37 4.0 -4.62
1.2 -1.63 5.0 -
7.2
Then the slope of the least squares fit, using Eq. 7-6 is
Process gain,
 1  13Stz  St S z
slope      (1)
h(5.0)  h(0) 6.52  5.50 min    13Stt  ( St )
2
K   0.339 2
qi 30.1 0.1 ft
where the datum at t = 5.0 has been ignored.

a) Output at 63.2% of the total change Using definitions,

= 5.50 + 0.632(6.52-5.50) = 6.145 ft St  18.0 Stt  40.4

S z  23 .5 Stz  51.1
Interpolating between h = 6.07 ft and h = 6.18 ft
Substituting in (1),
(0.8  0.6)
  0.6  (6.145  6.07) min  0.74 min
(6.18  6.07)  1
    1.213   0.82 min
 
Solution Manual for Process Dynamics and Control, 4th edition
Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III d)

6-38 7-1 7-2

Let z1 and z2 be the natural log of the fraction incomplete response for T1
6.8 and T2, respectively. Then, 28

26

 T (50)  T1 (t )  18  T1 (t ) 
6.6
z1 (t )  ln  1   ln  
 T1 (50)  T1 (0)  
24
8
6.4 22

 T (50)  T2 (t )   26  T2 (t ) 
z2 (t )  ln  2   ln  
20

2
 T2 (50)  T2 (0) 

, T
6.2  6

1
T
18

6 A plot of z1 and z2 versus t is shown below. The slope of the z1 plot is 16

– 0.333; hence (1/-1)= - 0.333 and 1 = 3.0

Experimental data 14 T1
Model a)
5.8 T2
Model b)
Model c)
From the best-fit line for z2 versus t, the projection intersects z2 = 0 at 12 T1 (experimental)
t  1.15. Hence 2 = 1.15. T2 (experimental)
5.6 10
0 2 4 6 8 10 12 14 16 18 20 22
T1 ' ( s ) 2.667 time
 (2)
5.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Q' ( s ) 3s  1 Figure S7.3b Comparison of experimental data and models for a step change.

T2 ' ( s) 0.75
Figure S7.2 Comparison between models a), b) and c) and the step response data.  (3) 7.4
T1 ' ( s) 1.15s  1
2 1.5
0.0 Y ( s)  G ( s) X ( s)  
0 5 10 15 20 (5s  1)(3s  1)( s  1) s
7.3 -1.0

-2.0
Taking the inverse Laplace transform,
-3.0
a)
y(t )  (-75/8) exp(-t /5) + (27/4) exp (-t /3) - (3/8) exp(-t ) + 3
z 1,z 2

(1)
T1( s) K1 T2( s ) K2 -4.0
 
Q( s) 1s  1 T1( s ) 2 s  1 -5.0
a) Fraction incomplete response
-6.0

T2( s) K1 K 2 K K e2 s -7.0

 y (t ) 
  1 2 (1) z (t )  ln 1 
Q( s) (1 s  1)(2 s  1) 1 s  1 3 
-8.0

time,t

where the approximation follows from Eq. 6-58 and the fact that 1>2, as
revealed by an inspection of the data. Figure S7.3a z1 and z2 as a function of t

T (50)  T1 (0) 18.0  10.0 b) Using Simulink-MATLAB, the following results are obtained:
K1  1   2.667
q 85  82

T2 (50)  T2 (0) 26.0  20.0

K2    0.75
T1 (50)  T1 (0) 18.0  10.0

7-3 7-4 7-5

 The models are compared in Fig. S7.4b: So t1  0.5355; t2  2.017;
0.0
-1.0 0 10 20 30 40 50 2.5   1.3t1  0.29t2  0.1112
-2.0   0.67  t2  t1   0.9926
-3.0
-4.0 2
1

z(t)
-5.0 True data
0.9 Approximate data
-6.0
1.5
0.8
-7.0
-8.0 z(t) = -0.1791 t + 0.5734 0.7
Third-order model

y(t)
First order model
-9.0 Second order model 0.6

Response
time,t 1
0.5

0.4
Figure S7.4a Fraction incomplete response; linear regression
0.5 0.3

0.2
From the plot: slope = - 0.179 and intercept  3.2 0.1
0
0 5 10 15 20 25 30 35 40
0
Hence, time,t
0 5 10 15
Time/s
Figure S7.4b Comparison of three models for a step input
-1/ = -0.179 and  = 5.6
Figure S7.5b Comparison of true data and approximate model
 = 3.2 7.5
2e 3.2 s Sum of squared error = 0.0232
G(s)  1
5.6s  1 For a first-order plus time-delay model G  e s , assume   1;  0.1,1,10 ,
 s 1 b)  /   1 :
b) In order to use Smith’s method, find t20 and t60: we have:
a)  /   0.1 : 1
X: 2.918
y(t20)= 0.2  3 =0.6 0.9 Y: 0.8531

1
0.8
y(t60)= 0.6  3 =1.8 0.9
0.7
0.8 X: 2.017
Y: 0.853 0.6
Using either Eq. 1 or the plot of this equation, t20 = 4.2 , t60 = 9.0

Response
0.7
0.5
0.6
Using Fig. 7.7 for t20/ t60 = 0.47 X: 1.436

Response
0.4 Y: 0.3537

0.5

= 0.65 , t60/= 1.75, and  = 5.14 X: 0.5355 0.3

0.4 Y: 0.3531
0.2
0.3
2 0.1
G( s) 
26.4s  6.68s  1 2 0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.1
Time/s

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time/s Figure S7.5c Plot of the true data;  /   1

Figure S7.5a Plot of the true data;  /   0.1 So t1  1.436; t2  2.918;

7-6 7-7 7-8

  1.3t1  0.29t2  1.021 So t1  10.44; t2  11.92; 1

G( s) 
  0.67  t2  t1   0.9929   1.3t1  0.29t2  10.12 (4.8s  1) 2
  0.67  t2  t1   0.9916
1 Nonlinear regression
True data
0.9 Approximate data 1 From Figure E7.5, we obtain these values (approximate):
0.8 True data
0.9 Approximate data
0.7 Table Output values from Figure E7.5
0.8
0.6
Time Output
Response

0.7 0.0 0.0

0.5
2.0 0.1
0.4 0.6 4.0 0.2
Response

5.0 0.3
0.3 0.5 7.0 0.4
8.0 0.5
0.2
0.4 9.0 0.6
11.0 0.7
0.1
0.3 14.0 0.8
0 17.5 0.9
0 5 10 15 30.0 1.0
Time/s 0.2

0.1
For the step response of Eq. 5-48, the time constants were calculated so as
Figure S7.5d Comparison of true data and approximate model to minimize the sum of the squares of the errors between data and model
0 predictions. Use Excel Solver for this Optimization problem:
0 5 10 15
Sum of squared error = 0.1050 Time/s
1 =6.76 min and 2 = 6.95 min
Figure S7.5f Comparison of true data and approximate model
1
c)  /   10 : G ( s) 
Sum of squared error = 4.3070 (6.95s  1)(6.76s  1))
1
The models are compared in Fig.S7.6:
0.9

0.8 X: 11.92
Y: 0.853
7.6
0.7

0.6
Response

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

0.4
X: 10.44
Y: 0.3528
intersection with y(t) = 0 line is at t  1 and with the y(t)=1 line at t  14.
Hence  =1 and  = 141=13
0.3
e s
0.2 G1 ( s ) 
13s  1
0.1

0
0 5 10 15
b) Smith’s method
Time/s

From the plot, t20 = 3.9 , t60 = 9.6 ; using Fig 7.7 for t20/ t60 = 0.41
Figure S7.5e Plot of the true data;  /   10
 = 1.0 , t60/= 2.0 , hence  = 4.8 and 1 = 2 =  = 4.8

7-9 7-10 7-11

 Assuming plug-flow in the pipe with constant-velocity, (b) Nonlinear regression
1

 p s 3 1 By using deviation variables, the first order tank can be expressed as

0.9 G pipe ( s )  e , p    0.1min
0.5 60 H ' s K
0.8 
Assuming that the thermocouple has unit gain and no time delay Q'  s   s 1
0.7

1 The inlet flow rate is quickly changed from 1.5 gallon/min to 4.8 gallon/min so it
GTC ( s )  since 2  1
0.6
is a step change, Q’(s) 3.3/s:
2 s  1

Output
0.5

H ' s 
K
Q'  s  
K  4.8  1.5  0.1337  K 0.4412
0.4 Then  s 1  s 1 s  s 1 s
0.3
3 s
10e Apply the inverse Laplace transform:
GHE ( s)  ,
1 s  1
h'  t   0.4412 K 1  et / 
0.2

0.1 Non linear regression model so that,

First-order plus time delay model
Second order model (Smith's method)
By using EXCEL, the estimated model is:
 10e  0.1s  1 
3 s
0
0 5 10 15 20 25 30 35 40 45 50
G ( s)  GHE ( s)G pipe ( s )GTC ( s )    (e )   H ' s
time

 1s  1   2 s  1  
K

46.31
Figure S7.6 Comparison of three models for unit step input Q'  s   s 1 2.65s  1

7.8 A comparison of the data and the two models is shown in Fig.S7.8.

7.7
(a) 63% response method
a) From the plot, time delay  = 4.0 min From inspection of the data, it is obvious that there is no time delay in the system
(=0).
Using Smith’s method,
Time constant  is estimated by the 63% response method:
from the graph, t20    5.6 , t60    9.1
h'    0.63  hss'  h0'   0.63* 20.3  12.78 ft
t20  1.6 , t60  5.1 , t20 / t60  1.6 / 5.1  0.314 h    12.78  10.4  23.18 ft

From Fig.7.7 ,   1.63 , t60 /   3.10 ,   1.645 From inspection at the data,  ≈ 270 min.
The process gain is calculated as:
Using Eqs. 5-45 and 5-46, 1  4.81 min , 2  0.56 min
hss'  h0' 20.3  0 20.3
K    46 min/ ft 2 Figure S7.8 A comparison of the step responses of the data and the two models.
q  q0  4.8  1.5   0.1337 0.4412
b) Overall transfer function
The estimated process model is:
10e4 s The Sum of Squared Errors for the two models are:
G( s)  , 1  2
(1s  1)(2 s  1) H ' s K 46
  SSE (63.2%) = 0.75
Q'  s   s 1 2.70s  1

7-12 7-13 7-14

SSE (NR) = 0.43 7.10 So the FOPTD model of the process is

As indicated in Fig.S7.8, both methods fit the data well. The NR model is  0.8 e2 s
preferred due to its smaller SSE value. G( s) 
Assume that T() = T(13) = 890 C. The steady-state gain K is the change in 3.46 s  1

output divided by the change in input:

7.11
7.9 890 – 850
K = 950 – 1000 = – 0.8 C/cfm
y 3  0 (a) For a SOPTD model shown in below,
K   0.75 Assume that the input change in air flow rate is made at t = 2+ min so that the
u 5  1 K
G s  2 2 e  s
=2 (by inspection) observed input first changes at t = 3 min ; the output first changes at t = 5 min.  s  2 s  1
Use Smith’s method to find 1 and 2. This means that the time delay is two sampling periods, i.e.,  = 2 min. Why is Based on visual inspection on the figure, it is an underdamped process, using
y20 = y(0) + (0.2) (y) = 0 + (0.2) (3) = 0.6 Equation (5-51) we have gain
From inspection of the data, =2 min, rather than 3 min? To understand this point, first consider a process
ln  OS  
2
t20 = 4 –  = 2 1.5  1
with no time delay (=0). For a step change at t = 2+ min, the first observed K  1,  2, OS   0.5     0.1572  0.16
Similarly, 1 0  2  ln  OS  
2
,
y60 = y(0) + (0.6) (y) = 0 + (0.6) (3) = 1.8 changes in the input and the output of this undelayed would occur at t = 3 min,
t60 = 7 –  = 5 t p  5.5    5.5  2  3.5s
because the output cannot change simultaneously due to the process dynamics.
Therefore, Based on Equation (5-50):
t20 2 But for our process, the first changes are observed at t = 5 min which implies that
  0.4 tp 1  2 5.5* 1  0.15722
t60 5  min    1.7289  1.73
From Fig. 7.7:  
  1.2 Time constant  can be obtained from the 63.2% response time: G s 
1
e2 s  2
1
e2 s
and 1.73 s  2*0.16*1.73s  1
2
3s  0.55s  1
t60 t 5 T63.2% = 850 C + (890 – 850 C)(0.632) = 875.3 C
 2.1   = 60   2.38 An alternative method is to use the Smith’s Method shown in Figure 7.7:
 2.1 2.1 Interpolating between t = 7 min and t = 8 min gives
Thus the transfer function can be written as:   2, K  1, t20  2.6, t60  3.1
0.75e 2 s  t(8) – t(7) 
G ( s)  t63.2% = T(8) – T(7)(T63.2% – T(7)) + t(7) The adjusted times are employed for the actual graphical analysis:
5.66 s 2  5.71 s  1   '
t20  t20    2.6  2  0.6
 8–7  '
 t60    3.1  2  1.1
= 878 – 873(875.3 – 873) + 7 t60
From (5-45) and (5-46) or by factoring (e.g., using MATLAB command roots)   '
t20
gives: = 7.46 min '
 0.54
t60
2 s
0.75e
G ( s)  Then t60 3.1
(4.44 s  1) (1.28 s  1) Based on Figure 7.7, we have   0.14,  1.3     2.38
 1.3
t63.2% =  +  + t(0)
1 1
G s  e 2 s
 e2 s
where t(0)=3, the time when the input first changes. Thus 2.382 s  2*0.14*2.38s  1 5.67 s 2  0.67 s  1

= t63.2% –  - t(0) = 7.46 – 1 - 3

(b) Because a damped oscillation occurs, this matches the features of SOPTD.
 FOPTD method does not allow for oscillation.

7-15 7-16 7-17

 t t 10 10
y (k )  (1 
5
) y(k  1)  6u(k  1)
5  y(k )y(k  1)  b  y(k  1)x(k  1) 1
a1  n 1
10
n 1

7.12 In the integrated results tabulated below for t = 0.1, the values are shown  y(k  1)
n 1
2

only at integer values of t, for comparison.

(a) For a FOPTD model shown in below: 10 10 10 10

G s 
Ke  s Table S7.13 Integrated results for the first order differential equation
 x(k  1)y(k ) y(k  1)   y(k  1)x(k  1) y(k  1) y(k ) 2

 s 1 y(k) y(k) y(k) b1  n 1 n 1 n 1 n 1

2
Based on visual inspection, the gain K  2;   2.5 ; when the response reaches
t 10 10
 10

0
(exact)
3
(Δt=1)
3
(Δt=0.1)
3  x(k  1)  y(k  1) 2 2
   y (k  1)x(k  1) 
 n 1 
63.2% complete, i.e., 2*0.632=1.264,   t63.2%  5s 1 2.456 2.400 2.451
n 1 n 1

(b) 2 5.274 5.520 5.296

Using the given data,
3 6.493 6.816 6.522
4 6.404 6.653 6.427
10 10
2.5
FOPTD
5
6
5.243
4.293
5.322
4.258
5.251
4.290
 x(k  1)y(k )  35.212
n 1
,  y(k  1) y(k )  188.749
n 1
Raw data 7 3.514 3.408 3.505
2
8 2.877 2.725 2.864 10 10

1.5
9
10
2.356
1.929
2.180
1.744
2.340
1.912
 x(k  1)
n 1
2
 14 ,  y(k  1)
n 1
2
 198.112

10

 y(k  1) x(k  1)  24.409

1
y

Thus t = 0.1 does improve the finite difference model making it a more
n 1
0.5
accurate approximation of the exact model.

7.14 Substituting into expressions for a1 and b1 gives

0

To find a1 and b1 , use the given first order model to minimize a1 = 0.8187 , b1 = 1.0876
-0.5
0 5 10 15 20 25 30
t,sec
10
The fitted model is y(k  1)  0.8187y(k )  1.0876x(k )
Figure S7.12 The response of the derived FOPTD model J   ( y (k ) a1 y (k  1)  b1 x(k  1)) 2

n 1
or y(k )  0.8187y(k  1)  1.0876x(k  1) (1)
(c) The inverse response at the initial state is caused by a right-half plan zero and where y(k) denotes the data.
is not captured by FOPTD model. Let the first-order continuous transfer function be,
J 10
  2( y (k )  a1 y (k  1)  b1 x(k  1))( y (k  1)  0 Y (s) K
a1 n1 
7.13
J 10 X ( s ) s  1
  2( y (k )  a1 y (k  1)  b1 x(k  1))( x(k  1))  0
b1 n1
For Eq. 7-34, the discrete model is
a) Replacing  by 5, and K by 6 in Eq. 7-25 Solving simultaneously for a1 and b1 gives
t / 5 t / 5
y (k )  e t /  y (k  1)  [1  et /  ]Kx(k  1) (2)
y (k )  e y (k  1)  [1  e ]6u (k  1)

b) Replacing  by 5, and K by 6 in Eq. 7-22 Comparing Eqs. 1 and 2, for t=1, gives

7-18 7-19 7-20

=5s and K = 6 volts e1/  0.911; K 1  e 1/   0.1329  K  1.49  a1   4 0 2.5 0   4 

a   4
An alternative way to calculate K is to set y  k   y  k  1  yss , u  k  1  uss  1 4 2.5 2.5  11 
Hence, the continuous transfer function is β̂   2  , X  , and Y   
 b1     
yss  0.911yss  0.1329  K 
0.1329
 1.49      
G( s) 
6
0.089  b2   213 211 2.5 2.5  214 
5s  1
Calculate parameter estimates using (2):
1
True value

βˆ  1.29 0.31 3.67 1.26 .

8 0.9 FOPTD T
actual data ARX model
fitted model 0.8

7 0.7 Next, we generate model predictions for the calibration data (dataset basal1)
0.6 using past inputs and past model predictions, but not past output data.
Response

6 0.5 Figure S7.16a compares the calibration data and the model predictions,
0.4 where y = y - y(0). Metric S denotes the corresponding sum of squared
0.3 errors,
5
y(t)

0.2 N
S    y (k )  yˆ (k ) 
2
0.1 (2)
4 1
0
0 1 2 3 4 5 6 7 8 9 10 ARX2 Cal
Time 50
3 basal1
2nd Order Discrete Time Model Prediction
Figure S7.15. Comparison of true data and model responses. 0 S = 1921

The result obtained using the ARX model is different from that obtained using an -50
0 1 2 3 4 5 6 7 8 9 10 FOPTD model, because the extra constraint “K=1” is not used. In other words,

 y (mg/dL)
time,t
the discrete time data do not include the final steady-state value, so the calculation
Figure S7.14 Responses of the fitted model and the data gives a different gain. If more data points are added on steady-state values, the -100

result obtained using ARX model will converge to K=1.

-150
7.15 7.16

y (k )  a1 y (k  1)  a2 y (k  2)  b1u (k  1)  b2u (k  2)
-200
a) FOPTD model: (1)
Since K=1, using linear interpolation to find times corresponding to the 35.3%
-250
and 85.3% of response: a) For the model in (1), the least squares parameter estimates are given by 0 25 50 75 100 125 150 175 200 225 250
Time (min)
t35.3%  2.89; t85.3%  8.66
βˆ =  X X  X Y
-1 Figure S7.16a Comparison of model predictions and calibration data for the 2nd
  1.3t35.3%  0.29t85.3%  1.24
T T
(2)
order discrete-time model (S is the sum of squared errors in Eq. 2).
  0.67  t85.3%  t35.3%   3.87
For the basal1 dataset: b) The comparison of the validation data (dataset basal2) and the corresponding
model predictions is shown in Figure S7.16b.
b) Discrete-time ARX model:
y  k   0.911y  k  1  0.1329u  k  1 (since u  k   u  k  1  1 )
Thus:

7-21 7-22 7-23

 20 Table S7.16. Average squared error for the model predictions.
basal2 Figures S7.16c and S7.16d show the model predictions for the calibration and
0 2nd Order Discrete Time Model Prediction
S = 60282 validation data, respectively.
Modelh
-20
Discrete-time Transfer
50
-40 basal1 function
1st Order Transfer Function Model Prediction Calibration data 1921 2428

 y (mg/dL)
-60 0 S = 2428 (basal1)
Validation data 60,282 48,194
-80
-50 (basal2)
-100

 y (mg/dL)
-100
The discrete-time model is more accurate than the transfer function model for
-120
the calibration data, which is not surprising because the former has more
-140 model parameters. Although, the transfer function model is more accurate for
-150
the validation dataset, neither model is very accurate for this dataset.
-160
0 50 100 150 200 250 300 350
Time (min) -200

Figure S7.16b Comparison of model predictions and validation data for the 2nd
order discrete time model (S is the sum of squared errors in Eq. 2). -250 7.17
0 25 50 75 100 125 150 175 200 225 250
Time (min)
b) Now, consider the first-order transfer function model
Figure S7.16c Comparison of the model predictions and calibration data for the a) Fit a first-order model:
Y (s) K 1st order transfer function (S is the sum of squared errors in Eq. 2).
 (3)
U (s)  s  1 20
Let y = hydrocarbon exit temperature, THC
basal2 u = air flow rate, FA
1st Order Transfer Function Model Prediction
First, determine the steady-state gain, K = Δy/Δu. The output finally reaches a 0
S = 48194
Note: There is a typo in the 1st printing. The step change in u should start at 17.9
new steady state of about 250 mg/dL. For dataset basal1, the input change is -20 m3/min, not 17.0 m3/min.
u=2.5 units/day. Thus,
-40 The step response data is shown in Fig. S7.17a. The step change in u from 17.9 to
250 mg day

y (mg/dL)
K   100 21 m3/min occurs at t = 14 min. By inspection of the noisy y data, the time delay
-60
2.5 dL units is approximately = 4 min.
-80
To identify time constant τ, determine the time at which 63.2% of the total
change has occurred. This corresponds to the time at which the output, Δy has a -100
value of - 250 × 63.2% = - 158. For inspection of the data, τ = 134 min when
y= 158 mg/dL. -120

The model predictions for the model in (3) can be calculated from the step -140
0 50 100 150 200 250 300 350
response for a first-order transfer function in (5-18) Time (min)

Figure S7.16d Comparison of the model predictions and validation data for the
y (t )  KM (1  e t / ) (5-18)) 1st order transfer function (S is the sum of squared errors in Eq. 2).

The input step sizes are M = 2.5 units/day for basal1 and 1.5 units/day for c) Discussion of results
basal2.
Table S7.16 lists the calculated values of S for the two models.

7-24 7-25 7-26

t20 1.5
  0.3 The model comparisons in Fig. S7.17b indicate that the two models are very
t60 5
similar and reasonably accurate. However, the low-order transfer function models
From Fig. 7.7: fail to capture the higher order dynamics of the physical furnace model that was
used to generate the step response data. The first-order model has a lower value of
  2.2 the least squares index, S:
and
t60 t 5 First order model: S = 1.71 x 104
 4   = 60   1.25 min
 4 4 Second-order model: S = 2.01 x 104

Thus the second-order transfer function can be written in standard form as:

THC ( s ) 1.5e 2 s 7.18


Figure S7.17a Step response data for the furnace module. FA ( s ) 5.66 s 2  5.71 s  1

From the step response data, the following information can be obtained: (a) Fit a FOPTD model to the column step response data:
THC ( s )  11.9e 4 s

y 25 K K FA ( s ) (1.25) 2 s 2  2(2.2)(1.25) s  1 Let y = distillate MeOH composition, xD
K   11.9 3
u 2.1 m3 /min m /min u = reflux ratio, R
From (5-45) and (5-46) or by factoring (e.g., using MATLAB command roots )
The step response data is shown in Fig. S7.18a with the step change in u from
y(0) = 609.5 K, y(∞) = 584.5 K; thus y = 609.5 – 584.5 = - 25 K gives:
1.75 to 2.0 occurring at t = 3950 s. By inspection of the noisy data, the time delay
THC 11.9e 4 s is ≈ 50 s.
y63.2 = 609.5 + (0.632)(-25) = 593.7 K 
FA (5.2 s  1) (0.3 s  1)
From the figure, t63.2 = 19.5 – 14 - 4 = 5.5 min. Thus, the transfer function model The following information can be obtained from the step response data:
is: c) Simulations
Y ( s )  11.9 xD(0) = 0.85, xD(∞) = 0.88;

U ( s ) 5.5s  1 thus
 xD = 0.88 – 0.85 = 0.03
b) Fit a second-order model: and
y 0.03
K   0.12
Use Smith’s method to find 1 and 2. u 0.25
Also,
y20 = y(0) + (0.2)(y) = 609.5 + (0.2)(-25) = 604.5 K
y63.2 = 0.85 + (0.632)(0.03) = 0.869
y60 = y(0) + (0.6) (y) = 609.5 + (0.6)(-25) = 594.6 K

From inspection of the data,

t20 ≈ 19.5 - 10 = 1.5 min

Similarly, Figure S7.17b Comparison of furnace step response data and model responses.
t60 = 23 –  = 5 min
Therefore,
d) Discussion

7-27 7-28 7-29

 t20 280
  0.31 First order model: S  9.014 104
t60 900
Second-order model: S  9.012 104
From Fig. 7.7:
t60 t
 4    60  225 s
 4
and
  2.0
The SOPTD model can be written as:

xD ( s ) Kes 0.12 e 50 s

 
R( s ) 2 s 2  2ts  1 4 x 104 s 2  800s  1
Figure S7.18a Step response data for the column module.
which can be factored using (5-45) and (5-46):
From the figure,  = t63.2 – t(0)= 5050 – 3950 - 50 ≈ 1050 s. Thus, one estimate of
the time constant is  = 1050 s. A second estimate can be obtained from the xD ( s ) 0.12 e50 s

settling time, ts ≈ 7600 – 3950 = 3650 s. Thus,  ≈ ts/4 = 912 s. Averaging these R( s) (769s  1)(54s  1)
two estimates gives:
1050  912
ave   981 s c) Simulations
2
Thus the identified transfer function is,

xD ( s) 0.12 e50 s

R( s ) 981s  1

(b) SOPTD model:

Use Smith’s method:
y20 = 0.85 + (0.2)(0.03) = 0.856
y60 = 0.85 + (0.6)(0.03) = 0.868
Figure 7.18b Comparison of column step response data and model responses.
From the step response data:
t20 ≈ 4280 – 3950 – 50 = 280 s d) Discussion
t60 ≈ 4900 – 3950 – 50 = 900 s
The model comparisons in Fig. S7.18b indicate that both models are reasonably
Thus, accurate. However, the second-order model is more accurate as indicated visually
and by its slightly lower S value:

7-30 7-31 7-32

K 2
Chapter 8 © 8.1
or 1
K1  K 2
The response of a PI controller to a unit step change in set point at t = 0 is shown in
Many of the problems in this chapter require determining whether a controller should be Fig. 8.6. The instantaneous change at t = 0 is Kc and the slope of the response is K1  K 2  K 2 
direct-acting or reverse-acting. The following chart can help guide the thinking process Kc/I. Now consider a more general step change in the set point of magnitude M.
for these problems when also considering the style of the valve. Note that the chart P( s) 1 K K1  K 2   K 2  K 2 (  1)
assumes all unmentioned gains are positive (measurement, I/P, etc.).  K c (1  )  Kc  c
E (s) Is Is
Substituting,
Table S8.1: Chart for determining if controller should be direct-acting or reverse-acting. M K1  ( K c  K1 )(  1)  (  1) K c  (  1) K1
E (s) 
Then the controller s
If the process gain is: And the valve is: KM KM
should be: P( s)  c  c 2 Then,
s Is   1
K1   K c
Positive
Fail Close (Air-to-Open) Reverse-Acting
p (t )  K c M 
Kc M
t for t  0   
Kv + Kc +
MV ↑, CV ↑ I
For MV ↓, want p ↓ For CV ↑, want p ↓ c) If Kc = 3 , D = 2 ,  = 0.1 then,
Kp + The instantaneous change at t = 0 is KcM and the slope of the response is KcM/I
Fail Open (Air-to-Close) Direct Acting From the data given in the table, the initial instantaneous change is -1.3 mA and the
For control, if CV ↑,  0.9
Kv – Kc – slope is -0.0335 mA/s for a step change of M = 2.5 mA. Thus, K1   3  27
want MV ↓ 0.1
For MV ↓, want p ↑ For CV ↑, want p ↑ 1.3 mA 1.3mA
KC     0.52
Fail Close (Air-to-Open) Direct Acting
M 2.5 mA K 2  3  (27)  30
Negative KC M
Kv + Kc –   0.0335 mA/s
MV ↑, CV ↓ I 1 = 0.1  2 = 0.2
For MV ↑, want p ↑ For CV ↑, want p ↑
Kp –
KC M 0.52(2.5mA)
Fail Open (Air-to-Close) Reverse-Acting I  = = 26 s Hence
For control, if CV ↑,  0.0335 mA/s  0.05 mA/s K1 + K2 = -27 + 30 = 3
Kv – Kc +
want MV ↑
For MV ↑, want p ↓ For CV ↑, want p ↓ Because KC is negative, we classify this controller as direct acting.
K 2 1 30  0.2
 2
K1  K 2 3

8.2
 2s  1 
Gc ( s)  3 
 K 2 1   0.2s  1 
P ( s ) K1 K  K 2 1 s  K 2  K  K s  1
a)   K2  1  ( K1  K 2 )  1 2
 8.3
E ( s ) 1 s  1 1 s  1  1 s  1 
 
 
b) Kc = K1 + K2  K2 = Kc  K1 a) From Eq. 8-14, the parallel form of the PID controller is :

1  D  1 
Gi ( s )  K c 1   D s 
 I s 
K 2 1 K 
D   2 D
K1  K 2 K1  K 2

[Type here] 8-1 [Type here] 8-2 8-3

 From Eq. 8-15, for  0, the series form of the PID controller is: b) System I: Flow rate too high  need to close valve  decrease
controller output  reverse acting controller Configuration (b):
 1 
 D s  1
As h increases, we want to increase q, the exit flow rate. For an air-to-close
Ga ( s )  K c 1  Or: Process gain +
 I s 
control valve, the controller output should decrease. Thus as h increases p
Valve gain + decreases  a reverse-acting controller.
 D 1  Controller gain must be + (which means reverse acting)
 K c 1     D s
 I I s  (b) Configuration (a) in Fig. E8.5:
System II: Flow rate too high  need to close valve  increase As h increases, we want to decrease qi, the inlet flow rate. For an air-to-open
  controller output  direct acting controller.
  control valve, the controller output p should decrease. Thus as h increases p
   D s decreases  a reverse-acting controller.
 K c 1  D  1  
1

  I    D    
Or: Process gain +
 1   I s 1  D   Valve gain – Configuration (b):
   I   I   Controller gain must be – (which means direct acting)
Comparing Ga(s) with Gi(s) As h increases, we want to increase q, the exit flow rate. For an air-to-open
c) System I: Kc > 0. control valve, the controller output should increase. Thus as h increases p
increases  a direct-acting controller.
   System II: Kc < 0.
K c  K c 1  D 
 I 
   8.5 8.6
I   I 1  D 
  I 

D a) From Eqs. 8-1 and 8-2, For PI control

D 
D  1
t

1 p(t )  p  K c  e(t )   e(t*)dt *
I 
p(t )  p  K c y sp (t )  y m (t )  (1)  I 0 
    t

b) Since 1  D   1 for all D, I, therefore 1
p (t )  K c  e(t )   e(t*)dt *
 I 
The liquid-level transmitter relation is
 I 0 
K c  K c ,  I  I and  D  D ym(t) = KT h(t) (2) Since
where: e(t) = ysp – ym = 0 - 2 = - 2
c) For Kc = 4, I=10 min , D =2 min h is the liquid level
KT > 0 is the gain of the direct acting transmitter. Then
K c  4.8 , I  12 min , D  1.67 min  1
t
  2 
p (t )  K c   2   (2)dt *  K c   2  t 
d) Considering only first-order effects, a non-zero value  will dampen all The control-valve relation is  I 0   I 
responses, making them slower.
q(t) = Kvp(t) (3) The initial response at t = 0 is  2 Kc
8.4 2K c
where The slope of the response is 
q is the manipulated flow rate I
Kv is the gain of the control valve. Substitute the numerical values of the initial response and slope from Fig. E8.6:
a) System I (air-to-open valve): as the signal to the control valve increases,
the flow through the valve increases  Kv > 0. - 2 Kc = 6  Kc = -3
(a) Configuration (a) in Fig. E8.5:
System II (air-to-close valve): as the signal to the control valve increases, As h increases, we want to decrease qi, the inlet flow rate. For an air-to-close 2K c
the flow through the valve decreases  Kv < 0. control valve, the controller output p should increase. Thus as h increases p  = 1.2 min-1  I = 5 min
I
decreases  a direct-acting controller.

8-4 8-5 8-6

 t 
8.7 p k i  p  K c r  (1  i ) r  , i = 1, 2, …
 I 

(a) The error signal can be described by:

e(t )  0.5t
0.5
E ( s) 
s2
D
Kc r
The PID controller transfer function is given by (Eq. 8-14): pk t

P' ( s)  1 
 K C 1    D s
E (s)  Is 
t
Substituting gives the controller output: K c r Kc r
I
p
0.5 K C  1 
P '( s )  1   Ds
s2   I s  k-1 k k+1 k+2 k+3
 1 2 
p '(t )  0.5 K C t  t   D S (t ) 
 2 I  Figure S8.7: PID controller output response a) To eliminate derivative kick, replace (ek – ek-1) in Eq. 8-25 by - (yk-yk-1).
(Note the minus sign.)
Substituting numerical values and adding p  12 mA gives:
8.8
1 8.9
pPID (t )  12  t 2  t  0.5S (t ) D
3 From inspection of Eq. 8-25, the derivative kick = K c r
t
(b) The equation for a PI controller is obtained by setting τD to zero. a) Proportional kick = K c r a) Let the constant set point be denoted by y sp . The digital velocity P
algorithm is obtained by setting 1/I = D = 0 in Eq. 8-27:
1
pPI (t )  12  t 2  t b) e1 = e2 = e3 = …. = ek-2 = ek-1 = 0
3 pk = Kc(ek – ek-1)
ek = ek+1 = ek+2 = …= r
(c) The plot of the controller response for both controllers is shown in Fig. S8.7. = K c ( y sp  y k )   y sp  y k 1 
p k 1  p
The two controllers have similarly-shaped responses. The difference is the
 t   = K c  y k 1  y k 
sudden jump at t=0 that occurs with the PID controller as a result of the p k  p  K c r  r  D r 
derivative term. When the set point begins to change with a constant slope, there  I t 
is a step change in the error derivative from 0 to 0.5. The derivative term in the The digital velocity PD algorithm is obtained by setting 1/I = 0 in Eq. 8-
controller gives it a jumpstart right when the setpoint begins to change that the 27:

PID controller does not have. pk = Kc [(ek – ek-1) + D (ek – 2ek-1 + ek-2)]
t

= Kc [ (-yk + yk-1) + D (-yk – 2yk-1 + yk-2) ]
t

8-7 8-8 8-9

 In both cases, pk does not depend on y sp . d 2 p (t ) dp (t )  de(t ) d 2 e(t ) 
 I  D  I  K c  e(t )  ( I   D )  I D 
dt 2 
8.12
dt 2 dt  dt
b) For both these algorithms pk = 0 if yk-2 = yk-1 = yk. Thus a steady state is
reached with a value of y that is independent of the value of y sp . Use of c) The simulation is performed for the following parameter values: a) False. The controller output could saturate or the controller could be in the
these control algorithms is inadvisable if offset is a concern. manual mode.
Kc  2 ,  I  3 ,  D  0.5 ,   0.1 , M = 1 b) False. Even with integral control action, offset can occur if the controller output
t saturates. Or the controller could be in the manual mode.
c) If the integral mode is present, then pk contains the term Kc ( y sp  y k ) .
I The Simulink-MATLAB results are shown in Figure S8.10.:
Thus, at steady state, pk = 0 and yk-2 = yk-1 = yk , yk = y sp , and the offset Step Response
22 8.13
problem is eliminated. Parallel PID with a derivative filter
20 Series PID with a derivative filter

18 First consider qualitatively how h2 responds to a change in q2. From physical

8.10 considerations, it is clear that if q2 increases, h2 will increase. Thus, if h2 is
16
increasing, we want q2 to decrease, and vice versa. Since the q2 control valve is air-
14 to-open, the level controller output p should decrease in order to have q2 decrease.
P ( s )  1 D s  In summary, if h2 increases we want p to decrease; thus a reverse-acting controller
a)  K c 1    p'(t) 12
E (s)   I s  D s  1  is required.
10

 Kc
 I s(D s  1)  D s  1  D sI s  8
8.14
I s(D s  1) 6
1  ( I   D ) s  (1  ) I  D s  2
4
 Kc   First consider qualitatively how solute mass fraction x responds to a change in steam
  I s ( D s  1)  2 flow rate, S. From physical considerations, it is clear that if S increases, x will also
0 2 4 6 8 10
Time increase. Thus, if x is increasing, we want S to decrease, and vice versa. For a fail-
Cross- multiplying open (air-to-close) control valve, the controller output p should increase in order to
Figure S8.10. Step responses for both parallel and series PID controllers have S decrease. In summary, if x increases we want S to decrease, which requires
(I  D s 2   I s) P ( s)  K c 1  ( I  D ) s  (1  ) I  D s 2  E ( s) with a derivative filter. an increase in controller output p; thus a direct-acting controller is required.

Taking inverse Laplace transforms gives, 8.11 8.15

d 2 p (t ) dp (t )  de(t ) d 2 e(t ) 
 I  D  I  K c  e(t )  ( I   D )  (1  ) I  D 
dt 2 dt  dt dt 2  The integral component of the controller action is determined by integrating the First consider qualitatively how exit temperature Th2 responds to a change in cooling
error between the measurement and the set point over time. As long as the sign on water flow rate, wc. From physical considerations, it is clear that if wc decreases, Th2
the error stays the same (i.e., if the measurement does not cross the set point), the
P ( s )   s  1   D s  will increase. Thus, if Th2 is decreasing, we want wc to decrease, and vice versa. But
b)  K c  I  
integral component will continue to change monotonically. If the measurement in order to specify the controller action, we need to know if the control valve is fail
E (s)   I s   D s  1  crosses the set point, the error term will change sign and the integral component open or fail close. Based on safety considerations, the control valve should be fail
will begin to change in the other direction. Thus, it will no longer be monotonic. open (air-to-close). Otherwise, the very hot liquid stream could become even hotter
Cross-multiplying and cause problems (e.g., burst the pipe or generate a two phase flow).

 I s 2 (D s  1) P ( s)  K c ( I s  1)(  D s  1)  E ( s) For an air-to-close control valve, the temperature controller output p should
increase in order to have wc decrease. In summary, if Th2 decreases we want wc to
decrease, which requires controller output p to increase; thus a reverse-acting
controller is required.

8-10 8-11 8-12

8.16 Chapter 9 ©
a) The safest conditions are achieved by the lowest temperatures and pressures
in the flash vessel.
Two pieces of information are needed to specify controller action:
i) Is the control valve fail open or fail close 9.1 VALVE 1.- Fail close (air-to-open)
ii) Is x1 > x2 or x1 < x2 VALVE 2.- Fail open (air-to-close)
VALVE 3.- Fail open (air-to-close)
If x1 > x2, then the mass balance is: a) Flow rate transmitter: VALVE 4.- Fail open (air-to-close)
 15 psig - 3 psig  VALVE 5.- Fail close (air-to-open)
x1w1  x2 w2  xw  x ( w1  w2 ) qm(psig)=   (q gpm - 0 gpm)  3 psig
 400 gpm-0 gpm 
x1  x2 Setting valve 1 as fail close prevents more heat from going to flash drum and
 psig  setting valve 3 as fail open to allow the steam chest to drain. Setting valve 3
 x1  x2   =  0.03  q(gpm)  3 psig
 gpm  as fail open prevents pressure build up in the vessel. Valve 4 should be fail-
( x2   ) w1  x2 w2  x( w1  w2 ) Pressure transmitter: open to evacuate the system and help keep pressure low. Valve 5 should be
x2 w1  w1  x2 w2  x ( w1  w2 ) fail-close to prevent any additional pressure build-up.
x2 ( w1  w2 )  w1  x ( w1  w2 )  20 mA - 4 mA 
Pm(mA)=   ( p in.Hg  10 in.Hg)  4 mA b) Vapor flow to downstream equipment can cause a hazardous situation
w1  30 in.Hg - 10 in.Hg 
x  x2 
( w1  w2 )  mA  VALVE 1.- Fail close (air-to-open)
=  0.8  p(in.Hg)  4 mA
 in.Hg  VALVE 2.- Fail open (air-to-close)
Since all the variables in the equation are positive, then x > x2 . The only way to Level transmitter: VALVE 3.- Fail close (air-to-open)
decrease x is to increase w2 (but x can never be less than x2). Therefore, w2 should be VALVE 4.- Fail open (air-to-close)
increased when x increases, in order to have x decrease. If the control valve is fail  5 VDC - 1 VDC  VALVE 5.- Fail close (air-to-open)
open (air-to-close), then the composition controller output signal p should decrease. hm(VDC)=   (h(m) - 0.5m)  1 VDC
Thus a reverse-acting controller should be selected. Conversely, for a fail close (air-  10 m - 0.5 m 
Setting valve 1 as fail close (air-to-open) prevents more heat from entering
to-open) control valve, a direct-acting controller should be used.  VDC  flash drum and minimizes future vapor production. Setting valve 2 as fail
=  0.421  h(m)  0.789 VDC
 m  open (air-to-close) will allow the steam chest to be evacuated, setting valve 3
If x1 < x2, then as fail close (air-to-open) prevents vapor from escaping the vessel. Setting
Concentration transmitter:
x1w1  x2 w2  xw  x ( w1  w2 )  10 VDC - 1 VDC  valve 4 as fail open (air-to-close) allows liquid to leave, preventing vapor
Cm(VDC)=   (C (g/L)-3 g/L)+1 VDC build up. Setting valve 4 as fail close (air-to-open) prevents pressure buildup.
x1  x2  20 g/L - 3 g/L 
 x1  x2    VDC  c) Liquid flow to downstream equipment can cause a hazardous situation
=  0.529  C (g/L)  0.59VDC
( x2   ) w1  x2 w2  x( w1  w2 )  g/L 
VALVE 1.- Fail close (air-to-open)
x2 w1  w1  x2 w2  x( w1  w2 ) VALVE 2.- Fail open (air-to-close)
b) The gains, zeros and spans are:
x2 ( w1  w2 )  w1  x ( w1  w2 ) VALVE 3.- Fail open (air-to-close)
w1 VALVE 4.- Fail close (air-to-open)
Flow Pressure Level Concentration
x  x2  VALVE 5.- Fail close (air-to-open)
( w1  w2 ) Gain 0.03 psig/gpm 0.8 mA/in.Hg 0.421 VDC/m 0.529 VDC/g/L
Zero 0 gpm 10 in.Hg 0.5 m 3 g/L Set valve 1 as fail close to prevent all the liquid from being vaporized (This
Since all the variables are positive, then x < x2 . If x increases, the controller will Span 400 gpm 20 in.Hg 9.5 m 17 g/L would cause the flash drum to overheat). Setting valve 2 as fail open will
need to decrease it to bring it back to the set point. The only way to decrease x is to
allow the steam chest to be evacuated. Setting valve 3 as fail open prevents
decrease w2 (although x can never be smaller than x1). If the control valve is fail
pressure buildup in drum. Setting valve 4 as fail close prevents liquid from
open (air-to-close), then the composition controller output signal p should increase
escaping. Setting valve 5 as fail close prevents liquid build-up in drum
in order to reduce w2. Thus the composition controller should be direct acting.
Conversely, for a fail close control valve, a reverse acting controller should be used. 9.2

[Type here] 9-1 9-2

8-13 [Type here]
9.3 qd  0.6 m3 / min  36 m3 / h
9.5
Psd  200 kPa
Note: This exercise is best understood after the material in Ch. 11 has been considered. Psd 200 kPa
K   0.154 kPa/(m3 /h) 2
qd 2 362 (m3 /h)2 Let Pv/Ps = 0.33 at the nominal q  320 gpm
a) Changing the span of the temperature transmitter will change its steady-state Now substitute (3) into (2) to get an expression for Pv in terms of q.
gain, according to Eq. 9-1. Because the performance of the closed-loop Ps = Pb+ Po = 40 + 1.953  10-4 q2
system depends on the gains of each individual element (cf. Chapter 11), Pv  P  Kq 2 (4)
closed-loop stability could be adversely affected. Substitute (4) into (1) to get:
Pv= P - Ps = (1 –2.44  10-6 q2)PDE – (40 + 1.953  10-4 q2)
q
Cv  (5)
b) Changing the zero of a transmitted does not affect its gain. Thus, this change P  Kq 2
Nf (l ) (1 - 2.44  10 -6  320 2 )PDE - (40 + 1.953  10 -4  320 2 )
will not affect closed-loop stability.
gs  0.33
(40 + 1.953  10 -4  320 2 )
c) Changing the control valve trim will change the (local) steady-state gain of The problem specifies that qd should be 2/3 of qmax (where qmax is the flow
the control valve, dq/dp. Because the performance of the closed-loop system rate through the valve when the valve is fully open). PDE = 106.4 psi
depends on the gains of each individual element (cf. Chapter 11), closed-loop 2
qd  qmax
stability could be adversely affected 3 Let qd= q  320 gpm
3 3
d) For this process, changing the feed flow rate could affect both its steady-state qmax  qd  36 m3 /h
2 2 For the rated Cv, the valve is completely open at 110% qd i.e., at 352 gpm or the
gain and its dynamic characteristics (e.g., time constant and time delay).
Because the performance of the closed-loop system depends on the gains of qmax  54m3 /h upper limit of 350 gpm
each individual element (cf. Chapter 11), closed-loop stability could be Now find the Cv that will give qmax = 54 m3/h. Substitute q = qmax and f (l)=1 1

adversely affected. (valve fully open) into (5).  p  2

qmax Cv  q  v 
Cv   gs 
P  Kqmax 2 
1
9.4 N  (1  2.44 106  3502 )106.4  (40  1.953  104  3502 )  2
gs Cv  350  
Now that all of the variables on the right hand side of the equation are  0.9 
Starting from Eq. 9-7: known, plug in to solve for Cv.
q kPa m3 Then using Eq. 9-27,
Cv  (1) P  450 kPa, K  0.154 3 , N  0.0865 ,
Pv (m / h) 2 h(kPa)1/2
Nf (l )  q  66.4  4.55  10  4 q 2  1 / 2 
gs g s  1.2, qmax  54 m /h 3
ln    
m3 101.6  0.9  
The pressure drop in the valve is: 54 l  1
h ln 50
Pv  P  Ps (2) Cv 
kPa
where 450kPa  0.154 54 2 (m 3 /h) 2 The plot of the valve characteristic is shown in Figure S9.5. From the plot of the
m3 (m3 /h) 2
Ps  Kq 2 (3) 0.0865 valve characteristic for the rated Cv of 101.6, it is evident that the characteristic is
h(kPa)1/2 1.2
Solve for K by plugging in the nominal values of q and Ps . First, convert reasonably linear in the operating region 250  q  350.
3 3
m m
the nominal value of q into units of m3/h to match the metric units version 54 54
 h  h The pumping cost could be further reduced by lowering PDE to a value that would
of N (the parameter N = 0.0865 m3/h(Kpa)1/2 when q has units of m3/h and
m3 make Pv/Ps = 0.25 at q  320 gpm. Then PDE = 100 and for qd = 320 gpm, the
pressure has units of KPa). 0.0865
h(kPa)1/2
 0.88(kPa)1/2
 0.076
m3
h rated Cv = 133.5. However, as the plot shows, the valve characteristic for this design
Cv  710.5 is only slightly more nonlinear in the operating region. Hence, the selected valve
coefficient is Cv = 133.5.

9-3 9-4 9-5

ii. Level controller: As the liquid level h increases, we want the product 9.8
400 flow rate B to increase. Since the control valve is air-to-open, this
means that the controller output signal to the control valve (via the
350
I/P) should increase. Thus, the controller should be direct-acting. Configuration I: This series configuration will not be very effective because a
iii. Pressure controller: As the pressure P increases, we want the solvent large flow rate has to pass through a small control valve. Thus, the pressure drop
300
flow rate D to increase. Since the control valve is air-to-close, this will be very large and flow control will be ineffective.
means that the controller output signal to the control valve (via the
250
I/P) should decrease. Thus, the controller should be reverse-acting. Configuration II: This parallel configuration will be effective because the large
q (gpm) control valve can be adjusted to provide the nominal flow rate, while the small
200
control valve can be used to regulate the flow rate. If the small valve reaches its
maximum or minimum value, the large valve can be adjusted slightly so that the
150
9.7 small valve is about half open, thus allowing it to regulate flow again.
------- Cv = 101.6
100

- - - - Cv = 133.5 Because the system dynamic behavior would be described using deviation 9.9
50
variables, the dynamic characteristic can be analyzed by considering that the input
terms (not involving x) can be considered to be constant, and thus deviations are
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 zero. The starting form is the linear homogeneous ODE: First write down the time-domain step response for a step change of 10°C. Then
l (valve lift) M d 2x dx solve the equation to find when y(t) is equal to 5 (since the variables are in deviation
 R  Kx  0 variables, this represents when TM will reach 30°C).
Figure S9.5. Control valve characteristics. gc dt 2 dt
Taking the Laplace transform gives,
ym (t )  KM (1  e  t / )
M 2 
9.6 X ( s)  s  Rs  K   0 where M  10o C, K  1, and   10s
 gc 
ym (t )  10(1  e  t /10 )
 M 2 R 
a) There are three control valves. The selection of air-to-close vs. air-to-open is X ( s)  s  s  1  0 5  10(1  e  ta /10
)
based on safety considerations:  Kgc K 
ta  6.93s
Calculate and ζ by comparing this equation to the standard form of the second-
i. Steam control valve: Air-to-open to prevent overheating of the order model in (5-39) (keeping in mind that gc = 32.174 lbm ft/(lbf s2)).
evaporator. Therefore, the alarm will sound 6.93 seconds after 1:10PM.
ii. Level control valve (that adjusts liquid flow rate B): Air-to-open to M
  0.00965s
prevent the steam coils from being exposed to the vapor space, Kg c
which could lead the coils to being burned out. 9.10
M R
iii. Pressure control valve (that adjusts solvent flow rate D): Air-to-close 2  2 
Kg c K 0.1 psig
to prevent over-pressurization of the evaporator.  precision =  0.5% of full scale
20 psig
R gc
  155.3  accuracy is unknown since the "true" pressure in the tank is unknown
b) For the three controllers: 2 KM
0.1 psig
 resolution =  0.5% of full scale
The valve characteristics are highly overdamped and can be accurately 20 psig
i. Concentration controller: As the product concentration xB increases, approximated by a first-order model obtained by neglecting the d2x/dt2 term.
we want the steam pressure, Ps to increase. Since the steam valve is ±0.1 psig
 repeatability = =±0.5% of full scale
air-to-open, this means that the controller output signal to the control 20 psig
valve (via the I/P) should increase. Thus, the controller should be
direct-acting.

9-6 9-7 9-8

9.11
Chapter 10 © even death. Therefore, the long detection time would not be acceptable if the
hazardous gas is CO.

Assume that the gain of the sensor/transmitter is unity (i.e. there is no steady-state
measurement error). Then, 10.2
10.1

Tm ( s ) 1 Assumptions: (a) Start with a mass balance on the tank. Then solve the equation to find how much

T ( s) ( s  1)(0.1s  1) 1. Incompressible flow. time it takes for the height to decrease from 1 m to 0.25 m.
2. Chlorine concentration does not affect the air sample density.
where T is the temperature being measured and Tm is the measured value. For the 3. T and P are approximately constant. dV (t )
 C h(t )
ramp temperature change: dt
The detection time, td, depends on the transportation time delay, , and the response
0.3 Adh(t )
T  (t) = 0.3t (C/s) , T  (s) = 2 time of the analyzer, tr = 10 s:  C h(t )
s dt
Tm ( s) 
1

0.3 td =  + tr (1) dh C
  h0.5
( s  1)(0.1s  1) s 2 dt A
Time delay can be calculated as the ratio of the volume of the tubing V divided C
h 0.5 dh   dt
Tm (t )  0.00333e10t  0.333et  0.3t  0.33 by the volumetric flow rate of chlorine q: A
0.25 tf
C
h dh   
V 0.5
The maximum error occurs as t : θ (2) dt
q 1 0
A
Maximum error = |0.3t  (0.3t  0.33)| = 0.33 C 2( 0.25[m]  1[m])  
C
(t f  0)
where q = 10 cm3/s and, A
If the smaller time constant is neglected, the time domain response is slightly π Di 2 L C
V (3) 1m0.5   tf
different for small values of t, although the maximum error (t) does not change. 4 A
where the inside diameter Di is: A
T °C tf  1[m0.5 ]
Di = 6.35 mm – 2(0.762 mm) = 4.83 mm = 4.83 x 10-3 m C
6  (0.5) 2 [m 2 ]
tf  [m0.5 ]
Substitute Di and L = 60 m into (3): 0.065[m 2.5 / min]
5
t f  12.1[min]
V = 1.10 x 10-3 m3
4
Substitute Di into (2): Therefore, the alarm will sound at 5:12:06AM
3
(b) To find how much liquid has leaked out of the tank, calculate the difference in
V  1.10 x10 m   100cm 
3 3
2 θ 
3

  110 s
 volume between the starting level and the alarm level.
q  10 cm3 /s   1 m 
1 2
1 
Substitute into (1): V  Vh 1m  Vh 0.25 m    [m]  1[m]  0.25[m]  0.59m3
0 Time s 2 
0 5 10 15 20 td =  + tr = 110 + 10 = 120 s = 2 min
Figure S9.11. Response for process temperature sensor/transmitter. Orange solid 0.59m3 of liquid has leaked out when the alarm sounds.
Carbon monoxide (CO) is one of the most widely occurring toxic gases, especially
line is T’(t), and purple dashed line is T’m(t). for confined spaces. High concentrations of carbon monoxide can saturate a
10.3
person’s blood in matter of minutes and quickly lead to respiratory problems or

9-9 10-2
[Type here] 10-1 [Type here]

The proposed alarm/SIS system is shown in Figure S10.4:  n

The key safety concerns include: where the notation,   , refers to the number of combinations of n objects taken r
1. Early detection of leaks to the surroundings r
 3
2. Over-pressurizing the flash drum at a time, when the order of the r objects is not important. Thus    3 and
3. Maintain enough liquid level so that the pump does not cavitate.  2
4. Avoid having liquid entrained in the gas. S  3
   1. From Eq. 1,
 3
These concerns can be addressed by the following instrumentation.
PSH
P(k  2)  0.135  0.857  0.992
1. Leak detection: sensors for hazardous gases should be located in the vicinity
of the flash drum. C 1
O See any standard probability or statistics book, e.g., Montgomery D.C and G.C. Runger,
2. Over pressurization: Use a high pressure switch (PSH) to shut off the feed L Applied Statistics and Probability for Engineers, 6h edition, Wiley, New York, 2013.
when a high pressure occurs. U
3. Liquid inventory: Use a low level switch (LSL) to shut down the pump if a M
N 10.6
low level occurs.
4. Liquid entrainment: Use a high level alarm to shut off the feed if the liquid
level becomes too high. Figure S10.4: Proposed alarm/SIS system
Solenoid switch: S = 0.01

This SIS system is shown in Fig. S10.3 with conventional control loops for
The solenoid-operated valve is normally closed. But if the pressure in the Level switch: LS = 0.45
column exceeds a specified limit, the high pressure switch (PSH) activates
pressure and liquid level. Level alarm: A = 0.3
an alarm (PAH) and causes the valve to open fully, thus reducing the
Vapor pressure in the tank. Notation:
PS = the probability that the solenoid switch fails
PSH
10.5 PLS = the probability that the level switch fails
PA = the probability that the level alarm fails
PT PC
LSH PI = the probability that the interlock system (solenoid & level switch fails)
Define k as the number of sensors that are working properly. We wish to calculate
the probability that k  2 , P (k  2) . We wish to determine,
S P
P = the probability that both safety systems fail (i.e., the original system
Feed and the additional level alarm)
Because k = 2 and k = 3 are mutually exclusive events (cf. Appendix F),
Because the interlock and level alarm systems are independent, it follows that (cf.
P(k  2)  P(k  2)  P(k  3)
LT LC
(1) Appendix F):
LSL P = PI PA (1)
These probabilities can be calculated from the binomial distribution 1
From the failure rates, the following table can be constructed, in analogy with
S
 3 Example 10.4:
P(k  2)     0.05 (0.95)2  0.135
1

 2
Liquid
Component  R P=1–R
 3
P(k  3)     0.05 (0.95)3  0.857   
0
Solenoid:
Figure S10.3: SIS system  3 Level switch: 0.45 0.638 0.362
Level alarm 0.3 0.741 0.259
10.4

10-3 10-4 10-5

 Assume that the switch and solenoid are independent. Then,
10.8 Chapter 11
PI = PS + PSW - PS PSW
PI = 0.01 + 0.362 – (0.01)(0.362)
Let P3 = the probability that none of the 3 D/P flowmeters are working properly.
PI = 0.368 Then P3 and the related reliability, R3, can be calculated as (cf. Appendix F):
Substitute into (1): 11.1
P3 = (0.82)3 = 0.551
P = PI PA = (0.368)(0.259) = 0.095 D1
R3 =1 – P3 = 1 – 0.551 = 0.449 Gd1

Mean time between failures, MTBF: To calculate the overall system reliability, substitute R3 = 0.449 for the reliability D2
value for two D/P flowmeters (R2=0.33) in the R calculation from Exercise 10.7: Gd2
X2 X1
From (10-6) through (10-8): 5
R   Ri  (0.449)(0.95)(0.61)(0.55)(0.64)
~
R = 1 – P = 1 – 0.095 = 0.905 Ysp Ysp E P U X3 + +
i 1 Km +- Gc + + Y
Gv Gp
 = - ln (0.905) = 0.0998 R  0.092
1
MTBF   10.0 years
 Thus, the addition of the third D/P flowmeter has increased the overall system
reliability from 0.067 (for Exercise 10.7) to 0.092.
B
10.7 Gm

10.9
Let P2 = the probability that neither D/P flowmeter is working properly. Then P2 11.2
and the related reliability, R2, can be calculated as (cf. Appendix F):
Assume that the switch and solenoid are independent. From the failure rate data,  1 
P2 = (0.82)2 = 0.672 the following table can be constructed, in analogy with Example 10.4: Gc ( s)  K c 1  
 I s 
R2 =1 - P2 = 1 - 0.672 = 0.33 The closed-loop transfer function for set-point changes is given by Eq. 11-36
Component  R P=1–R
 1 
To calculate the overall system reliability, substitute R2 = 0.33 for the reliability Pressure switch 0.34 0.712 0.288 with Kc replaced by K c 1   ,
value for a single D/P flowmeter, 0.18, in the R calculation of Example 10.4:  I s 
Solenoid switch/valve:   
 1  1
K c K IP K v K p K m 1  
5 H ( s)   I s  (s  1)
R   Ri  (0.33)(0.95)(0.61)(0.55)(0.64) Assume that the switch and solenoid are independent. Then, the overall reliability 
H sp ( s )  1  1
i 1
of the interlock system is, 1  K c K IP K v K p K m 1  
R  0.067   I s  (s  1)
R = (0.712)(0.657) = 0.468
Thus, the addition of the second D/P flowmeter has increased the overall system
where Kp = R = 1.0 min/ft2 ,
reliability from 0.037 (for Example 10.4) to 0.067.  = - ln (0.468) = 0.760
and  = RA = 3.0 min. Note also that 𝜏𝐼 = 𝜏 = 3.0 min.
1
MTBF   1.32 years

Solution Manual for Process Dynamics and Control, 4th edition
Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III

10-6 10-7
11-1

 psi   ft 3 / min   min  mA  20 ma  4 ma C’2(s)

KOL  K c K IP K v K p K m  (5.33)  0.75 Km   0.32 ma/  F
  0.2  1.0 2  4   3.2 50 F
Gd
 mA   psi   ft  ft  lb sol
ft3
X1
5
K v  1.2 , K IP  0.75 psi/ma ,  =5 min , Tsp ( s)  C’3sp(s)
^
C’3sp(s) E(s) P'(s) P’T(s) Q'2(s)
3𝑠 + 1 1 s X2 C’3(s)
+
𝐻 ′ (𝑠) 3.2 ( )( ) 3.2 1.0 Km +- Gc GIP Gv Gp +
= 3𝑠 3𝑠 + 1 = = lb sol ma ma ma psi USGPM
′ (𝑠)
𝐻𝑠𝑝 3𝑠 + 1 1
1 + 3.2 ( 3𝑠 ) (3𝑠 + 1) 3𝑠 + 3.2 0.94𝑠 + 1 T ( s) 
7.20K 2 ft3

s(5s  1  1.440K 2 )
C'3m(s) C'3(s)
(3  2) 1 T ()  lim sT ( s) 
7.20K 2 Gm
 ( s) 
For H sp  ma lb sol
s s s 0 (1  1.440K 2 ) ft3

h(t )  1  e 1.07  T ()  4.14 F K 2  3.34 F / psi

(b)
t  0.94ln 1  h(t ) c) From Fig. 11.7, since Ti  0 Gm ( s)  K m e   s assuming m = 0
m

h(t )  2.5 ft h(t )  0.5 ft t  0.65 min Pt() K v K 2  T () , Pt()  1.03 psi (20  4)ma 2 s  ma   2 s
Gm ( s )  e   2.67 e
lb sol  lb sol/ft 3 
(9  3) 3
h(t )  3.0 ft h(t )  1.0 ft t  and PtK v K 2  Ti K1  T , Pt  3.74 psi ft

Therefore, Pt ()  Pt  Pt()  4.77 psi  1 

Gc ( s)  K c 1  
h(t  0.65min)  2.5ft  I s 

G IP ( s )  K IP  0.3 psi/ma
h(t  )  3.0 ft
11.4
(10  20) USGPM USGPM
Gv ( s )  K v   1.67
11.3 (12  6) psi psi
(a) Controlled variable: c3
Manipulated variable: q2 For process and disturbance transfer function:
Disturbance variable: c2 (note: q1 and c1 are kept constant.)
Gc ( s )  K c  5 ma/ma If c2 changes, then q2 must be adjusted to keep c3 at the set point. Overall material balance for the tank,
Assume m = 0, v = 0, and K1 = 1, in Fig 11.7.
 USgallons  dh
 7.481 A  q1  q 2  q3
 ft 3  dt
a) Offset = Tsp ()  T ()  5 F  4.14 F  0.86 F
 K  As h is held constant at 4 ft by the overflow pipe:
K m K c K IP K v  2 
T ( s )  s  1 
b) 
Tsp ( s )  K  0  10  15  q3 (1)
1  K m K c K IP K v  2 
 s  1 
Thus q3  25
Using the standard current range of 4-20 ma,
Component balance for the solute,

11-2 11-3 11-4

 Therefore, these responses will change only if any of the transfer functions in the
d (c3 ) blocks of the diagram change.
7.481 Ah  q1c1  q2 c2  q3 c3 (2)
dt
i. c 2 changes. Then block transfer function G p (s ) changes due to K1. Hence Gc(s)
Linearize each term on the right hand side of Eq. 2 as described in Section 4.3: does need to be changed, and retuning is required.

q1c1  q1c1  q1c'1  q'1 c1 ii. Km changes. The close loop transfer functions changes, hence Gc(s) needs to be
q 2 c 2  q 2 c 2  q 2 c' 2  q' 2 c 2 (3) adjusted to compensate for changes in Gm and Km. The PI controller should be
retuned.
q 3 c3  q 3 c3  q 3 c ' 3  q ' 3 c3
iii. Km remains unchanged when zero is adjusted. The controller does not need to
At steady state: be retuned.
0  q1c1  q 2 c 2  q3 c3 (4)

Put (2) into deviation variable by considering (3) and (4): To verify the linearization results, the nonlinear model is used:
d (c 3 )
7.481 Ah  q1c1  q 2 c2  q3 c3
dc'3 dt
7.481 Ah  q 2 c ' 2  q ' 2 c 2  q 3 c ' 3  q ' 3 c3
dt q1  q 2  q3 The results agree with linearization.

As q1 is constant, q’3 = q’2: Step response of c3 to q2: (Gain 0.077 compared with linearized gain (Kp) 0.08 in 11.5
Eq. 6)
dc'3
7.481 Ah  q 2 c ' 2  q ' 2 c 2  q 3 c ' 3  q ' 2 c3
dt (a)

dc' From Eq. 11-26 we get the closed loop transfer function for set point changes
7.481 Ah 3  q 2 c' 2 q' 2 (c2  c3 )  q3 c'3 (5)
dt
𝑌 𝐾𝑚 𝐺𝑐 𝐺𝑣 𝐺𝑝
=
Taking Laplace transform and rearranging gives 𝑌𝑠𝑝 1 + 𝐺𝑐 𝐺𝑣 𝐺𝑝 𝐺𝑚

K1 K2 Substituting the information from the problem gives

C3 ( s)  Q2 ( s)  C 2 ( s) (6)
s  1 s  1 4
c  c3 lb sol/ft 3 q 7.481Ah 𝑌 𝑠(𝑠 + 4) 4 4
where K1  2  0.08 , K 2  2  0.6 and    15 min = = =
q3 USGPM q3 q3 𝑌𝑠𝑝 1 + 4 𝑠(𝑠 + 4) + 4 𝑠 2 + 4𝑠 + 4
𝑠(𝑠 + 4)
since A  D 2 / 4  12.6 ft 2 , and h  4 ft . 1
Or in standard form (Eq. 5-40), with 𝜏 = 2 and ζ= 1
0.08 0.6 Step response of c3 to c2: (Gain 0.6 compared with linearized gain (Kd) 0.6 in Eq.
Therefore, G p ( s)  and Gd ( s)  𝑌 1
15s  1 15s  1 6) =
𝑌𝑠𝑝 1 𝑠 2 + 𝑠 + 1
(c) 4
The closed-loop responses for disturbance changes and for setpoint changes can
be obtained using block diagram algebra for the block diagram in part (a). (b)

Given a unit step change in set point we obtain

11-5 11-6 11-7

4 The block diagram for this control system is the same as in Fig.11.8.
𝑌(𝑠) = Hence Eqs. 11-26 and 11-29 can be used for closed-loop responses to
𝑠(𝑠 2 + 4𝑠 + 4)
setpoint and load changes, respectively. 11.7
Using the Final Value Theorem we get
The transfer functions G p (s ) and G d (s ) are as given in Eqs. 11-66 and
4 4 11-67, respectively. Using block diagram algebra
𝑙𝑖𝑚 𝑠𝑌(𝑠) = 2 = =1
𝑠⟶0 𝑠 + 4𝑠 + 4 4
a) Substituting for Gc, Gm, Gv, and Gp into Eq. 11-26 gives Y  G d D  G pU (1)

Therefore y(∞) = 1  1 
Km Kc Kv   
 ~
U  Gc Ysp  Y  G pU   (2)
Y
  As   1
(c) As the step change is a unit step change, and we have shown that y(∞) = 1, s  1 GcYsp  GcY
Ysp  1 
we can say that offset = 0. This is consistent with the fact that the gain of 1 Kc Kv  K m From (2), U ~
the overall transfer function is 1, so no offset will occur. Normally  As  1 Gc G p
proportional control does not eliminate offset, but it does for this A
where    (1)
integrating process. Kc Kv Km Substituting for U in Eq. 1

(d) Using Eq. 5-50 or taking the inverse Laplace transform of the response
given above we get
For a step change in the setpoint, Ysp ( s)  M / s 1  G (G
c p
~
 ~
 G p ) Y  Gd (1  Gc G p ) D  G p Gc Ysp

M /s Therefore,
y  t   1  1  2t  e 2t
Y (t  )  lim sY ( s)  lim s  M
s 0 s 0
 s  1
Y G p Gc
Substituting the value of 0.5 for t gives  ~
Offset = Ysp (t  )  Y (t  )  M  M  0 Ysp 1  Gc (G p  G p )

y  t   0.264
and
b) Substituting for Gc, Gm, Gv, Gp , and Gd into (11-29) gives Gd (1  Gc G p )
Y
 1   1  
(e) We can tell from the response derived above that the response will not be    K K K  D 1  Gc (G p  G p )
Y ( s)
  As   c v m
oscillatory, since   1 . D( s)  1  s  1
1 Kc Kv   K m
 As 
11.8
where  is given by Eq. 1.
11.6 For a step change in the disturbance, D(s)  M / s
The available information can be translated as follows
  M /( K c K v K m )  M
For proportional controller, Gc ( s )  K c Y (t  )  lim sY ( s)  lim s   K K K 1. The outlets of both the tanks have flow rate q0 at all times.
s 0 s 0
 s(s  1)  c v m

Assume that the level transmitter and the control valve have negligible 2. To ( s )  0
 M 
dynamics. Then, Offset = Ysp (t  )  Y (t  )  0     0
 Kc Kv Km  3. Since an energy balance would indicate a first-order transfer function
Gm ( s)  K m between T1 and Q0 ,
Gv ( s )  K v Hence, offset is not eliminated for a step change in disturbance.
T (t ) 2
 1  e t / 1 or  1  e 12 / 1 , 1 = 10.9 min
T () 3

11-8 11-9 11-10

 a) 11.9
Therefore

T1 ( s) 3 F / (0.75 gpm) 4
  The given block diagram is equivalent to
Q0 ( s) 10.9s  1 10.9s  1
D
Gd
T3 ( s) (5  3) F /(0.75gpm)

2.67
  for T2(s) = 0 Ysp E E'
Q0 ( s)  2s 1  2s 1 +-
P + Y
+- Gc + G
~ ~
T ( s) (78  70)  F /(12  10)V 4 Y1- Y2 ~
4. 1   ~
G*(1-e-s)
V1 ( s) 10s  1 10s  1

T3 ( s) (90  85)  F /(12  10)V 2.5

 
V2 ( s) 10s  1 10s  1
For the inner loop, let
5. 52 =50 min or 2 = 10 min P Gc
b)  Gc  ~ ~
Since inlet and outlet flow rates for tank 2 are q0 and the volumes of E 1  Gc G * (1  e  s )
the tanks are equal,
In the outer loop, we have
T3 ( s) q0 / q0 1
  Y Gd G
T2 ( s)  2 s  1 10.0s  1 
V3 ( s) D 1  Gc G
6.  0.15
T3 ( s)
Substitute for G c ,
 30 
7. T2 (t )  T1  t    T1 (t  0.5)
 60  Gd G
Y
T2 ( s) 
 e 0.5 s D Gc G
T1 ( s) 1 ~ ~
c) The control configuration in part a) will provide the better control. As is 1  Gc G * (1  e  s )
evident from the block diagrams above, the feedback loop contains, in
 
Using these transfer functions, the block diagrams are as follows. ~ ~
addition to Gc, only a first-order process in part a), but a second-order-
Y G G 1  G G * (1  e  s )
plus-time-delay process in part b). Hence the controlled variable responds  d ~ * c ~s
D 1  Gc G (1  e )  Gc G
faster to changes in the manipulated variable for part a).

11.10

a) Derive CLTF:

Y  Y3  Y2  G3 Z  G2 P

11-11 11-12 11-13

Y  G3 ( D  Y1 )  G2 K c E a)
CF'(s) GD ( s)   K c  D s
GL
Y  G3 D  G3 G1 K c E  G2 K c E Kg/m3
with Kc >0 as the controller should be reverse-acting, since P(t) should
X1
Y  G3 D  (G3G1 K c  G2 K c ) E E   K mY ~ increase when Cm(t) decreases.
C'sp(s) C'sp(s) E P' Pv' Q'A X2
+ C'(s)
Km +- GPI + KIP Gv Gp + I/P transducer
Y  G3 D  K c (G3G1  G2 ) K mY Kg/m3 ma ma +
ma psig m3/min

GD (15  3) psig psig

K IP   0.75
Y

G3 (20  4) ma ma
D 1  K c (G3G1  G2 ) K m C'm(s) C'TL C'
Gm GTL Control valve
ma Kg/m3 Kg/m3
b) Characteristic Equation:
Kv
Gv ( s ) 
1  K c (G3G1  G2 ) K m  0 b) v s  1
 5 4 
1  Kc   0
 s  1 2s  1
Transfer Line: 5 v  1 ,  v  0.2 min

Volume of transfer line =  /4 (0.5 m)2(20m)= 3.93 m3

 5(2 s  1)  4( s  1)  dq
pv  3
1 Kc  0 Kv  A  0.03(1 / 12)(ln 20)(20) 12

 ( s  1)(2 s  1)  Nominal flow rate in the line = q A  q F  7.5 m 3 / min dpv pv  pv

( s  1)(2s  1)  K c 5(2s  1)  4( s  1)  0 3.93 m 3 pv  3

Time delay in the line =  0.52 min q A  0.5  0.17  0.03(20) 12
7.5 m 3 /min
2s 2  s  1  K c (10s  5  4s  4)  0
pv  3
GTL ( s )  e 0.52 s 0.03(20) 12
 0.5  0.17  0.33
2s 2  (14K c  1) s  ( K c  1)  0
Composition Transmitter: m 3 /min
Necessary conditions: K c  1 / 14 and K c  1 K v  (1 / 12)(ln 20)(0.33)  0.082
psig
(20  4) ma ma
Gm ( s )  K m   0.08
For a 2nd order characteristic equation, these conditions are also sufficient. (200  0) kg/m 3 kg/m 3
0.082
Therefore, K c  1 for closed-loop stability. Gv ( s ) 
0.2s  1
Controller
Process
From the ideal controller in Eq. 8-14
Assume cA is constant for pure A. Material balance for A:
 1 
 
11.11 ~
P ( s)  K c 1   E ( s)  K c  D s C sp ( s)  C m ( s)
 I s  V
dc
 q A c A  q F c F  ( q A  q F )c (1)
~
In the above equation, set C sp ( s )  0 in order to get the derivative on the dt
process output only. Then,
Linearizing and writing in deviation variable form
 1 
G PI ( s)  K c 1  
 I s 

11-14 11-15 11-16

 dc  11.13  When  I  10 , 50s  60s  10 1  Kc  s  Kc  0
3 2
V  c A q A  q F c F  (q A  q F )c   c q A
dt Set s  j :
Taking Laplace transform
(a) 1 
Kc
 0  5s 2  6 s  1  K c  0
 50 3
 10 1  K c    j  60 2  K c  0
Vs  (q A  q F )C ( s)  (c A  c )Q A ( s)  q F C F ( s) (2)  5s  1 s  1
Applying the quadratic formula yields the roots:
From Eq. 1 at steady state, dc / dt  0 , Re: −60𝜔2 + 𝐾𝑐 = 0 (1)
Im: −50𝜔3 + 10(1 + 𝐾𝑐 )𝜔 = 0 (2)
c  (q A c A  q F c F ) /(q A  q F )  100 kg/m 3 6  36  20 1  K c    0 : K cm  1.09
s
10 To have a stable system, we have:
Substituting numerical values in Eq. 2, To have a stable system, both roots of the characteristic equation must have Kc  0
negative real parts. Thus, 20 1  Kc   0  Kc  1 (c) Adding larger amounts of integral weighting (decreasing  I ) will destabilize
5s  7.5C ( s)  700 Q A ( s)  7 C F ( s)
 1  the system
K c 1  
0.67 s  1C ( s)  93.3 Q A ( s)  0.93 C F ( s)   I s   0   5s 3  6 s 2  s  K s  K  0
(b) 1  I  c 
 5s  1 s  1 c
11.14
When  I  0.1 , 0.5s  0.6s  0.11  Kc  s  Kc  0
93.3
G p ( s)  
3 2

0.67s  1 From the block diagram, the characteristic equation is obtained as

Using direct substitution, and set s  j :
Gd ( s ) 
0.93   2  
0.67s  1  (1)  s  3    2   1 
  
 0.5 3
 0.11  K c    j  0.6  K c  0
2
1  Kc    0
1  (1)  2    s  1   s  10 
−0.6𝜔2 + 𝐾𝑐 = 0  
  s  3  
Re: (1)
Im: −0.5𝜔3 + 0.1(1 + 𝐾𝑐 )𝜔 = 0(2)
that is,
11.12   0 : K cm  0.136
 2  2  1 
1 Kc  0
 s  5   s  1  s  10 
To have a stable system, we have:
Simplifying,
The stability limits are obtained from the characteristic Eq. 11-83. Hence 0  K c  0.136 s  14s  35s  (4 K c  50)  0
3 2

if an instrumentation change affects this equation, then the stability limits

Set s  j :
will change and vice-versa.
 When  I  1 , 5s  6s  1  Kc  s  Kc  0    35  j  14 2   4 K c  50   0
3 2 3

a) The transmitter gain, Km, changes as the span changes. Thus Gm(s) Set s  j : Re: −14𝜔2 + 4𝐾𝑐 − 50 = 0 (1)
changes and the characteristic equation is affected. Stability limits would Im: (𝑗) − 𝜔3 + 35𝜔 = 0 (2)
be expected to change.
 5  1  K    j  6
3
c
2
 Kc  0   0 : Kcm  135
b) The zero on the transmitter does not affect its gain Km. Hence Gm(s)   0 : Kcm  12.5
remains unchanged and stability limits do not change. 12.5  K c  135
Re: −6𝜔2 + 𝐾𝑐 = 0 (1)
3
−5𝜔 + (1 + 𝐾𝑐 )𝜔 = 0
Changing the control valve trim changes G  s  . This affects the
Im: (2)
c)
  0 : K cm  6
characteristic equation and the stability limits would be expected to
change as a result. To have a stable system, we have:
Kc  0

11-17 11-18 11-19

11.15  100 I  3  1  5K c  I   j  20 I  2  5K c  0

11.16
Re: −20𝜏𝐼 𝜔2 + 5𝐾𝑐 = 0 (1)
Substituting the transfer functions into the characteristic equation in (11-81)  1 
gives: Gc ( s)  K c 1  
Kp  I s  Im: (𝑗) − 100𝜏𝐼 𝜔3 + (1 + 5𝐾𝑐 )𝜏𝐼 𝜔 = 0 (2)
K c K v e  s
Y K mGcGvG p  ps 1 K c K v K p e  s
 1.3 25 K c
   Gv ( s ) 
Kv
   0 :  Im 
Ysp 1  GcGvG pGm  s
Kp  p s  1  K c K v K p e  s (10 / 60) s  1 0.167s  1 1  5K c
1  Kc Kve
 ps 1   0 : K cm  0
Y e s e s G p ( s)  
1

1
since A  3 ft  22.4
2 gal To have a stable system, we have:
Let Kc  Kv  K p   p    1 , we have  ; thus, GOL 
Ysp s  1  e  s 1 s As 22.4s ft 25 K c
K c  0, I 
Simulate the above relation through MATLAB, we have: 1  5K c
Gm ( s)  K m  4
The stability region is shown in the figure below:
Characteristic equation is
7

 1    1.3   1 
1  K c 1     (4)  0 6
Stability region
  I s   0.167s  1  22.4s 
5

(3.73 I ) s  (22.4 I ) s  (5.2 K c  I ) s  (5.2 K c )  0

3 2 I
4

Use direct substitution, and set s  j :

 3.73  I
3
 5.2 K c I   j  22.4 I  2  5.2 K c  0 3

Re: −22.4𝜏𝐼 𝜔2 + 5.2𝐾𝑐 = 0 (1) 2

Step response of a closed loop function Im: (𝑗) − 3.73𝜏𝐼 𝜔3 + 5.2𝐾𝑐 𝜏𝐼 𝜔 = 0 (2)
  0 :  cm  0.167
0.7
1

0.6 To have a stable system, we have: 0

K c  0, I  0.167
0 1 2 3 4 5 6 7
K
c
0.5

c) Find  I as K c  
0.4
Response

11.17  25K c   25 
  lim   5
0.3
lim 
  s  1  5  N (s) 1  5K c  Kc  1/ K c  5 
K c 
GOL ( s )  K c  I  
2 
  I s  (10s  1)  D( s )
0.2

0.1   I  5 guarantees stability for any value of Kc. Appelpolscher is

D( s)  N ( s)   I s(100s 2  20s  1)  5K c ( I s  1)  0 wrong yet again.
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time
 100 I s 3  20 I s 2  (1  5K c ) I s  5K c  0
Figure S11.15 Step response of a closed loop function
As shown in the figure, the time delay will not lead to an inverse response. Set s  j , we have:

11-20 11-21 11-22

11.18 11.20
1  0.5s
Gc ( s )  K c e s  The characteristic equation is
1  0.5s
KV 20s 3  47s 2  (14.5  0.053K c ) s  (1  0.106K c )  0 0.5 K c e 3 s
GV ( s)  1 0 (1)
V s  1 Substituting s=j in above equation, we have: 10s  1
a) Using the Pade approximation
dws 0.6 lbm/s 1  (3 / 2) s
Kv    0.106 47  1  0.106 K c   20  14.5  0.053K c  j  0
2 3
e 3 s 
dp p 12 2 12  4 ma 1  (3 / 2) s
5 v  20 s v  4 s Thus, we have: in (1) gives
 47 2  1  0.106 K c  0    0.587 15s 2  (11.5  0.75K c ) s  (1  0.5K c )  0
2.5e  s  
G p (s) 
20  14.5  0.053K c  0  K c  143.46
3
10s  1 Substituting s=j in above equation, we have:
(20  4) ma ma Therefore, the maximum gain, Kc = 143.46, is a satisfactory 15 2  1  0.5Kc  11.5  0.75Kc  j  0
Gm ( s )  K m   0.4  approximation of the true value of 139.7 in (a) above
(160  120)  F F Thus, we have:
15 2  1  0.5 K c  0    0.760
Characteristic equation is  
 11.5  0.75K c  0  K cm  15.33
11.19
 0.106  2.5e 
s
1  ( K c )  0.4  0 (1) b) Substituting s = j in (1) and using Euler's identity.
 4s  1  10s  1  4(1  5s )
a) G (s) 
(25s  1)(4 s  1)(2 s  1) e 3 j  cos(3)  j sin(3)
a) Substituting s=j in (1) and using Euler's identity Gc ( s )  K c gives
D( s)  N ( s)  (25s  1)(4s  1)(2s  1)  4 K c (1  5s )  0 10 j  1  0.5K c cos(3)  j sin(3)  0
e-j=cos – j sin 
200s 3  158s 2  (31  20K c ) s  1  4 K c  0 Then,
gives
Substituting s=j in above equation, we have: 1  0.5 K c cos(3)  0 (2)
-402 +14j + 1 + 0.106 Kc (cos – jsin)=0 158 2  1  4 K c   200 3  31  20 K c  j  0 and 10  0.5K c sin(3)  0 (3)
Thus Thus, we have: From (2) and (3)
-402 + 1 + 0.106 Kc cos = 0 (2) tan(3) = -10 (4)
 158 2  1  4 K c  0    0.191
  Eq. 4 has infinite number of solutions. The solution for the range
200  31  20 K c  0  K cm  1.19
3
and 14 - 0.106Kc sin =0 (3) /2 < 3 < 3/2 is found by trial and error to be  = 0.5805.
Then from Eq. 2, Kc = 11.78
From (2) and (3), b) (25s  1)(4s  1)(2s  1)  4 K c  0 The other solutions for the range 3 > 3/2 occur at values of  for which
14 200s 3  158s 2  31s  (1  4 K c )  0 cos(3) is smaller than cos(3  5.805). Thus, for all other solutions of ,
tan   (4)
402  1 Substituting s=j in above equation, we have:
Eq. 2 gives values of Kc that are larger than 11.78. Hence, stability is
ensured when
158 2  1  4 K c   200 3  31  j  0
Solving (4),  = 0.579 by trial and error. 0 < Kc < 11.78
Thus, we have: To solve Eqs. 2 and 3, another way is to use Newton’s method. With
Substituting for  in (3) gives 158 2  1  4 K c  0    0.394 initial guess Kc = 5,  = 0 ( steady state), the solution to Eqs. 2 and 3 is:
  Kc = -2,  = 0
 200  31  0  K cm  5.873
3

Kc = 139.7 = Kcm
c) Because Kc can be much higher without the RHP zero being present, With a different initial guess ( e.g., Kc = 5,  = 5), the solution is:
the process can be made to respond faster.
Frequency of oscillation is 0.579 rad/sec Kc = 11.78,  = 0.5805
Again, c = 0.5805 and the stability is ensured when
b) Substituting the Pade approximation into (1) gives: 0 < Kc < 11.78

11-23 11-24 11-25

(c). 11.21 0
Kc =-1, unstable

Kc =15.33, unstable -2
150
a) To approximate GOL(s) by a FOPTD model, the Skogestad approximation -4
100
technique in Chapter 6 is used.
50 Initially, -6

response
3K c e  (1.50.30.2) s 3K c e 2 s
0
GOL ( s)   -8

(60s  1)(5s  1)(3s  1)(2s  1) (60s  1)(5s  1)(3s  1)(2s  1)

response

-50
-10

-100
Skogestad approximation method to obtain a FOPTD model:
Time constant  60 + (5/2) -12

-150
Time delay  2 +(5/2) + 3 + 2 =9.5 -14
0 100 200 300 400 500 600 700 800 900 1000
-200 Then time

3K c e 9.5 s Kc =3.678, stable but with oscillations

GOL ( s ) 
-250
0 50 100 150 200 250 300 350 400 450 500 1
time
62.5s  1 0.9
Kc =14, stable but slow convergence with oscillations
150 0.8
(b) The characteristic equation is
0.7
3K c e 9.5 s
100

1 0 (1) 0.6
50
62.5s  1

response
0.5
0 Substituting s = j in (1) and using Euler's identity. 0.4
response

-50 0.3
9.5 j
e  cos(9.5 )  j sin(9.5 ) 0.2
-100
gives 0.1

3Kc cos  9.5   1  62.5  3Kc sin(9.5)  j  0

-150
0
0 100 200 300 400 500 600 700 800 900 1000
-200
Then, time

1  3K c cos(9.5 )  0
Kc =15, unstable
-250
0 50 100 150 200 250 300 350 400 450 500 (2) 250
time

Kc =6, stable and quick convergence

and 62.5  3K c sin(9.5 )  0 (3) 200

150 150

100 From (2) and (3) 100

50 tan  9.5   62.5 (4) 50

response
0
0
-50
Eq. 4 has infinite number of solutions. The solution for the range
response

/2 < 9.5 < 3/2 (to make sure 𝐾𝑐 is positive) is found by trial and error
-50 -100

-100
to be  = 0.1749. -150

-150 Then from Eq. 2, Kc = 3.678 -200

Hence, stability is ensured when -250

0 100 200 300 400 500 600 700 800 900 1000
-200
time
0< Kc < 3.678
-250
0 50 100 150 200 250 300 350 400 450 500
time
c) Conditional stability occurs when K c  K cu  3.678;   0.1749 Figure S11.21 Simulation results of different 𝐾𝑐 settings
Figure S11.20 Simulation results of different 𝐾𝑐 settings

11-26 11-27 11-28

 11.24 Kc can be arbitrarily large for this PI controlled second order system and
11.22 still maintain stability.

The open loop process transfer function is:

Characteristic equation is: Kp 2
Gp  
( 1s  1)( 2 s  1) (4s  1)(s  1) 11.25
5  as s 3  3s 2  3s  1  5K c  aK c s
1  Gc G p Gv G m  1  K c 
( s  1) 3 ( s  1) 3 The controller transfer function is:
s 3  3s 2  (3  aK c ) s  1  5K c s Gc  K c (1 
1
)  2
1 First we use Eq. 11-26 to get the closed loop transfer function

( s  1) 3 Is 2s
10
A necessary condition for stability is all the coefficients of the numerator (a) According to Eq. 11-26, the closed loop transfer function for set point 𝑌 (𝑠 + 1)(2𝑠 + 1) 10 10
= = =
are positive. tracking is: 𝑌𝑠𝑝 1 + 10 (𝑠 + 1)(2𝑠 + 1) + 10 2𝑠 2 + 3𝑠 + 11
When a < 3/Kc (Kc > 0), the coefficient of s becomes negative so the 2 1 (𝑠 + 1)(2𝑠 + 1)
 (2  )
control system becomes unstable. Y K m G p Gc Gv (4s  1)( s  1) 2s
  Or in standard form
Ysp 1  Gm G p Gc Gv 2 1
1  (2  )
(4s  1)( s  1) 2s
11.23 2 1 10⁄
 (2  ) 𝑌 11 3√22 √22
Y K m G p Gc Gv (4s  1)(s  1) 2s 1 = ζ= 𝜏=
   2 𝑌𝑠𝑝 2 2 3 44 11
(a) Ysp 1  Gm G p Gc Gv
1
2
 (2  )
1 s  s 1 11 𝑠 + 11 𝑠 + 1
Offset = hss – hfinal = 22.00 – 21.92 = 0.08 ft (4s  1)(s  1) 2s
(b) The closed loop transfer function is: The time at which the maximum occurs is given by Eq. 5-52
20  4 mA
Km   1.6 mA / ft Y 1
10 ft  𝑡𝑝 = 𝜋𝜏 𝑡𝑝 = 1.41
Ysp s 2  s  1 ⁄
15  3 √1 − ζ2
K IP   0.75 psi / mA
20  4 (b)
K V  0.4 cfm / psi K 5 The characteristic equation is the denominator of the closed loop transfer (b)
and c
We have: function, which is underdamped (ζ = 0.5):
s2  s 1 The response is given by
K OL  K m K c K IP KV G p  1.6  5  0.75  0.4 K p  2.4 K p
20
(c) 𝑌(𝑠) =
Offset equals to: 𝑠(2𝑠 2 + 3𝑠 + 11)
1
M 22  20 For stability analysis, Gc  K c (1  ) is substituted into 1  GmG p GcGv
offset    0.08 4s
1  K OL 1  2.4 K p The Final Value Theorem gives the steady state value as
and we get:
K p  10 ft / cfm 2 1
1  G m G p Gc Gv  1   4(1  ) 20
(4s  1)(s  1) 4s 𝑦(∞) =
11
(c)
2s( s  1)  Kc 2s 2  2s  Kc
Add integral action to eliminate offset.   Subtracting the steady state value from the set point change gives offset as
2s( s  1) 2s( s  1)
To find the stability region, the roots of the numerator polynomial should 2
be on the right half plane. For this 2nd order polynomial, this means: offset=
11
Kc  0

11-29 11-30 11-31

(c) Substituting the values from above and in part (a), we get, A sketch would look like this

The period of oscillation is given by Eq. 5-55 𝑌 𝐾𝑐 𝐸

=
𝑌𝑠𝑝 1 + 𝐾𝑐 𝐸
2𝜋𝜏
𝑃= 𝑃 = 2.83
Multiplying by a unit step change in set point gives
√1 − ζ2
𝐾𝑐 𝐸 1 𝐾𝑐 𝐸
𝑌(𝑠) = ⟹ 𝑦(𝑡) =
1 + 𝐾𝑐 𝐸 𝑠 1 + 𝐾𝑐 𝐸
(d)
A sketch might look like this (the step change at t = 5)

Figure S11.26b Step response to unit step change with integral control
Kc E
1  Kc E As evident in the sketch, there is no offset for this controller.

11.27

Figure S11.25 y(t) responses as a function of time. 8

 s  2
3
Y Gd 8
Hint: You do not need to obtain the analytical response y(t) to answer Figure S11.26a Step response to unit step change with   
D 1  GcGvG pGm 1  K 1 8 1 s 3  6s 2  12s  8  8K c
the above questions. Use the standard second order model expressed in proportional control.
 s  2
c 3
terms of ζ and τ (see Chapter 5).
As evidenced by the sketch, there is offset for this controller. The characteristic equation for above is shown as:
s 3  6s 2  12s  8  8K c  0
For part (b), we substitute the values into Eq. 11-26 to get Substituting s=j in above equation, we have:
11.26 6 2  8  8K c  12   3  j  0
𝑌 𝐸 1
= =
𝑌𝑠𝑝 𝜏𝐼 𝑠 + 𝐸 𝜏𝐼 𝑠 + 1 Thus, we have:
𝐸 6 2  8  8 K c  0   2 3
The closed loop transfer function for a set point change (Eq. 11-26), is  
 12    0  K cm  8
3
given by Multiplying by a unit step change in set point gives
So K c  1 is stable; K c  8 is marginally stable, and K c  27 is unstable
𝑌 𝐾𝑚 𝐺𝑐 𝐺𝑣 𝐺𝑝 1 1 𝐸
𝑌(𝑠) = 𝜏 ⟹ 𝑦(𝑡) = 1 − exp (− 𝑡) 1
= 𝐼 𝑠 𝜏𝐼 For a step change D  , applying the final value theorem:
𝑌𝑠𝑝 1 + 𝐺𝑐 𝐺𝑣 𝐺𝑝 𝐺𝑚 𝐸𝑠+1 s

11-32 11-33 11-34

  1  So the answers to parts (a)-(c) are: 3 2  1  K c  3   3  j  0
 8  1
Offset: lim y  t   lim  s 3 s
 (a) stable
Thus, we have:
 s  6s  12s  8  8K c  1  K c
t  s 0 2
3 2  1  K c  0    3
   
 3    0  K c ,max  8
3
(b) unstable
(b)
We conclude that the system will be stable if
(c) unstable
8
Kc  8
 s  2 (3)
3
Y Gd 8 I s The following plot shows the Simulink responses and confirms the above
   Simulation results are in Figure S11.29a.
D 1  GcGvG p Gm  1   I s  s  2   K c  I s  1
3
8 answers:
1  K c 1    1 1
 Is   s  2
3

3
1 Kc=7.9
For a step change D  , applying the final value theorem:
s 2.5 Kc=8

 1  Kc=8.1
 8 Is 
2
Offset: lim y  t   lim  s s
0
 I 
    c I
  
t  s 0 3
s s 2 K s 1 1.5
 
1

Y
So there is no offset for PI controller.
0.5

0
11.28
-0.5

The closed loop transfer function for set point changes is given by -1
0 50 100 150
Time (min)
𝑌 𝐾𝑚 𝐺𝑐 𝐺𝑣 𝐺𝑝 Figure S11.28 y(t) responses with different Kc
=
𝑌𝑠𝑝 1 + 𝐺𝑐 𝐺𝑣 𝐺𝑝 𝐺𝑚 Figure S11.29a: System response to a unit step setpoint change. Note that the
11.29 system is stable at Kc=7.9, marginally stable at Kc=8, and unstable at Kc=8.1.
Substituting the information in the problem gives
b) PD controller:
𝑌 𝐾𝑐 (𝑠 + 3) 𝐾𝑐 (𝑠 + 3) We derive the transfer function as follows:
= = a) Proportional controller:
𝑌𝑠𝑝 (𝑠 + 1)(0.5𝑠 + 1)(𝑠 + 3) + 3𝐾𝑐 0.5𝑠 3 + 3𝑠 2 + 5.5𝑠 + 3 + 3𝐾𝑐 We derive the transfer function as follows:
Gc  K c 1   D s 
Y K mGc GvG p
So the characteristic equation is  Y K mGc Gv G p
YSP 1  GmGc Gv G p 
YSP 1  GmGc Gv G p
0.5𝑠 3 + 3𝑠 2 + 5.5𝑠 + 3 + 3𝐾𝑐 = 0 1
Kc (1)
 s  1  K c 1   D s 
1
3
Y Kc Kc
Substituting s=j in above equation, we have:    s  1
3

 s  1  K c s3  3s 2  3s  1  K c Y
3
YSP 1  K 1 
3 2  3  3K c  5.5  0.5 3  j  0
 s  1 c
YSP 1  K 1   s
D 
1
c 3

 s  1
3
Thus, we have:
The characteristic equation of (1) is the following:
3 2  3  3K c  0    11 Kc  D s  1
   , (4)
 5.5  0.5  0  K c ,max  10 s3  3s 2  3s  1  K c  0
3
(2) s3  3s 2  3s  1  K c   D K c s
Substituting s=j in above equation, we have: where Kc=10.

11-35 11-36 11-37

Chapter 12 ©
The characteristic equation of (4) is the following:
12.2
s3  3s 2  10 D  3 s  11  0 (5)
4(1  s )
Substituting s=j in above equation, we have: G  Gv G p Gm 
s
3 2  11  10 D  3    3  j  0
a) Let G  G . Factor the model as
Thus, we have: 12.1
 3  11  0 G  GG
2
1
   D ,min   0.0667
10 D  3     0
3
15 with:
For K = 1, 1=10, 2=5, and =0, the PID controller settings are obtained using
4
1.8
\tau_d=0.1, stable
Eq.12-14 as G  1  s, G 
s
1.6 The controller design equation in (12-20) is:
1 1   2 15 1  2
1.4 Kc   , I = 1+2=15 , D   3.33 Gc
* 1
f
1.2
K c c 1   2 G
with a given first-order “filter”,
response

1
The characteristic equation for the closed-loop system is
1
f 
0.8

    
cs 1
0.6
1 1.0  
1   K c 1    D s   0 Substitute,
  I s   (10s  1)(5s  1) 
0.4

1 s
0.2
G *c 
0 Substituting for Kc, I, and D, and simplifying gives 4 cs 1
0 10 20 30 40 50 60 70 80 90 100
time b) The equivalent controller in the classical feedback control configuration in
(i) 𝜏𝐷 > 𝜏𝐷,𝑚𝑖𝑛  c s  (1  )  0 Fig. 12.6(a) is:
G *c
\tau_d=0, unstable
In order for the closed loop system to be stable, the coefficients of this first-order Gc 
1500
1  G *c G
polynomial in s must be positive. Thus,
1000
Substitute to give,
c > 0 Gc 
1
500 and 4( c  1)
(1+) > 0   > 1. Thus Gc is a proportional-only controller.
response

Results:
-500
12.3
a) The closed loop system is stable for  > 1
-1000

-1500
b) Choose c > 0 For the FOPTD model, K = 2,  = 1, and  = 0.2.
0 10 20 30 40 50 60 70 80 90 100

Using entry G in Table 12.1 for c = 0.2

time
c) The choice of c does not affect the robustness of the system to changes in a)
(ii) 𝜏𝐷 < 𝜏𝐷,𝑚𝑖𝑛
. For c  0, the system is unstable regardless of the value of . For
c > 0 , the system is stable if  >1, regardless of the value of c .  1
Kc    1.25
Figure S11.29b Simulation results of different 𝜏𝐷 settings K ( c   ) 2(0.2  0.2)
I  1
b) Using entry G in Table 12.1 for c = 1

[Type here] 12-1 12-2

11-38 [Type here]
 (b) Controller settings using AMIGO method
 1 12.4
Kc    0.42
K ( c   ) 2(1  0.2) The model parameters are: K  4,  =3
I  1 For this model, use the right-hand column of Table 12.5.
The process model is,
c) From Table 12.4 for a disturbance change 4e3s 0.35 0.35
G( s)  (assume the time delay has units of minutes) (1) Kc    0.029
s K 12
KKc = 0.859(/)-0.977 or Kc = 2.07
 I  13.4  13.4(3)  40.2
/I = 0.674(/)-0.680 or I = 0.49 (a) Proportional only control, Gc(s) =Kc. The characteristic equation is:

d) From Table 12.4 for a set-point change 1  K c G( s )  0 12.5

KKc = 0.586(/)-0.916 or Kc = 1.28
Substitute and rearrange,
/I = 1.03 0.165(/) or I = 1.00
Assume that the process can be modeled adequately by the first-order-plus-time
s  4 K c e3s  0
e) Conservative settings correspond to low values of Kc and high values of I. delay-model in Eq. 12-10. The step response data and the tangent line at the
Clearly, the IMC method (c = 1.0) of part (b) gives the more conservative inflection point for the slope-intercept identification method of Chapter 7 are shown
Substitute the stability limit conditions from Section 11.4.3: s = j,  = u, and in Fig. S12.5.
settings; the ITAE method of part (c) gives the least conservative settings. Kc.= Kcu:
The controller setting for (a) and (d) are essentially identical.
ju  4 K cu e3u j  0 (2)
f) A comparison for a unit step disturbance is shown in Fig. S12.3 18

17
Apply Euler’s identity, eθj  cos(θ)  j sin(θ) :
16
e3u j  cos(3u )  j sin(3u )

Output
15

Substitute into (2), 14

ju  4Kcu [cos(3u )  j sin(3u )]  0

13

12
0 2 4 6 8 10 12
Collect terms for the real and imaginary parts:
Time
4Kcu cos(3u )  0 (3)
Figure S12.5. Step response data and tangent line at the inflection point.
u  4Kcu sin(3u )  0 (4)
This estimated model parameters are:
For (3), because Kcu  0 , it follows that:
 psi  psi  16.9  12.0 mA 
K = KIP Kv(KpKm) =  0.75  0.9   = 1.65
cos ( 3u ) = 0  3u 
π
(5)  mA  psi  20  18 psi 
2  = 1.7 min
π
 u   0.5236 rad/min (6)  +  = 7.2 min   = 5.5 min
6 1
a) Since / > 0.25, a conservative choice of c =  is used. Thus, c = 2.75
From (4) – (6), 2
min. From Case H in Table 12.1:
 0.5236
Figure S12.3. Comparison of PI controllers for a unit step disturbance. 0.5236  4 K cu sin ( )  0  K cu   0.130
2 4

12-3 12-4 12-5

θ  3
1
  4 12.7
Kc  2  1.76 Kc  2  2  0.24
K  θ  3
c K ( c  ) 5(3  ) a.i) The model reduction approach of Skogestad gives the following
2 2 2
approximate model:
3
θ θ I    4
 5.5
I    6.35 min, D   0.736 min 2 2
2 2  θ e 0.028s
 4(3) G(s) 
D    1.1 ( s  1)(0.22s  1)
b) From Table 12.6, the AMIGO tuning parameters are: 2   2(4)  3
Since / < 0.25, an aggressive choice of c =  = 0.028 is made. From
1  1  5.5  The simulated process for a step change in the set point is plotted below for both
Case I in Table 12.1 with 3 = 0, the IMC settings are:
Kc   0.2  0.45   0.2  0.45  1
K   1.65  1.7  the PI and PID controllers. Note that the PID controller was implemented in the
proper form to eliminate derivative kick (see chapter 8). 1 1   2
0.4  0.8 0.4(1.7)  0.8(5.5) Kc   21.8
I   (1.7)  3.8 min K c  θ
  0.1 1.7  0.1(5.5)
 1 2
0.5 0.5(1.7)(5.5)  I   1   2  1.22, D   0.180
D    0.78 min 1   2
0.3   0.3(1.7)  5.5

c) From Table 12.4, the ITAE PID settings for a step disturbance are a.ii) To use the AMIGO tuning relations in Table 12.6, the model reduction
method of Skogestad can be used to reduce the model to a FOPDT model.
KKc = 1.357(/)-0.947 or Kc = 2.50 The time constant in the resulting FOPDT model is the largest time constant
/I = 0.842 (/)-0.738 or I = 2.75 min in the full-order model plus one half of the next biggest time constant, 1 +
D/ = 0.381 (/)0.995 or D = 0.65 min 0.5(0.2) = 1.1. The time delay in the resulting FOPDT model is half of the
second-biggest time constant in the full-order model, 0.5(0.2) = 0.1. The
other smaller time constants are neglected.
d) The most aggressive controller is the one from part c, which has the
highest value of Kc and smallest value of τI
e0.1s
G( s) 
1.1s  1
12.6
The AMIGO rules for a PID controller in Table 12.6 give:

The model for this process has K=5, τ=4, and θ=3. The PI controller parameters 1   1.1 
Kc   0.2  0.45    0.2  0.45   5.15
for an FOPDT model using IMC tuning are given by entry G in Table 12.1: K   0.1 
Figure S12.6: Responses to a step change in the set point at t = 1for PI and PID 0.4  0.8 0.4(0.1)  0.8(1.1)
 controllers. I   (0.1)  0.44
Kc  
4
 0.13   0.1 0.1  0.1(1.1)
K ( c   ) 5(3  3) The PID controller allows the controlled variable to reach the new set point more 0.5 0.5(0.1)(1.1)
D    0.049
I   4 quickly than the PI controller, due to its larger Kc value. This large Kc allows an 0.3   0.3(0.1)  1.1
initially larger response from the controller during times from 1 to 4 minutes. The
The parameters for a PID controller are given by entry H in Table 12.1: reason that the Kc can be larger is that, after the controlled variable begins to change b) The simulation results shown in Figure S12.7 indicate that the IMC
and move toward the set point, the derivative term can “put on the brakes” and slow controller is superior for a step disturbance due to its smaller maximum
down the aggressive action so the controlled variable lands nicely at the set point. deviation and lack of oscillations. This result makes sense, given that we
made an aggressive choice for τC for the IMC controller.

12-6 12-7 12-8

 Closed-loop responses are compared for four values of β: 1, 0.7, 0.5 & 0.3. parallel form, the closed-loop system becomes unstable for D  5.4; for the
series form, it becomes unstable for D  4.5.

3
Parallel form
Series form

2.5

y(t) 1.5

Figure S12.8. Closed-loop responses for different values of β. 0.5

Figure S12.7. Closed-loop responses to a unit step disturbance at t=1. As shown in Figure S12.8, as β increases the set-point response becomes faster
but exhibits more overshoot. The value of β =0.5 seems to be a good choice. The 0
0 50 100 150 200 250 300
disturbance response is independent of the value of β. Time
12.8
Figure S12.9. Closed-loop responses for parallel and series controller forms.
12.9
From Eq. 12-40 (with γ=0):
1 dy  12.10
p (t )  p  K c   ysp (t )  ym (t )   K c   e(t*)dt *  D m 
 I dt  a) From Table 12.2, the controller settings for the series form are:

This control law can be implemented with Simulink as follows: a) Block diagram
  
: K c  K c 1  D   0.971 X1'
 I  Gd

CONTROLLER
 I  I  D  26.52
X'sp X'sp E P' W'2 X'
+
I D Km +- GC Gv Gp +
WEIGHTING FACTOR
D   2.753 (mA) (mA) (mA) (Kg/min)
SET-POINT PROPORTIONAL CONTROLLER
I  D
+
K

ß +- ACTION C
OUTPUT
+
Closed-loop responses generated from Simulink are shown in Fig. S12.9. X'm
The series form results in more oscillatory responses; thus, it produces more Gm
INTEGRAL
+ ACTION
aggressive control action for this example. (mA)
-
b) By changing the derivative term in the controller block, the Simulink results
show that the system becomes more oscillatory as D increases. For the
CONTROLLER
INPUT

12-9 12-10 12-11

b) Process and disturbance transfer functions: 15  3 300 / 1.2 187.5 d) Figure S12.10d indicates that  c  1.57 min gives very good results.
Gv ( s )   
20  4 0.0833s  1 0.0833s  1
Overall material balance: 0

w1  w2  w  0 (1) Controller:
Solute balance:
-0.02

w1 x1  w2 x 2  wx  V
dx 187.5 2.6  10 4
(2) Let G  Gv G p G m  32e  s
dt 0.0833s  1 1  4.71s
Substituting (1) into (2) and putting into deviation variables: -0.04

1.56e  s
dx  then G
w1 x1  w2 x2  w1 x   w2 x  w2 x  V (4.71s  1)(0.0833s  1) y(t) -0.06
dt
For process with a dominant time constant,  c   dom / 3 is recommended.
Taking the Laplace transform: -0.08
Hence .  c  1.57 min. From Table 12.1
w1 X 1(s)  ( x2  x )W2(s)  (w1  w2  Vs) X (s)
Kc = 1.92 and I = 4.71 min -0.1

Finally:
x2  x c) Simulink results: -0.12
0 2 4 6 8 10 12 14 16 18 20
X ( s) x2  x w  w2
G p (s)    1
Time
0.04
W2( s) w1  w2  Vs 1  s Figure S12.10d. Closed-loop response for set-point change.
0.035
w1
X ( s) w1 w  w2 e) Improved control can be obtained by adding derivative action:  D  0.4 min.
Gd ( s)    1 0.03

X 1 ( s ) w1  w2  Vs 1  s
0
0.025

V
where  
w1  w2 y(t) 0.02 -0.02

Substituting numerical values: 0.015

-0.04

2.6  10 4 0.01
G p ( s) 
y(t)

1  4.71s -0.06

0.005

0.65
Gd ( s)  -0.08
1  4.71s 0
0 5 10 15 20 25 30 35

Time
Composition measurement transfer function: -0.1

Figure S12.10c. Closed-loop response for the step disturbance.

20  4  s
Gm ( s )  e  32e  s -0.12
0 2 4 6 8 10 12 14 16 18 20
0.5
Time

Final control element transfer function: Figure S12.10e. Closed-loop response after adding derivative action.

12-12 12-13 12-14

 f) For  =3 min, the closed-loop response becomes unstable. It is well known that 12.13
the presence of a large time delay in a feedback control loop limits its 12.12
performance. In fact, a time delay adds phase lag to the feedback loop, which
adversely affects closed-loop stability (cf. Ch 13). Consequently, the controller From the solution to Exercise 12.5, the process reaction curve method yields,
gain must be reduced below the value that could be used if a smaller time delay 2e s
The process model is given as: G( s)  K = 1.65,  = 1.7 min,  = 5.5 min
were present. 3s  1

a) From Table 12.1, the IMC settings are: a) IMC method:

0.6

1  From Table 12.1, Controller G with c = /3:

Kc   0.75
0.4
K c  θ
1  1 5.5
 I    3min Kc    0.94
0.2
K  c   1.65 (5.5 / 3)  1.7
0
b) Cohen-Coon tuning relations: I =  = 5.5 min

1
y(t)

Kc  [0.9  θ /12 ]  1.39 b) Ziegler-Nichols settings:

-0.2
Kθ
1.65e1.7 s
θ[30  3(θ /  )] G(s) 
I 
-0.4
 1.98 min 5.5s  1
9  20 (θ /  )
-0.6
First, determine the stability limits; the characteristic equation is:
The IMC settings are more conservative because they have a smaller Kc
-0.8
0 5 10 15 20 25 30 35 value and a larger I value. 1 + GcG = 0
Time

Figure S12.10f. Closed-loop response for  = 3 min. c) The Simulink simulation results are shown in Fig. S12.12. Both controllers Substitute the Padé approximation,
are rather aggressive and produce oscillatory responses. The IMC controller
is less aggressive (that is, more conservative). 1  0.85s
e s 
12.11 1.8 1  0.85s

1.6 Cohen-Coon into the characteristic equation:

IMC
The controller retuning decision is based on the characteristic equation, which
1.4
takes the following form for the standard feedback control system. 1.65K c (1  0.85s )
0 = 1 + GcG  1 
1.2 4.675s 2  6.35s  1
1 + GcGI/PGvGpGm = 0 y Rearrange,
1

The PID controller may have to be retuned if any of the transfer functions, GI/P, 0.8 4.675s2 + (6.35 –1.403Kc)s + 1 + 1.65Kc = 0
Gv, Gp or Gm, change.
0.6
Substitute s=ju at Kc = Kcu:
a) Gm changes. The controller may have to be retuned. 0.4

b) The zero does not affect Gm. Hence the controller does not require 0.2  4.675 u2 + j(6.35  1.403Kcu) u + 1 +1.65Kcu = 0 + j0
retuning. 0
0 10 20 30 40 50 60 70 Equate real and imaginary coefficients,
c) Gv changes. Retuning may be necessary. Time

Figure S12.12. Controller comparison. (6.35  1.403Kcu)u = 0 ,

d) Gp changes. The controller may have to be retuned.
1+ 1.65Kcu  4.675 u2 = 0

12-15 12-16 12-17

c) Simulation results
Ignoring u= 0, the approximate values are: 12.16
The closed-loop responses for the IMC and RAT controller settings and a step
Kcu = 4.53 and u= 1.346 rad/min change in feed composition from 0.5 to 0.55 are shown in Figs. S12.16a and
a) IMC design: S12.16b, respectively.
2
Pu   4.67 min From Table 12.1, Controller H with c = /2 =3.28 min is:
u

Oxygen Exit Concentration (mol/m 3)

620 1
The Z-N PI settings from Table 12.7 are:

Hydrocarbon Temperature (K)

1 τθ/ 2 1 6.5  2 / 2
Kc   0.9
Kc = 2.04 and I = 3.89 min (approximate) K τ c  θ / 2 220 3.25  2 / 2 610

0.8
Note that the values of Kcu and u are approximate due to the Padé
K c  0.00802 600
approximation. By using Simulink, more accurate values can be obtained 0.7
by trial and error. For this case, no Padé approximation is needed and: θ 2 590
τI  τ   6.5   7.5 min 0.6
Kcu = 3.76 Pu = 5.9 min 2 2
580 0.5
0 20 40 60 80 0 20 40 60 80
The Z-N PI settings from Table 12.7 are: τθ (6.5)(2)
τD    0.867 min t (min) t (min)
2τ+θ 2(6.5)  2
Kc = 1.69 I = 4.92 min (more accurate) 1.6 19

Fuel Gas Flowrate (m3/min)

Compared to the Z-N settings, the IMC method setting gives a smaller Kc b) Relay auto tuning (RAT) controller

Air Flow Rate (m 3/min)

18.5
and a larger I, and therefore provides more conservative controller settings. 1.5

From the documentation for the RAT results, it follows that: 18

1.4
12.14 17.5
a = 54, d = 0.5
1.3
17
From (12-46),
Eliminate the effect of the feedback loop by opening the loop. That is, operate
temporarily in an open loop mode by switching the controller to the manual mode. 16.5
4d 4(0.5) 0 20 40 60 80 0 20 40 60 80
This change provides a constant controller output and a constant manipulated input. K cu    0.0118
If oscillations persist, they must be due to external disturbances. If the oscillations  a  (54) t (min) t (min)

vanish, they were caused by the feedback loop.

Fig. S12.16a. Performance of the IMC-PID controller for a step change in
Pu  14 min hydrocarbon flow rate from 0.035 to 0.040 m3/min.
12.15
From Table 12.7, the Ziegler-Nichols controller settings are:

The sight glass has confirmed that the liquid level is rising. Because the controller K c  0.6 K cu  0.0071
output is saturated, the controller is working fine. Hence, either the feed flow is
higher than recorded, or the liquid flow is lower than recorded, or both. Because Pu Pu
the flow transmitters consist of orifice plates and differential pressure transmitters, I   7 min, D   1.75 min
2 8
a plugged orifice plate could lead to a higher recorded flow. Hence, the liquid-flow-
transmitter orifice plate would be the prime suspect.

12-18 12-19 12-20

 From Table 12.7, the Ziegler-Nichols controller settings are:
12.17
K c  0.6 K cu  27.5

a) IMC design:
Pu P
I   425 s,  D  u  119 s
From Table 12.1, Controller H with c = /2 =381 is: 2 8

1 τθ/ 2 1 762  138 / 2

Kc  
K τ c  θ / 2 0.126 381  138 / 2

K c  14.7

θ 138
τI  τ   762   831min
2 2

τθ (762)(138)
τD    63.3min
2τ+θ 2(762)  138

b) Relay auto tuning (RAT) controller

Fig. S12.16b. Performance of the RAT controller for a step change in The distillation column model includes an RAT option for the xB control loop,
hydrocarbon flow rate from 0.035 to 0.040 m3/min. but not the xD control loop. Thus, the Simulink diagram must be modified by
copying the RAT loop for xB and adding it to the xD portion of the diagram.
The RAT controller is superior due to its smaller maximum deviation and shorter
settling time. Also, the parameters for the relay block must be changed. The new Simulink
diagram and appropriate relay settings are shown in Fig. S12.17a. The results
d) Due to the high noise level for the xD response, it is difficult to obtain from the RAT are shown in Fig.S12.17b.
improved controller settings. The RAT settings are considered to be
satisfactory. From the documentation from the RAT results, it follows that:

a = 5.55 x 10-3, d = 0.2 Fig. S12.17a. Modified RAT Simulink diagram and relay settings.

From (12-46),

4d 4(0.2)
K cu    45.9
a  (5.55 x103 )

Pu  950 s

12-21 12-22 12-23

Fig. S12.17c. Performance of the IMC-PID controller for a step change in feed
composition from 0.5 to 0.55.

Fig. S12.17b. Results from RAT.

c) Simulation results Fig. S12.17d. Performance of the RAT controller for a step change in feed
composition from 0.5 to 0.55.
The closed-loop responses for the IMC and RAT controller settings and a step
change in feed composition from 0.5 to 0.55 are shown in Figs. S12.17c and
S12.17d, respectively.

The RAT controller provides a somewhat better response with a smaller

maximum deviation and a shorter settling time.

d) Due to the high noise level for the xD response, it is difficult to obtain
improved controller settings. The RAT settings are considered to be
satisfactory.

12-24 12-25 12-26

 Chapter 13 © Three flow-head relations: q3  CV 1 h1
For the flooded condenser in Fig. E13.4, the area available for heat transfer
changes as the liquid level changes. Consequently, pressure control is easier q5  CV 2 h2
when the liquid level is low and more difficult when the level is high. By q4  f (h1 , h2 )
13.1 contrast, for the conventional process design in Fig. 13.2, the liquid level
has a very small effect on the pressure control loop. Thus, the flooded Two conservation of mass equations:
condenser is more difficult to control because the level and pressure control
According to Guideline 6, the manipulated variable should have a large loops are more interacting, than they are for the conventional process design
effect on the controlled variable. Clearly, it is easier to control a liquid level dh1
in Fig. 13.2. A1  (q1  q6  q3  q 4 )
by manipulating a large exit stream, rather than a small stream. Because R/D dt
>1, the reflux flow rate R is the preferred manipulated variable. dh
A2 2  (q2  q4  q5 )
13.5 dt

13.2 Conclude: NE = 5
(a) The larger the tank, the more effective it will be in “damping out”
disturbances in the reactor exit stream. A large tank capacity also provides Degrees of freedom: = NF = NV – NE = 8  5 = 3
Exit flow rate w4 has no effect on x3 or x4 because it does not change the a large feed inventory for the distillation column, which is desirable for
relative amounts of materials that are blended. The bypass fraction f has a periods where the reactor is shut down. Thus a large tank is preferred from
dynamic effect on x4 but has no steady-state effect because it also does not a process control perspective. However a large tank has a high capital cost, Disturbance variable: q 6  ND = 1
change the relative amounts of materials that are blended. Thus, w2 is the so a small tank is appealing from a steady-state, design perspective. Thus,
best choice. the choice of the storage tank size involves a tradeoff of control and design NF = NFC + ND
objectives. NFC = 3  1 = 2
13.3
(b) After a set-point change in reactor exit composition occurs, it would be 13.7
desirable to have the exit compositions for both the reactor and the storage
Both the steady-state and dynamic behaviors need to be considered. From a tank change to the new values as soon as possible. But concentration in the
steady-state perspective, the reflux stream temperature TR would be a poor storage tank will change gradually due to its liquid inventory. The time Consider the following energy balance assuming a reference temperature
choice because it is insensitive to changes in xD, due to the small nominal constant for the storage tank is proportional to the mass of liquid in the tank of Tref = 0:
value of 5 ppm. For example, even a 100% change in from 5 to 10 ppm (cf. blending system models in Chapters 2 and 4). Thus, a large storage tank
would result in a negligible change in TR. Similarly, the temperature of the will result in sluggish responses in its exit composition, which is not
top tray would be a poor choice. An intermediate tray temperature would be Heat exchanger:
desirable when frequent set-point changes are required. In this situation, the
more sensitive to changes in the tray composition but may not be storage tank size should be smaller than for case (a).
representative of xD. Ideally, the tray location should be selected to be the Cc (1  f )wc (Tc 0  Tc1 )  Ch wh (Th1  Th 2 ) (1)
highest tray in the column that still has the desired degree of sensitivity to
composition changes. Overall:
13.6
The choice of an intermediate tray temperature offers the advantage of early Cc wc (Tc 2  Tc1 )  Ch wh (Th1  Th 2 ) (2)
detection of feed disturbances and disturbances that originate in the
stripping (bottom) section of the column. However, it would be slow to Variables : q1, q2,…. q6, h1, h2 Nv= 8 Mixing point:
respond to disturbances originating in the condenser or in the reflux drum.
But on balance, an intermediate tray temperature is the best choice. wc  (1  f )wc  fwc (3)

Equations :
13.4 Thus,

[Type here] 13-1

[Type here] 13-2 13-3

Since three flow rates ( fw1 , w2 and w3 ) can be independently adjusted, it be demonstrated by a more complicated theoretical analysis or by
NE =3 , NV = 8 ( f , wc , wh , Tc1 , Tc 2 , Tc 0 , Th1 , Th 2 ) would appear that there are three control degrees of freedom. But the bypass simulation studies.
flow rate, fw1 , has no steady-state effect on x4. To confirm this assertion,
NF =NV  NE = 8  3 = 5 (b) Consider the steady-state model in (1) and (2). Substituting (1) into
consider the overall steady-state component balance for the tank and the (2) and solving for T gives:
mixing point:
NFC = 2 (f, wh)
also x1 w1  x2 w2  x4 w4 wcTc  whTh
(5)
T (5)
wc  wh
ND = NF  NFC = 3 (wc, Tc1, Tc2)
This balance does not depend on the fraction bypassed, f, either directly or
indirectly, Since w does not appear in (5), it has no steady-state effect on T.
The degree of freedom analysis is identical for both cocurrent and Consequently, the proposed multiloop control strategy is not
countercurrent flow because the mass and energy balances are the same for Conclusion: NFC = 2 (w2 and w4) feasible.
both cases.

13.9 13.10
13.8

(a) In order to analyze this situation, consider a steady-state analysis. (a) Model degrees of freedom, NF
The dynamic model consists of the following balances:
NF = NV - NE (13-1)
Assumptions:
Mass balance on the tank:
1. Steady-state conditions with w, Th, and Tc at their nominal NV = 11 (xF, TF, F, wL, L, wV, V, T, P, h, VT)
dh values. where TF is the feed temperature and VT is the volume of the
A  (1  f ) w1  w2  w3 (1) 2. Constant heat capacities
dt flash separator.
3. No heat losses
Component balance on the tank: 4. Perfect mixing NV = 7:
Mass balance
d (hx3 ) Steady-state balances: Component balance
A  (1  f ) x1 w1  x 2 w2  x3 w3 (2) Energy balance
dt
wc  wh  w (1) Vapor-liquid equilibrium relation
Valve relations (2)
Mixing point balances: wcTc  whTh  wT (2) Ideal gas law
w4 = w3 + fw1 (3)
Assume that T=Tsp, where Tsp is the set point. Thus, NF = 11 – 7 = 4
x4w4 = x3w3 + fx1w1 (4)
wc  wh  w (3) (b) Control degrees of freedom, NFC
Thus,
wcTc  whTh  wTsp (4) NF = NFC + ND (13-2)
NE = 4 (Eqs. 1- 4)
Equations (3) and (4) are two independent equations with two Typically, some knowledge of the feed conditions would be
NV = 9 (h, f , w2 , w3 , w4 , x1 , x2 , x3 , x4 ) unknown variables, wh and wc. For any arbitrary value of Tsp, these available. We consider two cases:
equations have a unique solution. Thus the proposed multiloop
control strategy is feasible. Case 1: xF and TF are disturbance variables
NF = NV  NE = 5
This simple analysis does not prove that the liquid level h can also Here ND =2 and:
be controlled to an arbitrary set point hsp. However, this result can

13-4 13-5 13-6

 NFC = NF - ND = 4 – 2 = 2
Chapter 14 For a first-order Pade approximation

The two degrees of freedom can be utilized by manipulating two of s

the three flow rates, for example, V and L, or F and V. 1
2
e s 
s
1
Case 2: xF, TF, and F are disturbance variables 2
14.1
from which we obtain
Here ND =3 and:
3 G1 ( j)
NFC= NF - ND = 4 – 3 = 1 AR  G ( j)  AR Pade  1
G2 ( j) G3 ( j)
  
The single degree of freedom could be utilized by manipulating one  Pade  2 tan 1  
of the exit flow rates, either V or L.  2 
3 () 2  1 3 2  1 Both approximations represent the original function well in the low
 
 (2) 2  1  4 2  1 frequency region. At higher frequencies, the Padéapproximation matches
the amplitude ratio of the time delay element exactly (ARPade = 1), while
the two-term approximation introduces amplification (ARtwo term >1). For
From the statement, we know the period P of the input sinusoid is 0.5 min
the phase angle, the high-frequency representations are:
and, thus,

2  2 two term  90

   4 rad/min
P 0.5  Pade  180

Substituting the numerical value of the frequency: Since the angle of e  j is negative and becomes unbounded as    ,
we see that the Pade representation also provides the better approximation
3 16  1 2
to the time delay element's phase angle, matching  of the pure time delay
Aˆ  AR  A   2  0.12  2  0.24
4 64 2  1 element to a higher frequency than the two-term representation.
Thus the amplitude of the resulting temperature oscillation is 0.24 degrees.

14.3
14.2

First approximate the exponential term as the first two terms in a truncated 128 F  120 F
Nominal temperature T   124 F
Taylor series 2
1
Aˆ  (128 F  120 F)  4 F
e s  1  s 2
  5sec.,   2(1.8 / 60 sec)  0.189 rad/s
Then G( j)  1  j
and ARtwo term  1  () 2  1  2  2 Using Eq. 13-2 with K=1,

two term  tan 1 ()   tan 1 () A  Aˆ  

 2 2  1  4 (0.189)2 (5)2  1  5.50 F

Solution Manual for Process Dynamics and Control, 4th edition Actual maximum air temperature = T  A  129.5 F
Copyright © 2016 by Dale E. Seborg, Thomas F. Edgar, Duncan A. Mellichamp,
and Francis J. Doyle III Actual minimum air temperature = T  A  118.5 F

13-7 14-1 14-2

14.4 14.7
Bode Diagram

G(s)
Tm ( s ) 1 Using MATLAB G(s) w ith Pade approx.


Magnitude (abs)
T ( s ) 0.1s  1 0
10
5
Bode Diagram
T ( s)  (0.1s  1)Tm ( s) 10

amplitude of T  =3.464 (0.1 ) 2  1  3.465

Magnitude (abs)

0
phase angle of T  =  + tan-1(0.1) =  + 0.02
-1
10 10

Since only the maximum error is required, set  = 0 for the comparison of T  and -5
10
0

Tm . Then -50

Phase (deg)
-100
-10
Error = Tm  T  =3.464 sin (0.2t) – 3.465sin(0.2t + 0.02) 10
-150
-45
= 3.464 sin(0.2t) –3.465[sin(0.2t) cos 0.02 + cos(0.2t)sin 0.02]
-200
= 0.000 sin(0.2t)  0.0693 cos(0.2t) -90
-250
Phase (deg)

-135
Since the maximum absolute value of cos(0.2t) is 1, -2 -1 0 1
10 10 10 10
-180
maximum absolute error = 0.0693 Frequency (rad/sec)
-225 Figure S14.8. Bode diagram for G(s) and G(s) with Pade approximation.
As we can see from the figures, the accuracy of Pade approximation does not
-270
-1 0 1 2 change as frequency increases in magnitude plot, but it will be compromised in
10 10 10 10
14.5 the phase plot as frequency goes higher.
Frequency (rad/sec)

Figure S14.7. Bode diagram of the third-order transfer function.

st
(a) No, cannot make 1 order closed-loop system unstable. 14.9

(b) No, cannot make 2nd order overdamped system unstable for closed-loop. The value of  that yields a -180 phase angle and the value of AR at that =2f where f is in cycles/min
frequency are: For the standard thermocouple, using Eq. 14-13b
(c) Yes, 3rd order system can be made unstable.
 = 0.807 rad/sec 1 = -tan-1(1) = tan-1(0.15)
(d) Yes, anything with time delay can be made unstable. AR = 0.202
Phase difference  = 1 – 2
14.8
Thus, the phase angle for the unknown unit is
14.6
Using MATLAB,
2 = 1  
Engineer A is correct. and the time constant for the unknown unit is
Second order overdamped process cannot become unstable with a proportional 1
controller. 2 = tan(  2 )

FOPTD model can become unstable with a large Kc due to the time delay.

14-3 14-4 14-5

 using Eq. 14-13b . The results are tabulated below Aˆ h  / hmax
   KA / 4 2  2  1 (1) 14.11

f  1  2 2 Using Eqs. 14-28 , 14-13, and 14-17,

0.05 0.31 -2.7 4.5 -7.2 0.4023 Aˆ q   0.1337 A / 4 2  2  1 (2)
0.1 0.63 -5.4 8.7 -14.1 0.4000
2 2  a  1
2
0.2 1.26 -10.7 16 -26.7 0.4004 AR=
for the one-tank system, and
0.4 2.51 -26.6 24.5 -45.1 0.3995 100 2  1 4 2  1
0.8 6.03 -37 26.5 -63.5 0.3992
1 6.28 -43.3 25 -68.3 0.4001 Aˆ h1 / h1max   KA /  2  2  1 (3)
2 12.57 -62 16.7 -78.7 0.3984  = tan-1(a) – tan-1(10) – tan-1(2)
4 25.13 -75.1 9.2 -84.3 0.3988
Aˆ h2 / h2 max   KA / (   1)
2 2 2
(4) The Bode plots shown below indicate that
That the unknown unit is first order is indicated by the fact that 0 as ,
Aˆ q 2   0.1337A / ( 2  2  1) 2
so that 21-90 and 2-90 for  implies a first-order system. This is i) AR does not depend on the sign of the zero.
(5)
confirmed by the similar values of 2 calculated for different values of , ii) AR exhibits resonance for zeros close to origin.
iii) All zeros lead to ultimate slope of –1 for AR.
implying that a graph of tan(-2) versus  is linear as expected for a first-order for the two-tank system.
system. Then using linear regression or taking the average of above values, 2 = iv) A left-plane zero yields an ultimate  of -90.
0.40 min. v) A right-plane zero yields an ultimate  of -270.
Comparing (1) and (3), for all  vi) Left-plane zeros close to origin can give phase lead at low .
vii) Left-plane zeros far from the origin lead to a greater lag (i.e.,
Aˆ h1 / h1max   Aˆ h / hmax
  smaller phase angle) than the ultimate value. u  90 ºwith a left-
14.10 plane zero present.
Hence, for all , the first tank of the two-tank system will overflow for a smaller
value of A than will the one-tank system. Thus, from the overflow consideration,
Bode Diagram
From the solution to Exercise 5-19, for the two-tank system the one-tank system is better for all . However, if A is small enough so that 10
1

overflow is not a concern, the two-tank system will provide a smaller amplitude in Case i
H 1 ( s) / h1max 0.01 K the output flow for those values of  that satisfy Case ii(a)
 

Magnitude (abs)
0 Case ii(b)
Q1i ( s) 1.32s  1 s  1 10
 
Case iii
H 2 ( s) / h2 max Aˆ q 2   Aˆ q 
0.01 K
 
Q1i ( s) (1.32s  1) 2 (s  1) 2
-1
10
0.1337A 0.1337A
or 
Q2 ( s) 0.1337 0.1337 (   1)
2 2 2
4   1
2 2

 
-2
10
Q1i ( s) (1.32s  1) 2 (s  1) 2 90
or   2 /  = 1.07
and for the one-tank system 0
Therefore, the two-tank system provides better damping of a sinusoidal

Phase (deg)
H ( s) / hmax
 0.01 K disturbance for   1.07 if and only if
  -90
Q1i ( s) 2.64s  1 2s  1
1.32 2  2  1
Aˆ h1 / h1max   1 , that is, A  -180
Q ( s) 0.1337 0.1337 0.01
 
Q1i ( s) 2.64s  1 2s  1 -270
-2 -1 0 1
10 10 10 10
For a sinusoidal input q1i (t )  A sin  t , the amplitudes of the heights and flow
Frequency (rad/sec)
rates are
Figure S14.11. Bode plot for each of the four cases of numerator dynamics.

14-6 14-7 14-8

14.12 14.13 14.15

a) From Eq. 8-14 with I = 4D (a)

2 0.6 4
1  GOL  1  Gm G p Gc Gv  1  Kc
( 4 D s  1  4 D s 2 )
2
(2 D s  1) 2 s  1 50s  1 2s  1
Gc ( s )  K c  Kc
4 D s 4 D s
2
Characteristic equation:
 4 2  2  1  (s  1)(50s  1)(2s  1)  K c (4)(2)(0.6)  0 (1)
  K 4 D   1
2 2

Gc ( j)  K c 
D

4 D  4 D 
c
For a third order process, a Kc can always be chosen to make the process unstable.
A stability analysis would verify this but was not required.
b) From Eq. 8-15 with I = 4D and  = 0.1
Substitute s = jω into Eq. (1), we have:
(4 D s  1) D s  1 ( j  1)(50 j  1)(2 j  1)  K c (4)(2)(0.6)  0
Gc ( s)  K c
4 D s0.1 D s  1 For τ = 1, we have:
(−100𝜔3 + 53𝜔)𝑗 + (1 + 4.8𝐾𝑐 − 152𝜔2 ) = 0 (2)
 16 2 2  1   2 2  1  Thus, we have ωc = 0.53 and Kcu = 16.58.
Gc ( j)  K c   D 
D

For τ = 0.4, we have:

4 D  0.01 D 2  1
2
(−40𝜔3 + 52.4𝜔)𝑗 + (1 + 4.8𝐾𝑐 − 120.8𝜔2 ) = 0 (3) Figure S14.15a Bode plot of GOL. (Kc = 10)
So we have:
The differences are significant for 0.25 < D < 1 by a maximum of 0.5 Kc ωc = 1.31 and Kcu = 41.28 1
at D = 0.5, and for D >10 by an amount increasing with D . GOL  GGc  Kc
(4 s  1)(2 s  1)
The second measurement is preferred because of a larger stability region of Kc.
10
2 Cannot become unstable – max phase angle 2nd order overdamped process
(GOL) is -180 degree.
Series controller with filter (asymptote)
14.14 (b)
(a) Always true. Increasing the gain does speed up the response for a set point
change. Care must be taken to not increase the gain too much or oscillations will
10
1 result.

(b) False. If the open loop system is first order, increasing Kc cannot result in
c
AR/K

oscillation.

0
(c) Generally true. Increasing the controller gain can cause real part of the roots of
10 the characteristic polynomial to turn positive. However, for first or second order
processes, increasing Kc will not cause instability.
Parallel controller (asymptote)
(d) Always true. Increasing the controller gain will decrease offset. However, if
Parallel controller (actual) the gain is increased too much, oscillations may occur. Even with the oscillations
Series controller (actual)
10
-1 the offset will continue to decrease until the system becomes unstable.
-2 -1 0 1 2
10 10 10 10 10
 D
Figure S14.12. Nominal amplitude ratio for parallel and series controllers

Figure S14.15b Bode plot of GOL. (Kc = 10)

14-9 14-10 14-11

 1  1 / 5s (5s  1) K c (d) e s
GOL  GGc  Kc  GOL  GGc  Kc
(4s  1)(2s  1) 5s(4s  1)(2s  1) (4s  1)

Cannot become unstable – max phase angle (GOL) is -180 degree while at Can become unstable due to time delay at high frequency.
low frequency the integrator has -90 degree phase angle.

(c)
14.16

By using MATLAB,

2
Bode Diagram
10

1
10

Magnitude (abs)
0
10
Figure S14.15d Bode plot of GOL. (Kc = 10)
-1
10
1 s (1  s) K c
GOL  GGc  Kc 
(4s  1)(2s  1) (4s  1)(2s  1) -2
10
0
Can become unstable – max phase angle (GOL) is -270 degree. Parallel
-45
Series w ith filter
Figure S14.15c Bode plot of GOL. (Kc = 10) -90

Phase (deg)
(e)
s 1 2s  1 ( s  1) K c -135
GOL  GGc  Kc 
(4s  1)(2s  1) s s(4s  1) -180

-225
Cannot become unstable – lead lag unit has phase lag larger than -90,
-270
integrator contributes -90 degree; the total phase angel is larger than -180. 10
-2
10
-1
10
0

Frequency (rad/sec)
Figure S14.16 Bode plot for Exercise 13.8 Transfer Function multiplied by
PID Controller Transfer Function. Two cases: a)Parallel b) Series with Deriv.
Filter (=0.2).
.
Amplitude ratios:

Ideal PID controller: AR= 0.246 at  = 0.80

Series PID controller: AR=0.294 at  = 0.74

There is 19.5% difference in the AR between the two controllers.

Figure S14.15e Bode plot of GOL. (Kc = 10)

14-12 14-13 14-14

14.17 Bode Diagram Bode Diagram

1
50 10

0
a) Method discussed in Section 6.3: 10
0

Magnitude (abs)
Magnitude (dB)

-50
12e 0.3s -1

Gˆ 1 ( s) 
10
G(s) G(s)
(8s  1)(2.2s  1) -100
G1(s) G'(s)
-2
-150 G2(s) 10
Method discussed in Section 7.2.1: G3(s)
4
-200 x 10 0
0
Step response of G(s) -360
12
-720
-1.152

Phase (deg)
-1080
Phase (deg)

10 -1440
-2.304
-1800
X: 10.09
8 Y: 7.2 -3.456 -2160
-2520
Value

6 -4.608 -2880
-3 -2 -1 0 1 2 3 -2 -1 0 1
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/sec)
4
X: 4.034
Inflection Figure S14.17b Bode plots for the exact and approximate models. Figure S14.18 Bode plots for the exact and approximate models.
Y: 2.4
point
2

As seen in Fig.S14.18, the approximation is good at low frequencies, but

0
0 10 20 30 40 50 60 70 80 90 100 14.18 not that good at higher frequencies.
Time

Figure S14.17a Step response of G(s) 14.19

Based on Figure S14.17, we can obtain the time stamps corresponding to The original transfer function is
𝑡
20% and 60% response: 𝑡20 = 4.034; 𝑡60 = 10.09; 20⁄𝑡 = 0.4. Based
60
10(2s  1)e 2 s (a)
G( s) 
𝑡
on Figure 7.7, we have 60 = 2.0; 𝜁 = 1.15, so we have 𝜏 = 5.045.Using (20s  1)(4s  1)(s  1) 2e 1.5 s 0.5e 0.3s 3e 0.2 s 3e 2 s
𝜏 G  G p Gv G m  
the slope of the inflection point we can estimate the time delay to be 0.8. (60s  1)(5s  1) 3s  1 2s  1 (60s  1)(5s  1)(3s  1)(2s  1)
So we have: The approximate transfer function obtained using Section 6.3 is:
12e 0.8 s ωc occurs where φ = -180:
Gˆ 2 ( s )  10  2s  1 e5 s  c  0.152 AR ( c )  0.227
25.45s 2  11.60 s  1 G'  s  
 22s  1 K cu 
1
 4.41
b) Based on Figure S14.17a, we can obtain θ = 0.8; τ = 15 − 0.8 = 14.2 AR ( c )
12e 0.8 s
Gˆ 3 ( s) 
14.2 s  1

Comparison of three estimated models and the exact model in the

frequency domain using Bode plots:

14-15 14-16 14-17

 3e 9.5 s The gain of Gp = 1.0 for all .
G
62.5s  1
At the critical frequency (c), a sine wave is formed with period
0.916
Using Table 12.3, K c  0.586(9.5 / 62.5) / 3  1.10 ;
2 2 rad rad
62.5 Pm  10 min s  , so  c   0.628
I   62.19 c 10 min min
 0.165(9.5 / 62.5)  1.03
Then, (a) The critical gain is easily found from
1 K c KV K p K m  1 at    c
Gc  1.10(1  )
62.19s , GOL  GGc
K cu (0.5)(1)(1)  1 , or K cu  2.0
ωc occurs where φ = -180:
 c  0.153 AR ( c )  0.249
(b) The phase angle of Gc G  Gcu = phase angle of e s , or Φ = - θ (rad)
1
K cu   4.02 (Eq. 14-33)
Figure S14.19a Bode plot of to find ωc. AR ( c )
when Φ = -180°= - π= -c θ
Simulation results with different Kc are shown in Fig. S14.19b. Kc > Kcu, the
system becomes unstable as expected.
2 rad 10
Because  c  then    5 min .
10 min 2

14.21

a) Using Eqs. 14-56 and 14-57

 1  5  1 
AR OL   K c  1  

(1.0)

 25 2  100  1    1 
2 2

 = tan-1(-1/5) + 0 + (-2  tan-1(10)) + (- tan-1())

Figure S14.19b Step response of closed loop system with different Kc.
Figure S14.19c Bode plot of FOPTD model.

(b)
Use Skogestad’s half rule
14.20
  60  0.5  5  62.5
  2.5  3  2  2  9.5
Using the Bode plot, at a phase angle of -180°, we require that K c KV K p K m  1
The approximated FOPTD model:
G p ( s)  e s GV  0.5 Gm  1.0

14-18 14-19 14-20

From part b), AR OL  c

= 1.025 Kc
R1
Bode Diagram Gp(s) =
( A1 R1 A2 R2 ) s 2  ( A1 R1  A2 R1  A2 R2 ) s  1
2
10
Therefore 1.025 Kc = 1/1.7 or Kc = 0.574
1
10
2
Bode Diagram For R1=0.5, R2 = 2, A1 = 10, A2 = 0.8
10
AR/Kc

0
10 10
1
0.5
Gp(s) = (1)
8s 2  7 s  1
AR/Kc

0
-1 10
10
0.5
-1 For R2 = 0.5, Gp(s) = (2)
-2
10
2s 2  5.8s  1
10
10
-2 a) For R2=2
-100
-150

  7 c   
|Gp| =  
-150 -180
0.5
Gp= tan-1  2 
,
Phase (deg)

 2 
Phase (deg)

1  8 c 
-200

 (1  8 c )  (7 c ) 
2 2
-200
-250
-250 -300
Kcu and c are obtained using Eqs. 14-7 and 14-8:
-300 -350
-2 -1 0 1 2
10 10 10 10 10

-350
Frequency (rad/sec)
  7 c 
10
-2
10
-1
10
0 1
10
2
10 Figure S14.21b Solution for part b) using Bode plot -180 = 0 + 0 + tan-1  2 
 tan-1(0.5c)
Frequency (rad/sec) 1  8 c 
Bode Diagram
Figure S14.21a Bode plot 10
2

b) Set  = 180 and solve for  to obtain c = 0.4695 Solving, c = 1.369 rad/min
1
10

  
AR/Kc

1  ( K cu )(2.5) 
Then AR OL  c
= 1 = Kcu(1.025) 10 0.5 1.5 
  
 (1  8 c )  (7 c )   (0.5c )  1 
-1 2 2 2 2
10

Therefore, Kcu = 1/1.025 = 0.976 -2

10
-150 Substituting c = 1.369 rad/min, Kcu = 10.96, cKcu = 15.0
System is stable for Kc  0.976
-180

For R2=0.5
Phase (deg)

-200

c) For Kc = 0.2, set AROL = 1 and solve for  to obtain g = 0.1404 -250

  5.8 c   
|Gp| =  
-300 0.5
Gp = tan-1  2 
,
Then g =   = -133.6 2 
1  2 c   (1  2c )  (5.8c ) 
-350 2 2
g -2 -1 0 1 2
10 10 10 10 10
Frequency (rad/sec)

From Eq. 14-61, PM = 180 + g = 46.4 Figure S14.21c Solution for part c) using Bode plot For Gv = Kv = 2.5, v=0, |Gv| = 2.5

1.5 1.5
d) From Eq. 14-60 14.22 For Gm = , m= -tan (0.5) ,
-1
|Gm| =
0.5s  1 (0.5c ) 2  1
1 1
GM = 1.7 = = From modifying the solution to the two tanks in Section 6.4, which have a
Ac AR OL 
c slightly different configurations,

14-21 14-22 14-23

   5.8 c  Therefore, gain margin GM =1/Ac = 1.358
-180 = 0 + 0 + tan-1  2 
 tan-1(0.5c) 2
 
1  2 c   1   2 
Solving Eq.(14-16) for g AR OL  1  Kc   1
  g  0.5   
 g   g 2  1 
Solving, c = 2.51 rad/min  
AROL|=c = 1 at g = 0.925

Substituting c = 2.51 rad/min, Kcu = 15.93, cKcu = 40.0 Substituting for g gives Kc = 1.34. Then from Eq. 14-8
Substituting into Eq. 14-57 gives g=|=g = 172.7

Therefore, phase margin PM = 180+ g = 7.3  1 

180 = tan-1    0.2c  tan (c)
-1

 0.5c 
b) From part a), for R2=2,

c = 1.369 rad/min, Kcu = 10.96

Solving, c =7.19 rad/min
2 14.23 From Eq. 14-56,
Pu = = 4.59 min
c
 1 
2
 
Ac  AR OL  1.34    1  2   0.383
K=2 ,  = 1 ,  = 0.2 , c=0.3  
 0.5 c 
  c
a)
 c  1 
2
Using Table 12.4, the Ziegler-Nichols PI settings are

Kc = 0.45 Kcu = 4.932 , I= Pu/1.2 = 3.825 min Using Eq. 12-11, the PI settings are
From Eq. 14-60, GM = 1/Ac = 2.61
1 
Using Eqs. 13.63 and 13-62 , Kc  1 , I =  = 1 min,
K   c
c) By using Simulink-MATLAB, these two control systems are compared for
c= -tan-1(-1/3.825) a unit step change in the set point.
Using Eq. 14-58 ,
2 1.4
 1  part a)
|Gc| = 4.932   1  1 part b)

 3.825  -180 = tan-1    0.2c  tan-1(c) = -90  0.2c 1.2

 c 
1
Then, from Eq. 14-56
/ 2
or c = = 7.85 rad/min 0.8

-1 
1  -1   7  c 
0.2

Output
-180 = tan   + 0 + tan  2 
 tan-1(0.5c)
 3.825c  1  8 c 
0.6

Using Eq. 14-57,

0.4

Solving, c = 1.086 rad/min   2

1  
1 2
Ac  AR OL   0.255
0.2

  c
c
2   
 c  1 
Using Eq. 14-57 2
c
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time
Ac = AROL|=c =
From Eq. 14-60, GM = 1/Ac = 3.93 Figure S14.23 Close-loop response for a unit step change in set point.
  
2    
 1
  1 (2.5)
0.5  1.5 
=  4.932 
 3.825c    2    b) Using Eq. 14-61, The controller designed in part a) (Direct Synthesis) provides better
  (1  8c )  (7c )   (0.5c )  1 
2 2 2
  performance giving a first-order response. Part b) controller yields a large
g = PM  180 =  140  = tan-1(-1/0.5g)  0.2g  tan-1(g) overshoot.
= 0.7362
Solving, g = 3.04 rad/min

14-24 14-25 14-26

AR OL AR OL
14.24 From the plot of vs , = 0.144 Bode Diagram
Kc Kc   g 10
2

1
a) Using Eqs. 14-56 and 14-57 Since AR OL   g
=1 , Kc = = 6.94 0
10
0.144

AR/Kc
Ysp  4 2  1  2  0.4  c) From the plot of  vs. ,  = -180 at c = 4.05 rad/min -2

  Kc
10
AR OL    (1.0)
   
Ym  0.01 2  1  0.25 2  1   25 2  1 
AR OL AR OL -4

vs ,
10
From the plot of = 0.0326 -90
= tan-1(2)  tan-1(0.1)  tan-1(0.5) – (/2)  tan-1(5) Kc Kc   c
-135
Ac = AR OL  c
= 0.326

Phase (deg)
Bode Diagram From Eq. 14-60, GM = 1/Ac = 3.07 -180
2
10
-225
2
Bode Diagram
10
0 -270
-2 -1 0 1 2
10 10 10 10 10 10
0
Frequency (rad/sec)
AR/Kc

10
AR/Kc

Figure S14.24c Solution for part c) using Bode plot

-2
10 10
-2

-4
-4 10
10 -90
14.25
-90
-135
Phase (deg)

-135 -180
Phase (deg)

a) Schematic diagram:
-180 -225

-270
-2 -1 0 1 2 TC
-225 10 10 10 10 10
Frequency (rad/sec)
Hot fluid
-270
Figure S14.24b Solution for part b) using Bode plot
10
-2 -1
10 10
0
10
1 2
10 TT
Frequency (rad/sec)
Figure S14.24a Bode plot
Cold fluid
b) Using Eq.14-61 Mixing Point Sensor

g = PM – 180 = 30 180 = 150

From the plot of  vs. , g = -150 at g = 1.72 rad/min

14-27 14-28 14-29

 Block diagram: D = Pu/8 = 16/8 = 2 s c = 8.443 for F = 0
Mixing Transfer c = 5.985 for F = 0.1
Controller Valve Process line c)
TR T
Gc Gv Gp GTL Then, from Eq. 14-57,
+
- 1.4
  
1  K cu   
2 1
  
  c  1   F c  1 
Gm 1.2 2 2 2

b) GvGpGm = Km = 6 ma/ma 1 Solving for Kcu gives,

GTL = e-8s Kcu = 4.251 for F = 0

y
0.8 Kcu = 3.536 for F = 0.1
GOL = GvGpGmGTL = 6e-8s
PID control
PI control Therefore,
0.6
If GOL = 6e-8s cKcu = 35.9 for F = 0
0.4 c Kcu= 21.2 for F = 0.1
| GOL(j) | = 6
Since cKcu is lower for F = 0.1, filtering the measurement results in
 GOL (j) = -8 [rad] 0.2 worse control performance.

Find c: The critical frequency corresponds to an open-loop phase angle

of  180 phase angle =   radians 0
0 30 60 90 120 150
t
-8c = - or c = /8 rad/s 14.27
Figure S14.25 Set-point responses for PI and PID control.
2 2 0.047 5.264
Find Pu: Pu = 
 16s Note: The MATLAB version of PID control uses the following controller a) Gv ( s)   112 
c  / 8 0.083s  1 0.083s  1
settings: ki=Kc/I and kd= KcD.
1 1
Find Kcu: Kcu =   0.167 2
| G p ( jc ) | 6 d) Derivative control action improves the closed-loop response by reducing the G p ( s) 
settling time, at the expense of a more oscillatory response. (0.432s  1)(0.017s  1)
[ Note that for this unusual process, the process AR is independent of frequency]
0.12
Gm ( s) 
Ziegler-Nichols ¼ decay ratio settings: (0.024s  1)

PI controller: 14.26 Using Eq. 14-61

Kc = 0.45 Kcu = (0.45)(0.167) = 0.075 Kcu and c are obtained using Eqs. 14-56 and 14-57. Including the filter GF into -180= 0  tan-1(0.083c)  tan-1(0.432c)  tan-1(0.017c)
I = Pu/1.2 = 16 s/1.2 = 13.33 s these equations gives  tan-1(0.024c)

PID controller: -180 = 0 + [-0.2c  tan-1(c)]+[-tan-1(Fc)] Solving , c = 18.19 rad/min

Kc = 0.6 Kcu = (0.6)(0.167) = 0.100 Solving, Using Eq. 14-60
I = Pu/2 = 16/2 = 8 s

14-30 14-31 14-32

    Using Eq. 14-60, GM = 1/Ac = 1.54 By using MATLAB, the Nyquist diagram for this open-loop system is
1  ( K cu )   
5.624 2
 (0.083 ) 2  1   (0.432 ) 2  1 (0.017 ) 2  1 
 c   c c  Solving Eq. 14-56 for g gives Nyquist Diagram
1
 
 
0.12
AR OL =1 at g = 11.78 rad/min 0.5
 (0.024 ) 2  1    g
 c  Substituting into Eq. 14-57 gives 0

Substituting c=18.19 , Kcu = 12.97 -0.5

g =   =  (/2) + tan (0.288g)  tan (0.083g)  tan (0.432g)

-1 -1 -1
-1
Pu = 2/c = 0.345 min

Imaginary Axis
g

-1.5
Using Table 12.4, the Ziegler-Nichols PI settings are  tan-1(0.017g)  tan-1(0.024g) = -166.8
-2

Kc = 0.45 Kcu = 5.84 , I=Pu/1.2 = 0.288 min Using Eq. 14-61 -2.5

-3
PM = 180 + g = 13.2 
-3.5
b) Using Eqs.14-39 and 14-40
-4
-3 -2.5 -2 -1.5 -1 -0.5 0
c =  Gc = tan-1(-1/0.288)= -(/2) + tan-1(0.288) Real Axis

14.28 Figure S14.28a The Nyquist diagram for the open-loop system.
2
 1 
|Gc| = 5.84   1
 0.288  a) From Exercise 14.28, b) Gain margin = GM =
1
AR c
Then, from Eq. 14-57, 5.264
Gv ( s ) 
0.083s  1 where ARc is the value of the open-loop amplitude ratio at the critical
2 frequency c. By using the Nyquist plot
- =  (/2) + tan-1(0.288c)  tan-1(0.083c)  tan-1(0.432c) G p ( s) 
(0.432s  1)(0.017s  1)
Nyquist Diagram
0.12
 tan-1(0.017c)  tan-1(0.024c) Gm ( s)  1

(0.024s  1) 0.5
 1 
The PI controller is Gc ( s )  51   0
Solving, c = 15.11 rad/min.  0.3s 
-0.5
Hence the closed-loop transfer function is
Using Eq. 14-56 -1

Imaginary Axis
GOL  Gc Gv G p Gm -1.5
  
2   
1 5.264
Ac  AR OL  5.84    1    Rearranging,
-2

 c
 0.288c    (0.083 ) 2  1 
   c  -2.5

    6.317s  21.06
2 0.12 GOL  -3
   1.46  10 5 s 5  0.00168s 4  0.05738s 3  0.556s 2  s
 (0.432c ) 2  1 (0.017c ) 2  1   (0.024c ) 2  1  -3.5

-4
-3 -2.5 -2 -1.5 -1 -0.5 0
Real Axis

= 0.651 Figure S14.28b Graphical solution for part b)

14-33 14-34 14-35