Sei sulla pagina 1di 13

DE!

WINATION
Desalination 133 (2001) 271-283

ELSEVIER

www.elsevier.com/locate/desal

Advanced process control of a B-9 Permasep@permeator


desalination pilot plant
Andrew C. Burden, Pradeep B. Deshpande, James C. Waiters
Chemical Engineering Department, Universityof Louisville,Louisville, KY 40292, USA
Tel. +I (502) 852-6347; Fax +I (502) 852-6355
Received 26 October 2000; accepted 13 November 2000

Abstract
An experimental application of advanced control and optimization on a hollow-fiber membrane module (B-9
Permasep@permeator by DuPont) is presented. The objective of the study was to compare the performance of standard
proportional-integral (PI) control with the performance of a constrained model predictive control (CMPC). A proper
control strategy, whether PI or CMPC, should allow for the manipulation (servo control) of the product flow rate while
maintaining product quality. In doing so, a plant can adjust the production of water to meet demand. Several PI control
experiments involving set point changes in product flow rate and conductivity (a measure of quality) were conducted.
It was found that PI control was unable to properly control the quality of the product by means of manipulating the pH
ofthe feed. The PI controller over-compensated for offset in product conductivity. In contrast, CMPC displayed superior
performance in the control of the pilot plant by holding the process outputs within specified bounds; especially the feed
pH which prevented the conductivity PI control loop from becoming unstable. Furthermore, CMPC was able to
maximize the product flow rate by 13.6% while improving the conductivity (quality) by 1.1%.
Keywords: Constrained model predictive control; Reverse osmosis; Avanced control

1. Introduction

Due to an ever-changing demand for potable


water and advances in microprocessor technology, interest has risen in the reverse osmosis
(RO) industry in recent years. The computational
power of todays microprocessors has fueled an
*Corresponding author.

area of research

desalination

on advanced

control

plant is the system pressure.

001 l-9164/01/$- See front matter 0 2001 Elsevier Science B.V. All rights reserved
PII:SOOll-9164(01)00107-2

focused

strategies for RO plants. The hope from such


research is to find a means of proper process
control that will allow the adjustment of the plant
water production rate to meet the changing
demand while maintaining quality within desired
limits.
The main operating v,ariable of an RO
The

272

A.C. Burden et al. /Desalination

system pressure has a direct effect on the product


water flux as well as the quality of the product, as
described by the principle of RO. Applications of
advanced control strategies incorporate the
system pressure as well as other operating variables (e.g., temperature, pH, and feed composition). For instance, in 1989 Alitiqi et al. [l]
suggested that the feed water pH could be used as
a means of controlling product quality. In a more
resent study, Singh and Tembrock [2] discussed
the complexities involved in the effective control
of various RO system arrangements.
Previous research at the University of
Louisville [3] involved the application of
advanced control strategies to an RO desalination
pilot plant that utilized four 2 spiral-wound
cellulose acetate membrane modules in series. In
this paper we present an experimental application
of a commercial constrained model predictive
control (CMPC) package. For the purpose of this
research, the RO desalination pilot plant was
modified to incorporate a hollow-fiber module,
specifically, a B-9 Permasep@ permeator by
DuPont. The performance of the CMPC strategy
is compared to that of conventional proportionalintegral (PI) control.
The organization of this paper is as follows.
We begin with a brief overview of the two
control strategies applied to the pilot plant, PI
and CMPC. Next, the pilot plant and the
computer control system are described. Then the
results of the two control strategies are presented
and discussed. Finally, concluding remarks and
recommendations are offered.

2. Overview

of conventional
integral control [4]

proportional-

The PI control strategy applied to the pilot


plant is derived from the basic proportionalintegral-derivative (PID) control law described
by Eq. (1). The PID law offers several options
that can be applied to various control situations.

133 (2001) 271-283

Each of the three modes of PID corresponds to a


degree of control and error minimization.

(1)
+ j&z,

de(t)
dt

where m(t) is the controller output (CO), %CO;


Kc the controller gain, %CO/%TO (TO stands for
transmitter output); mthe bias value, %CO (the
value of the controller output when there is zero
error); e(t) the error, the difference between
controlled variable C(t) and controller set point
R(t); T, is the integral (or reset) time, min; and T,
is the derivative (or rate) time, min.
As Eq. (1) illustrates, the proportional(P) part
of the equation contains the bias value, my and
the &e(t) term. The last two terms of Eq. (1) are
the integral (I) and derivative (D) portions of the
control law, respectively. The PID control law
has three tuning constants - Kc, r,, and ~~ that must be adjusted to achieve satisfactory
control.
PID controllers are single-input/single-output
(SISO) devices. The PID control law only allows
for one input, which is the error, e(t), and one
output which is the adjustment of the manipulated variable, m(t).
Due to the simplicity of
the control law, a control computer may use
multiple PID controllers. However, since the
controllers are SISO, a PID control strategy can
only handle square systems, that is, systems with
the same number of inputs and outputs.
The derivative mode of the PID control law
has many advantages such as the reduction of
overshoot and oscillations around the set point.
The derivative mode, however, should only be
used for processes that are free of noise. In the
presence of fast changing noise, the derivative of
the error is very large; consequentIy, the
derivative action only amplifies the noise. Due to

A.C. Burden et al. /Desalination 133 (2001) 271-283

the process noise of the pilot plant, the derivative


time of the PID control law was set to zero, thus
eliminating the derivative term from the control
law. Without the derivative mode, the PID
control law is simply called PI.

3. Overview of constrained

model predictive

control [5]
CMPC is a model-based control strategy that

applies feedba&feedforward control for multivariable processes. In addition, CMPC utilizes an


optimization routine that allows constraint
handling. Whether it is the cost of the raw
materials or the conditions for safe operation, all
processes have limitations and economic
constraints. For this reason, CMPC has grown in
popularity in the chemical process industry.
CMPC is applicable to non-square systems. In
the case where a process has more outputs than
inputs, the outputs may be held within specified
upper and lower bounds. On the other hand, if a
process has more inputs than outputs, an
economic optimization can be performed to
allocate the manipulated variable to reduce
operating costs.
The principle of CMPC can best be explained
with a general example of a linear multi-variable
system schematically shown in Fig. 1. The
process outputs, Y,are related to the manipulated
variables, M, the measured disturbances, L, and
unmeasured disturbances, D, according to
Eq. (2):

Y = KpiU+KLL+D

subject to the following constraints for the ti


manipulated variable (mJ and output (y,):
Kgmj-y,

Process
e

Fig. 1. Schematicof open-loopprocess.

-bi-di-KLi 1

(4)

j=l

yi +

sy-viu=yiu

yi+s;-v,L

(5)

= y;

(6)

u
miLs m,s m,

(7)

where M?ZG,4is a very large number, e.g., 106;b,


is the bias, initial steady-state value for output i;
4 is the effect of unmeasured disturbances and

Unmeasured
Disturbances, D
+

Manipulated
Variables, M

(2)

where Y, M, L, and D are in deviation form and


Kp and KLrepresent the matrices of steady-state
gains for the process and load dynamics,
respectively.
For the process depicted in Fig. 1, a general
CMPC algorithm using linear programming (LP)
may use an optimization index as described in
Es. (3):

Load
Dynamics

Measured
Disturbances, L

273

+.

+
+1

Outputs,Y

274

A.C. Burden et al. /Desalination

modeling errors for output i; Wj the weight for


output i; Ci the cost for manipulated variable i; U
the upper bound, L the lower bound; K the
violation of the bounds for output i; S, slack/
surplus variable for output i; M the number of
manipulated variables and N is the number of
outputs.
The optimization algorithm seeks to minimize
the objective function Jdefined in Eq. (3). The
algorithm may be chosen to maximize J by
making the function negative. The constant
coefficient M.GA is a very large number of the
order of lo6 that directs the focus of the
optimization algorithm to minimize the violation
variables. In order for the algorithm to minimize
the violation variables, the output variables must
be maintained within their respective bounds.
The constraint Eqs. (5)-(7) contain user-specified
upper and lower bounds of the controlled and
manipulated variables.
If the violation variables cannot be reduced to
zero, then the algorithm seeks to minimize the
violation variables as much as possible. Conversely, if the optimization algorithm is able to
find a solution within the specified bounds of the
process, the violation variables become zero and
the entire MEGA term drops from the equation.
When the MEGA term zeros, the algorithm then
focuses on optimizing the manipulated variables
at the least amount of cost (determined by the
cost coefficient, Ci).
The CMPC optimization algorithm discussed
above is a simplistic explanation of this type of
control technology. Commercial CMPC algorithms predict process outputs at several sampling
time instances into the future. This ability allows
for the compensation of unmeasured disturbances
while maintaining future process outputs within
specified bounds. For a CMPC algorithm to make
predictions, a dynamic model of the process must
be available. Dynamic models are usually loworder transfer functions fit to experimental data
or a set of impulse-response coefficients describing the experimental data. Impulse-response

133 (2001) 271-283

coefficients represent the difference, or change,


between the discrete values of the sampled
instances of a process step-response.

4. Description

of the pilot plant

and the

control system

The present study uses a variation of the RO


desalination pilot plant described in previous
research [3]. The present study replaced the
spiral-wound membranes used previously with a
4 hollow-fiber membrane module. The pilot
plant is shown schematically in Fig. 2.
The hollow-fiber module is a B-9 Permasep@
permeator initially capable of producing
1400gaVd at an operating pressure of 3504OOpsig from a brackish water feed solution
containing 1500 mg/l of salt. The permeator is
manufacturer-rated at 94% salt rejection [6]. The
hollow-fiber module is operated at a much lower
pressure than the specified range given by
DuPont because the pilot plant pump can only
generate a maximum of 250psia of feed water
pressure. The feed water pressure is, however,
sufficient to desalinate a dilute feed solution
containing 0.5% weight sodium chloride.
The reject (concentrate) and acid-inlet valves
are pneumatic and computer controlled. The
control valves are air-to-open valves; in other
words, if there is a system power or air supply
failure, the valves will close. The positions of
these two valves are the manipulated variables of
the system. The valve positions are regulated by
pneumatic signals produced by I/P. The I/P
transducers convert 4 to 20 mA signals, produced
from the pilot plant controller, to 3 to 15 psig
pneumatic signals. The pilot plant has four
outputs: (1) permeate flow rate, (2) permeate
conductivity (a measure of salt content), (3)transmembrane pressure, and (4) inlet pH. Various
transmitters, also seen in Fig. 2, are connected to
the plant controller to monitor the four process
outputs.

A.C. Burden et al. /Desalination

133 (2001) 271-283

275

Fig. 2. Schematic of the pilot plant.

The plant controller is a Dell computer


equipped with a 400MHz Intel Pentium II
processor with 128 MB of RAM operated by
Microsoft Windows 98. The computer has a 32bit VESA local bus IDE operating at 100 MHz.
The plant controller utilizes a data acquisition
and control software called FIX DMACS version
7.0, which stands for Fully Integrated Distributed
Manufacturing Automation and Control Software, from the Intellution Company.
FIX DMACS is a computer program that
allows graphical user interface (GUI) between
the process and the control engineer. The GUI
allows a control engineer to schematically
display the process with real-time data and
graphs, as well as manipulate process variables
(e.g., control valve positions). Furthermore, FIX
DMACS has the ability to save data for later
review or to export data to files that can be used
in such programs as Microsoft Excel. FIX
DMACS contains the conventional PID control
algorithm. For the current study, only PI control
was used due to process noise. The FIX DMACS
PID algorithm is basically the same as the PID

algorithm discussed in previous sections; however, the sampling time is referred to as scan
time, which is the scanning (sampling) period of
the database.
The CMPC software used in this study is
called ONY..,?? which was developed by
Simulation and Advanced Controls Incorporated.
ONLLW? is an external program that optimizes
the pilot plant in conjunction with the normal
operating management of FIX DMACS.

5. Results and discussion

To develop an appropriate control strategy for


the pilot plant, the system must be characterized
by open-loop step-responses. The open-loop stepresponses describe the dynamics of the system
when small changes in the manipulated variables
are
occur. The open-loop step-responses
performed from a desired steady state that
mimics normal operating conditions so that
proper control may be achieved. The change in
the manipulated variable should be large enough

276

A.C. Burden et al, /Desalination 133 (2001) 271-283


5% decrease in
Reject Valve Position

1O?hincrease in
Acid-Inlet Valve Position

Permeate
Flow Rate
(gP@

Pemreate
Conductivity
Wcm)

TransMembrane
RCSSUre

(psia)

Inlet pH
0

100

200

300

25

50

75

100

125

Fig. 3. Open-loop step-responses (deviations from steady state).

to distinguish system changes from noise and


small enough to ensure stability. The open-loop
step-responses are expressed in deviation form;
that is, deviations from steady state. Since the
system has two manipulated variables, two openloop system step-responses were conducted.
The first open-loop step-response conducted
was of a small change in the reject valve position.
The system was brought to a steady state with a
reject valve position of 50% open. Then, the
reject valve position was decreased to 45% open.
The acid-inlet valve position was closed throughout the test. The second open-loop response test
conducted was for a change in the acid-inlet
valve position. Once again, the process was
allowed to reach steady state with an acid-inlet

valve position of 40% open. Then, the acid-inlet


valve was adjusted to 50% open. The reject valve
was held at a constant position of 50% open
throughout the test. Fig. 3 depicts the stepresponses ofthe four process outputs in deviation
form. Smooth curves were drawn through the
data points using Polymath, a computer program
that performs data regression analysis and least
square-error trending.
5.1. PI control
Before the PI control law could be applied to
the pilot plant, suitable tuning constants had to be
found. The method of obtaining tuning constants
for this study is the ultimate gain analysis

277

A.C. Burden et al. /Desalination 133 (2001) 271-283

Table2
Processdata for PI controltest

Table 1
PI controllertuning constants
Flow
controller

Conductivity
controller

p, s
K,, Yd%

-25.5
25
- 11.475

-1.7
150

21,s

20.83

Ziegler-Nichols:
Ku, %I%

Adjusted:
Kc, o/d%

- 11.475
20.83

?, s

- 0.765
125
- 0.765
65

originally proposed by Ziegler and Nichols. This


method is described in detail by Deshpande and
Ash [7]. The ultimate gain analysis was performed using a Microsoft Excel simulation of the
pilot plant. The simulation utilizes the open-loop
step-response coefficients obtained at 5-s sampling intervals from the smooth curves drawn
through the open-loop step-response data.
The impulse-response models describe the
process outputs at set intervals of time. Eq. (8)
describes the calculation of an output, y, used in
the simulation [7].
1
yk+l

= 2

i=l

hi k+ 1 -i

(8)

where yk+1is the output, y, at the next sampling


instance; hi the impulse-response coefficients
derived from the open-loop response tests; and
u,,_,=is the position of the manipulated variable.
Table 1 lists the ultimate gains and periods, as
well as the tuning constants. Using the ZieglerNichols tuning constants, the simulation produced an acceptable response except for a minor
offset of conductivity. To compensate for the
minor offset, the integral time of the conductivity
controller was decreased from 125 to 65s, as
shown in Table 1.

Controlled variables:
Permeateflow rate, gph
Permeateconductivity,
pS/cm
Associatedvariables:
Transmembranepressure
psia
Feed pH
Manipulated variables:
Rejectvalve position
% open
Acid-inletvalve position
% open

Initial
values

Final Set
values point

6.97
483.52

6.50
488.8

126.08
5.4

135.95 2.99
-

49.84

52.32

40.0

60.49

6.5
483.0

Once the controller tuning constants were


obtained, the normal start-up procedure was
followed with the controllers set in manual. Once
the system reached the desired steady state, the
controllers were set to hold the process at steady
state by placing them in automatic.
For the PI control test, a set point change in
permeate flow rate, from 7 to 6Sgph, was
introduced to the pilot plant. Fig. 4 shows the
closed-loop response of the PI control test and
the accompanying process data are listed in
Table 2.
As seen in Fig. 4, the PI control strategy failed
to control the pilot plant properly for a set point
change in permeate flow rate. A vertical dashed
line was drawn on the first two graphs of Fig. 4 at
the 53-min mark, the point at which the
conductivity PI control loop became unstable.
The reason for the failure was due to a nonlinearity in the effect of feed pH as a means of
controlling permeate conductivity.
Due to the decrease in permeate flow rate and
transmembrane pressure, the permeate conductivity increased from a set point of 483 pS/cm. To
compensate for the deviation, the conductivity

278

A.C. Burden et al. /Desalination

Permeate

7.5

133 (2001) 271-283

Conductivity

7'
I
I

-'-480

f!

Permeate

8
5.5

Flow

-470

Rate

I
=

,
I

I
,

-- 460

I
I

450
0

10

15

20

25

30

Tim0

35

40

45

50

55

60

65

(mInutea)

200
180
150
170
%

160
150
140
I
,,

130
0

10

15

20

25

30

tlmo

35

40

45

50

55

-0
60

65

(mlnutrr)

90
80
70

Acid-Inlet

Valve

POeltlOtI

Reject

Valve

Posltlon

80
B
ur

50
40
30
20

-r

IO

15

20

25

30

36

40

45

50

55

60

65

Tlm 0 (m Inutrr)

Fig. 4. Unsatisfactory

PI control.

controller opened the acid-inlet valve, which in


turn lowered the feed pH and permeate
conductivity. However, the permeate conductivity did not return to the set point and there was
offset. The feed pH exhibited a maximum effect

on the permeate conductivity at a value of 4.5 in


the time interval of 43 to 53 min of the test. This
non-linearity
was encountered
in previous
research conducted by Light et al. [S], who found
that the maximum chlorine ion rejection for

279

A.C. Burden et al. /Desalination 133 (2001) 271-283

99.0

;ofkedvitH

Fig. 5. The pH effect from Light et al. [8].

spiral-wound
thin
composite
membranes
occurred around a pH of 5. Fig. 5 illustrates the

non-linear pH effect.
The PI controller for the permeate conductivity only acts upon the error associated with the
set point and the controlled variable as described
previously by the PI control law. Therefore, the
PI controller continues to open the acid-inlet
value to reduce permeate conductivity when
actually the opposite occurs as seen past the
vertical dashed line in Fig. 4.

5.2. CMPC control


The study used a commercial CMPC software
package called ONLLN. developed by Simula-

tion and Advanced Controls Incorporated. The


purpose of the CMPC control strategy is to
operate the pilot plant by maintaining the four
process variables within specified ranges, or
bounds. Unlike the PI control strategy, the CMPC
control strategy can account for all process
outputs. In the previous PI control test, the feed
pH and transmembrane pressure were not
specifically controlled by a controller. Therefore,
one of the primary focuses for implementing the
CMPC control strategy was to properly control
the feed pH, to prevent run away control as well
as improve the performance of the pilot plant.
Before the CMPC control strategy could be
applied to the pilot plant, simulations were
performed using ONLINE to determine proper
tuning parameters such as the move penalties
(suppressions) of the manipulated variables and
the weights of the controlled variables. The first
ONLINE test discussed utilized proper tuning
constants that were obtained from the simulations. Moreover, the purpose of the first test
was to determine if ONLLVP could hold the
permeate flow rate at a desired value while
holding the other process variables within the
specified bounds, as well as to find the best state
of the system. The second CMPC test discussed
presents the maximization ability of the CMPC
control strategy by which the reject valve
position is minimized (which maximizes the

Table 3
Process data for the first CMPC test
Initial value

Final value

Weight

Upper limit

Lower limit

7.04
476.1
114.2
5.0

7.01
476.8
111.4
4.5

15
3
1
10

7.05
480
170
6

6.95
470
100
4.5

39.88
40.65

41.81
40.93

15
20

70
50

30
0

Controlled variables:

Permeate flow rate, gph


Permeate conductivity, @/cm
Transmembrane pressure, psia
Feed pH
Manipulated variables:

Reject value, % open


Acid-inlet valve, % open

280

A.C. Burden et al. /Desalination

133 (2001) 271-283


520

if!

-3

-I

510

Permeate

Flow

500

Rate

490

4 -

450

470

3
2

Permeate

%0

450

Conductlvlty

450
1 --

440

07:::::::::::
0
4
5

12

15

20

24

Tlmr

25

(m

32

35

40

:45

44

430

Inutra)

155
5

145

135

Tranr-Mem

-I:
0

brane

Pressure

::::::::
4

12

20

tlm

24

25

0 (m

32

35

I1

40

44

45

Inutr8)
Acid-Inlet

:l,

::
15

Valve

Poetlon

~ctvalve,Poet,ion,

10
0
5

12

15

20

Tlmr

24

25

32

35

40

44

45

(mlnutrr)

Fig. 6. First CMPC test.

permeate flow rate) while holding the process


outputs within their specified bounds.
Table 3 lists the tuning parameters developed
from the simulations performed in UNLINK for
the first CMPC test. The pilot plant was brought
to a steady state in which the permeate flow rate
was about 7 gph; tight bounds of ho.05 gph were
placed upon the permeate flow rate. The bounds

of the pH were specified to prevent run away


control, with a lower bound of 4.5. The limits of
the reject valve were specified to hold the pilot
plant within modeling range; outside of this
range, the process step-response models were
invalid. The limits of the acid-inlet valve were set
to protect the membrane module. The results of
the first CMPC test are presented in Fig. 6.

A. C. Burden et al. /Desalination

281

133 (2001) 2 71-283

the CMPC control law only utilized the first term


of Eq. (3). For the second CMPC test, an
arbitrary cost value of 1 was assigned to the
reject valve. Using a positive value for cost, the
CMPC algorithm minimizes the reject valve
position, which in turn maximizes the permeate
flow rate. When the controlled variables are
within their respective bounds, the first term in
Eq. (10) becomes zero, and the CMPC subroutine
focuses on minimizing the reject valve position.
In order, to illustrate the maximization feature of
ONLINE@, the tight bounds placed upon the
permeate flow rate were relaxed. Table 4 lists the
adjusted tuning parameters for this CMPC test.
In this test, the pilot plant was brought to an
initial steady state at a permeate flow rate of
6.0gph. Once ONLINE was implemented, the
pilot plant went through a transition period
during the first 10 min of the test before moving
toward a final steady state. This test did not
exhibit the series of oscillations observed in the
first CMPC test. This is due to the relaxation of
the bounds on the permeate flow; the flow had
more freedom to deviate from the initial value
before ONLINE had to perform corrective
actions. The results of the second CMPC tests are
illustrated in Fig. 7.
As seen in Fig. 7, ONLIiVi minimized the
reject valve position to the lower bound after the
first couple of minutes. Throughout the rest of

Fig. 6 shows that the plant variables oscillated


for about 35 min before reaching a steady state.
For the first 35 min, ONLlNp manipulated the
control valves to find the best values, or state, of
the process outputs within their specified bounds.
Oscillations occurred because of the tight bounds
placed on the flow rate and the control action of
ONLLV...
Unlike PI control, ONLINE prevented runaway control due to the high weight placed upon
the feed pH. The bounds on the controlled
variables can be broken, as seen in the pH
response (the second graph of Fig. 6); however,
due to the high weight of the feed pH, ONLINE@
corrects its action and returns the feed pH back
within the proper bounds. The most obvious
point at which the feed pH broke the specified
bounds was 18 min into the test. The feed pH
dropped to almost 3 before ONLLVE@returned
the pH back within the lower bound of 4.5. In
fact, the pH oscillated around and finally reached
a steady state value of 4.5. In light of this observation, ONLLYIi@was able to bring the pilot plant
to a steady state within specified bounds. Moreover, for that steady state ONLINE held the pH
at the lower bound of 4.5.
The second CMPC test demonstrates the
maximization ability of ONLINE@. The first
CMPC test only regulated the pilot plant; the cost
variables Cj of Eq. (3) were zero. This means that
Table 4
Process data for the second CMPC test
-

Initial value

Final value

Weight

Upper limit

Lower limit

6.05
487.5
92.4
5.48

6.87
482.1
110.7
4.51

15
3
1
10

7.0
505
170
6

5.0
470
90
4.5

51.72
27.15

35.42
45.26

15
20

70
50

35
0

Controlled variables:

Permeate flow rate, gph


Permeate conductivity, pS/cm
Transmembrane pressure, psia
Feed pH
Manipulated variables:

Reject value, % open


Acid-inlet valve, % open

282

A.C. Burden et al. /Desalination

133 (2001) 271-283


520
610

Permeate

Flow

Rate

500
430
8

430

Prrmerto

Conductivity

470
4eo
450

12

13

Ttm

24

20

23

(m Inutor)
9

120

100

Trans.Mom

brrnr

.- 5

Prrrruro

--

30

30

40

20

I L-=y7y~~q#j
0

12

20

15

tlm

0 (m

24

23

Inutrr)

60

Acid-Inlet

Valve

Potltlon

60

40

30

20
to
0

12

Tim

16

20

24

2.8

32

0 (m Inulor)

Fig. 7. Second CMPC test.

the test, ONLIiV~ maintained the reject valve


position at the lower bound. Due to the minimization of the reject valve position, the flow rate
was maximized to a final value of 6.87 gph. This
is a 13.6% increase in permeate flow from the
initial steady-state value of 6.05 gph. Furthermore, the permeate conductivity was reduced by

1.1% to a final steady-state value of 482.1 @/cm,


which is an improvement in quality.
As observed from the CMPC tests, the pilot
plant can be controlled properly to incorporate
pH as a means of adjusting the quality of the
permeate. Finally, CMIT was able to maximize
the permeate flow while maintaining the other

A. C. Burden et al. /Desalination

process outputs within bounds, particularly the


feed pH.

I33 (2001) 2 71-283

283

References
PI I.M. Alitiqi, A.H. Ghabris and S. Ebrahim, Desalination, 75 (1989) 119.

6. Conclusions
The PI control strategy was unable to properly
control the permeate conductivity by means of
adjusting the feed pH. When large deviations in
permeate conductivity resulted from changes in
permeate flow, the PI control loop became
unstable. This occurrence is due to the fact that
the feed pH has a maximum effect at a value of
4.5 at which the chloride ion rejection is at a
maximum.
The CMPC control strategy was implemented
using ONLLV.., by which proper control of the
pilot plant was achieved. By using proper output
weighting, the feed pH was held within bounds at
the point of maximum chloride ion rejection.
Furthermore, the permeate flow rate and
conductivity as well as the transmembrane
pressure were also held within bounds. Finally,
ONLINE was able to maximize the throughput
of the pilot plant by a 13.6% increase in the
permeate flow rate with a 1.1% decrease in the
permeate conductivity, indicating an improvement in quality.

PI R. Singh and J. Tembrock, Chemical Engiueering


Progress, 95 (1999) 57.
[31 J.Z. Assef, J.C. Watters, P.B. Deshpande and I.M.
Alatiqi, J. Process Control, 7(4) (1997) 283.
141C.A. Smith and A.B. Corripio, Principles and Practice
of Automatic Process Control, 2nd ed., Wiley, New
York, 1997.
[51 P.B. Deshpande, J.A. Caldwell, S.S. Yerrapragada
and M.A. Bhalodia, Chemical Engineering Progress,
91 (1995) 65.
DuPont
Company, Permasepe reverse osmosis
PI
products, Product Bulletin No. 2 13, Wilmington, DE,
1994.
P.B.
Deshpande and R.H. Ash, Computer Process
E71
Control with Advanced Control Application, Instrument Society of America, North Carolina, 1988.
PI W.G. Light, Z.B. Taylor and A.B. Riedinger, in:
Reverse Osmosis and UltrafIltration, S. Sourirajan
and M.T. Ma&mm, eds., American Chemical
Society, Washington, DC, 1985, pp. 247-260.
Intellution.
FIX32 System Development. Intellution
PI
Inc., Norwood, MA, 1994.
WI ONLINE Users Manual, Advanced Controls Inc.,
Louisville, KY, 1999.

Potrebbero piacerti anche