You are on page 1of 9

DYNAMIC MODELLING OF AN OPERATING ALUMINIUM REDUCTION CELL

C. Y. Cheung, C. Menictas, J. Bao, M. Skyllas-Kazacos and B. J. Welch

School of Chemical Engineering The University of New South Wales UNSW, Sydney NSW 2052, Australia j.bao@unsw.edu.au ABSTRACT The operating conditions in the Hall-Hroult process used for aluminium production have become more critical in recent years due to need to optimise current efficiency and minimise the energy consumption per kilogram of aluminium produced. The tightening of operating parameters makes it more difficult for existing control systems employing global control practices to prevent the increase in the frequency of abnormal conditions occurring. Therefore, the development of control systems with fault diagnosis capabilities becomes more important. It is desirable for such system to be based on dynamical mathematical models so that the root causes of anomalies can be determined and more targeted control action can be made. This paper presents a thermal model that can estimate regional thermal behaviours of the aluminium reduction process based on anode current distribution in an operating cell. Its capability of predicting local thermal variations is important for effective fault diagnosis of localised abnormalities. The thermal model can then be combined with mass and voltage model for the future development of fault diagnosis system for the Hall-Hroult process. The dynamic model employs heat balances to determine local bath temperature and global ledge thickness. The mathematical model is intended to be employed in a fault detection system that can be used to patch existing controllers. Such a modified control system is aimed to have the capability of providing early indications of impending abnormal conditions on a localised level. INTRODUCTION Process Description The Hall-Hroult process is the major commercial method for the industrial scale production of aluminium (Habashi, 2002, Brooks et al., 2007, Choate et al., 2005). The main reaction produces liquid aluminium via an electrochemical reaction at a high operating temperature, described by the following chemical equation (Thonstad et al., 2003).
Al 2 O 3(dissolved) + C Al (l) + CO 2(g)

This reaction is carried out in a molten cryolite base bath. The bath is highly corrosive at the high operating temperature and thus no commercially available refractory materials can offer prolonged resistance against such corrosion (Tikasz et al., 1994). In

C. Y. Cheung et. al. order to extend the service life of a reduction cell, the cell operating condition allows a layer of frozen cryolite to form along the sidewall of the reduction cell to protect it against the corrosive bath. This layer of frozen cryolite is commonly referred as the ledge. Its thickness is determined by a thermal driving force in the bath called the superheat and is defined by: T = Tbath Tliq The bath temperature, Tbath depends on the amperage, the cell voltage and the path resistance. The liquidus temperature, Tliq is the melting point of the bath which is determined by the bath chemistry.

Coupled mass and thermal balance Changes in either or both bath and liquidus temperature can cause variation in the superheat. Its variation leads to freezing or melting of the ledge and consequently a change in thickness. The change determines the amount of cryolite removed from or released into the bath. As a result, the bath composition is altered and hence so is the liquidus temperature (Grjotheim & Kvande, 1993, Kan et al., 2007). The change in superheat then again affects the ledge thickness, resulting in a different bath composition. This cycle is initiated when the mass and/or thermal balance in the reduction cell is disturbed. It continues until the system reaches a new steady state when the heat flow from the electrolyte is eventually balanced by the thermal resistance of the side ledge.
The interdependent nature between the mass and the thermal balance is one of the distinctive features in the Hall- Hroult process and only exists in the presence of the ledge (Iffert et al., 2005). Although the system can self-regulate if either of them is disturbed, their influences to the cell operation and performance are significant. The imbalance can induce abnormalities such as anode effect that cause an abrupt increase of cell voltage, more frequently and intensify their impacts. A prolonged presence of the mass and heat imbalance will inevitably deteriorate current efficiency, power consumption and productivity further (Kvande & Haupin, 2000, Berezin et al., 2007, Kan et al., 2007, Bearne, 1999). Thermal models built for control systems often take into account the coupled feature, however they are not capable of showing local conditions caused by regional input and localised abnormalities. It is therefore important to develop a thermal model that can capture the coupled dynamics between the mass and thermal balance, and more importantly the local variations for effective fault diagnosis.

Limited process data and continuous measurements The corrosive nature of the bath also introduces challenge to cell monitoring and control. The harsh environment restricts the number of measurements that can be carried out for most of the cell parameters, resulting lack of available data (Bearne, 1999, Tikasz et al., 1994, Kols, 2007, Sorheim & Borg, 1989). Cell monitoring and control therefore mostly rely on cell voltage and amperage as they are viable to be measured continuously (Kvande & Haupin, 2000, Homsi et al., 2000). Bath temperature and superheat are also measured for reference purposes but are often carried out at one location intermittently as labour and equipment costs are too high for frequent

C. Y. Cheung et. al. measurements (Bearne, 1999). They are often used to represent the thermal condition of the entire reduction cell for a period of time before the next measurements. This assumes the condition is invariant regardless of the cell location and time. In order words, this assumption ignores localised conditions that can arise in different part of the cell. As those conventional measurements only reflect the overall dynamics of a process, they have limited ability to observe local variations. Signals of individual anode currents are incorporated with the model development presented in this paper in order to obtain spatial information for the development of fault diagnosis system. The introduction of anode current measurement is intended to compensate for insufficient data results from traditional non-continuous measurements. This paper will focus on the development of the thermal dynamic model that can provide the foundation for fault diagnosis capabilities for the Hall Hroult process.

Mathematical models Thermal models of the aluminium reduction cell have been developed for research and cell control purposes. They capture different levels of detail and process dynamics to fit their own applications. Distributed parameter models are usually developed for the development and retrofit of the cell technology (Dupuis, 2002). Partial differential equations with a set of specified boundary conditions are solved to compute the conditions of the process at steady state. This type of model provides high level of detail and accuracy yet substantial simulation time and computational power are often required for one analytical solution. This limitation therefore restricts its application for real time processing.
Lumped parameter models simplify the partial differential equations of the system by discretising the infinite number of states into a finite number of parameters. They are capable of computing transient thermal behaviours of the process by solving ordinary differential equations with a fast response time. The level of detail in such model and the simulation time required can be compromised by selecting a suitable finite increment as needed. Most of the thermal mathematical models for the Hall- Hroult process found in literatures belong to this category (Tikasz et al., 1994, Taylor et al., 1996, Kols & Stre, 2009). They usually consist of several governing ordinary differential equations to describe the global variations in the process. Although models that capture fairly comprehensive aspects of the process on thermal balance are found in literature, transient models that compute local variations of conditions within the cell by utilising individual anode current signals for fault diagnosis process has not been found. The development of the model presented in this paper is therefore inspired by the potential of fault diagnosis for the process on a localised level, and also lack of reporting in literature.

MATHEMATICAL MODELLING Theoretical Background The bath temperature in the Hall-Hroult process is solely supplied and maintained by the heat generated, Qgen from electricity via the Joule effect. Its heat equation can therefore be described by;

C. Y. Cheung et. al.

C pV

dTbath = Qout Qin + Qgen dt

where , C p and V are the density, the specific heat capacity and the volume of the element. As the dynamic change of ledge thickness is driven by heat in, Qin and heat out, Qout to/from the ledge, the mass balance of the ledge is also included in the thermal model. The temperature at the interface between the ledge and the bath is assumed to be always at the liquidus temperature. The change of its mass, M ledge and thus the thickness can therefore be calculated by;

dM ledge dt

Qout Qin H f

where H f is the heat of fusion of cryolite.

The structure of the model In order to gain an insight of thermal conditions within the bath at a localised level, the bath is discretised into an array of elements corresponding to anode locations. It is further divided into several horizontal layers according to the cell lining and structure as they vary vertically along the reduction cell. The topology of one of the layers in the model for a cell consisting of 10 anodes is shown in Figure 1. It can be expanded according to the number of anodes implemented in a particular cell technology.

Fig. 1 Topology of the discretised bath for thermal modelling This paper presents the development of a two-dimensional model based on the proposed topology shown in Figure 1 above. It takes global inputs including liquidus temperature and ambient temperature to determine temperature distribution across the bath by incorporating the signals of individual anode currents inputs. The fundamental principle behind the computation of heat exchange for each zone is the same but energy terms

C. Y. Cheung et. al. included in each equation are slightly different. Each zone is characterised by surrounding environments and therefore its heat balance is described by a specific thermal dynamic equation. As a result, the model consists of 10 different types of ordinary differential equations at the first horizontal layer. This first layer locates in the most upper part of the bath surrounded by the thinnest cell sidewall. A sample equation of heat transfer in one of the bath zone surrounded by the ledge is given below. The last two terms account for the heat convection that exists between the bath and the ledge. The temperature, Tzone1 is calculated as a function of both global ledge thickness and local bath temperatures, Tx1, Tx2, Ty1 and Ty2 based on a specific geometry, Vzone 1 and the cross-sectional area A;

C pVzone 1

dTzone1 kA1 (Ty1 Tzone 1 ) kA2 (Tx1 Tzone 1 ) = + hA1 (Tzone 1 Ty 2 ) hA2 (Tzone 1 Tx 2 ) dt y x

where x and y are the increments of that particular zone in x and y directions with k and h being the thermal conductivity and the heat transfer coefficient respectively. The heat equation at each zone has a similar structure yet some equations have extra terms to take into account regional variations. One example is the heat equation of the zone where the point feeder located. It has extra terms to include the effect of local feeding and dissolution and thus the localised effects caused by regional inputs can be simulated.

RESULTS AND DISCUSSION


In order to show the impact of anode current distribution on bath temperature, two different sets of current distribution were employed as model inputs to compute temperature distribution under the same and unchanged simulation condition. The first set of current inputs for all anodes was set to be 11500A, remained unchanged during the simulation. The resulting cell conditions at the first layer of the cell are shown in Figures 2 and 3 respectively. A cell consists of twenty anode blocks has been used to study in the simulation. Blocks that represent anodes are labelled with their corresponding number.

C. Y. Cheung et. al.

Fig. 2 Temperature distribution computed based on first set of model inputs


960 959 958 Bath temperature (degree C) 957 956 955 954 953 952 951 950 0 2 4 6 8 10 Simulated time (hour) 12 14 16

Fig. 3 Dynamics of bath temperature at the corner of the cell (Simulation case 1) The second set of inputs to the model was changed to simulate uneven current distribution with the amperage being the same as in the first simulation. The current running down both anode 3 and anode 18 becomes 25000A while the rest is set at 10000A. The simulated results are shown in Figures 4 and 5 respectively.

C. Y. Cheung et. al.

Fig. 4 Temperature distribution computed based on second set of model inputs


955 954.5 954 953.5 953 952.5 952 951.5 951 950.5

Bath temperature (degree C)

3 4 5 Simulated time (hour)

Fig. 5 Dynamics of bath temperature at the corner of the cell (simulation case 2) The temperature profiles obtained from both of the simulations illustrate that the change in current distribution can cause local variations on the bath temperature even though the amperage remains unchanged. Its impact becomes more significant when the

C. Y. Cheung et. al. situation persists as shown in Figure 5. Note that heat generation from bubble resistance via Joule effect has not yet been included in the two-dimensional thermal simulation; otherwise a greater difference in the local temperatures would result. This model does not take into account the change of bath composition involved during the freezing and melting of ledge. Impacts of ledge dynamics to the bath composition and the thermal balance in the bath will be included in future work.

CONCLUSION AND FUTURE WORK


The model presented in this paper demonstrates the ability to simulate local variations of thermal conditions in an aluminium reduction cell imposed by current distribution. Future development of the model will be in the direction of improving the existing model, combined with mass and voltage models, and develop a fault diagnosis system based on an integrated thermal, voltage and mass balance model.

ACKNOWLEDGEMENT
This project is supported by the CSIRO Cluster on Breakthrough Technologies for Aluminium Reduction.

REFERENCE
Bearne, G. 1999. The development of aluminum reduction cell process control. JOM Journal of the Minerals, Metals and Materials Society, 51, 16-22. Berezin, A. I., Piskazhova, T. V., Gritsko, V. V., Tarakanov, A. V., Volokhov, I. N. & Polyakov, P. V. 2007. Bath superheat to control electrolysis process. Brooks, G., Cooksey, M., Wellwood, G. & Goodes, C. 2007. Challenges in light metals production. Mineral Processing and Extractive Metallurgy : IMM Transactions section C, 116, 25-33. Choate, W. T., Aziz, A. & Friedman, R. Year. New technology will sustain the U.S. primary aluminium industry. In: TMS Light Metals, 2005. 495 - 500. Dupuis, M. Year. Using ANSYS to model aluminum reduction cell since 1984 and beyond. In: International ANSYS conference, 2002. Grjotheim, K. & Kvande, H. 1993. Introduction to aluminium electrolysis: understanding the Hall-Heroult process, Aluminium-Verlag, Dusseldorf. Habashi, F. 2002. Hall and Heroult and the production of aluminium. CIM Bulletin, 95, 109-113. Homsi, P., Peyneau, J.-M. & Reverdy, M. 2000. Overview of process control in reduction cells and potlines. TMS Light Metals. Iffert, M., Skyllas-Kazacos, M. & Welch, B. 2005. Challenges in mass balance control. Kan, H., Wang, Z., Shi, Z., Ban, Y., Cao, X., Yang, S. & Qiu, Z. 2007. Liquidus temperature, density and electrical conductivity of low temperature electrolyte for aluminum electrolysis. Kols, S. 2007. Defining and verifying the correlation line in aluminum electrolysis. JOM Journal of the Minerals, Metals and Materials Society, 59, 55-60. Kols, S. & Stre, T. 2009. Bath temperature and AlF3 control of an aluminium electrolysis cell. Control Engineering Practice, 17, 1035 - 1043. Kvande, H. & Haupin, W. 2000. Cell voltage in aluminum electrolysis: A practical approach. JOM Journal of the Minerals, Metals and Materials Society, 52, 3137. Sorheim, E. A. & Borg, P. 1989. Dynamic model and estimator for online supervision of the alumina reduction cell. TMS Light Metals. 8

C. Y. Cheung et. al. Taylor, M. P., Zhang, W. D., Wills, V. & Schmid, S. 1996. A dynamic model for the energy balance of an electrolysis cell. Chemical Engineering Research and Design, 74, 913 - 933. Thonstad, J., Fellner, P., Haarberg, G. M., Hives, J., Kvande, H. & Sterten, A. 2003. Aluminium electrolysis : fundamentals of the Hall-Hroult process, Dsseldorf : Aluminium-Verlag. Tikasz, L., Bui, R. T. & Potocnik, V. 1994. Aluminium electrolytic cells: A computer simulator for training and supervision. Engineering with Computers, 10, 12-21.

BRIEF BIOGRAPHY OF PRESENTER


C. Y. Cheung received a B. E. (first class honours) in Industrial Chemistry in 2008 from the University of New South Wales. She is currently a PhD candidate at the same university. Her research areas focus on mathematical modelling on aluminium smelting processes and fault diagnosis.

You might also like