Sie sind auf Seite 1von 6

st

Proceedings of the 21 International Conference on


Automation & Computing, University of Strathclyde,
Glasgow, UK, 11-12 September 2015

Self-Tuned Fuzzy Logic Control of a pH


Neutralization Process
Parikshit Kishor Singh1, Surekha Bhanot2, Hare Krishna Mohanta3, Vinit Bansal4
Dept. of Electrical & Electronics & Engineering1,2, Dept. of Chemical Engineering3, Application Engineering Dept.4
BITS Pilani, Pilani campus1,2,3; National Instruments, India4
Pilani, Rajasthan, India1,2,3; Bengaluru, Karnataka, India4
pksingh.bitspilani@gmail.com , surekha0057@gmail.com2, harekrishna.bits@gmail.com3, vinitbansal2010@gmail.com4
1

AbstractOn-line implementation of self-tuning mechanism Laguerre polynomial techniques, and neural networks [7],
based adaptive fuzzy logic control of a pH neutralization [8], [9], [10], [11], [12], [13], [14]. In addition, popular
process which takes care of steady state error and time conventional adaptive techniques such as gain-scheduling,
taken to reach steady state under varying operating model reference adaptive control and self-tuning
conditions has been presented in this paper. The pH regulators are also realized using advanced identification
neutralization system is Armfield pH Sensor Accessory and control techniques [15], [16], [17].
(PCT42) in conjunction with Process Vessel Accessory
(PCT41) and Multifunction Process Control Teaching Dynamic pH model ability to replicate the actual
System (PCT40). The proposed adaptive scheme updates the nonlinear behavior of neutralization processes limited the
normalized universe of discourse of output fuzzy accuracy and ability of the controller. Thus few
membership functions with varying scaling factors based on researchers used concept of model-free intelligent control
error and change of error values. The speed of response of [18], [19], [20]. The fuzzy logic control is based on
the adaptive controller is taken care by use of coarse control intelligent methodology of human thinking and decision
technique whereas amount of deviation under steady state is making mechanisms. Self-organizing based adaptive
accounted with the help of fine control technique. The control has been implemented using fuzzy logic [21], [22].
performance of adaptive scheme is tested for pH control at This paper discusses on-line application of fuzzy logic
equivalence point. LabVIEW software is used for online based self-tuned pH control of strong acid-strong base
communication, control and display.
neutralization process stream.
Keywords-pH neutralization; nonlinear process; fuzzy II. DESCRIPTION OF NEUTRALIZATION PLANT
logic control; self-tuning control; adaptive control; on-line
control The Armfield pH neutralization system is shown in
Figure 1. The pH probe PCT42 calibration against buffer
I. INTRODUCTION pH solutions of 4, 7 and 9.2 results in a linear relationship
Control of pH has vital significance in our daily life. between sensor voltage and equivalent pH. The pH
Modern process industries such as food processing units, neutralization process takes place in PCT41 with perfect
biopharmaceutical manufacturing plants, iron & steel mixing and constant maximum volume (V). The PCT40
industry and thermal power plants run various operations has two peristaltic pumps A and B which regulate flow of
where pH monitoring and controlling are critical. More hydrochloric acid (HCl) and sodium hydroxide (NaOH)
importantly, vast and rapid globalization necessitated having concentrations Ca and Cb respectively. Although
enormous focus on pH control of industrial effluents using speed of pumps A and B, Sa and Sb respectively, can vary
wastewater treatment so that ecological and environmental from 0 to 100% of maximum speed, both pumps do not
balance could be maintained in our Mother Nature. pH start below a minimum value of 18%. During pump
control is often taken as benchmark problem for nonlinear operation within useful range of 18 to 100%, pump
control because of highly nonlinear nature of flowrate varies almost linearly with speed. A brief and
neutralization reaction. Time varying nature of pH important specifications of pH neutralization process is
neutralization process essentially makes pH as a moving given in Table I.
target whose precise control is almost impossible to Using standard universal synchronous bus interface
achieve. Therefore, although pH control is conventional the PCT40 communicates with LabVIEW software
problem but it still fascinates many young researchers. installed on a personal computer having Windows XP
Early works on pH control were based on design of Professional 2003 operating system. The PCT40 interface
adaptive techniques using dynamic process models device driver contains a dynamic link library (DLL) file
developed using first-principle approaches such as laws of which stores various input-output analog and digital
conservation, physical and chemical laws, reaction control signal values. The analog signals between 0 V or
invariants, and strong acid equivalent [1], [2], [3], [4], [5], 0% to 5 V or 100% are stored in 12-bit signed-magnitude
[6]. Rapid advances in modern process control led to representation as 000000000000 to 011111111111 in
development of nonlinear model based predictive control binary or 0 to 2047 in decimal whereas the digital signals
schemes. Many popular model predictive control either 0 V or 5 V are stored in 1-bit representation as 0 or
strategies incorporated nonlinear process models including 1 in binary, respectively. LabVIEW software accesses the
those based on Wiener, Hammerstein, Volterra series, DLL file for following functionality: read analog input,
in the operating conditions. In this paper, we have
designed a self-tuned FLC, shown in Figure 4, whose
output scaling factor K3 are as per entries in Table III. The
input variables e and ce are divided into following seven
identical regions: e1, ce1 [-6,-1); e2, ce2 [-1,-0.5); e3,
ce3 [-0.5,-0.1); e4, ce4 [-0.1,0.1]; e5, ce5 (0.1,0.5];
e6, ce6 (0.5,1]; e7, ce7 (1,6]. As evident from Table III
entries, K3 has been assigned a larger value if pH is away
from pHSP and K3 has been assigned a smaller values as
pH approaches pHSP. Using large K3 when pH is at
distance from pHSP will ensure reduced settling time and
using small K3 when pH is near pHSP will ensure steady-
state response within settling band. Therefore Table III
entries validates use of coarse control and fine control
techniques.
NL NM NS ZE PS PM PL
Figure 1. Armfield pH neutralization system (PCT42 plus 1
PCT41 plus PCT40)

Degree of membership
TABLE I. NEUTRALIZATION PROCESS SPECIFICATIONS

Quantity Specification

PCT41 volume (V) 2000 mL

Concentration of HCl (Ca) 0.0174 mol/L


0
Concentration of NaOH (Cb) 0.0138 mol/L
-1.00 -0.67 -0.33 0.00 0.33 0.67 1.00
Speed range of pumps A and B 18 to 100%
Error / Change of error (pH)
Equivalent flowrate of pump A (Fa) 0.2021 to 5.1139 mL/s
Figure 2. Fuzzy inputs membership functions
Equivalent flowrate of pump B (Fb) 0.2989 to 5.8749 mL/s
NL NM NS ZE PS PM PL
Voltage range of pH sensor 0 to 5 V 1

Equivalent pH reading 0.1868 to 13.2438


Degree of membership

Sampling period 1s

pH probe value from channel 11 (Ch11); write analog


outputs, pump A and B speed values to digital-to-analog
converters DAC0 and DAC1 respectively; write digital
output, stirrer signal value to digital output line 7 (DO7).
0
III. DESIGN OF ADAPTIVE FUZZY LOGIC CONTROL -1.00 -0.67 -0.33 0.00 0.33 0.67 1.00
The fuzzy logic control (FLC) of pH neutralization Change of output (%)
process is based on Mamdani fuzzy inference system
(FIS). The input variables used for FLC are e*(k) = Figure 3. Fuzzy output membership functions
e(k)/K1 = (pHSP(k) pH(k))/K1 and ce*(k) = ce(k)/K2 =
(e(k) e(k-1))/K2, where pHSP is setpoint, pH is controlled TABLE II. FUZZY RULE BASE
variable, e is error, ce is change of error, k is sampling ce
instant, and K1, K2 are scaling factors. The output variable
e
used for FLC is co*(k) where co* is normalized change of NL NM NS ZE PS PM PL
output. The normalized membership functions for the NL NL NL NL NL NM NS ZE
input variables (e*, ce*) and output variable (co*) are
shown in Figure 2 and Figure 3 respectively. The rule base NM NL NL NL NM NS ZE PS
for the input and output variables are shown in Table II. NS NL NL NM NS ZE PS PM
For a nonlinear process, the FLC needs to be ZE NL NM NS ZE PS PM PL
optimized for better performance. However, often
operating conditions of the nonlinear process changes PS NM NS ZE PS PM PL PL
which means repeated application of optimization PM NS ZE PS PM PL PL PL
procedure. Self-tuning mechanism allows adaptive FLC to
alter its input and/or output scaling factors as per changes PL ZE PS PM PL PL PL PL
TABLE III. DETERMINATION OF K3
Start %% Begin LabVIEW implementation
Range for ce for n = 1 to No. of Iterations %% Begin initialization
Range for e write in DLL to start the stirrer
ce1 ce2 ce3 ce4 ce5 ce6 ce7
read from DLL to obtain pH sensor voltage
e1 8 8 8 8 6 4 2 estimate pH
e2 8 8 8 6 4 2 4 while initial pH is not within range [pHLB pHUB]
if pH < pHLB
e3 8 8 6 4 2 4 6 write in DLL to set Sa = 35 and Sb = 40
e4 8 6 4 2 4 6 8 end
if pH > pHUB
e5 6 4 2 4 6 8 8
write in DLL to set Sa = 35 and Sb = 35
e6 4 2 4 6 8 8 8 end
e7 2 4 6 8 8 8 8
end %% End initialization
initialize Sa = 35, Sb = 38, ISE = 0, final pHSP
1/K1 pHSP for k = 1 to Sampling Duration %% Begin control
estimate pH(k), e(k), and ce(k)
e obtain e*(k) = e(k)/K1, ce*(k) = ce(k)/K2
+
e* obtain co*(k) using FIS based on Table II
obtain K3(k) using Table III
co* multiply K3(k) by factor M to obtain K3'(k)
ce scale co*(k) by factor K3'(k) to obtain co(k)
ce* +
z-1
obtain Sb(k) = Sb(k-1) + co(k)
Mamdani Fuzzy Unit time delay write in DLL to update Sa and Sb
Inference System
1/K2
update ISE(k) = ISE(k-1) + (e(k))2
update sampling time k = k + 1
Row end %% End control
K3 end %% End LabVIEW implementation
Column
ON
Figure 5. Pseudocode of self-tuned FLC
Table III for K3
K3'
given by e3, e4, e5 and ce3, ce4, ce5 respectively,
Stirrer magnification of K3 results in faster pH response.
M
Sa Pump A (b) For M = 1, magnitude of first overshoot is largest as
pH compared to the same for M = 2, 3, 4. A magnified K3
co
provides better neutralization of acidic process stream
Sb
+ Pump B using basic manipulated stream.
(c) For M = 1, 2, 3, pH response remains within 70.2 pH
z-1 PCT42 + PCT41 + PCT40 settling band in a better way as compared to the same for
M = 4. A magnified K3 sometimes drives the pH response
Figure 4. Self-tuned fuzzy logic based pH control outside of the settling band.

IV. RESULTS AND DISCUSSIONS (d) For M = 4, pH response indicates self-controlling


property of the adaptive FLC. Figure 6 shows that during
Figure 5 shows pseudocode of the proposed control sampling instants 85 to 130 seconds, pH response is
algorithm which consists of two parts for sequential showing tendency to go unbound. Subsequently, the self-
execution: first process initialization and second adaptive tuned fuzzy controller adjusts the value of K3 so that pH
FLC. Process initialization ensures that initial pH value response comes within the desired settling band.
falls within user specified lower bound (pHLB) and upper
bound (pHUB). In order to evaluate performance of Similarly, for Cases 25 to 28, Figure 9, Figure 10 and
adaptive intelligent controller, pHSP is changed from 6 to 7 Figure 11 indicates graphical performance comparisons in
with [pHLB (= 5.9), pHUB (= 6)], and from 8 to 7 with terms of pH response, pumps speed and magnified output
[pHLB (= 5.9), pHUB (= 6)]. Table IV summarizes ISE scaling factor K3' respectively. From the system response,
performance of the proposed controller for various values it is observed that a magnified K3 results in less
of scaling factors K1, K2 and M. In particular, graphical undershoot, faster response and reduced settling time.
performance comparison in terms of pH response, pumps Table IV shows that, under similar operating conditions
speed and magnified output scaling factor K3' has been and parameters setting, ISE is largest for M = 1 and ISE
shown in Figure 6, Figure 7 and Figure 8 respectively for improves for M = 2, 3, 4.
Cases 1 to 4. Following observations can be made from
the obtained results. V. CONCLUSION
Self-tuned fuzzy logic control has been implemented
(a) For M = 1, pH response is slower than the same for M
on a laboratory scale pH neutralization systems from
= 2, 3, 4. Since mostly e and ce attains values within range
TABLE IV. PERFORMANCE SUMMARY OF SELF-TUNED FLC [3] T.K. Gustafsson and K.V. Waller, Dynamic modeling and
reaction invariant control of pH, Chemical Engineering Science,
Case Initial pHSP Final pHSP K1 K2 M ISE vol. 38, pp. 389-398, 1983.
1 6 7 10 0.5 1 23.75 [4] T.K. Gustafsson, An experimental study of a class of algorithms
for adaptive pH control, Chemical Engineering Science, vol. 40,
2 6 7 10 0.5 2 18.91 pp. 827-837, 1983.
3 6 7 10 0.5 3 14.72 [5] R.A. Wright and C. Kravaris, Nonlinear control of pH processes
using strong acid equivalent, Ind. Eng. Chem. Res., vol. 30, pp.
4 6 7 10 0.5 4 17.42
1561-1572, 1991.
5 6 7 10 1 1 58.67 [6] R.A. Wright, M. Soroush, and C. Kravaris, Strong acid
6 6 7 10 1 2 30.98 equivalent control of pH processes: An experimental study, Ind.
Eng. Chem. Res., vol. 30, pp. 2437-2444, 1991.
7 6 7 10 1 3 35.07
[7] Y.-K. Yeo and T.-I. Kwon, A neural PID controller for the pH
8 6 7 10 1 4 13.79 neutralization process, Ind. Eng. Chem. Res., vol. 38, pp. 978-
9 6 7 20 0.5 1 24.58 987, 1999.
[8] B.M. kesson, H.T. Toivonen, J.B. Waller, and R.H. Nystrm,
10 6 7 20 0.5 2 21.74
"Neural network approximation of a nonlinear model predictive
11 6 7 20 0.5 3 18.90 controller applied to a pH neutralization process," Computers &
Chemical Engineering, vol. 29, pp. 323-335, 2005.
12 6 7 20 0.5 4 18.22
[9] S.J. Norquay, A. Palazoglu, and J.A. Romagnoli, "Application of
13 6 7 20 1 1 26.94 wiener model predictive control (WMPC) to a pH neutralization
14 6 7 20 1 2 14.03 experiment," IEEE Transactions on Control System Technology,
vol. 7, pp. 437-445, 1999.
15 6 7 20 1 3 16.55
[10] S. Oblak and I. krjanc, "Continuous-time Wiener-model
16 6 7 20 1 4 14.88 predictive control of a pH process based on a PWL
17 6 7 30 0.5 1 34.95 approximation," Chemical Engineering Science, vol. 65, pp. 1720-
1728, 2010.
18 6 7 30 0.5 2 31.65
[11] H.C. Park, S.W. Sung, and J. Lee, "Modeling of Hammerstein-
19 6 7 30 0.5 3 28.82 Wiener processes with special input test signals," Ind. Eng. Chem.
20 6 7 30 0.5 4 30.14 Res., 45, 1029-1038, 2006.
[12] K.P. Fruzzetti, A. Palazolu, and K.A. McDonal, "Nonlinear
21 6 7 30 1 1 35.99 model predictive control using Hammerstein models," Journal of
22 6 7 30 1 2 21.63 Process Control, vol. 7, pp. 31-41, 1997.
23 6 7 30 1 3 19.15 [13] S. Mahmoodi, J. Poshtan, M.R. Jahed-Motlagh, and A. Montazeri,
"Nonlinear model predictive control of a pH neutralization process
24 6 7 30 1 4 15.40 based on WienerLaguerre model," Chemical Engineering
25 8 7 10 0.5 1 20.55 Journal, vol. 146, pp. 328-337, 2009.
26 8 7 10 0.5 2 15.34 [14] R. Daz-Mendoza and H. Budman, "Structured singular valued
based robust nonlinear model predictive controller using Volterra
27 8 7 10 0.5 3 13.73 series models," Journal of Process Control, vol. 20, pp. 653-663,
28 8 7 10 0.5 4 12.69 2010.
[15] R.H. Nystrm, B.M. kesson, and H.T. Toivonen, "Gain-
29 8 7 20 0.5 1 29.26
scheduling controllers based on velocity-form linear parameter-
30 8 7 20 0.5 2 22.86 varying models applied to an example process," Ind. Eng. Chem.
Res., vol. 41, pp. 220-229, 2002.
31 8 7 20 0.5 3 21.56
[16] M.C. Palancar, J.M. Aragn, J.A. Miguns, and J.S. Torrecilla,
32 8 7 20 0.5 4 20.12 "Application of a model reference adaptive control system to ph
control. effects of lag and delay time," Ind. Eng. Chem. Res., vol.
35, pp. 4100-4110, 1996.
Armfield. LabVIEW software has been used for on-line [17] M. Alpbaz, H. Hapolu, G. zkan, and S. Altuntas, "Application
signal processing. Adaptive fuzzy controller performance of self-tuning PID control to a reactor of limestone slurry titrated
has been evaluated for step change in pH setpoint from 6 with sulfuric acid," Chemical Engineering Journal, vol. 116, pp.
to 7 and 8 to 7. In almost all test results, pH response 19-24, 2006.
finally settles within 70.2 pH band. In some cases, when [18] Z. Zheng and N. Wang, "Model-Free Control based on Neural
Networks," in Proc. International Conference on Machine
pH response occasionally overshoots and undershoots the Learning and Cybernetics, 2002, Beijing, pp. 2180-2183.
above band, controller adjusts its output universe of
[19] S. Syafiie, F. Tadeo, and E. Martinez, "Macro-actions in model-
discourse and again brings the pH response back within free intelligent control with application to pH control," in Proc.
the desired band. The self-tuned fuzzy controller gives 44th IEEE International Conference on Decision and Control, and
better performance in terms of higher speed of response, the European Control Conference, 2005, Spain, pp. 2710-2715.
less deviation from setpoint and better settling within band [20] S. Syafiie, F. Tadeo, and E. Martinez, "Model-free learning
for magnified fuzzy output scaling factor. control of neutralization processes using reinforcement learning,"
Engineering Applications of Artificial Intelligence, vol. 20, pp.
REFERENCES 767-782, 2007.
[21] E.H. Mamdani, Application of fuzzy logic to approximate
[1] T.J. McAvoy, E. Hsu, and S. Lowenthals, Dynamics of pH in reasoning using linguistic synthesis, IEEE Trans. on Computers,
controlled stirred tank reactor, Ind. Eng. Chem. Process Des. vol. 26, pp. 1182-1191, 1977.
Develop., vol. 11, pp. 68-70, 1972.
[22] T.J. Procyk and E.H. Mamdani, A lingustic self-organizing
[2] T.J. McAvoy, Time optimal and Ziegler-Nichols control. process controller, Automatica, vol. 15, pp. 15-30, 1979.
Experimental and theoretical results, Ind. Eng. Chem. Process
Des. Develop., vol. 11, pp. 71-78, 1972.
7.8 pHSP pH (M = 1) pH (M = 2) pH (M = 3) pH (M = 4)

7.4

7
pH

6.6

6.2

5.8
1 101 201 301 401 501
Sample number

Figure 6. pH response for K1 = 10, K2 = 0.5, pHSP (initial) = 6, pHSP (final ) = 7

Sa Sb (M = 1) Sb (M = 2) Sb (M = 3) Sb (M = 4)
65

60

55

50
Pump speed (%)

45

40

35

30

25

20
1 101 201 301 401 501
Sample number

Figure 7. Pump B response for K1 = 10, K2 = 0.5, pHSP (initial) = 6, pHSP (final ) = 7
37 K3' (M = 1) K3' (M = 2) K3' (M = 3) K3' (M = 4)

33

29

25

21

17
K3'

13

1
1 101 201 301 401 501
Sample number
Figure 8. K3' for K1 = 10, K2 = 0.5, pHSP (initial) = 6, pHSP (final ) = 7
8.2 pHSP pH (M = 1) pH (M = 2) pH (M = 3) pH (M = 4)

7.8

7.4
pH

6.6
1 101 201 301 401 501
Sample number
Figure 9. pH response for K1 = 10, K2 = 0.5, pHSP (initial) = 8, pHSP (final ) = 7

50 Sa Sb (M = 1) Sb (M = 2) Sb (M = 3) Sb (M = 4)

45
Pump speed (%)

40

35

30

25
1 101 201 301 401 501
Sample number
Figure 10. Pump B response for K1 = 10, K2 = 0.5, pHSP (initial) = 8, pHSP (final ) = 7

37 K3' (M = 1) K3' (M = 2) K3' (M = 3) K3' (M = 4)

33

29

25

21
K3'

17

13

1
1 101 201 301 401 501
Sample number
Figure 11. K3' for K1 = 10, K2 = 0.5, pHSP (initial) = 8, pHSP (final ) = 7

Das könnte Ihnen auch gefallen