Prediction of Low-Voltage Tetrafluoromethane EmissionsBased on the Operating Conditions of an AluminiumElectrolysis Cell

LUKAS DION ,1,3 LASZLO I. KISS,1 SANDOR PONCSAK,1

and CHARLES-LUC LAGACE2

1.—Universite du Quebec a Chicoutimi, 555 boulevard de l’Universite, Chicoutimi, QC G7H 2B1,Canada. 2.—Aluminerie Alouette Inc., 400 chemin de la Pointe-Noire, C.P. 1650, Sept-Iles,QC G4R 5M9, Canada. 3.—e-mail: lukas.dion@uqac.ca

Greenhouse gas (GHG) generation is inherent in the production of aluminiumby a technology that uses carbon anodes. Most of those GHG are composed ofCO2 produced by redox reaction that occurs in the cell. However, a significantfraction of the annual GHG production is composed of perfluorocarbons (PFC)resulting from anode effects (AE). Multiple investigations have shown thattetrafluoromethane (CF4) can be generated under low-voltage conditions inthe electrolysis cells, without global anode effect. The aim of this paper is tofind a quantitative relationship between monitored cell parameters and theemissions of CF4. To achieve this goal, a predictive algorithm has beendeveloped using seven cell indicators. These indicators are based on the cellvoltage, the noise level and other parameters calculated from individual anodecurrent monitoring. The predictive algorithm is structured into three differentsteps. The first two steps give qualitative information while the third onequantitatively describes the expected CF4 concentration at the duct end of theelectrolysis cells. Validations after each step are presented and discussed.Finally, a sensitivity analysis was performed to understand the effect of eachindicator on the onset of low-voltage PFC emissions. The standard deviation ofindividual anode currents was found to be the dominant variable. Cell voltage,noise level, and maximum individual anode current also showed a significantcorrelation with the presence of CF4 in the output gas of an electrolysis cell.

INTRODUCTION

In primary aluminium reduction, continuousefforts are taken by the industry to minimize thetonnes of CO2 equivalent produced per ton ofaluminium. Perfluorocarbon (PFC) emissions, com-posed essentially of tetrafluoromethane (CF4) andhexafluoroethane (C2F6) are key elements to con-sider in this process. The global warming potentialof these two gases is 6630 and 11,100 times greaterthan CO2, respectively.1 They are generated duringan undesired event in the cell called an anode effect.This event is usually associated with an importantincrease in the cell voltage and is easily identifiable.Hence, these ‘high-voltage PFC’ emissions are wellknown in the industry and specific guidelines2 exist

to quantify the amount of gas generated during thisevent. For this reason, smelters have optimisedtheir process worldwide over the years and the totalamount of PFC emissions from the aluminiumindustry has been significantly reduced between1990 and 2010.3 However, by lowering the AEfrequency and duration, another source of emissionshas become more apparent in recent years, charac-terized as ‘low-voltage PFC’ emissions. This partic-ular type of emission can occur for a significantperiod of time with little or no indication of misbe-haviour in the electrolytic cell, for instance, anincrease in cell the noise or voltage. For this reason,low-voltage PFC are important to take into consid-eration but, up to now, no available method otherthan real-time PFC monitoring exists to account for

JOM, Vol. 68, No. 9, 2016

DOI: 10.1007/s11837-016-2043-6� 2016 The Author(s). This article is published with open access at Springerlink.com

2472 (Published online August 2, 2016)

these emissions.4 This paper investigates the corre-lation between specific cell variables and the level ofCF4 emissions at the duct end of the electrolysis cell.

STATE OF THE ART

Numerous authors (see Table I) have previouslyinvestigated the occurrence of low-voltage PFC.There is general agreement in the scientific com-munity that the basic mechanisms generating low-voltage PFC are similar to the standard AE mech-anism, which is very well documented. AE occursdue to the privation of dissolved alumina in alocalized region of the bath. If it happens, transportof the electric charges is no longer supported by thestandard electrolysis Reaction 1. This will lead to anincrease in the anodic overvoltage, and subsequentReactions 2–3 will occur in the cell, leading to theelectrolysis of the cryolite and the generation ofPFC; therefore, an AE.5

2Al2O3ðdissÞ þ 3CðsÞ ! 4Al lð Þ þ 3CO2ðgÞ

E0 ¼ �1:18 Vð1Þ

4Na3AlF6ðlÞ þ 3CðsÞ ! 4AlðlÞ þ 3CF4ðgÞ þ 12NaFðdissÞ

E0 ¼ �2:58 V

ð2Þ

2Na3AlF6ðlÞ þ 2CðsÞ ! 2AlðlÞ þ C2F6ðgÞ þ 6NaFðdissÞ

E0 ¼ �2:80 V

ð3Þ

Once an AE occurs in the cell, the localised areawhere PFC are produced usually becomes stronglyresistive to the passage of current (increase in ohmicresistance) and the current will be redistributedtoward the other anodes in the cell. This redistri-bution generally provokes the same problem else-where and the AE propagates from one anode to theother until terminated, meanwhile significantlyincreasing the global cell voltage. The detectionlimit of an AE varies from one smelter to another.However, a generally accepted value is when thecell voltage exceeds 8 V.6 For this reason, general-ized AE, or high-voltage anode effects (HVAE), areeasily identifiable and are very well documented inthe literature.6–8

However, if the AE phenomenon occurs onlylocally in the cell without propagating to all theother anodes, it can lead to a continuous generationof PFC while the cell voltage still remains under theAE detection limit. This makes the detection of low-voltage emissions difficult without continuous mon-itoring of the output gas composition. This eventcan either self-terminate due to alumina or cur-rent redistribution or, eventually, it can lead to aHVAE. Historically, this phenomenon was called

‘‘background or non-AE PFC emissions’’. Recently,the International Aluminum Institute (IAI) hasadopted the term ‘‘low-voltage AE’’ (LVAE).9 Thispaper will use the latest terminology.

Some authors10–14 refer to Reactions 4–6 as thedominant mechanism for the generation of PFCunder low-voltage conditions as these reactions canoccur without a significant increase in the anodicovervoltage. Furthermore, it could also explain whymost of the LVAE measurements only indicate verysmall traces of C2F6 above the detection limit of theinstrument. In contrast, during HVAE, the ratio ofC2F6 to CF4 can change with time and cell tech-nologies but the typical value is approximately 0.1.

2AlF3ðdissÞ þ Al2O3ðdissÞ þ 3CðsÞ ! 4AlðlÞ þ 3COF2ðgÞ

E0 ¼ �1:88 V

ð4Þ

2COF2ðgÞ þ CðsÞ ! 2COðgÞ þ CF4ðgÞ

K ¼ 94:8ð5Þ

3COF2ðgÞ þ2CðsÞ ! 3COðgÞ þC2F6ðgÞ

K ¼ 1:2�10�3ð6Þ

Past researchers have demonstrated that multipleparameters or events could be linked to the occur-rence of low-voltage PFC emissions. Their conclu-sions are summarized in Table I and the mostrelevant points are discussed briefly afterward.

Most authors5,7,10–12,16–19 agree that aluminaconcentration, anode current density and anodicovervoltage have a significant impact on the onset oflow-voltage PFC emissions. All of these parametersare interrelated and will dictate if the localisedovervoltage eventually exceeds the threshold neces-sary to generate PFC. Similarly, LVAE related tothe feeding strategy have been observed at the endof underfeeding periods. This observation confirmsthat low alumina concentration in the bath is morelikely to generate CF4 even during normal celloperation.

Some authors11,12,15,16 have explored the influ-ence of bath temperature, bath chemistry andsuperheat on the PFC generation. These variableswill have an influence on the maximum solubility ofthe alumina in the bath as well as on the kinetics ofits dissolution. Hence, keeping these variables in anappropriate range will minimize the alumina con-centration gradients in the electrolytic bath.

The new, high-amperage cells that are becomingmore prevalent in the aluminium industry arecomposed of a greater number of anodes with largersurface areas in order to preserve anode currentdensity. For this reason, a local AE is less likely topropagate towards the other anodes and thusprobably can last longer.7 Moreover, as each anodeare connected in parallel electrically, the effect of

Prediction of Low-Voltage Tetrafluoromethane Emissions Based on the Operating Conditionsof an Aluminium Electrolysis Cell

2473

Table

I.Overview

ofposs

ible

correlationsbetw

een

cell

variablesorsp

ecifi

cevents

inth

ecell

with

low-voltageemissionsofPFC

Year

Auth

ors

Cell

parameters

Alu

min

aconcentration

Feedin

gstrategy

Anode

current

density

Anodic

overvoltage

Bath

temperatu

re

Bath

chemistrySuperheat

Tota

lnumber

ofanodes

Anode–

cath

ode

dista

nceCell

stability

(noise)

Cell

voltage

2011

Li

etal.

15

?�

�?

2013

Ch

enet

al.

10

++

+2013

Won

gan

dM

ark

s16

++

��

2013

Zaro

un

iet

al.

17

++

++

2014

Ash

eim

etal.

5+

2014

Won

get

al.

7+

++

++

2015

Ash

eim

etal.

11

++

++

2015

Dan

do

etal.

18

++

2015

Jass

imet

al.

12

++

++

++

+2016

Bati

sta

etal.

19

++

+S

um

ofth

ep

osit

ive

fin

d-

ings

83

66

10

11

22

1

Year

Auth

ors

Events

Anodechange

Meta

lta

ppin

gOverly

aggress

iveAE

treatm

ent

Blocked

feeder

2011

Li

etal.

15

??

2013

Ch

enet

al.

10

2013

Won

gan

dM

ark

s16

++

2013

Zaro

un

iet

al.

17

+2014

Ash

eim

etal.

5

2014

Won

get

al.

7+

2015

Ash

eim

etal.

11

2015

Dan

do

etal.

18

++

+2015

Jass

imet

al.

12

+2016

Bati

sta

etal.

19

+S

um

ofth

ep

osit

ive

fin

din

gs

51

12

+a

corr

elati

onw

as

obse

rved

,�

no

corr

elati

onw

as

obse

rved

,?

the

rela

tion

ship

was

un

clea

r,em

pty

cell

sin

dic

ate

that

this

vari

able

was

not

men

tion

edby

the

au

thor

s

Dion, Kiss, Poncsak, and Lagace2474

gas passivation under a single anode will be lesssignificant on the global cell voltage if the totalnumber of anodes increases.

PFC emissions are frequently observed up tomultiple hours after an anode change.12,16–19 Thisis due to the redistribution of current towards anodesthat increases the current density locally. The non-uniformity of the current distribution can be ampli-fied if the anode–cathode distance is reduced in orderto operate with lower energy consumption. In such acell, a small imbalance in the anode setting will havea greater impact on the anode current distribution,thus increasing the risk of generating PFC.

EXPERIMENTAL SETUP

Data used in this study were collected during twomeasurement campaigns of gas emissions at Alu-minerie Alouette from prebaked AP40LE cells usingpoint feeders and operating above 390 kA. They wereequipped with on-line monitoring of individual anodecurrents using a measurement frequency of 1 Hz.Cell current, voltage and pseudo-resistance weremeasured by the cell control system with the samefrequency, and all the data were recorded. Therefore,only one cell at a time was monitored, and there wasno dilution other than the gases entering the cellthrough the hooding. Inspection of the hooding wasperformed frequently to make sure it remained insimilar condition throughout the whole campaign.

During these periods, gas composition was mea-sured with a GASMETTM DX-4000 Fourier trans-formed infrared spectrometer (FTIR) using a Peltiercooled mercury–cadmium–telluride detector (sam-ple cell path: 9.8 m, volume: 0.5 L, resolution:7.8 cm�1). A stainless steel sampling probe waslocated at the duct end of the electrolysis cell andgas was continuously fed to the analyzer at avolumetric rate of 2.5 L min�1. The gas streamwas sent sequentially through a 15-l filter, adesiccant, activated alumina, a 5-l filter and finallya 2-l filter to remove dust, traces of water andhydrogen fluoride for the protection of the measur-ing equipment. The gas went through a line heatedat 120�C before entering the FTIR and concentra-tion measurements were performed at a rate of 10scans per second. Average values for 5-s periodswere recorded. The background spectrum was rede-fined using high purity nitrogen every 24 h.

Gas composition was measured for a total of2 weeks. The collected data were then classified into15 scenarios. Thirteen of them were selectedbecause the CF4 concentration remained withinthe range of interest for a significantly long period.Two extra scenarios represent stable periods with-out LVAE. CF4 emissions issued from HVAE werenot considered in this study. Additional informationon the preparation of the data has been publishedpreviously.20 After the data selecting process,22,000 data points remained. Half of them corre-sponded to LVAE. This relatively high number of

points was sufficient to develop the six artificialneural networks required for the predictive algo-rithm described below, but a further increase in thetotal number of data points could still improve itsperformance.

DEVELOPMENT OF THE PREDICTIVEALGORITHM

A predictive algorithm was developed to predictthe CF4 concentration at the duct end of anelectrolysis cell using continuously measuredparameters. The range of interest of the CF4

concentration for this study is between 10 ppb and2000 ppb. The lower limit was set according to thethreshold detection of the FTIR, and the upper limitwas determined by the highest CF4 concentrationthat was measured under LVAE conditions. C2F6

was not considered as most of the data remainedunder the detection limit of the FTIR (20 ppb).

The strategy of the algorithm, as well as thechoice of the input variables, was developed itera-tively. Multiple combinations have been examinedincluding different strategies and/or differentinputs. It is important to mention that furtherrefinement is still possible but it would first requireadditional measurement campaigns.

Seventy percent of the selected data were used forthe learning process to develop the artificial neuralnetworks and the remaining 30% was used forvalidation to evaluate the accuracy of the predic-tions. One of the main applications of this algorithmis the sensitivity analysis that has been performedsubsequentially. This analysis clearly indicates thevariables with the strongest influence on the emis-sions of LVAE as well as the expected variationsover the entire range of each entry variable.

Artificial neural networks (ANN) behind thealgorithm were developed using the data miningpackage offered with STATISTICA 12�.

List of Indicators

The first selection of the potential indicators (inputvariables) was based on the results of the literaturereview indicating which parameters were most likelyto correlate with the presence of low-voltage CF4

emissions. To introduce an input to the algorithm, itwas necessary to have data collected with relativelyhigh frequency (0.2–1 Hz) for each respective input.As alumina concentration could only be measuredintermittently, it was not included as an input. Afteroptimisation, seven variables were retained as inputsfor the predictive model. Most of these are related toindividual current monitoring as it offers local dataon the cell behaviour. More importantly, it suppliesindirect information regarding the alumina distribu-tion21 and the influence of changing anodes. The listof the seven indicators:

� Cell voltage (volts) Average cell voltage com-puted for 5 consecutive seconds.

Prediction of Low-Voltage Tetrafluoromethane Emissions Based on the Operating Conditionsof an Aluminium Electrolysis Cell

2475

� Noise (temporal stability) indicator (nanoohms)The difference between the maximum and theminimum pseudo-resistance measured for thecell within the last 6 s.

� Maximum current driven through individualanodes (amps) The highest current value thathas been observed among the individual anodes.

� Standard deviation between individual anodecurrents (amps) For every second, the standarddeviation was calculated using individual anodecurrent measurements from the entire cell.

� Absolute difference between upstream and down-stream current averages (amps) the averagecurrent value is calculated for both halves ofthe cell and the absolute difference between thetwo sides is calculated.

� Absolute difference between tap hole and ductend side current averages (amps) Same as theprevious variable but the division of the cell wasperformed across the other axis.

� Range of measured individual currents (amps)Difference between the maximum and the min-imum individual anode currents measured.

Description of the Algorithm Strategy

The algorithm is divided into three steps as shownin Fig. 1. The first step is performed by an ANNdesigned to indicate if the conditions are met for thegeneration of CF4 in the electrolysis cell. The outputfrom this ANN can either be positive or negative. A

positive output indicates that CF4 emission isexpected at the duct end of the electrolysis cellunder those conditions. In contrast, a negativeoutput indicates that the measured CF4 concentra-tion would be below the limit of detection. Negativeoutputs are considered as 0 ppb concentration.During the learning phase of the ANN, it wasnecessary to use weighting factors to favor thepredictions of ‘‘no emissions’’. This minimizes therisk of amplifying the error caused by a wrongprediction in the following steps, which could lead todivergent predictions. Therefore, weighting factorsof 2:1 were used in favor of the negative predictions.

The second step of the algorithm is performed byanother ANN that aims to classify its output inspecific concentration ranges that are to be expectedwithout assigning a quantitative value. A correctionalgorithm is applied after this ANN to minimise therisk of wrong classification. This correction consid-ers the previous temporal prediction to assureconsistency. The resulting output can be dividedinto four different categories:

A. 10 ppb–49.99 ppbB. 50 ppb–199.99 ppbC. 200 ppb–499.99 ppbD. 500 ppb–2000 ppb

The last step of the algorithm assigns a quantitativeCF4 concentration to the specific entry conditions. Ituses four artificial networks working in parallel

Fig. 1. The predictive algorithm strategy.

Dion, Kiss, Poncsak, and Lagace2476

depending on the respective category that wasassigned in the previous steps. Once each valuehas been defined, a post-treatment is applied to theprediction to take into account the evolution of theCF4 concentration over time. In this case, the post-treatment is a mobile moving average over a 5-minperiod.

RESULTS AND DISCUSSION

Validation of the Algorithm

Thirty percent of the collected data was usedexclusively for the validation of the model. Thesedata were fed into the model and the results wereexamined after each step of the algorithm toevaluate its performance.

Figure 2a clearly indicates the ability of themodel to predict the presence or absence of CF4

based on the input variables. The percentage ofcorrect predictions after step #1 rises above 83%.Moreover, the effect of the weights discussed previ-ously is clearly visible in the incorrect predictionscolumn. Hence, when no CF4 is present in theoutput gas composition, the model rarely predictsotherwise, which increases the performance in thefollowing steps of the algorithm. Figure 2b indicatesthat 69% of the data is correctly classified after thesecond step. The results indicate that most incorrectpredictions are offset only by 1 category. Furtherinvestigation revealed that the majority of incorrectclassifications are due to concentrations that arenear the limits of each category (i.e. 10 ppb, 50 ppb,200 ppb and 500 ppb). This is more important forclassifications from the category A, where thepercentage of misclassifications exceeds the numberof correct predictions. However, more than 80% ofthese incorrect predictions were within the range of10 ppb–15 ppb. Therefore, it might be relevant toreconsider the lower limit of prediction of the modelin the future to avoid being too close to the detectionlimit of the FTIR. For this reason, it is unclearwhether the errors came from the noise of the FTIR

or if the variations within cell variables are just toosmall in this range of concentration to be detectedabove the normal noise level of each respectivevariable.

Final validation was performed by calculating theoverall mass of CF4 emitted during each specificperiod based on the measurements and comparing itto the overall mass obtained using the predictedvalues for the same periods. The mass of CF4 wascalculated using integration according to the trape-zoid rule for each respective scenario and by mul-tiplying the resulting (ppb*s) by the flow rate at theduct end of the cell as well as by the CF4 density forthe corresponding temperature and pressure. Illus-trative results for all scenarios are presented inFig. 3.

In most cases, the overall behaviour of the CF4

prediction is in good accordance with the corre-sponding measurements, including the cases whereno emissions are present (Fig. 3-14 and 15). Itindicates the consistent behaviour of the algorithmto quantitatively predict the concentration of CF4

from an electrolysis cell based exclusively on someof the cell parameters. The results in Fig. 4 are alsoconsistent with this statement as they indicate thatthe model correctly predicts the total mass of CF4

within a ±25% error margin in two-thirds of thecases. If we consider that to our knowledge, no otherpredictive model to account for low-voltage PFCemissions has been developed in the open literature,added to the fact that the average error for theentire set of data is 8%, the algorithm’s performancecan be considered as good. Henceforth, it is possibleto proceed with a sensitivity analysis representativeof the inputs’ effect.

Sensitivity Analysis: Individual Effect of theIndicators on the Low-Voltage Emissions ofCF4

A sensitivity analysis was performed based on aseven-level full factorial design22 including all theseven indicators. Henceforth, it was possible to

Fig. 2. (a) Percentages of correct and incorrect predictions after Step #1. (b) Percentages of correct predictions along with the different offsets inincorrect predictions after Step #2.

Prediction of Low-Voltage Tetrafluoromethane Emissions Based on the Operating Conditionsof an Aluminium Electrolysis Cell

2477

Fig. 3. Comparison between predicted CF4 concentrations (filled circles) and measured concentration (open circles).

Dion, Kiss, Poncsak, and Lagace2478

examine the effect of each input variables on theresulting predictions obtained with the algorithm.For the purpose of identifying the dominant inputvariables, most of the useful information is availableafter the first step of the algorithm. Therefore, onlythese results are presented, as they are morerelevant.

The exploration limits of each variable weredefined using their respective data distributioncollected during the measurement campaign. Thecorresponding lower and upper limits for eachvariable were defined as the 1st and the 99thcentiles in order to eliminate the atypical values.The impact of the cross-effects between the differentparameters was investigated but no significantinteraction was observed, hence it is not presentedin this study.

An investigation regarding the probability of CF4

emissions as a function of cell parameters wasperformed. For each individual variable, the totalnumber of predicted emissions has been normalizedfor easier interpretation. The reference (0%) indi-cates the point where the variable has no influence.

Hence, a positive value indicates that the presenceof CF4 is more likely. In contrast, a negative valueindicates that CF4 emission is predicted by themodel less frequently. Moreover, a threshold valuehas been added to each figure. This threshold isbased on the actual measurements and representsthe transition point where the probability of occur-rence of CF4 emissions gets higher than the prob-ability of having no emissions. Figure 5 illustratesthe change in CF4 emissions with respect to eachinput variable resulting from the sensitivity analy-sis. A clear correlation can be observed between avariable and the CF4 emissions if the slope is steepand the trend is uniform along most of the rangestudied.

A positive correlation was observed between anincreasing cell voltage and the occurrence of CF4

emissions. It is important to consider that thevoltage values from Fig. 5a are representative of aspecific cell technology. However, the upper range ofvalues can be representative of the variation thatoccurs after an anode change or when higher noiseis observed in the pot. Interestingly, as the cell

Fig. 4. Radar chart illustrating the absolute value of the error percentages for all 15 scenarios.

Prediction of Low-Voltage Tetrafluoromethane Emissions Based on the Operating Conditionsof an Aluminium Electrolysis Cell

2479

Fig. 5. Influence level of each indicator on the frequency of predictions of CF4 based on a full factorial design sensitivity analysis. The vertical linerepresents the measured threshold value for each variable.

Dion, Kiss, Poncsak, and Lagace2480

voltage decreases below the threshold value, theslope of influence is steeper. This indicates thatvariations in this range have a greater effect on CF4

emissions. It can be interpreted as a range ofvariation, which is more plausibly associated withan increase in overvoltage. In contrast, a higherincrease in voltage is more likely associated with achange in the total resistance of the cell.

The noise level (Fig. 5b) has a positive correlationwith the occurrence of CF4 emissions. However, animportant variation in the stability of the cell willonly generate a small increase in the probability ofCF4 emissions.

The maximum current measured among theanodes (Fig. 5c) appears to have a positive influenceon the generation of CF4 emissions up to 35 kA(�1.75 times the normal level). Up to a certainpoint, it can be representative of the currentredistribution following an anode change or whenexcessive gas passivates the anodes. Consequently,higher current locally consumes the alumina morerapidly which eventually leads to PFC generation inthat region of the cell. Further investigations arerequired to see if those conditions are maintainedfor the same anodes or if the current jumps from oneanode to the another over time. Moreover, very highcurrent in a single anode, maintained for a longperiod, can only be explained by a short-circuitgenerating other negative impacts for the cell. Thisphenomenon explains the drop in the CF4 occur-rence when the maximum current in an anodereaches more than 40 kA.

The results from Fig. 5d indicate a clear andstrong relationship between the standard deviationamong individual anode currents and the predic-tion of CF4 emissions. These results are consistentwith the literature as well as with the mechanismof PFC emissions. Therefore, when a disruption ofthe current uniformity starts generating LVAE,the local resistivity is expected to increase underspecific anodes. It will lower the current from theseanodes and redistribute a part of the currenttoward other anodes, hence amplifying the currentnon-uniformity in the cell. The behaviour appearsto be linear up to a certain limit (7.5 kA).Afterward, there is no more significant increaseof the influence of this variable on the emissions ofCF4.

Figure 5e shows a negative correlation but thiseffect is mainly due to a permanent offset ofapproximately 1.75 kA between the average indi-vidual anode currents from the upstream anddownstream side under normal operation. Hence-forth, it has no real correlation with LVAE.

No significant correlation can be observed onFig. 5f and g between the occurrence of CF4 emis-sions and those variables. However, even if thesevariables are not as useful in predicting the occur-rence of LVAE, they are relevant in the subsequentsteps of the algorithm to predict the level of theseemissions when these are detected.

CONCLUSION

A study has been performed to determine whethercertain measurable indicators permit the predictionof low-voltage PFC emissions from aluminum elec-trolysis cells. Inspired by a literature review and agood understanding of the mechanisms, seven indi-cators were selected and used to develop a predic-tive algorithm based on measurements carried outon selected aluminium electrolysis cells. The modelis able to successfully predict the emissions of CF4

at the duct end of an electrolysis cell using onlythose seven inputs. A sensitivity analysis wasperformed using the algorithm to understand theeffect of each variable on the occurrence of low-voltage CF4 emissions.

The sensitivity analysis clearly demonstratedthat inhomogeneity among individual anode cur-rents is the best indicator to predict low-voltage CF4

emissions. Cell voltage and maximal anode currentshow also a significant and positive correlation withthe emissions. The noise level has a positive, but notsignificant, correlation. No other measurable cellvariable was found to have a direct and significantcorrelation with the occurrence of low-voltage PFCemissions.

The model described in this paper shows promis-ing results as a predictive method, but furtherimprovements are still required before it can beused as a robust quantitative tool integrated intothe cell control system. Moreover, due to the varietyof the reduction technologies and the limited acces-sibility of individual anode current monitoringacross smelters worldwide, the proposed algorithmcannot be easily applied outside of the cell technol-ogy from which it was developed. However, theinvestigation and results described in this paper canlead to refinements that would be applicablethroughout the entire aluminium industry.

Finally, the primary objective of this study wasreached, namely, a new tool was developed, usingcertain key indicators to predict the generation ofCF4 under low-voltage conditions.

ACKNOWLEDGEMENTS

This work was financed by a BMP-Innovationgrant from a partnership composed of ‘‘Fonds deRecherche du Quebec – Nature et Technologies’’(FRQNT), ‘‘National Science and Engineering Re-search Council of Canada’’ (NSERC) and Alu-minerie Alouette Inc. The authors would like tothank Aluminerie Alouette Inc. for the permissionto present these results. The support and recom-mendations provided by Dr. Jerry Marks during themeasuring campaign and in the interpretation ofthe results are acknowledged and greatly appreci-ated.

OPEN ACCESS

This article is distributed under the terms of theCreative Commons Attribution 4.0 International

Prediction of Low-Voltage Tetrafluoromethane Emissions Based on the Operating Conditionsof an Aluminium Electrolysis Cell

2481

License (http://creativecommons.org/licenses/by/4.0/),which permits unrestricted use, distribution, and re-production in any medium, provided you give appro-priate credit to the original author(s) and the source,provide a link to the Creative Commons license, andindicate if changes were made.

REFERENCES

1. G.Myhre, D. Shindell, F.-M. Breon, W.Collins, J. Fuglestvedt,J. Huang, D. Koch, J.-F. Lamarque, D. Lee, B. Mendoza,T. Nakajima, A. Robock, G. Stephens, T. Takemura, andH. Zhang, Anthropogenic and natural radiative forcing. Cli-mate Change 2013: The Physical Science Basis. Contributionof Working Group I to the Fifth Assessment Report of theIntergovernmentalPanel onClimateChange, ed. T.F. Stocker,D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung, A.Nauels, Y. Xia, V. Bex, and P.M. Midgley (Cambridge, UK andNew York, NY, USA: Cambridge University Press, 2013).

2. IPCC 2006, 2006 IPCC Guidelines for National GreenhouseGas Inventories, Prepared by the National Greenhouse GasInventories Programme, H.S. Eggleston, L. Buendia, K.Miwa, T. Ngara, K. Tanabe (eds.) (Japan: IGES, 2006).

3. M. Fischedick, J. Roy, A. Abdel-Aziz, A. Acquaye, J.M.Allwood, J.-P. Ceron, Y. Geng, H. Kheshgi, A. Lanza, D.Perczyk, L. Price, E. Santalla, C. Sheinbaum, and K. Ta-naka, Industry. Climate Change 2014: Mitigation of Cli-mate Change. Contribution of Working Group III to theFifth Assessment Report of the Intergovernmental Panel onClimate Change, ed. O. Edenhofer, R. Pichs-Madruga, Y.Sokona, E. Farahani, S. Kadner, K. Seyboth, A. Adler, I.Baum, S. Brunner, P. Eickemeier, B. Kriemann, J.Savolainen, S. Schlomer, C. von Stechow, T. Zwickel, andJ.C. Minx (Cambridge, UK and New York, NY, USA:Cambridge University Press, 2014).

4. D.S. Wong, P. Fraser, P. Lavoie, and J. Kim, JOM 67, 342(2015).

5. H. Asheim, T.A. Aarhaug, A. Ferber, O.S. Kjos, and G.M.Haarberg, Light Metals 2014, ed. J. Grandfield (Hoboken:Wiley, 2014), pp. 535–539.

6. A. Tabereaux (Paper presented at the Eight AustralasianAluminium Smelting Technology Conference and Work-shops, Yeppoon, Australia, 2004).

7. D. Wong, A. Tabereaux, and P. Lavoie, Light Metals 2014,ed. J. Grandfield (Hoboken: Wiley, 2014), pp. 529–535.

8. G. Tarcy and A. Tabereaux, Light Metals 2011, ed. S.Lindsay (Hoboken: Wiley, 2011), pp. 329–332.

9. International Aluminium Institute, Workshop on PFCEmissions in the Aluminium Industry (London, UK,2015).

10. X. Chen, W. Li, S. Qiu, Y. Zhang, and C. Bayliss, LightMetals 2013, ed. B. Sadler (Hoboken: Wiley, 2013), pp. 877–881.

11. H. Asheim, T.A. Aarhaug, E. Sandnes, O.S. Kjos, A. Solheim,and G.M. Haarberg, ECS Trans. 69, 1 (2015).

12. A. Jassim, S. Ahkmetov, B.J. Welch, M. Skyllas-Kazacos, J.Bao, and Y. Yao, Light Metals 2015, ed. M. Hyland(Hoboken: Wiley, 2015), pp. 545–550.

13. H. Asheim, T.A. Aarhaug, A. Ferber, O.S. Kjos, G.M.Haarberg, 7th Kyoto International Forum for Environmentand Energy (Kyoto, 2014).

14. B.J. Welch (Paper presented at the 10th AustralasianAluminium Smelting Technology Conference and Work-shops, Launceston, Tasmania, 2011).

15. W. Li, Q. Zhao, J. Yang, S. Qiu, X. Chen, J. Marks, and C.Bayliss, Light Metals 2011, ed. S. Lindsay (Hoboken: Wi-ley, 2011), pp. 309–314.

16. D. Wong and J. Marks, Light Metals 2013, ed. B. Sadler(Hoboken: Wiley, 2013), pp. 865–870.

17. A.A. Zarouni, M. Reverdy, A. Zarouni, and K.G.Vankatasubramaniam, Light Metals 2013, ed. B. Sadler(Hoboken: Wiley, 2013), pp. 859–863.

18. N.R. Dando, N. Menegazzo, L. Espinoza-Nava, N.Westendorf, and E. Batista, Light Metals 2015, ed. M.Hyland (Hoboken: Wiley, 2015), pp. 551–555.

19. E. Batista, N.R. Dando, N. Menegazzo, and L. Espinoza-Nava,LightMetals2016, ed. E. Williams (Hoboken: Wiley, 2016), pp.545–550.

20. L. Dion, L.I. Kiss, S. Poncsak, and C. Lagace, Light Metals2016, ed. E. Williams (Hoboken: Wiley, 2016), pp. 545–550.

21. L. Dion, L.I. Kiss, C. Lagace, R. Victor, and J.W. Evans,Light Metals 2015, ed. M. Hyland (Hoboken: Wiley, 2015),pp. 723–728.

22. A. Saltelli, M. Ratto, T. Andres, F. Campolongo, J. Cariboni,D. Gatelli, M. Saisana, and S. Tarantola, Global SensitivityAnalysis: The Primer, 1st ed. (Chischester: Wiley, 2008), pp.53–108.

Dion, Kiss, Poncsak, and Lagace2482