Mark Zirnsak and Dirk Stegink

Comalco Research and Technical Support

PO Box 316

Thomastown, Victoria

Australia, 3074

Alfredo Teodosi

EurAllumina s.p.a

PO Box 64

09010 Portoscuso (CA), Italy


The EurAllumina refinery (Italy) had a problem with predesilication tanks becoming filled with stagnant thick mud in a period of approximately six months. A new multiple impeller agitation system was installed in an attempt to extend the time needed between cleaning the tanks. This did not address the problem, with the tanks subsequently requiring cleaning within a period of one to two months. The design criteria were reviewed and measurements made of the rheological characteristics of the slurry. The slurry rheology was found to be highly time dependent and it is believed this behaviour contributed to the performance problems.

Operational procedures were also examined and were identified as having contributed to the problem. Further, these procedures were not considered in designing the multiple impeller agitation system. In particular, a tank being brought on-line would be filled over a 24 hour period, resulting in a significant period of time in which the predesilication slurry in the tank was not agitated nor pumped forward to the digesters. Rapid filling of the tank at start up, changing the position of the inlet and outlet pipes, varying tank levels regularly and increasing the agitation speed has resulted in the tanks being relatively clean after a period of three months operation. This project has highlighted the caution needed when implementing an experimentally tested design and the need to consider operational practice in modifications to a plant.


predesilication, agitation, rheology, slurry, tank service life



Mark Zirnsak, Dirk Stegink and Alfredo Teodosi


The EurAllumina refinery in Italy have had ongoing problems with predesilication tanks becoming filled with thick stagnant mud. The original design of the tanks included a single impeller agitation system, which allowed stagnant thick slurry to build up during operation and within a period of six months fill most of the tank volume. The tank would then need to be taken out of service and cleaned. The cleaning operation was both costly and hazardous.

To solve the problem EurAllumina worked with Comalco Research & Technical Support (CRTS) to identify a means of maximizing the operational time between cleaning cycles. The CSIRO Division of Building, Construction and Engineering was engaged to model the flow within predesilication tanks and recommend an improved agitator design. Operational aspects of the predesilication tanks, such as tank start-up, were not considered at this stage of the project. The predesilication tanks were assumed to be well-stirred reactors with constant rheology slurry for the purposes of the physical modelling work. A three impeller agitation system (Figure 1) was found to give thorough mixing throughout the tank with minimal stagnant zones for a model fluid representative of the highest rheology slurry expected during operation. The impellers were predicted to be capable of operating at a lower speed than for the one impeller arrangement, which would deliver the added benefit of a power saving.


Schematic diagram of three impeller agitation system

EurAllumina installed the three impeller agitation system in one of the predesilication tanks on the refinery in April 1996. After three months of operation, it was concluded that the new agitation system had been successful because only a 0.7 m high compacted layer of thick mud was detected in the bottom of the tank. On that basis the three impeller agitation system was installed in a second tank, which was brought on line in October 1996.

After six months the first tank with the three impeller agitation system had to be taken out of service, and it was found to be full of thick slurry. The second tank was found to have poor agitation after only four days of operation and was taken off-line after 70 days for cleaning. It too was found to be full of thick stagnant mud.


An initial assessment of the problem identified two possibilities for the poor performance of the newly installed agitation system:

  • the impellers installed were slightly smaller in diameter (by 2.4%) than those recommended by the CSIRO study; and
  • the rheology of the predesilication slurry in the tanks was higher than the rheology the agitation system design had been based on.

In support of the latter possibility was data from plant measurements made in 1989, which showed the predesilication slurry at that time had a rheology up to an order of magnitude higher than the model fluid the agitation design had been based on (Figure 2).

Figure 2

Comparison of design rheology of predesilication slurry to 1989 data

To investigate these possible causes, EurAllumina and CRTS conducted an extensive series of rheology measurements on the predesilication slurries at EurAllumina in May and June 1997. At the same time CRTS conducted trials in a model tank to determine if the slightly smaller impeller size contributed to the poor agitation in the predesilication tanks.

A portable Haake VT550 viscometer was used to conduct the rheology measurements. The Haake VT550 had wide-gap cup-and-bob geometry to measure viscosity of slurries, with the predesilication slurries having a particle top size of 1 mm. The viscometer also had the ability to measure yield stresses of slurries directly by using a vane. The yield stress is defined as the stress corresponding to transition from elastic to plastic deformation (Barnes et al. 1996), or in practical terms the stress that must be applied to cause slurry to flow. The vane test is detailed in Nguyen and Boger (1983).

The measurements revealed that the rheology of predesilication slurry at EurAllumina varied considerably from day to day or even hour to hour, but rarely was higher than the rheology of the fluid used to design the three impeller agitation system (Figure 3).

Flow visualisation tests in a 1:25 scale model tank showed that top to bottom fluid flow existed in the tank, with only small stagnant regions (Figure 4), even with the smaller diameter impellers installed at EurAllumina. The impeller Reynolds numbers used in the model were much lower than the impeller Reynolds numbers the full-scale tanks were operating at. The impeller Reynolds number is defined as:


where r is the fluid density, N is the impeller speed, D is the impeller diameter and h is the fluid viscosity. The viscosity of predesilication slurries and model fluids was fitted to a power-law model, given by:


where is the shear rate, K is the empirically determined consistency factor and n is the empirically determined shear-thinning index. The average shear rate is determined from:


where ks is an empirically determined constant for the impeller and ni is the number of impellers. The ks value is determined from the method used by Metzner and Otto (1957) to have a value of 13. The impeller Reynolds numbers in the model tank reached 1.0 103 before surging occurred in the tank. The full-scale tanks operating at 10 rpm have an impeller Reynolds number of 2.6 103, with no sign of surging.

Figure 3

Comparison of viscosity versus shear rate design data with that measured for

Predesilication slurries in 1997


Figure 4

Schematic of the flow patterns observed in the physical modelling

The rheology results of predesilication slurries at EurAllumina and physical modelling work both led to the conclusion that neither the slurry rheology nor the impeller diameter were causing the agitation problems.


With the initial hypothesis for the cause of the agitation problems proving negative, alternative possibilities for the short operating period of the tanks before they required cleaning were considered. Possibilities considered were placed into three categories:

  • related to slurry properties;
  • related to agitation; and
  • operations related.

Those related to slurry properties were:

  • The time dependent rheology of the slurry. Physical modelling work had been conducted with a time independent fluid, whereas it was recognized that the rheology of predesilication slurry changed with time.
  • Settling of larger particles in the slurry. It was postulated that larger particles were settling out at the bottom of the tank over time and building up the layer of thicker stagnant mud.
  • Shear migration of particles. It has been reported that particles can migrate in flow due to the application of shear (Graham et al. 1991, Hampton et al. 1997, Hogg 1994, Krishnan et al. 1996, Philips et al. 1992, Seifu et al. 1994). It was postulated that over time such migration, were it to occur in the tanks, could result in the build up of stagnant regions.

Possible causes related to agitation were:

  • Too much swirl in the tank, resulting in poor impeller performance and poor top-to-bottom flow throughout the tank. The fact that there are no baffles in the tank could contribute to swirl in the tank.
  • Inappropriate impeller type.

Operations related issues were:

  • Slow start-up procedure in the tanks, taking up to 24 hours to fill an empty tank to the point where agitation would commence.
  • Variation in rheology of the slurry coming into the tanks. Such variation may be the cause of poor mixing and formation of regions of different rheology in the tank.
  • Variation in the level of the tanks, which can be responsible for poor mixing at the top of tanks. Failure to maintain the fluid level within a certain operating range has been demonstrated to result in poor mixing in the top of tanks (Pullum et al. 1994).


After consideration of the factors listed above, a decision was made to investigate time dependent rheology of the slurry, which was seen to impact mainly on agitation and operation issues.

Bauxite from Weipa was ground and mixed with synthetic liquor. The size distribution of ground bauxite was coarser than the size distribution measured from mills at EurAllumina. The chemical composition of the two liquors is given in Table 1. The liquors had only minor differences.

Table 1

Chemical composition of the liquors used to make up simulated predesilication slurries


Liquor 1

Liquor 2

CS (g/l as Na2CO3)



Al2O3 (g/l)









TS (g/l)



The slurry was made up to 50% w/w in 250 and 500 ml plastic bottles. The bottles were mixed in a water bath with a rotating carousel at 100 C.

Bottles were withdrawn at different times from the bath and the viscosity versus shear rate behaviour and yield stress of the material was measured. The last sample was removed from the water bath 163 hours after the test began. Rheological properties were measured using a Haake RS 150 Rheometer. The viscosity of the simulated predesilication slurries was fitted to the power-law model (Equation (2).

Over the shear rate range examined, the consistency factor, K, can be used as an indication of the overall magnitude of the slurry viscosity. In fact K is the slurry viscosity at a shear rate of 1 s-1. It was found that K and the yield stress increased almost linearly with time for simulated predesilication slurries. A plot of yield stress with time is shown in Figure 5.

Figure 5

Plot of yield stress versus time for a 50% w/w solids predesilication slurry agitated at 100C

This result indicated that slurry trapped within stagnant and recirculation zones within EurAllumina’s predesilication tanks will quickly thicken to the point where the rheology exceeds the design capabilities of the tank agitation system. Thickened zones are highly likely to grow with time.

There were also clear implications for filling a predesilication tank. If slurry is fed to an empty tank, and it takes a significant period of time before the tank can start to be agitated, then the stagnant material in the tank will thicken. The yield stress of the material is likely to then be at a level where much of the material will remain stagnant after agitation is commenced. At the time of testing, it was taking up to 24 hours to fill a tank to the point where slurry could be agitated.

Furthermore, if predesilication slurry is left in a tank for a significant period before being fed through to digestion it will also thicken. The rheology of slurry may then exceed design. This situation is also important during filling of a tank.


The effect of a layer of stagnant slurry forming at the bottom of predesilication tanks, due to slow filling, was investigated in a model tank. Fluid velocities were measured in a 1:25 scale model of the predesilication tanks. A solution of 0.097%w/w Carbopol 940 polymer in de-ionized water was used to simulate predesilication slurry. The model fluid rheology corresponded to the thicker predesilication slurries measured at EurAllumina.

The model tank was operated with the ratio of the clearance between the bottom impeller and the tank bottom to the tank diameter set at values of 0.19, 0.16 and 0.063. The level in the tank was lowered so that clearance between the top impeller and the fluid surface was constant.

The tank was operated at 223 rpm (3.7 Hz). The impeller Reynolds number at that speed was 739, whereas the full-scale tank operated at an impeller Reynolds number of approximately 2.6 103 for a full-scale speed of 10 rpm.

Velocity profiles were measured in the model tank using Particle Image Velocimetry (PIV). The PIV system used a Spectra Physics PIV 400 laser system consisting of two Nd:YAG lasers with an output light wavelength of 532 nm (green light) to illuminate the tank. A Kodak Digital Camera capable of cross-correlation is used to capture images and a Standford Research signal generator was used to control firing of the lasers and the camera capturing the images. The PIV system operates by determining tracer particle displacement over time using a double-pulsed laser technique. The signal generator induces the first laser fire a pulsed light sheet to illuminate a plane in the flow such that the camera captures the image. A fraction of a second later, the signal generator induces the second laser to illuminate the same plane and the camera captures the second image. By cross-correlation, computer software calculates particle displacement between two images and, through a knowledge of the time between images, it calculates the particle velocity for each tracer particle. The tracer particles used in this work were 80 to 90 m m alumina. PIV images were taken in the regions under each impeller to capture the axial and radial velocities of the fluid.

By integration of the velocity profiles measured it was possible to calculate flowrates under the impellers.

It was found that as the clearance between the bottom impeller and the tank bottom decreased the amount of pumping by the bottom impeller decreased (Figure 6). The result indicated that allowing a stagnant layer of thick slurry to form during the process of filling tanks could be detrimental to flow patterns in the tank for the entire operating period.

Figure 6

Plot of flowrate under each impeller against the ratio of the clearance between the bottom impeller and the tank bottom to the tank diameter


Following the finding of a strong time-dependence of predesilication slurry rheology and that a stagnant layer in the bottom of the tank has a negative impact on the flow within the tank, EurAllumina made a number of changes to operation of their tanks:

  • the impeller speed was increased from 10 rpm to 16 rpm;
  • feed and recycle slurry inlets were placed at the tank bottom, instead of at the top, in a way such that entering fluid sweeps the tank bottom;
  • regular significant variation of tank level was implemented to stop build up of stagnant material at the slurry surface; and
  • filling tanks from empty was conducted within 12 hours.

These changes have meant that tanks have been operated for up to three months at a time and when taken out of service are relatively clean of thick stagnant slurry (a layer of 0.5 to 0.6 m of stagnant slurry in the bottom). Thus, the loss of active tank volume to stagnant regions has been decreased and cleaning operations can be conducted quickly and with far less risk to safety than in the past.


Poor agitation in predesilication tanks at EurAllumina was caused by a failure to include the time dependent rheology of the slurry and operational aspects when considering a new agitator design. The project has highlighted the need to consider the limitations of experimental modelling and the importance of operational practice when implementing design modifications to a plant.

Further work is still needed to optimise the agitation system and tank operations to maximize time required between cleaning of tanks.


The authors would like to acknowledge contributions made to the experimental work and analysis by Lucia Caddeo, Emanuele Contu, Tassy Grosdanis, Richard Hassall, Franco Provenzale and Graham Stanton.

The authors would also like to acknowledge the contribution of CSIRO Division of Building, Construction and Engineering to the earlier design work.


Barnes, H.A., Hutton, J.F., and Walters, K. (1996), An Introduction to Rheology, Elsevier, Amsterdam.

Graham, A.L., Altobelli, S.A., Fukushima, E., Mondy, L.A., and Stephens, T.S. (1991), Note: NMR imaging of shear-induced diffusion and structure in concentrated suspension undergoing Couette flow, J. Rheol. 35(1), pp. 191-201.

Hampton, R.E., Mammoli, A.A., Graham, A.L., and Tetlow, N. (1997), Migration of particles undergoing pressure-driven flow in a circular conduit, J. Rheol. 41(3), pp. 621-640.

Hogg, A.J. (1994), The inertial migration of non-neutrally buoyant spherical particles in two-dimensional shear flows, J. Fluid Mech. 272, pp. 285-318.

Krishnan, G.P., Beimfohr, S., and Leighton, D.T. (1996), Shear-induced radial segregation in bidisperse suspensions, J. Fluid Mech. 321, pp. 371-393.

Metzner, A.B., and Otto, R.E. (1957), Agitation of Non-Newtonian Fluids, A.I.Ch.E. J. 3(1), pp. 3-9.

Nguyen, Q.D., and Boger, D.V. (1983), Yield Stress Measurement for Concentrated Suspensions, J. Rheol. 27(4), pp. 321-349.

Philips, R.J., Armstrong, R.C., Brown, R.A., Graham, A.L. and Abbott, J.R. (1992), A constitutive equation for concentrated suspension that accounts for shear-induced particle migration, Phys. Fluids A 4(1), pp. 30-40.

Pullum, L., Welsh, M.C., Hamilton, N., Bailie, K., and Kam, P. (1994), The Use of a non-Newtonian Fluid to Visualise the Mixing of a Psuedo-Homogeneous Slurry, Proc. 1994 ASME Fluids Eng. Div. Meeting, Lake Tahoe, NV, USA, pp. 207-214.

Seifu, B., Nir, A., and Semiat, R. (1994), Viscous dissipation rate in concentrated suspensions, Phys. Fluids 6(9), pp. 3189-3191.