Jomar Thonstad*, Ove Kobbeltvedt** and Sverre Rolseth**

*Department of Electrochemistry, Norwegian University of Science and Technology,

7034 Trondheim, Norway

**SINTEF Materials Technology, 7034 Trondheim, Norway


A laboratory study was made of the dissolution characteristics of alumina in cryolite melts. The most important feature is the tendency of alumina to agglomerate, which is associated with the conversion from gamma alumina (g ), into alpha alumina (a ). The dissolution process is mass transfer controlled, and dispersed alumina dissolves very rapidly (seconds), while agglomerates dissolve slowly (minutes). Dissolution tests were designed to simulate the conditions in industrial cells with point feeding. High superheat and preheating of the alumina promoted rapid dissolution. To study the agglomeration phenomenon loose alumina of commercial low-a quality was placed in a steel frame lined with aluminium foil, and immersed into alumina-saturated cryolite bath. It was found that sintering of the alumina particles took place while heating between 700 C and 950 C. When the alumina was permeated by molten bath, the alumina body still retained its shape. With alpha alumina this was not the case. This shows that the g a phase transformation plays an important role in the dissolution process.


alumina, dissolution, agglomeration, sintering, phase transformation


Jomar Thonstad*, Ove Kobbeltvedt** and Sverre Rolseth**


When a batch of alumina is being fed to aluminium cells, rapid and complete dissolution is desirable in order to control the concentration of alumina in the bath and to avoid anode effects. Undissolved alumina may settle at the bottom of the cell forming so-called sludge, which is unwanted for several reasons. The introduction of point-feeding represented a major improvement with respect to alumina dissolution, compared to the centre or side bar-breaker feeding systems. Point-feeding is today the preferred feeding technique for prebake cells, and it has also been adopted for Soderberg cells (Paulsen et al., 1997, Jensen et al., 1998).

However, the problems of incomplete alumina dissolution and sludge formation are not eliminated by conversion to point feeding. Even a point feeder drops a fairly large batch of alumina (of the order of 1 kg) on to a restricted area of bath surface. When the cold alumina hits the bath, the alumina spreads out, and some bath freezes on to the alumina. This results in the formation of flake-shaped agglomerates. The formation of such agglomerates strongly reduces the contact area between alumina and bath compared to what would be the case if all the alumina grains were dispersed in the bath. A large contact area between alumina and bath is obviously important in order to ensure rapid heating and dissolution of the alumina, since the process has been found to be mass transfer controlled (Thonstad et al., 1988).


So-called sandy, low-calcined aluminas, with a low content of the a phase (corundum), are today preferred for use in aluminium production. As first reported by Less (1977), alumina with an initial low a content converts to a and tends to form a network of interlinked a -alumina platelets upon addition to cryolite melts. In the following this phenomenon will be called interlinking or sintering.

The formation of such networks impedes the liberation of single alumina grains as the molten bath penetrates the alumina. It is well established that aluminas with a low content of the a phase form harder top crusts compared to aluminas with a high a content (Winkhaus, 1970, Keller, 1984, Rolseth et al., 1986). It is therefore generally accepted that the process of sintering of the alumina grains is related to the transition of the metastable g phases to the stable a phase. It is also well known from many studies that fluorides catalyze this phase transformation (Kachanovskaya et al., 1971, Oedegard et al., 1985). With fluorides present the g a conversion can take place at temperatures down to 800 C (Oedegard et al., 1985) or even lower, as shown in the following. The fluorides can be present as adsorbed fluorides in secondary aluminas or in the bath which permeates into the alumina.

The presence of alumina lumps infiltrated with bath, here named alumina agglomerates, is the most important factor limiting the dissolution rate of alumina in cryolitic baths. The purpose of the present measurements was to elucidate the mechanisms of the formation and dissolution of such agglomerates. In the first part of this paper experiments are described where the dissolution of alumina in cryolite melts was studied in a laboratory apparatus designed to simulate the conditions in industrial cells with point feeding. In the second part of the paper the agglomeration process was studied in a special apparatus where the agglomeration was followed as a function of heating rate and temperature.



As mentioned above the process of alumina dissolution in cryolite melt involves the formation and break-up of agglomerates. In a laboratory cell for studies of alumina dissolution, the convection pattern in the melt should be close to that in industrial cells. This is difficult to attain on a laboratory scale. In industrial cells the anode gas sets up strong convection, with bath rising close to the sides of the anode and flowing down again in the middle of the channel between two anodes. At the same time the bubble release generates an undulating bath surface.

The laboratory apparatus was made to combine the effects of bath convection and the undulating bath surface. Convection in the bath was accomplished by mechanical stirring with a graphite impeller. Bubbling of argon gas through the bath simulated the surface undulations caused by bubble release. A description of the gas stirrer arrangement is given elsewhere (Rolseth et al., 1994).

Figure 1 shows a sketch of the experimental arrangement. The inner diameter of the crucible was 20 cm. The amount of bath was 6500 g, and the bath composition was 10 wt% AlF3, 5 wt% CaF2 and 2 wt% Al2O3 (initial concentration), the balance being cryolite (Na3AlF6). The liquidus temperature was measured before each run by recording the cooling curve of the bath while the heating elements of the furnace were turned off and the bath was stirred vigorously.


Figure 1

Sketch of the crucible used for dissolution experiments. Inner diameter of the graphite crucible:

200 mm

The alumina was added in one batch, since the intention was to simulate the operation of a point feeder. The batch size was 0.45 g/(cm2 bath surface) corresponding to an increase in alumina concentration in the bath of 2.2 wt%. In some experiments the alumina was preheated so that it had a temperature of approximately 600 C when it was added to the bath (Kobbeltvedt et al., 1996).

The concentration of alumina was measured in situ by a modified linear sweep voltammetric method, as well as by analyzing bath samples which were collected during the dissolution experiment and analyzed for alumina on a LECO RO-336 Oxygen Determinator. The principle of the voltammetric method is to correlate the so-called critical current density with the alumina concentration in the bath. The method has been described by Haverkamp et al. (1992), and it has been used extensively for dissolution studies by Professor Welch and his group (e.g. Kushel and Welch, 1991). The voltammetric method allows the alumina concentration to be monitored as frequently as every second. The temperature was measured during the experiment using a thermocouple (Pt/Pt10%Rh) protected by a steel tube immersed in the melt.

2.1 Studies of Agglomerate Formation

A procedure to study the temperature response in alumina after being immersed in the bath, was developed by Walker (1993). He formed cylindrical agglomerates in the bath of an industrial cell and recorded the temperature response within and outside the agglomerate.

In the present experiments the rig shown in Figure 2 was used to study agglomerate formation. The alumina rested in a steel cup where half of the side wall was replaced by aluminium foil. Three K-type thermocouples were located inside the alumina, positioned 2.5, 1.5 and 0.5 cm from the outer surface of the cylinder, and one thermocouple was located just outside the cylinder. The rig was immersed in a graphite crucible (20 cm i.d.) containing 7700 g bath, giving a bath height of 12 cm. The bath composition was cryolite with 10 wt% AlF3 , 5 wt% CaF2 , Al2O3(sat). Alumina saturation was preferred in order to avoid any dissolution of the agglomerate. The temperature measured in each thermocouple was logged every second.

Figure 3 shows a horizontal view of the rig. The cylinder was 8 cm high and had a diameter of 5 cm. The aluminium foil quickly melted after immersion of the cylinder in the bath, exposing the alumina to the bath. Convection in the melt was provided by a graphite impeller (2.5 cm dia., with four blades 3 cm high) which was positioned at a distance of 7 cm from the outer surface of the agglomerate. The stirring rate was either 200 or 400 rpm.

Figure 2

Illustration of the equipment used to record the thermal response in alumina after being immersed in a cryolite bath


Figure 3

Horizontal cross section of the rig for agglomerate formation

In the treatment of the results from these measurements it was convenient to define a new liquidus temperature, i.e. the temperature at which both Na3AlF6 and Al2O3 precipitate (i.e. the eutectic), defined as Tliq*. Prior to the experiment a cooling curve of the melt was recorded in the same way as described above. A change in slope was observed at the liquidus point (in this case the eutectic).

135 g primary alumina was kept at 200 C for 3 to 4 hours to reduce the moisture content before it was poured into the cylinder. The rig was lowered into the melt and kept there until all thermocouples had reached the temperature of the bulk of the bath. The rig was then removed from the bath, and the content of the cylinder was inspected for dimensional stability.


3.1 The Dissolution of Alumina Added Batch-Wise

When alumina was added to the cell in Fig. 1, the dissolution process could be divided into at least two stages, as indicated in Fig. 4. The figure shows the content of alumina dissolved in the bath as a function of time after addition of a batch of alumina. In this particular experiment the alumina had been preheated to 600 C, the stirring rate of the impeller was 200 rpm., the gas bubbling intensity was 2.5 Nml/min(cm2 bath surface) and the initial superheat was 10 C (temperature above the liquidus).

Figure 4

Typical dissolution curve for batch-wise feeding of alumina to a cryolite melt. Both sweep data and LECO analysis data are given. The straight lines indicate the changes in dissolution rate during the dissolution process. Average rates of dissolution are indicated

As can be seen from the figure, the initial stage of the dissolution process exhibits a very high dissolution rate. This has been interpreted as rapid dissolution of dispersed alumina grains (Rolseth et al., 1994). Both measurements (Thonstad et al., 1972) and calculations (Thonstad et al., 1988) have shown that when the alumina is present in the bath as dispersed grains, it dissolves within approximately 10 seconds.

The remaining part of the batch dissolves at a remarkably slower rate, because of agglomeration of the alumina. Formation of agglomerates drastically reduces the contact area between alumina and bath. There is, however, still a significant change in the dissolution rate from the beginning to the end of this period, which probably reflects a diminuition of the contact area between bath and alumina..

3.1.1 Interruption of the Gas Induced Stirring

A series of experiments was conducted where the alumina was added to a quiescent bath surface with no stirring and no gas bubbling. After a given time the gas bubbling was resumed.

Primary as well as secondary aluminas were used. The time the stirring was interrupted was varied between 0, 0.25 and 3 minutes. It was observed that primary alumina was more affected by the lack of agitation than secondary alumina. When primary alumina was allowed to rest on a quiescent bath surface, stronger agglomerates were formed compared to the case when the surface was subjected to wave action. This resulted in significantly slower dissolution.

These differences are probably related to agitation caused by the release of volatiles in secondary alumina. Small "volcanoes" were observed at the surface of the alumina during the first seconds after addition. This agitation of the alumina grains in the period when the phase transition from g to a phase takes place, may prevent a firm interlocking of the newly formed grains of a alumina.

A high superheat in the bath prior to addition of the batch turned out to be most beneficial for the total dissolution process. The increasing dissolution rate with increasing superheat probably reflects the importance of heat transport for the process of agglomerate disintegration, as discussed in a previous publication (Kobbeltvedt et al., 1996).

3.2 Agglomeration Studies

It has tacitly been assumed that the grains have time to sinter together during the time the alumina grains are cemented together by the frozen bath, but this has so far not been verified. However, the dissolution studies have given strong indications of agglomeration of alumina grains taking place when alumina was being fed to cryolite melts, but the extent of agglomeration and the factors affecting it were not clear. Therefore, two sets of experiments were undertaken, firstly agglomeration (sintering) of secondary alumina upon heating was studied, and secondly agglomeration of alumina immersed in alumina-saturated melts was investigated.

Agglomeration of alumina upon heating

A series of simple experiments was performed by heating secondary alumina in the absence of bath to temperatures in the range 600 to 950 C, where interlinking of alumina grains was expected to take place.

Alumina was placed in a crucible made of sintered alumina (i.d. 2.5 cm, height 4 cm) in an open furnace which was heated to a predetermined temperature. Secondary alumina was used to ensure the presence of fluorides to catalyse the phase transformation reaction (Oedegard et al., 1985). A K-type thermocouple was immersed into the alumina, and the temperature was recorded every second. After a predetermined time the thermocouple was taken out and the crucible was removed from the furnace. The alumina was then inspected to assess the extent of aggregation.

In order to determine the lowest temperature at which interlinking could occur, a series of experiments was run at low temperatures (500 to 725 C). Below 660 C long term testing was performed, with exposure times of hours or days. In Table 1 the times of exposure at the various temperatures are listed. The extent of aggregation is also given in the table, where "no aggregation" means that the alumina could be poured out of the crucible as a free-flowing powder. "Traces of aggregation" means that extremely weak interlinking could be found, i.e. the "lump" disintegrated by a light touch.

        Table 1

        Times of exposure of secondary alumina at various temperatures in low temperature tests

        Test no

        Temperature (C)

        Time for exposure

        Extent of aggregation



        3 days

        No aggregation



        2 - 3 hours

        No aggregation



        1 hour

        Traces of aggregation



        10 minutes

        No aggregation



        10 minutes

        Traces of aggregation



        8 minutes

        Traces of aggregation

        Below 650 C no interlinking was found. The lowest temperature at which any interlinking was found for exposure times of the same order of magnitude as the times of alumina dissolution (Fig. 4), was 700 C.

        In the high temperature tests the furnace was preheated to a temperature in the range 900 to 1100 C. The temperature in the furnace determined at which rate the alumina would be heated. The times elapsed in heating from one temperature level to another are listed in Table 2. The extent of aggregation is classified by using Roman numerals. Grade IV means that the alumina stuck to the crucible as one coherent lump, so that a tool was needed to remove it. Grade II means that the alumina near the bottom and the walls of the crucible had agglomerated, but with lesser mechanical strength than for grade IV. At the top of the crucible the agglomerate had very little mechanical strength for grade II, and it crumbled into loose alumina when touched.

        Table 2

        Times of exposure of secondary alumina at various temperatures in the series of high temperature tests. The heading "total exposure time" encompasses the total time above 700 C

        Test no

        Temperature cycles (C)

        Total exposure

        time (s)

        Extent of aggregation


        700 940 900




        700 815




        700 930




        700 910



        In test no 7 the temperature in the alumina peaked at 940 C before it settled at the furnace temperature, which was 900 C. This was attrributed to the extra heat generated by the phase transformation (g a ). Gerlach (1983) has made a similar observation.

        To summarise the results from the high temperature tests, loose aggregates were formed when secondary alumina was exposed for 3 to 4 minutes in the temperature range of 700 to 940 C. Hard agglomerates were formed at exposure times of 10 minutes or more in this temperature range.

      1. Agglomeration of Alumina Immersed in Cryolite Melts

Figure 5 shows the recorded temperature response in the alumina when the rig shown in Fig. 2 was immersed in a bath kept at 962 C, which was 10 C above Tliq*. The stirring rate of the impeller was 200 rpm. The alumina retained its dimensional stability after removal from the bath, meaning that thermocouples no 2, 3 and 4 were firmly embedded in alumina agglomerate.

Figure 5

Temperature response in a cylinder of alumina after being immersed in the bath. The numbers on the curves refer to the thermocouples (Figure 2) recording the temperature. The bath temperature before immersion was 962 C and the stirring rate of the impeller was 200 rpm

The figure shows that the heat transfer into the alumina body was slow in this case. The bath temperature was not reached in all thermocouples until 15 minutes after immersion. As thermocouple no 2 reached 740 C, the slope of the curve increased markedly. This can probably be attributed to the start of infiltration of bath at this position. The thermal conductivity of the body will increase as bath displaces air between the alumina grains (Rye, 1992). Furthermore, as bath starts penetrating the alumina powder, heat will be transferred by convection in addition to conduction.

The sudden rise in temperature seen in thermocouple no 4 after 200 s is probably connected to some movement of the alumina as air or water vapour is being expelled. Thermocouple no 4 shows an overshoot in temperature, which can be attributed to the extra heat generated when the exothermic phase transformation from g to a phase takes place (Gerlach, 1983).

The fact that the alumina agglomerate retained its dimensional stability at a temperature where all bath was liquid, can only be explained by interlinking or sintering of the alumina grains. As shown above the lowest temperature at which interlinking of "pure" secondary alumina was observed with similar exposure times as in this study, was 700 C. In studies of crust formation Rye (1992) found that the temperature at the bath front was approximately 700 C. This implies that the phase transformation and interlinking of grains are initiated already when the first bath penetrates into the alumina. The interlinking is a process involving mass transfer, and it requires that the alumina grains must remain in physical contact for some time for interlinking to occur. Therefore, the time of heating from the temperature where interlinking can start to the time when all bath is liquid, Tliq*, should be important for the mechanical strength of the agglomerate.

The results from these experiments can be used to find the heating time between these two temperature levels (700 C and Tliq*) at three locations in the agglomerate. In the experiment described above the heating time between these temperatures was more than 8 minutes for thermocouple no 2. Thus, this is sufficient time for the formation of an interlinked network of alumina grains with adequate mechanical strength to withstand disintegration in the melt under these conditions.

As a control experiment, a test was made with 100% alpha alumina. This alumina was prepared by keeping the alumina at 1150 C in an open furnace for six hours prior to the experiment. The bulk bath temperature before immersion of the rig was 30 C above Tliq* , and the stirring rate of the impeller was 400 rpm. In this case Tliq* was reached after 6.5 minutes at the innermost thermocouple. As expected, no alumina was attached to the rig when it was lifted out of the melt 10 minutes after immersion, i.e. the alumina had not sintered at all.

In Table 3 are listed the times elapsed in heating between 700 C and Tliq*, where all bath should be liquid. The rate of temperature rise was practically constant in this range. It shows that when the bath which penetrated into the alpha alumina had melted, even for a retention

Table 3

The time needed for heating from 700 C to Tliq* for three locations in the alumina body. The thermocouple numbers refer to Figure 2

Type of




Time for heating 700 C Tliq* (s)

Thermocouple encapsulated

Low alpha




Low alpha




Low alpha




100% alpha




100% alpha




100% alpha




time of 219 s, the alumina body fell apart. For this alumina there was nothing to hold the grains in place after the solid bath phase had melted. For the low alpha alumina, however, a retention time as short as 58 s between these temperatures was sufficient to agglomorate the alumina. The results then show that one minute is sufficient time to form an interlinked network of alumina grains with adequate coherence to withstand disintegration.


    1. By dissolution of alumina in cryolite melts one part of the almina grains get dispersed and dissolves very rapidly (seconds), while the remainder forms agglomerates which dissolve very slowly (minutes).

2 Agglomeration is associated with the g a conversion.

3. This conversion occurs at temperatures above 700 C, and about one minute is sufficient time to form a coherent agglomerate.


The authors gratefully acknowledge financial support from the Research Council of Norway and from the Norwegian aluminium industry. They are also indebted to Rune Hovland and Lisbet Sten for participating in the experimental work.


Gerlach, J. (1983). Dissolution and Interaction between Alumina and Cryolite Melts. 1st Internat. Symp. Molten Salt Chem. Techn., Kyoto, Japan, Apr. 20-22, 1983, pp. 89-92.

Haverkamp, R.G., Welch, B.J., Metson, J.B. (1992). An Electrochemical Method for Measuring the Dissolution Rate of Alumina in Molten Cryolite. Bulletin of Electrochemistry, Vol. 8, pp. 334-340.

JANAF Thermochemical Tables (1975). 2nd ed., US Dept. Commerce, Nat. Bur. Stand., Washington 1971.

Kachanovskaya, I.S., Osovik, V.I., Kukhotkina, N.T. (1971). Phase Transformation and Alumina Dissolution in Cryolite. Sov. J. Non Ferrous Met., vol. 44, no. 10, pp. 40-42.

Kobbeltvedt, O., Rolseth, S., Thonstad, J. (1996). On the Mechanism of Alumina Dissolution with Relevance to Point Feeding Aluminium Cells. Light Metals 1996, pp. 421-427.

Keller, R. (1984). Alumina Dissolution and Sludge Formation. Light Metals 1984, pp. 513-518.

Kushel, G.I., Welch, B.J. (1991). Further Studies of Alumina Dissolution under Conditions Similar to Cell Operation, Light Metals 1991, pp. 299-305.

Less, L.N. (1977). The Crusting Behaviour of Smelter Aluminas. Met. Trans B, 8B, pp. 219-225.

Oedegard, R. Roenning, S. Rolseth, S., Thonstad, J. (1985). On Alumina Phase Transformation and Crust Formation in Aluminium Cells. Light Metals 1985, pp. 695-709.

Paulsen, K.A., Rolland, W.K., Larsen, A. (1997). Implementation of Point Feeding in the Soderberg Lines at Hydro Aluminium Karmy. Light Metals 1997, pp. 195-199.

Pedersen, T.B., Bentzen, H.J., Jensen, M., Larsen, W., Olsen, T. Pedersen, R (1998). Results from Implementation of Point Feeders and Complete Hooding on VS Soderberg Pots. Light Metals 1998, pp. 221-226.

Rolseth, S., Rnning, S., Solheim, A., Thonstad, J., degrd, R. (1986). Non-Conventional Method for Alumina Testing. Aluminium, Vol. 62, pp. 286-290.

Rolseth, S., Hovland, R., Kobbeltvedt, O. (1994). Alumina Agglomeration and Dissolution in Cryolitic Melts. Light Metals 1994, pp. 351-357.

Rye, K.. (1992). Crust Formation in Cryolite-Based Baths, Ph.D. thesis, Norwegian Inst. Technology, Trondheim, Norway.

Thonstad, J., Nordmo, F., Paulsen, J.B. (1972). Dissolution of Alumina in Molten Cryolite. Met. Trans. , vol. 3, pp. 403-408.

Thonstad, J., Solheim, A., Rolseth, S., Skaar, O. (1988). The Dissolution of Alumina in Cryolite Melts. Light Metals 1988, pp. 655-661.

Walker, D. (1993). Alumina in Aluminum Smelting and its Behaviour after Addition to Cryolite-Based Electrolytes, Ph.D. thesis, University of Toronto.

Winkhaus, G. (1970). On the Dissolution of Alumina in Cryolite Melts. Light Metals 1970, pp. 16-19.