BAYER PROCESS CONTROL AT ALCAN VAUDREUIL WORKS - 2007 - IFAC Proceedings Volumes
BAYER PROCESS CONTROL AT ALCAN VAUDREUIL WORKS - 2007 - IFAC Proceedings Volumes
BAYER PROCESS CONTROL AT ALCAN VAUDREUIL WORKS - 2007 - IFAC Proceedings Volumes
Abstract: Alcan Vaudreuil Works is the only alumina plant in Canada. The plant
produces about 1.4 milion tons of alumina per year, using the Bayer process to treat
bauxite ore. In addition to regulatory control, this complex extraction process involves
many applications of advanced process control. Techniques such as adaptative control,
multivariable control (model based predictive control), fuzzy logic, on-line models and
expert systems are used. This paper will first introduce the Bayer process and present the
different types of control commonly used. Three examples of closed loop advanced
process control applications will be presented. Copyright © 2007 IFAC
INFRARED
SIGNAL
LAB. ANALYSIS SOFT SENSORS
PROCESS
In this plant area, the red mud residue is washed prior MANUAL INPUTS MODEL NaOH
A/C Ratio
DISTURBANCES
to disposal in a multi-stage counter current circuit . FEEDFORW ARD
SolidS
MODEL
(NaOH & A/C ratio) LAB
SOFT SENSOR
2.5 Precipitation
Precipitation is the most complex area of the plant Fig. 1. Bauxite Charge Control
and the most advanced one in process control
application. Two reasons motivated the rapid
development of process control in this area. First, 3.2 Vacuum Heat Exchange Optimization
precipitation process is chemically complex. Second,
the Vaudreuil plant is relatively old (about 70 years This process is one of the two major heat recovery
old) and uses a batch process which requires a lot of systems in the Vaudreuil plant Bayer process. The
manual operations. Many advanced control purpose of the vacuum heat exchange network is to
techniques are used in this area such as: recover energy from a hot stream (pregnant liquor) by
transferring it to a cold stream (spent liquor). Figure
- Batch operation scheduling and management; 2 shows a simplified graphical representation of the
- Virtual sensors; heat exchange network system. A particular feature
- Volume monitoring and control; of this area is the use of 3 parallel lines of heat
- Constraints monitoring and production set point; exchangers fed by a common liquor cold feed divided
- Multivariable control; in two surge tanks and collected in 2 heated stream
- On-line mass balances; surge tanks. Similarly, the heat supplying streams are
- Discreet control; fed by liquor coming from a common surge tank and
- On-line diagnosis; the cooled steams are collected in 3 surge tanks.
- Model-based control;
- Sequence control.
HEAT EXCHANGE
NETWORK
F T
CONTROLLED
VARIABLES
MANIPULATED
volume control algorithm, based on wash flow rate,
Hot
10
11
VARIABLES
COOLIN L
has to be respected.
supply
stream
12
G TANK
tank L #1
TC T F MUD CIRCUIT FLOC FLOC
DEBIT
E E E USINE
X X X COLD
C C C STREA
Heate L
H H H SUPPL
M L Decanters
strea
d T
YANK U/F
collectin
m
#1
gtank T F T Drain
#1 EB2
EB1
16 COOLIN
G TANK L
17
18 ON ON
#2
F Filtration area
TC T E E E
X X X 7
C C C
H H H
Water
F T
24
COOLIN FLOC
FLOC FLOC
23
22
G TANK L
Heate #3 Washing
L Water
strea
d
collectin
m TC T F
COLD
gtank E E E
T STREA
#2 X X X L
C C C SUPPL
M
T
YANK EB3 CLARTE
EB4
H H H 25% EB5
#2 water
ON disposal
LIMITATION
- Long delays and reaction time of the circuit;
ON FILTRATE
FLOW S - Frequent operator changes;
LIMITATION
ON SPENT
- Interaction between stages;
LIQUOR
FLOW S
PROCESS BALANCE
ACCOUNT
FOR W ASH
- Mud surges, muddy O/F;
THE FLOW
COMING FROM
THE VOLUME
BETW EEN THE
3 FILLING
W ATER
TRANSFERT
- Instability in the amount of mud leaving the circuit;
THE FILTRATION
PROCESS TANKS
TO FILTRATE
- Unoptimized additive dosages;
Fig. 3. Control Objectives and Constraints -Difficulties to diagnose equipment break down
(tanks, probes, rake, etc.).
The selected technique for this system is a Model It was found that the control technique that best
Based Predictive Control algorithm, commonly called addresses these problems is the Rule Based Control
MPC or MBPC. One of the commercial package (RBC) algorithm which is mainly based on human
available on the market is used for the core of this reasoning. This system relies on a set of rules that
application. It is also supported by custom logic and define problems, causes and actions.
on-line model. The described application has 14
control variables, 9 manipulated variables and 2 The rules are based on combinations of conditions
feedforward variables. looking at variables absolute level and trending. It is
composed of approximately 100 rules for each washer
(thickener).
3.3 Mud-washnig Circuit Optimization
In order to keep the reasoning free of numbers, all
As described in section 2.4, mud is washed in a input variables are converted into fuzzy evaluation
counter-current washing circuit involving five using fuzzy sets. These sets define what is very low,
thickeners equipped with mud level detectors and low, normal, high or very high using instantaneous
clarity probes. Rake torque, underflow rate and values and averages. It also includes slopes that
settling additive flow rate are also measured. define if trending goes up and down slowly, normally,
rapidly or if it is stable. Similarly, manipulated
The major objectives of the system are to maximize variables are generated in fuzzy evaluation and then
the recovery of the process chemicals, to minimize transformed into numbers through fuzzy sets.
the additive consumption and to stabilize the mud
flow leaving the area. At the same time, two In the design of this type of application, the
conditions also have to be met: all the mud generated determination of delays before repeating actions and
by the digestion step has to be processed and the plant the priority between problems are important. These
are key aspects of the design to consider in order to
avoid over-reacting on manipulated variables.
4. CONCLUSION
REFERENCES