Encyclopedie de la recherche sur l'aluminium au Quebec - Edition 2014 | Page 42

40

Titre – langue première
PRODUCTION D’ALUMINIUM // ALUMINIUM PRODUCTION
Prediction of anode recipe by the artificial
Titre – langue seconde
neural network (ANN) method

PRÉDICTION DE LA RECETTE D'ANODE PAR
(Prédiction de la recette
LA MÉTHODE DE RÉSEAUX DE NEURONES ARTIFICIELS d’anode

par la méthode
de réseaux de neurones artificiels )

PREDICTION OF ANODE RECIPE BY THE
ARTIFICIAL NEURAL NETWORKArunima Sarkar1, Duygu Kocaefe1, Yasar Kocaefe1, Dipankar
(ANN) METHOD
Bhattacharyay1, Dilip Sarkar1, Brigitte Morais2

1Université

2Aluminerie

1er des sciences nom2 et 3 boul. de l’Université, Chicoutimi, Québec, Canada G7H 2B1
du Québec à Chicoutimi, Département nom1, 2eappliquées, 555, e nom3
Alouette Inc., 400, Chemin de la Pointe-Noire, C.P.1 1650, Sept-Îles, Québec, Canada, G4R 5M9
Affiliation 1
2 Affiliation 2
CHAIRE DE RECHERCHE UQAC/AAI SUR LE CARBONE
3 Affiliation 3

CHAIRE DE RECHERCHE UQAC /
AAI SUR LE CARBONE

Methodology

Introduction
In the last few years, the available anode-grade petroleum
coke supply cannot meet the demand, and thus different cokes
of varying quality are mixed for anode production. Therefore,
the anode recipe should be optimized according to the
availability of the raw materials so that the anode quality could
be maintained or even improved.

Green
Anode
Density

The anode production plant uses many different fractions in
order to obtain the best particle packing to achieve high anode
density. In the plant, the granulometric fractions are blended in
such a way that strong aspects of individual coke fractions can
be used to maximize the anode performance.

Tapped Bulk Density

Laboratory Scale Anode Production
Anode Butt
+
Rejected
Anodes

Coke

The artificial neural network (ANN) method is a mathematical
tool specially designed to analyze complex data. In this study,
a model based on ANN has been developed to adjust the
granulometry of the raw materials for anode production.

Pitch

Characterization
of green anodes

Dry Aggregates

Binder

Objectives:
1. To investigate the impact of coke granulometry on anode
properties: to understand the effect of different fractions and
recycled butt materials on the green anode density.
2. To develop an artificial neural network model based on the
tapped bulk density of dry aggregates to predict the anode
paste recipe and the green anode density.

Pitch Preheating

Coke Preheating
and Mixing

Vibro-compactor

Green
Anode

, g/cc

Results
Table 1. Effect of different paste recipes on green anode density
(for the same pitch content)
Coarse
(%)

Medium
(%)

Fine
(%)

Green anode density
(g/cc)

25

2.5

11

25

64

1.587

2

25

2.5

11

26

63

1.609

3

25

2.5

13

26

61

1.568

4

25

2.5

15

31

54

1.621

5

0

2.5

15

28

56

1.574

6

0

2.5

9

21

69

1.586

7

15

2.5

9

23

68

1.584

8

25

2.5

9

23

67

1.587

9

35

2.5

9

24

67

1.568

, g/cc

Figure 1. Predicted and experimental
values of dry aggregate density

(a)

(b)

Under
Good
Over
Under
Good
Over
prediction prediction prediction prediction predictionprediction

Figure 3. Prediction capability of the ANN model:
(a) Dry aggregate density, (b) Ratio of green
anode density to dry aggregate density

Brigitte Morais
Aluminerie Alouette Inc.

Figure 2. Predicted and experimental values
of the ratio of green anode density to dry
aggregate density

No. of
cases

Rejects
(%)

1

Arunima Sarkar
Duygu Kocaefe
Yasar Kocaefe
Dipankar Bhattacharyay
Dilip Sarkar
Université du Québec
à Chicoutimi

Butt
(%)

No. of
cases

Anode
No.

Conclusions
1. The artificial neural network is proven to be an useful tool for the prediction of anode density from the
bulk density of dry aggregates.
2. It can be seen that the customized neural network model is able to predict the output for the test data
with a higher level of accuracy.
3. For the same pitch content, if the butt content in the anode is increased up to a certain level, the
green density does not vary significantly, and the density starts to decrease after this level. However,
the ratio of green anode density to dry aggregate density decreases with increasing butt content.
4. The packing of different fractions is an important parameter for anode density.
5. ANN, a powerful tool for artificial intelligence, can help aluminum industry improve anode quality and
decrease environmental impact, energy consumption, and production cost.

Figure 4. Effect of recycled butts on the
ratio of green anode density to dry
aggregate density

Acknowledgements
The technical and financial support of Aluminerie
Alouette Inc. as well as the financial support of the
Natural Sciences and Engineering Research Council of
Canada (NSERC), Dévelopment économique SeptÎles, the University of Québec at Chicoutimi (UQAC),
and the Foundation of University of Québec at
Chicoutimi (FUQAC) are greatly appreciated.

Journée des étudiants – REGALanode-grade petroleum coke supply has not
For the last few years, the available

Durant les dernières années, le stock disponible du coke de pétrole de « qualité
anode » n’est pas suffisant pour la demande : on mélange donc des cokes de
qualités diverses pour la production d’anodes. Par conséquent, la recette
des anodes doit être optimisée en fonction de la disponibilité des matières
premières afin de maintenir ou même d’améliorer la quali 0