Copyright © IF AC Adaptive Control of

Chemical Processes, Frankfurt am Main, 1985

ADAPTIVE CONTROL OF A LIQUID-LIQUID

EXTRACTOR
N. Yoswathana, G. Casamatta and K. Najim
Institut du Genie Chimique (UA-CNRS 192), Greco Systemes Adaptatifs, Chemin de
la loge, 31078 Toulouse, Cedex, France

Abstract, This paper presents experimental results of an application of self-

tuning P. I. D, controller to a liquid-liquid extraction column. The main
control objective is to maintain the pulse extraction columoperating at the
flooding point by controlling the pulser frequency. A simplified model of the
complex dynamics of the process was assumed. The controller derived from a
discrete P I D regulator is based on pole assignment. The controller uses a
recursive parameter estimator with constant trace, to estimate the plant para-
meters. The extractor was interfaced by an Apple 11 micro-computer. The results
obtained illustrate the key features of self-adaptive control.

Keywods. Adaptive control; data-acquisition; P I D control time-varying

systems ; liquid-liquid extraction column.

INTRODUCTION

During the last decade, the designsof regulators which are drilled 1/5 cm diameter holes, on a tri-
used in chemical process industry are the P, PI,PID angular pitch to give approximatively 19% free
types. Many rules have .been proposed for adjusting area ; plate spacing is 5 cm. In this study, the
these regulators, the most commonly used of them dispersed phase is the light solvent phase, so that
having been developped by Ziegler and Nichols (1943) the distributor is located at the bottom of the
column.
It has been shown that fitting the parameters of
the regulator revealed often itself a tedious and The interface level, in the settling zone is con-
uncertain task. Moreover, even in the case of a trolled by means of a capacity probe: this device
given process, these settings have to be modified affords a voltage proportional to the interface
for any changes that may affect the dynamic res- level; this voltage is compared to two reference
ponse of this process. values, corresponding to the highest and lowest
desired interface levels. By means of a relay, the
In the late seventies, Wittenmark (1979) has pro- comparator has an on-off action on an electroma-
posed a self-tuning PID controller algorithm gnetic valve at the output of the continuous phase.
consisting in two parts, estimating recursively the
parameters of a second order model, and then tuning The pulsing unit and both feed pumps are driven by
the regulator parameters to obtain a desired closed D. C. electrical motors which via a speed control
loop behavior. box can be commanded by a 0-10 V continuous current.
Therefore, both the pulsation frequency and the
The objective of this paper is to describe the feed flowrates might be automatically controlled.
application of such a regulator in order to keep
the operating point of a pulsed perforated plate External equipment: the flowrates are measured by
extraction column within the optimal region. means of electromagnetic flow-m~er. A conductivity
electrode is set below the light phase distri-
The description of apparatus, experimental techni- butor and connected to a conductivity meter: this
ques and procedure are given in section 1. On-line device affords a determination of the amount of
data acquisition and microcomputer interfacing are dispersed organic phase contained in the aqueous
described. In section 2, it is stated upon the hy- continuous phase. The reason why it has been selected
drodynamic behavior and flooding exhibition of a as the controlled variable will be discussed in
pulsed sieve plate column. In section 3, confirming section 2. The concentration of solute in the raf-
the previous work of Carrier (1981), it is demons- finate aqueous phase is measured by a continuous
trated that the efficiency of the column is optimal differential refractometer, deminerized water being
in the transition zone between the emulsion region used as a reference liquid.
and the flooding region. The self-tuning PID regu-
lator algorithm is presented in section 4. All of this apparatus are interfaced with an
Finally, some practical experiments and results are Apple 11 microcomputer. Both flowrates signals, the
discussed in section 5. conductivity signal and refractive index signal are
inputted through a 12-bit A/D converter. Before
EXPERIMENTAL DEVICE being sent to the A/D converter, the conductivity
signal is filtered by a low-pass filter with a cut-
The column: as shown on fig. 1, the column is cons- off frequency of 0.5 Hz. The feed flowrates and the
tructed from 5 cm diameter glass tubing ; wider pulse frequency D. C, motors are controlled from
glass sections are fitted to the 1.1 m high main the microcomputer by using a 12-bit D/A converter.
column body. A 5 cm diameter pulse line is fitted It can be summarized that all these connexions have
to the lower section. The sieve plates in the column been settled in order to allow the microcomputer
are made from stainless steel of 5 cm diameter, to perform several fonctions.

53
54 N. Yoswathana, G. Casamatta and K. Najim

- on-line acquisition of the flowrates of both feed being much lower than the conductivity of the
phases, of the concentration of solute at the outlet aqueous continuous one, the appearance of dispersed
of heavy aqueous phase and of the conductivity phase below the distributor results in a lowering
measured at the bottom of the column. of the conductivity.

- numeric command of both inlet flowrates and of It may be seen from fig. 4 that the conductivity
pulsing frequency decreases rapidly when the condition of operation
changes from emulsion region to cyclic local floo-
- programming of the controller algorithms. ding region. This transition can be obtained either
by increasing one of the feed flowrates, or by
The principal program used for controlling has been increasing the intensity of pulsation, the rate of
written in Pascal UCSD. It calls external assembly change depending on working conditions. As shown on
routines performing D/A and A/D conversions, and fig. 4, in this work the transition corresponds to
the clock reading for real-time control and on a value of the condictivi ty close to 0.4m Siemens/cm.
line-data sampling period.
During this hydrodynamic preliminary study, it has
Materials used : deminerized water,pure acetone been pointed out that the behavior of the column at
and commercial grade of toluene were employed. the transition point is not stable, very tiny
Before using, the equilibrium diagram of these changes in the operating conditions, e.g. the feed
three liquids was determined at 25 +- 0.5 0 C and flowrates or the pulsation intensity, leading to
the results agreed closely with those given in over-flooded or under-flooded operation. Even when
"recommended systems for liquid extraction studies trying to acheive flooding by means of very smooth
of European Federation of Chemical Engineering". changes of the pulsation intensity, it has always
revealed impossible to obtain stable operation at
Procedure : For all of our experiments, the ampli- the transition point.
tude of pulsation has been fixed at 1.5 cm and
both deminerized water and toluene were circulated OPTIMAL EFFICIENCY SCRUTINY
through the column until they were mutuallysatured.
Pure acetone was then added to the aqueous phase In this work, all of the experimental data have
in a storage tank until the concentration of ace- been obtained at fixed extraction factor L = 1.0
tone was 5% by weight. This was determined by means and a distribution coefficient m = 0.85.
of a manual refractometer. The column was then
filled with the aqueous phase and the frequency of For each operating point, mass transfer efficiency
pulsation was set to the required value. When may be evaluated by means of the concentration
the aqueous phase reached the interface level, the measurement in the aqueous effluent, whereas the
organic phase was fed into the column. The phases determination of the transition zone has been des-
flowrates were adjusted and the run continued until cribed in the former section. The effeciency of a
steady-state conditions had been obtained. pulsed column can be defined as the overall height
of a transfer unit based on continuous phase,
HYDRODYNAMIC BEHAVIOR AND FLOODING OF A (H.T.U.)R, or the ratio of solute extracted during
PULSED SIEVE PLATE COLUMN operation.

It was shown that the hydrodynamic behavior of a As indicated by fig. 5 and fig. 6, the percentage
pulsed column may be defined by the flow characte- of extracted solute increases very rapidly
ristics and the pulsation intensity (Smoot and Co- (respectively the (H.T.U.)R decreases) when
workers, 1959). Generally, as shown in fig. 2 the increasing the pulsation frequency and therefore
performance of pulsed column can be divided into as cyclic local flooding is approached. The simul-
five different regions (Sege and Woodfield, 1954). taneous measurement of the conducti vi ty has allowed
1 - flooding region owing to unsufficient pulsation to point out that for every feed flowrate the effi-
ciency is maximum at the transition zone. Further
2 - mixer-settler region increase in the degree of pulsation, leads to
cyclic local flooding and the efficiency of column
3 - emulsion region
decreases rapidly. From the results shown in
4 - cyclic local flooding region sections 2 and 3, it may be concll'd:ed that the
transition zone is the optimal one but the opera-
5 - flooding due to too large pulsation velocity
tion in this hydrodynamic condition is not stable.
It seems that the optimum condition of pulsed sieve Carrier (1981) has established that a classical PID
plate extraction columns can be obtained when the controller, as shown in fig. 7 below was able to
maintain the column within the optimal operating
column is operated at the transition zone between
emulsion region and cyclic local flooding region. range. In this work, the PID controller has been
replaced by a self-tuning PID controller
Flooding may cause the light-liquid be entrained
by the heavy-liquid effluent or vice versa. In the SELF-TUNING PID REGULATOR ALGORITHM
case where a control of the interface position is
To represent the column behavior about its nominal
exerced, flooding can only result in dispersed
operating point, the following single input-single
phase entrainment by the continuous phase. Carrier
(1981) shown that a liquid-liquid, fluidized bed output discrete linear model was aoolpted
like swarm of drops then spreads in the region A(Z-l) yet) = B(Z-l) u(t-1) (1)
setted between the dispersed phase distributor and with
-1
the continuous phase outlet (fig. 3). Coalescence 1 + a Z +
1
being intense, this swarm develops the formation of
-1
great diameter drops that travel back in the column, b + b Z
O 1
initiating transient flooding conditions.
In the Eq. (1), yet) is the output signal (the
Therefore, the transition zone can be detected,
conductivity of solution under solvent feed distri-
either by a scrutinity of the large and of irregular
butor) ; u(t) is the input signal (the frequency of
shape drops appearing in the column, or by the
pulsation) .
appearance of dispersed phase under the bottom
plate. A quantitative appreciation of these pheno-
The general structure of a discrete controller can
menum is afforded by the conductivity electrode
be written as follow:
the conductivity of the organic dispersed phase
 Control of a Liquid-Liquid Extractor 55

(2) Sege G. and Woodfield F. W. (1954). Pulse column

variables. Chemical Engineering Progress, 50,
where u(t), Yr(t) and y(t) represent the control 396 - 402.
signal, the reference signal and the system ouput
Smoot L. D., Mar B. W. and Babb A. L. (1959).
signal respectively. This may be described by the
Flooding characteristics and separation effi-
block diagram fig. 8.
ciencies of pulsed sieve-plate extraction
columns. Ind. Engng. Chem., 51, 1005 - 1010.
The different types of PID controllers depend on
the expression of the polynomials T(Z-l) and S(Z-l). Wittenmark B., (1979). Report of department of
The transfer function of the closed loop system automatic control, Lund Institute of Technology,
will be : October.
Ziegler J. G. and Nichols N. B. (1943). Trans. ASME,
65, 433 - 444.
(3)

NOMENCLATURE FOR FIG. 1

The PID controller parameters can be adjusted accor-
ding to the closed loop system characteristic equa-
1 - pulsed extraction column
tion as follow :
2 - sieve-plate
A(Z-l) R(Z-l) + Z-l B (Z-l) S(Z-l)~p(Z-l) (4)
3 - interface controller
4 - distributor
5 - conductivity sensor
where p(Z-l) represents the desired dynamics of the
6 - electromagnetic valve
closed loop system. The plant parameters are
7 - pulsing unit
recursively estimated using the recursive least
8 - continuous phase inlet
squares method.
9 - continuous phase outlet
10 - dispersed phase inlet
These estimate are used to solve the Eq. (4) and
11 - dispersed phase outlet
consequently to tune the PID parameters. The algo-
12-13 - rotameters
ri thm of the self-tuning PID controller is described
14-15 - electromagnetic instantaneous flowmeters
in the following steps :
16 - conductivity meter
Step 0 assign the initial value of the parameters 17 - refractometer
18-19 - continuous and dispersed phase pumps
Step 1 estimate the column parameters
20 - microcomputer
Step 2 solve the Eq. (4)
Step 3 calculate the control by Eq. (2) using the
NOMENCLATURE
parameters estimated in step 2.

EXPERIMENTS AND RESULTS extraction factor

m distribution coefficient
First of all, the pulsed column is manually brought (H.T.U. )R overall height of a transfer unit based
near the desired operating point as mentioned in on continuous phase, cm
section 1, the parameters of the second order concentration of solute in raffinate
linear discrete model are initially estimated. Then phase, wt% of acetone
the control of the pulsed column is switched to the Cr set point conductivity ms/cm
microcomputer. Many sampling periods (T) have been C(t) measured conductivity at time t, ms/cm
checked to obtain the desired response. In our case F(t) controller ouput at time t, Hz
the sampling petio.d was 10 seconds. Some of the re- Qc flowrate of continuous phase, l/h
suI ts are shown in fig. 9 and fig. 10 for differents Qd flowrate of dispersed phase, l/h
working conditions. It may be seen from fig. 9 and T sampling period, sec
fig. 10 that a simple single - input - single - output A(Z-l) polynomial in Z-l with aI, a2 as
self-tuning PID controller can be used to keep a parameters
B(Z-l) polynomial in Z-l with bO, b1 as
pulsed extraction column working at the transition
zone with a moderate disturbance on feed flowrates. parameters
p(Z-l) polynomial in Z-l, represents the desired
CONCLUSION dynamics of the closed system
R(Z-l) polynomial in Z-l
S(Z-l) , T(Z-l) polynomial in Z-l depend on the
This paper shows that a new idea of self-tuning PID
regulator works successfully in the field of liquid- types of PID controller
liquid extraction column. Despite the nonlinear and u(t) input signal (the frequency of pulsation)
time-varying behavior at the transition point, a at time t
self-tuning regulator is able to track the process y(t) output signal (the conductivity of solu-
nonlinearities and give reasonable conductivity tion under feed distributor) at time t
control when the deviations of the feed flowrates Yr(t) reference signal at time t
from the nominal operating point are small.
Further work will be studied by means of a new
method of recursive parameters estimation and con-
trol strategy.

RilFERENCES

Carrier J. (1981). These de Docteur-Ingenieur,

1. N. P. Toulouse.
European Federation of Chemical Engineering (1978).
Recommended systems for Liquid Extraction Studies.
Naj im K. and Nuratet G. (1983). Pratique de la
Regulation Numerique des Processus Industriels.
Nasson, Paris - 112 pp.

ACCP-E
 <.J1
O'l

r- - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- 8
.11
Qc Z
-<:
o
<A
~
I»
S-
I»
::I
I»

Cl
(')
I»
<A
I»
a
~
Si
'0-"
::I

r::
I z
I ~.
r-----..J
I 3'
~ 10

16
L ________________ _
I

./'...,....-- 9
6

FIG. 1 - DIAGRAMMATIC ARRANGEMENT OF APPARATUS

 Control of a Liquid-Liquid Extractor 57

Amplitude = constant
I
/
/
/

frequency, cycles/min

FIG. 2 - DIFFERENT TYPES OF PULSE EXTRACTION COLUMN

flUIDIZED
OD,in BED

CONDUCTIUITY

OD,out

FIG. 3 - ~~L FLUIDIZED BED OF LIQUID-LIQUID

0.8

0.7

0.6
E
0 0.5
"-
en
E
0.4
.c
0">
...,
0" 0.3
0
::s
'0
c0 0.2
0
0.1

0.0

0 5 10 15 20
T:i.r.ie, Inin
58 N. Yoswathana, G. Casamatta and K. Najim

Q)
s::
o
+'
Q)
()
tIl

60
'0
Q)
+'
()
tIl
,0><
><
Q)

Q)
55
+'
;0

'"o"
Ul
c,.;
o
~ 50
+'
s::Q)
()
015.0
><
Ql
0.
016.0
~ 45
0.5 lO 1.5 2.0
The in:oensity of pulsation, cm/sec

FIG. J - TH!': PERCENTAGE OF SOL'JTE EXTRACTED AS FU;,CTION OF THE

I'ITENSITY OF PULSATION FOR JI?FERENT FEED FLOWRATE Qc

Q (I/h)
c
• 10.0
80
/:::,. 14.0

o 15.0
70 0 16.0

60
E
()

~50

:c 40
0.5 lO 1.5 2.0
The intcn3ity of pulsation, cm/sec

FIG. 6 - (H. T. U')R;S FUNCTION OF THE INTENSITY OF PULSATION

FOR DIFFERENT FEED FLOWRATE Q

c
 Control of a Liquid-Liquid Extractor 59

Cl'

PID
CONTROLLER

c t)

FIG. 7 - A CLASSICAL PID CONTROLLER

yr(t) 1 u(t) 6(Z-I) y(t)

:~
T{Z -I) Z-I
R(Z-l ) A(Z-I)

5(Z -1)

FIG. 8 - BLOCK DIAGRAM OF A DISCRETE CONTROLLER

60 N. Yoswathana, G. Casamatta and K. Najim

I I T
0.5 r- - 0.5
T
...!:LO.4
If)
~- - . .,; 0.4

lo.3 f- - 0.3

--
U
.j..l 0.2

0.1
r-

r-
-
-
0.2

0.1

0.0 I I I 0.0

I I I 1 1 ,
2.0 r- - 2.0 r- -
1.8 r- - 1.8 r- -
N
~1.6 ~ -::: 1.6 I- -
-::;-- 1.4 I- - 1.4 ~ -
LL
1.2 r- - 1.2 f- -
I I I 1.0 I I I
1.0

I I I I I I
2.0 I- - 2.0 I- -
.--.. 1.8
~
I- - 1.8 I- -
..2-
X
0: 1.6 I- - 1.6 I- -
1.4 I- - 1.4 I- -
1.2 I- - 1.2 I- -
1.0 I I I 1.0 I I I

10 20 30 min 10 20 30 min
?::::c. 9 .- THE SIGNAL Ote CG;lJUCTIVITY, FIG. 10 - THE SIGNAL OF CONDUCTIVITY,
FRE;YiJsNGY OF PULSATION AND FTIEQUENCY OF PULSATION AND
C::;NCE:~Tr.A'i'iON OF ACET01~;;; IN CONCENTRATION OF ACETONE IN
~!5FINA'I'E AT ;; = ;J):::10% l/h, RAFFINATE AT Q =12~10% l/h,
Qd=l1.e:':lO~~ Ilg +
Qd=14.LlO% 1/ h