You are on page 1of 7

Aluminum Smelting

Overview

The Development of Aluminum


Reduction Cell Process Control
Geoffrey P. Bearne

It was primarily the improved under- and other additives must be con- are some routine tasks, such as anode
standing and control of magnetic fields, trolled to maintain the bath proper- setting, metal tapping, and a few mea-
assisted more recently by the use of computer ties and keep the bath liquidus tem- surements on each cell, but manual in-
modeling, that made modern aluminum perature constant. tervention would not be expected unless
reduction cells more stable and has enabled • Heat balance. This balance must a cell was identified as having a special
them to grow larger. However, the im- be controlled to maintain constant need. It is, therefore, a key function of
plementation of improved process control bath temperature and superheat. the process-control system to flag excep-
systems has not only underpinned the success • The anode cathode distance (the tion cells. Figure 1 is an example of a live
of new cell designs, but has also enabled distance between the anodes and exception screen from the New Zealand
many older technologies to remain the metal pad), which is maintained Aluminium Smelter Ltd. The black
competitive. While there is still much room roughly constant to maintain cell squares represent cells that the process
for improving reduction-cell control systems, performance. It is important not to computer has identified as needing at-
this task is increasingly difficult due to the go below a minimum value. tention. Cell 329, for example, has re-
diminishing performance gains available. • Liquid-bath and liquid-metal vol- cently had multiple anode effects, indi-
umes. These should be kept con- cating an alumina feeding problem.
INTRODUCTION
stant to prevent variation in such
IMPROVEMENT THROUGH
The control of aluminum reduction factors as alumina concentration,
PROCESS CONTROL
lines 40 years ago, before the application heat balance, etc.
of computers, was simple; experienced In addition to these issues, it is also criti- Improvement through process con-
operators were revered for their ability cal to maintain an even electrical current trol requires three factors.
to treat cells from observation of the distribution in the anodes. • A change to the control system to
flame. Even in the 1970s, aluminum The complexity of cell control is in- reduce the process variation. This
smelting was widely acclaimed as an art. creased dramatically by the fact that could be a major change, such as
Today, most of the secrets of the process hundreds of individual cells must be conversion to alumina point feed-
appear to have been uncovered, but managed collectively. Individual cell ing, or simply a control algorithm
applying the wealth of new knowledge control is, therefore, integrated into a improvement.
to improve process control has been system to manage the whole line. This • A change in operating target to a
difficult. To improve a process, one has system should automatically manage the more beneficial point (e.g., reduc-
to measure it; however, the highly cells, flag those that need manual inter- ing the cell power input, resulting
corrosive and distributed nature of the vention and identify the problem, and in lower bath superheat, facilitated
reduction process has prevented the identify and manage the changes that by the smaller mass of alumina fed
application of any new sensors for are common to all cells. Although con- into the bath).
routine control. Instead, what has trol systems are quite good at the first • Maintenance of the process at the
changed is the way that existing sensors requirement, there has been little devel- new target despite disturbances
have been used to improve the control. opment in problem diagnosis or auto- such as alumina quality problems.
For a given reduction cell technology, mated reduction-line management. Sometimes the performance improve-
there is a set of process parameters that A key feature of the reduction process ment from the first step will be sufficient
provide optimal performance. The ob- is that it is controlled by exception. There to financially justify the change if there is
jective of process control is to a substantial reduction in varia-
keep the cell at its targets at all tion. Generally, however, the
times or to return the cell to tar- improvement comes from the
get following a disturbance. In target change. It is risky to change
practice, it is very difficult to keep the operating point without first
even one of the parameters on improving the control. For ex-
target, so the real objective is to ample, lowering the bath tem-
minimize the variation. Specifi- perature without first improv-
cally, operators wish to control ing the heat-balance control may
the following parameters: lead to operational problems
• Bath chemistry (i.e., the con- with very cold cells and negate
stituents of the bath). Alu- the benefit of the target change.
mina is a special case be-
BARRIERS TO
cause its concentration
EFFECTIVE PROCESS
changes quickly and the im-
CONTROL
plications of poor control are
large—anode effects or There are a number of barriers
sludge formation. Excess to the effective control of the re-
aluminum fluoride (AlF3) Figure 1. A typical live exception screen display. duction process. By nature, the

16 JOM • May 1999


process is difficult to observe; the liquid current-efficiency im-
metal and bath are very aggressive and provement of 0.3% to
will attack immersed sensors. The pro- payback in four years.
cess is highly distributed, so equipment Another example is use
has to be moved large distances or linked of a consumable sensor
by expensive wiring. Measurements can costing $10 and used ev-
typically only be made at the ends of the ery day per cell must cre-
cell and may not be representative of the ate a voltage saving of
whole. This is particularly the case if 130 mV per cell or a cur-
there has been a disturbance (e.g., a cold rent efficiency improve-
Breakeven OPEX $1,000 $2,000 $3,000 $4,000 $5,000 $6,000
anode set close to the measurement ment of 0.9% to recoup ($US/cell/year)
point). The cells vary independently so the cost of the sensor CAPEX for 4 year $3,500 $7,000 $10,500 $14,000 $17,500 $21,000
each requires its own sensors, control alone. Note that modern payback ($US/cell)

hardware, and individual attention. cells typically operate at Figure 2. Performance improvement/cost relationships for a
Also, as the production intensity is low 90–96% current effi- medium amperage cell installation.
(typically only 0.5–2.5 tonnes of metal ciency and 4.2–4.8 V.
per day), there is limited scope for in- In reality, analyzing the benefit is more uted control equipment, frequent manual
vestment per cell. complex for a number of reasons. measurements/actions, or costly con-
This last point can be demonstrated • Most control equipment will have sumable sensors will be difficult to jus-
by analyzing the financial benefit of pro- capital and operating costs. tify unless the resultant performance
cess improvements. This is best done by • Some control improvements can improvement is high. In contrast, the
cell as most process-control changes will reduce operating costs through return on investment for a control algo-
require either a device mounted on each elimination of some measurements, rithm that creates a modest improve-
cell or additional actions on each cell. reduced staffing, etc. ment for negligible capital or operating
Figure 2 gives an example of relation- • There may be other benefits, such as cost will be very high.
ships derived for a medium amperage improved plant security or reduced
PROCESS-CONTROL
technology (100–200 kA) to relate pro- environmental impact.
MEASUREMENTS
cess improvement to allowable expen- • The performance benefits are diffi-
diture. The x axis has two scales: the cult to assess without long trials. Measurements In Widespread Use
breakeven operating expenditure (the An example of this type of analysis Table I is a list of the control measure-
additional operating expenditure was given by Bonnardel and Homsi1 in a ments used in the industry. The typical
[OPEX] per cell that will equal the ben- paper describing the development of an measurement frequency is also given.
efit from the performance improvement) automatic bath height and bath tem- The only universal automatic mea-
and capital expenditure (CAPEX) for a perature measurement device. Use of surements are cell voltage and line cur-
four-year payback (the capital expendi- this device on high amperage and cell rent. Visual inspection has become ex-
ture per cell that will give a four-year technology was demonstrated to give a ceptional rather than routine for mod-
payback for a particular performance number of benefits, including a 0.64% ern technologies, although it is still con-
improvement). increase in current efficiency. A two- sidered crucial at many plants. Bath and
The application of these relationships year return on investment was predicted metal depth are typically measured
can be demonstrated with an example. for the application of this technology in manually at a frequency determined by
Installation of new equipment on each a new reduction plant. the work cycle (i.e., especially before
cell (e.g., an individual microprocessor Improved control generally requires metal tapping). An automatic bath-depth
controller) costing $4,000, must create a more or better measurements; however, sensor has been patented2 but is not in
voltage saving of 40 mV per cell or a it is evident that sophisticated distrib- general use. The determination of metal
depth from anode position (e.g., from
Table I. Types and Typical Frequencies of Control Measurements the anode-changing crane) or from the
Measurement Frequency anode-beam position3 are utilized in
many plants. The weight of the tapped
Widespread Use
Cell Voltage Continuous (automatic)
metal per cell is commonly determined
Line Current Continuous (automatic) from weight scales on the crane or tap-
Observation (Flame, Crust, Etc.) Several times/day to daily ping vehicle; in some plants, this infor-
Bath Depth Daily to once in two days mation is transmitted automatically to
Metal Depth Daily to once in two days the process computer.
Tapped Metal Weight Daily to once in two days Bath temperature and chemistry mea-
Bath Temperature Daily to weekly surement frequencies vary greatly from
Bath Chemistry (AlF3, LiF, CaF2, MagF2) Twice per week to two weekly plant to plant. This is not surprising as
Metal Purity (Fe, Si) Daily to weekly there are numerous variations of cell
Cathode Voltage Drop Weekly to annually
Alumina-Feed Dump Weight Weekly to occasionally
heat balance control strategy in use in
Anode-Beam Position Continuous (automatic) on some the industry. Bath temperatures are in-
technologies variably measured manually at one end
Individual Anode Currents (by Rod Voltage Drop) Daily to weekly or on exception of the cell using thermocouples that last
Continuously on one technology for several hundred measurements. A
Limited Use superstructure-mounted intermittent
Crust Breaker Position Continuous on point feeders bath temperature measurement probe
Aluminum-Fluoride Feed Dump Weight On exception has recently been developed to permit
Sludge/Ridge Depth Daily (with metal height) more frequent measurements without
Sidewall Ledge Position5 Twice/week
Direct Alumina Concentration Daily
manual intervention;1 however, continu-
Superheat Daily ous bath temperature measurement is a
Sidewall Temperature Continuously long way off, with thermocouple lives
Cathode (Collector Bar) Temperature Continuously greater than a few weeks seeming un-
Sidewall Heat Flux Continuously likely.4 Laboratory-based bath chemis-

1999 May • JOM 17


that at least one smelter (bar and point-feed break cells), AlF3
has tried routine alu- additions (generally only point-feed
mina-concentration cells) and bath additions (only cells
analysis for control pur- equipped with crushed bath feeders).
poses. 7 However, be- The few reported variations from this
cause of the time delay include the control of individual anodes
and cost, the absolute on the Alcoa P-225 cell10 and automatic,
measurement of alumina compressed air suppression of anode
concentration is gener- effects on Söderberg lines.11
ally only used for refer- Hence, there are significant limitations
ence or trial purposes. to the application of additional informa-
Meters for in-situ alu- tion for improved control decisions. The
Figure 3. Applications of pseudo-resistance for cell control. mina concentration have heat input to the cell can be varied auto-
also been developed and matically by changing the current (all
try analysis (generally by x-ray diffrac- tested,8 but the process improvement cells) or the cell resistance (by altering
tion) of ground bath samples, taken, at obtained does not appear to justify the the beam position). The alumina feed
most, every two days, appears to be the labor and maintenance costs. rate can be altered, the quantity of alu-
widespread industry practice. To date, A commercial superheat probe capable minum fluoride added can be adjusted,
no commercial in-situ sensors for bath of simultaneous bath temperature and and anode effects can generally be sup-
chemistry have been developed. liquidus temperature determination to pressed automatically. Reports can be
The use of sensors (in particular, lin- within ±1°C is available.9 This device generated to instruct operators to tap
ear transducers) to indicate anode-beam requires a manual measurement be taken more or less metal or to adjust bath depth,
position is becoming more widespread at the cell using a consumable probe; and alarms can be issued to indicate that
due to the availability of reliable, low- hence, the cost of this measurement is the process requires manual interven-
cost devices. It is not clear whether the relatively high. While useful for research tion. The option of individual anode
investment is fully justified by perfor- purposes, it has yet to prove viable for positional control has not been embraced
mance improvement, but accurate indi- routine control purposes. by the industry because the additional
cation of beam position is useful for metal Other attempts at developing mea- cost does not appear to be justified.
height measurement, tapping, beam rais- surements for control purposes have also
PROCESS-CONTROL
ing, anode-effect termination, and bath- been directed at improving cell-heat bal-
DEVELOPMENTS
resistivity determination. ance. Known to have been investigated
The determination of individual an- are sidewall temperature, sidewall heat There has been much evolution and
ode currents by measurement of the volt- flux, and cathode (or collector bar) tem- improvement in reduction-cell process
age drop over a short distance of anode perature. The first two were unsuccess- control.12 The key control functions are
rod or stem has been a standard practice ful as it was determined that changes in alumina feed, noise, and heat-balance
on prebake anode cells for many years. sidewall temperature and heat flux ap- control. The calculation and use of
Due to the manually intensive nature of preciably lagged changes in bath tem- pseudo-resistance is also an essential part
the work, it is mainly limited to a quality perature. A reliable measurement of cath- of the control.
assurance (QA) tool or for problem diag- ode temperature is very difficult as the Originally, reduction cells were main-
nosis on exception cells. Some technolo- thermocouples are rapidly attacked by tained at a target voltage by manual
gies have been equipped with probes sodium and bath; however, thermo- adjustment of the position of individual
built into their superstructures to allow couples installed inside the collector bars anodes or the anode beam. Anode-beam
continuous anode current measurement. are believed to have survived for years. adjustment was later mechanized and
However, it would appear that the addi- It is not known if these temperatures subsequently automated to provide con-
tional information from this source, al- have been used for control purposes. stant voltage control. Following the in-
though useful in identifying problem troduction of computers in the mid-
REDUCTION PROCESS
anodes, metal-pad instability, and the 1960s, resistance control (now called
CONTROLS
onset of anode effects, has not been suf- pseudo-resistance control) was devel-
ficient to justify the cost and complexity Utilizing the measurements described oped. This was necessary as it was not
of installation and use. Hence, its appli- above for effective control is difficult. practical, or necessarily desirable, to keep
cation remains very limited. Table II lists the control actions used. It the line current constant at all times. The
can be seen that the level of automation pseudo-resistance is calculated from
Measurements in Limited Use
is very low. On an individual cell, auto-
R = 1,000 (U – Vext)/I
Table I also provides a list of control matic control is limited to anode-beam
measurements that have been tested in position (all cells), alumina feed volume where R is pseudo-resistance (µΩ), U is
trials or are in limited use in the indus-
try. Sensors have been developed for Table II. Actions Used to Control Reduction Cells
point-feeder crust breakers to indicate Control Usual Control Mode
the position of the breaker shaft when Intensity of Electrical Current (Line Only) Auto or manual
lowered. This is to identify instances Anode Beam or Söderberg Anode Position Auto
when the breaker fails to penetrate the Individual Anode Position (Prebake Anode Cells) Manual (auto on one technology)
crust, leading to alumina feeding prob- Alumina Feed Volume/Mass Auto or manual
lems and anode effects. Commercially AlF3 (and Other Modifier Additions) Volume/Mass Auto or manual
available options use either a full stroke Bath Removal (Tapping) and Addition Manual removal
of the pneumatic cylinder as an indica- Auto or manual addition
tor of success or the detection of an elec- Metal Removal (Tapping) and Addition Manual
Addition or Removal of Anode/Crust Covering Material Manual
trical circuit when the breaker contacts Adjustment of Composition of Anode/Crust Manual
liquid bath.6 Covering Material (Bath/Alumina Blend)
Equipment for the accurate labora- Skimming Carbon/Removing Crust/Sludge, Etc. Manual
tory analysis of alumina concentration Anode-Effect Suppression (Green Pole/Compressed Air) Manual, limited automatic
in the bath is available, and it is thought air suppression

18 JOM • May 1999


point crustbreakers and feeders. Bar- concentration corresponding to the mini-
breaker cells have feeders that typically mum resistance.19 Industrial measure-
deliver 10–30 kg of alumina. In practice, ments indicate that the minimum occurs
the alumina is fed to reform the center between 3.5 wt.% and 5.5 wt.%. It is
channel crust, so the quantity dumped accepted, however, that the alumina con-
at each break and feed may be as much centration corresponding to the mini-
as 100 kg. Hence, there may be a delay of mum resistance does not vary with ACD,
more than a hour between breaks. temperature, or excess aluminum fluo-
It is believed that the Alcoa P-155 cell ride content of the bath.
at Rockdale (1961) was the first full re- Around the time of Welch’s publica-
duction line implementation of point- tion, several companies were develop-
feed technology.14 Point-fed cells have ing practical ways of utilizing the phe-
Figure 4. A pseudo-resistance versus alu- feeders that typically deliver 1–3 kg of nomenon for alumina-concentration con-
mina concentration relationship. alumina. Two main design variations trol. U.S. patents filed in 1968 indicate
exist: an integral breaker/feeder, where the successful development of resis-
the cell voltage (V), Vext is the voltage the alumina feeds as the crust breaker is tance-based underfeed/overfeed20 and
extrapolated to zero current (typically operating, and an independent breaker demand feed.21 Because the change in
taken as 1.65 V), and I is the line current and feeder, where the intent is to deliver pseudo-resistance with alumina concen-
(kA). Although not technically correct, the alumina into the open hole left by the tration is relatively small, it is necessary
Vext is often referred to as the back EMF crust breaker. The latter method appears to operate on the left-hand side of the
of the cell. A detailed explanation of this preferable from an alumina dissolution minimum (i.e., at low alumina concen-
issue is given by Haupin.13 perspective,15 however, little informa- tration) or at least to return there peri-
It is fortuitous that the cell pseudo- tion is available on the impact of feeder odically to obtain a reference. Fortu-
resistance can be readily determined type on operational performance. De- nately, this appears to be the region of
automatically because it provides key velopment of the point feeder made it highest current efficiency.
information for cell-control purposes. possible to dissolve most of the alumina Today, there have been refinements to
Typical control applications of pseudo- directly upon addition but it was still the alumina concentration control strat-
resistance are shown in Figure 3. necessary to rely on the incidence of egies, but the principles have not really
Anode effects are generally detected anode effects to return to a known alu- changed in 30 years. For bar-breaker
from the cell voltage (typically when the mina concentration. cells, demand feed, a commonly used
voltage has exceeded 8 V for a few sec- Around the same time that alumina- strategy, has been very successful. For
onds), at which time anode-effect sup- feed control was being automated, it point-fed cells, there appear to be two
pression is initiated. The line current was observed that the anode polariza- main variations: underfeed/overfeed
and voltage are sampled at frequencies tion and, hence, cell pseudo-resistance and three-phase feed strategy.
between 0.03 Hz and 20 Hz, depending increased with decreasing alumina con-
Demand Feed
upon the technology, to produce an centration.16 Alumina concentration con-
unsmoothed or raw pseudo-resistance trol strategies were then developed that Demand feed works on the principle
for control purposes. Raw values are used this change in pseudo-resistance. that if the scheduled break and feed is
used for tapping control (where rapid The discovery has been applied with stopped, the resulting reduction in alu-
anode beam moves are required to main- success to most feeding technologies. In mina concentration will cause an increase
tain anode-cathode distance as the metal some side-break plants, the pseudo-re- in pseudo-resistance, which can be de-
height decreases) and for noise control. sistance increase with reducing alumina tected and used to trigger the next feed—
Filtering is used to produce a smooth concentration is used to alert the opera- hence, feeding on demand. The control
pseudo-resistance for resistance control tor, who manually breaks and feeds the strategy may use a critical change in the
purposes. Resistance control refers to cell.17 Clearly, however, it is the tech- pseudo-resistance (often referred to as
selecting an appropriate pseudo-resis- nologies with mechanized feeders that delta R) to trigger the change in feed rate,
tance target for the cell and maintaining have been able to utilize resistance track- a critical resistance slope (dR/dt), or
it via adjustments of the anode-beam ing to the best effect.
position. Only minor developments have Welch18 identified in 1965
been made to the way that the resistance that “for each operating
is measured and adjusted, generally to current density there ex-
improve the sensitivity of the measure- ists a discrete alumina
ment (through more frequent sampling) concentration that de-
and the accuracy of the anode-beam fines a minimum reduc-
movements. What has changed and dif- tion cell voltage” and
fers between technologies is how the published theoretical
target pseudo-resistance is modified for voltage/alumina con-
control purposes. centration curves. These
Further smoothing of the pseudo-re- curves, now generally
sistance signal may also occur, and used in pseudo-resis-
pseudo-resistance slope or “rate of tance form, are the basis
change of resistance” calculations (dR/ for modern alumina-
dt) are also made. These are used indi- feed strategies. Figure 4
vidually or together in various technolo- shows the typical rela-
gies for alumina-feed control. tionship for pseudo-re-
sistance versus alumina
Alumina-Feed Control
concentration at constant
Alumina-feed control was mecha- anode-cathode distance.
nized and automated in the early 1960s It should be noted that
by the introduction of break and feed in differences of opinion
the center of the cell using either bar or exist as to the alumina Figure 5. Underfeed/overfeed strategy characteristics.

1999 May • JOM 19


both. This strategy can then be used to pseudo-resistance to determine when to tance may provide a voltage saving (per-
either periodically track the cell to a change the feed rate. When fed at the haps up to 50 mV for a typical point-
known alumina concentration after a base feed rate, the pseudo-resistance of feeder technology). However, caution is
number of scheduled feeds or to trigger the cell will remain virtually constant, required as it will result in a less respon-
all feeds so that no alumina is added and anode-beam adjustments are made sive control. If the alumina-concentra-
other than by demand. A more detailed at this time. The pseudo-resistance will tion variation were to increase as a re-
description of an early strategy is given then rise during underfeed (if the alu- sult, this could nullify the benefit. It is
by Wilson and Tabereaux.22 mina concentration is in the desired suspected that modern point-feed strat-
Demand-feed strategies are generally range), and overfeed will be triggered egies have developed to a stage where
used in conjunction with scheduled an- once a critical resistance change and/or there is more to be gained from improv-
ode effects (routine stopping of feeding slope have been reached. A detailed ex- ing the ability to recover quickly from
until an anode effect occurs). This is planation of this strategy is given by disturbances than from refinements to
because there are control-reliability is- Robilliard and Rolofs.24 the basic strategy. Identifying and rank-
sues that can result in overfeeding and ing the sources of alumina-concentra-
Future Directions
excess sludge formation. Bar-breaker tion variation is a valuable exercise. The
cells, therefore, typically have average Industrial results indicate that cells major contributors are unlikely to be the
anode-effect frequencies of 0.5–1.5 an- using either of the point-feed strategies control strategy.
ode effects per day. Refinement of the described above are capable of current As the alumina concentration can drift
strategy may permit a dramatic reduc- efficiencies above 95% and specific en- out the desirable control range in min-
tion in frequency. Individual trial cells ergy consumptions of 13 DC kWh/kg. utes, an alternative alumina concen-
with bar-breaker technology have been Certainly, alumina-concentration varia- tration sensor will have to be cell
operated for long periods with anode- tion of less than 1% would be expected mounted and automated to be used for
effect frequencies below 0.1 anode ef- for stable cells. It is believed that most control purposes. Sensors have been tried
fects per day, indicating that this is prob- point-feed control strategies are varia- experimentally, but are not readily
ably ultimately achievable for entire re- tions of these two options. Probably the automated and are unlikely to meet the
duction lines. main difference between implementa- financial justification requirements. For
tions relates to the feed rates used for this reason, it is probable that the pseu-
Underfeed/Overfeed
underfeed. Very stable cells with reli- do-resistance change with alumina
This refers to a type of point-feed con- able feeding are able to underfeed closer concentration will continue to be used
trol strategy that sequentially adds alu- to the consumption rate (e.g., at 75%) as for control for many years, and therefore,
mina at a rate lower than consumption, the small change in pseudo-resistance the concentration will be required to
then higher than consumption and uses can be detected effectively. Less stable oscillate about the target. The cost of this
the change in pseudo-resistance slope cells are likely to be underfed at a lower oscillation, in voltage terms, is small,
(dR/dt) to determine when to change rate (typically 50% of consumption or contributing perhaps 20–40 mV that
the feed rate. To prevent anode moves below) so that the change is faster and, could potentially be saved.
from interfering with determination of hence, more easily detected. The main constraint to reducing the
the alumina-related pseudo-resistance As stated, the objectives must be to bath temperature appears to be the
changes, the control deadbands are set reduce the variation and then change to freezing of sludge on the cathode sur-
outside of the normal range of variation. a more beneficial operating point. Clearly face. This causes an imbalance in the
Hence, anode-position adjustments are the industry would like to operate at the electrical current distribution and, ul-
limited to the times when the feed rate alumina concentration for maximum timately, metal-pad instability (the bath
changes. Figure 5 shows a simple under- productivity and lower bath tempera- component of sludge invariably has a
feed/overfeed strategy; a detailed ex- ture as a result of lower bath-liquidus lower excess AlF3 concentration than the
planation of this type is given in an temperature and/or superheat. bulk bath and, thus, has a higher freezing
Aluminium Pechiney patent.23 The optimal alumina concentration point).25 Therefore, it is necessary to
has yet to be determined, as noted by substantially reduce or eliminate sludge
Three-Phase Feed Strategy
Kvande et al.,19 probably because it is formation in order to significantly reduce
An alternative for point-feed cells is governed by many factors, including cell the bath temperature. This may in the
the three-phase feed strategy, which technology, control strategy, and work first instance require conversion to point
cycles the feed rate from base or con- practices. Shifting the operating point to feed, but even with point-feed operation,
sumption feed rate to underfeed and the alumina concentration that corre- sludge may be formed by
then overfeed, using the change in sponds to the minimum pseudo-resis- • Feeding at too fast a rate or too large
Table III. A Comparison of Four Heat-Balance Control Strategies
Control Strategy A–Entner26 B–Peyneau27 C–Madsen28 D–Desclaux29
Measures Bath temperature Bath temperature Bath temperature Bath temperature
Excess AlF3 Excess AlF3 Excess AlF3 Cell age
AlF3 feed rate Cell age
Cell voltage Fluorine content of
reacted alumina
Cathode voltage drop
Basis Equations whose coefficients AlF3 addition determined Resistance target change based AlF3 additions based
are calculated with mainly by latest excess AlF3 on latest bath temperature on deviation of
regression models and latest bath temperature (to maximum of 360 mV) bath temperature
from measured data Resistance target based on AlF3 addition rate modified from target (two-day
excess AlF3 concentration based on excess AlF3 weighted average).
and cathode voltage drop analysis and temperature
Variables Cell resistance target Cell resistance target Cell resistance target AlF3 additions
AlF3/Na2CO3 additions AlF3 additions AlF3/Na2CO3 additions
Performance Excess AlF3 sd. 1.1% Excess AlF3 sd. 1.0% Excess AlF3 sd. 1.6% Excess AlF3 sd. 2.0%
Temperature sd. 5.5°C Temperature sd. 5°C30 Temperature sd. 9°C Temperature sd. 10°C
sd—standard deviation.

20 JOM • May 1999


to rapid changes in anode current are established, pattern-recognition soft-
density and localized disturbances ware could be used to identify the wave
in the metal pad from the formation forms. It is believed that some compa-
and release of bubbles in the ACD. nies have progressed wave recognition.
• Unstable cells display (in addition Most causes of cell instability will re-
to the bubble noise) another noise quire operator intervention at the cell, so
component with a lower frequency. full automation of noise control is not
The shape and frequency of this possible. But, with the drive toward re-
component will vary depending duced staff, there will be significant ben-
upon the cause of the instability and efit in improved diagnosis of causes of
Figure 6. Noise signals from stable and un- the characteristics of the cell. instability. Instability information, with
stable cells. Most, if not all, process-control sys- alumina feed and process event data,
tems will identify and respond to the could be analyzed by an expert system
a quantity to dissolve directly on signal noise of a lower frequency. The for guidance.
addition. sophistication of these approaches dif-
Heat-Balance Control
• Feeding when not required due to fers markedly between smelters. In their
poor reliability of the control strat- simplest form, noise-control strategies Heat-balance control is another area
egy (e.g., during instability). detect the amplitude of the noise, and if where there has been significant devel-
• Blockage and subsequent clearing this instability remains for a reasonable opment; it is probably fair to say that
of feeder holes, resulting in large time (e.g., ten minutes) will increase the every company has its own method, and
doses of alumina. cell resistance by raising the anode beam. some companies have many different
• Uncontrolled feeding from other This action will be repeated a few times methods. The objective of heat-balance
sources, such as anode setting, an- if the noise remains, and if unsuccessful control is to maintain the bath chemistry
ode-effect treatment, or crust col- in stabilizing the cell, an alarm will be (generally ignoring the alumina concen-
lapse. raised. It is, however, well known that tration) and the bath temperature con-
All four points are serious constraints different frequencies of noise indicate stant and on target, hence, maintaining a
to lowering the bath temperature. They different physical problems in the cell. A constant superheat. In its simplest form,
are also constraints to reducing bath su- component with a frequency of 0.05– this involves adding aluminum fluoride
perheat as the energy available to heat 0.001 Hz is due to a standing wave in the to compensate for losses, the largest of
and dissolve the alumina will be less. metal-bath interface (often referred to as which is usually the reaction with so-
The use of smaller dump weights or a metal roll); whereas, a component with a dium oxide in the alumina to form bath.
continuous feeder should eliminate point frequency above 0.2 Hz is usually attrib- Pseudo-resistance changes are also made
one even at lower superheats and in uted to an anode problem, such as a to compensate for changes in cell heat
baths with low alumina solubility, but spike or low-set anode. These observa- balance, especially when the cells are
will have no impact on the other causes tions have led to control routines that considered to be cold.
of sludge. These will give limited benefit differentiate between high- and low-fre- Many papers have been published on
without other changes to the control or quency noise, either by filtering or calcu- aspects of heat-balance control (gener-
operation. lating the period of the dominant signal, ally under bath ratio control or tempera-
There is scope for increasing the reli- and enact different responses. Figure 6 ture control), and there is clearly still a
ability of the feed control so that the shows examples of typical metal- and lot to learn. However, some basic prin-
frequency of spurious overfeeds is re- anode-related noise. ciples are evident.
duced, and the ability to cope with cell The logical progression is to relate • Consistency of actions on individual
instability is improved. The availability noise-wave shapes to specific cell prob- cells is critical to minimize varia-
of increased computer-processing power lems and, hence, provide more appro- tion. This requires well specified
close to the cell means that more sophis- priate control responses or information. procedures that are faithfully fol-
ticated pseudo-resistance-based algo- For example, a particular shape might lowed.
rithms can be utilized without making indicate a low bath level, triggering a • Consistency of inputs is also criti-
the control itself overly complex (which message to the operator to check the cal, particularly the current, the fluo-
can be a major problem if people are to bath depth. Development of this form of ride loading on secondary alumina,
interact with it). control requires considerable effort to and the bath content of the anode
monitor live resistance traces, link them cover material.
Noise Control to operational problems, and then estab- • Accurate measurements are vital.
The raw voltage or pseudo-resistance lish repeatability. Once the relationships Taking temperature measurements
signals from the reduction cell contain
fluctuations that can be used to identify
problems with the cell operation. Identi-
fication and use of this information to
take corrective action is generally re-
ferred to as noise control. Prior to the use
of computers in the reduction lines, cell
operators were aware that problem pots
could be identified by variations in cell
voltage that created a swinging or kick-
ing needle on the volt meter (hence, the
terminology swinging or kicking cell).
More detailed analysis identified that
• Stable (i.e., normally operating cells)
display a background noise that is
essentially random but has a fre-
quency of 0.5–2 Hz and an ampli- Figure 7. The reduction in process variation required for a 5°C target temperature change using
tude of less than 30 mV. This is due Strategy A from Table III.

1999 May • JOM 21


or bath samples at inappropriate was in continuous operation on two cells power supplier.
times (e.g., after an anode effect) or for a year.32 Predicted temperatures were At present, it would appear that more
using poorly calibrated instruments found to be within 4°C of manual mea- frequent measurements are the most ef-
will make control very difficult. surements over a six-day run. It is not fective way to improve the control, and
• The more frequent the measure- known whether this model was subse- this will generally mean more emphasis
ments, the better (taking into ac- quently used for control. No other con- on bath temperature. Automating and
count the previous point), as the trol applications have been identified. increasing the frequency of bath-tem-
dynamics of the process dictate that Thus, although several companies are perature measurement appears to have
the cell is constantly changing state, known to have worked on model-based potential to improve control for large
and it is beneficial to identify the control, there is little evidence of suc- cell technologies.1 Direct superheat
trends. cess. This is probably because the mod- measurement looks attractive for con-
• There is probably no one optimal els have a tendency to drift and require trol if it can be done cheaply and reliably.
control strategy for all technologies regular correction from measurements, As an experimental tool, it will certainly
due to different hardware, raw ma- thus negating their benefit. Also, the be useful in helping to understand
terials, operating points, and work higher complexity means that they will why some strategies work better than
cultures. However, there appears to not be accepted unless their performance others.
be a trend toward control based is significantly better than conventional
References
more on bath temperature than bath- systems.
chemistry analysis. In 1990, Tikasz and Potocnik predicted 1. O. Bonnardel and P. Homsi, Light Metals 1999, ed. C.E.
Eckert (Warrendale, PA: TMS, 1999), pp. 303–309.
Table III compares heat-balance con- that dynamic models would make a 2. “Bath Temperature and Level Measurement Device and
trol strategies from literature; they are comeback as part of expert systems.33 It Process,” French Patent FR2727985 (December 1994).
3. G.A. Guomundsson, Light Metals 1999, ed. C.E. Eckert
attempts by different companies at con- was proposed that an expert system (Warrendale, PA: TMS, 1999), pp. 297–302.
4. Qiu Zhuxian et al., JOM, 46 (8) (1994), pp. 28–30.
trolling the same factor. Comparing the would interface with the existing con- 5. J.J. del Campo and F. Blanco, Light Metals 1995, ed. J.W.
strategies, it can be seen that all use bath trol and a process simulator. The simula- Evans (Warrendale, PA: TMS, 1995), pp. 413–421.
6. D. Stefanidis and D. Georgantonis, Light Metals 1993, ed.
temperature as a control measure. Three tor would be updated on a regular basis S.K. Das (Warrendale, PA: TMS, 1993), pp. 271–276.
strategies use the measurement of ex- with process data and would provide 7. “Process for Measuring and Controlling the Concentration
of Alumina in Electrolyte Cells for Aluminium Production,”
cess aluminum-fluoride concentration; predictions to the expert system. So far, European Patent 901,226,134 (27 November 1990).
whereas, strategy D relies on tempera- it appears that the only reduction line 8. N.E. Richards et al., Light Metals 1995, ed. J.W. Evans
(Warrendale, PA: TMS, 1995), pp. 391–404.
ture alone. The base aluminum-fluoride application of expert systems was 9. P. Verstreken, JOM, 49 (11) (1997), pp. 43–46.
additions are determined either by long- HALDRIS, a knowledge-based sys- 10. G.T. Holmes et al., Light Metals 1980, ed. C.J. McMinn
(Warrendale, PA: TMS, 1980), pp. 401–410.
term consumption (strategy A) or by cell tem (without a process model) devel- 11. S.R. Brandtzaeg and K.A. Paulsen, Light Metals 1998, ed.
age. The control bases vary markedly oped by Hydro Aluminium and used at B.J. Welch (Warrendale, PA: TMS, 1998), pp. 603–606.
12. G. Bearne, Proceedings of the 6th Australasian Technology
between the strategies, both in principle a Söderberg plant.34 Workshop, ed. B.J. Welch and M. Skyllas-Kazacos (Sidney,
Australia: U. of NSW, 1998), pp. 91–130.
and by degree of complexity; some use
Future Directions 13. W. Haupin, Light Metals 1998, ed. B.J. Welch (Warrendale,
regression equations, others use the lat- PA: TMS, 1998), pp. 531–537.
14. G.T. Holmes, Light Metals 1995, ed. J.W. Evans
est temperature or bath chemistry mea- The benefits of improved heat-balance (Warrendale, PA: TMS, 1995), pp. 371–373.
surement or both. As one might expect, control will come from reduced varia- 15. B.J. Welch, Second International Alumina Quality Workshop
(Perth, Australia:1990), pp. 15–22.
the control variables are similar—add- tion directly35 and from the ability to run 16. W.E. Haupin J. of the Electrochem. Soc., March (1956), pp.
ing aluminum fluoride, and more rarely, at a more beneficial operating point. 174–78.
17. H.J. Meyer and D.G. Earley, Light Metals 1986, ed. R.E.
sodium carbonate, and altering the cell- Control at lower temperatures is desir- Miller (Warrendale, PA: TMS 1986), pp. 365–370.
resistance target. able, but is inherently more difficult be- 18. B.J. Welch, Austr. Institute of Min. & Met. Proceedings No.
214 (Carlton South, Australia: AusIMM, 1965), pp. 1–19.
There are major differences between cause the allowable variation is less, and 19. H. Kvande et al., Light Metals 1997, ed. R. Huglen
the aggressiveness of the control actions, the rate of change of the bath-liquidus (Warrendale, PA: TMS, 1997), pp. 403–408.
20. “Reduction Cell Control System,” U.S. patent 3,622,475
with some responding slowly and cau- temperature increases with increasing (1968).
tiously, while others make large changes AlF3. 21. “Alumina Feed Control,” U.S. patent 3,625,842 (1968).
22. C.A. Wilson and A.T. Tabereaux, Light Metals 1983, ed.
to try to correct deviations from the tar- The average bath temperature for cells E.M. Adkins (Warrendale, PA: TMS, 1983), pp. 479–493.
using strategy A in Table III is 962°C. 23. “Process and Apparatus for Accurately Controlling the
get. Strategy C, for example, will in- Rate of Introduction and the Content of Alumina in an
crease the cell-resistance target by up to Assume that the cells are operating at Igneous Electrolysis Tank in the Production of Aluminum,”
U.S. patent 4,431,491 (filed 1981).
an equivalent of 360 mV if the last tem- the lowest average temperature that the 24. K.R. Robilliard and B. Rolofs, Light Metals 1989, ed. P.G.
perature result was very low. In terms of capability of the control strategy will Campbell (Warrendale, PA: TMS, 1989), pp. 269–273.
25. X. Liu, Proceedings of the 5th Australasian Technology Work-
performance, there is some variation, permit. This implies that the minimum shop (Sidney, Australia: U. of NSW, 1995), pp. 619–625.
with the best reported being a 1% stan- acceptable temperature is 945°C. In or- 26. P.M. Entner, Light Metals 1996, ed. W. Hale (Warrendale,
PA: TMS, 1996), pp. 445–449.
dard deviation in excess aluminum fluo- der to operate at an average temperature 27. J.M. Peyneau, International Symposium on Reduction and
ride and 5°C in bath temperature. The 5°C lower, an improvement in tempera- Casting (NewYork: Pergaman Press, 1988), pp. 189–195.
28. D.J. Madsen, Light Metals 1992, ed. E. Cutshall (Warrendale,
cell technology will affect the perfor- ture standard deviation from 5.5°C to PA: TMS, 1992), pp. 453–456.
mance. For example, Comalco has de- about 4°C would be required. This con- 29. P. Desclaux, Light Metals 1987, ed. R.D. Zabreznik
(Warrendale, PA: TMS, 1987), pp. 309–313.
veloped a temperature-based control cept is shown in Figure 7. Clearly there 30. C. Vanvoren, Aluminium Pechiney, best industrial result
that, in contrast to strategy D, achieves are benefits in further improving heat- (private communication 1999).
31. T. Saksvikroenning et al., Light Metals 1976, ed. S.R.
excess AlF3 and bath-temperature stan- balance control but it will not be easy to Leavitt (New York: AIME, 1076), pp. 275–286.
achieve. 32. E. Sorheim and P. Borg, Light Metals 1989, ed. P.G.
dard deviations of 1.0% and 6°C, respec- Campbell (Warrendale, PA: TMS, 1989), pp. 379–384.
tively. It is likely that different strategies The application of dynamic models 33. L. Tikasz and V. Potocnik, Light Metals 1990, ed. C.M.
Bickert (Warrendale, PA: TMS, 1990), pp. 197–202.
can achieve a similar performance. may ultimately be able to improve bath 34. W.K. Rolland et al., Light Metals 1991, ed. E.L. Rooy
Dynamic models of reduction cell en- heat-balance control by predicting the (Warrendale, PA: TMS, 1991), pp. 437–442.
35. F.J. Stevens et al., Light Metals 1992, ed. E. Cutshall
ergy and mass balance have been devel- longer term trends; whether this will (Warrendale, PA: TMS, 1992), pp. 541–547.
oped since the 1970s. Although origi- happen remains to be seen, as consider-
able research money has already been Geoffrey P. Bearne is a manager in the Reduction Divi-
nally used as research tools, efforts have sion of Comalco Research and Technical Support.
been made to apply these, generally in spent with little success. Model-based
simplified form, for control purposes. systems certainly have the potential to For more information, contact G.P. Bearne,
improve the control at plants subject to Comalco Research and Technical Support, 15
Earlier linear models were not success- Edgars Road, Thomastown, Victoria 3074, Aus-
ful.31 More success was reported in the current modulation as a result of peak tralia; telephone 61-39-9469-0769; fax 61-3-9462-
1980s using a nonlinear model, which shaving or other arrangements with the 2700; e-mail geoff.bearne@riotinto.com.

22 JOM • May 1999

You might also like