PLANT IMPURITY BALANCES AND IMPURITY INCLUSION IN DSP

Geoff Riley1, Peter Smith2, David Binet1 and Russell Pennifold2

Alcoa of Australia Limited

PO Box 161, Kwinana, WA 6167

AJ Parker Cooperative Research Centre for Hydrometallurgy
CSIRO Minerals PO Box 90, Bentley, WA 6102

Abstract

The effect of liquor composition on the incorporation of sulphate, carbonate and chloride in desilication product (DSP) has been determined in a comprehensive series of laboratory tests. The DSPs were produced by reacting kaolinite and gibbsite in a wide range of synthetic liquors at temperatures and holding times corresponding to both Bayer predesilication and low temperature digestion conditions. The methodology used and the principal outcomes are described. The impurity incorporation depended on the desilication conditions; the DSPs produced under predesilication conditions contained less sulphate and aluminate but more chloride than those produced from the same liquors under digestion conditions. However the Na2O:SiO2 ratios for the digestion DSPs corresponded to full occupancy of the sodalite cages; the ratios for the predesilication DSPs were slightly lower. Sulphate incorporation was best described by a Langmuir type relationship with respect to sulphate concentration and with negative contributions from carbonate and chloride concentrations. Carbonate and chloride inclusions were better fitted by regression models involving linear dependence on the predicted sulphate content. Aluminate incorporation was insensitive to variations in alumina content of the liquors within either the set of predesilication or the set of digestion DSPs, but the average incorporation for the predesilication DSPs was slightly below that for the digestion DSPs. Fluoride and hydroxide levels in the DSPs were below their quantitation limits. The relationships obtained have a sound scientific basis and are now being used in predictive models for plant impurity concentrations.

Key Words:

desilication product, sodalite, impurity incorporation, slurry storage, digestion

 

PLANT IMPURITY BALANCES AND IMPURITY INCLUSION IN DSP

Geoff Riley1, Peter Smith2, David Binet1 and Russell Pennifold2

1.0 Introduction

1.1 Importance of DSP as a Sink for Bayer Liquor Impurities

Sodium carbonate, sulphate and chloride and to a lesser extent, fluoride are common inorganic impurities in Bayer liquors which can build up to appreciable levels sufficient to affect alumina solubility, reduce yield and increase density and viscosity, all of which contribute to reduced liquor productivity (Teas and Kotte, 1980; Lectard and Nicolas, 1983). Each of these impurities can be incorporated into DSP, and their removal by this route can account for a significant part of their exits from the liquor circuit. For example, DSP formation may account for around 75% of all the Na2SO4 exits from a typical Bayer circuit. Bauxite reactive silica levels are sometimes purposely increased for short campaigns, as a means of reducing liquor Na2SO4 concentrations.

Seimiya (1963) studied incorporation of inorganic impurities in DSPs produced under Bayer conditions, but his data was of limited value in developing quantitative relationships. The current work was undertaken to provide the quantitative data needed for mass balance calculations. To be able to predict steady state concentrations of inorganic impurities in plant liquors, DSP composition must be accurately quantified as a function of liquor composition and formation conditions.

1.2 DSP Fundamentals

DSP is a sodium alumino-silicate which precipitates from liquor following dissolution of reactive silica (mainly kaolinite) from bauxite. This occurs under slurry storage or digestion conditions. When formed under the conditions applying in Alcoa’s Western Australian refineries, DSP can be expected to have a sodalite-type structure with the following general formula:

Na6(Al6Si6O24).mNa2X.nH2O

where X=SO42-, CO32-, 2Cl-, 2F-, 2Al(OH)4- or 2OH- , 0 m 1 and n 8.

The sodalite crystal structure consists of a three dimensional network of alternating SiO4 and AlO4 tetrahedra. Lowenstein’s rule (Catlow et al., 1996, and references therein) states that, for hydrothermally produced zeolites, the framework Al:Si ratio  1. In the ideal sodalite structure, deduced from XRD studies (Hassan, 1983) the framework Al:Si ratio = 1. The framework encloses two cages per formula unit, with each cage containing 3+m sodium ions, to balance the negative charge of the framework and included anions. The cage structure of ideal sodalite, where X=2Cl- , m=1 and n=0, is illustrated in Figure 1.

Figure 1

Location of Na+ and Cl- in the classical sodalite structure (modified from Stroud et al, 1979)

At full occupancy (m=1), charge balance requires a total of 4 sodium ions per cage (Na2O:SiO2 = 4:6), with one singly charged anion (e.g. X= Cl-) in every cage or one doubly charged anion (e.g. X=SO42-) in every second cage. Cage dimensions limit the choices as to which anions X will fit within the cages and lead to an expectation that SO42-, CO32- and Cl- (but not OH- ) will be trapped once incorporated.

2.0 experimental methods and plans

2.1 DSP Formation and Washing Conditions

As well as firming up analytical procedures (see Section 3.0), the early parts of the project employed a series of preliminary experiments to decide details of raw materials, the conditions to use in DSP formation (temperatures, times and charge rates) and how to wash and prepare DSPs for analysis (avoiding leaching of impurities, or problems with CO2 loss/pickup in drying). Key decisions from this early work were:

    1. To use Eckalite kaolin (an English China Clay, very high in reactive silica) as our silica source, at a charge rate of 5gL-1 ReSiO2 (chosen as a compromise between the mass of DSP produced for analysis and the need to maintain small differences between start and finish liquor compositions).
    2. To use synthetic, organics-free, liquors (based on the minimal composition differences found between DSPs formed from Kwinana plant and matched synthetic liquors).
    3. To perform parallel DSP preparation runs, under digestion and slurry storage ("predesilication") conditions, using kaolin, gibbsite (if necessary) and liquor:

Predesilication = maintain temperature at 95 C for 18 hours in a bottle roller, then immediately filter, wash and dry.

Digestion = maintain temperature at 150 C in a gas fired reactor, for 60 minutes, then immediately quench, filter, wash and dry.

    1. To use a single batch wash of each experimental DSP, at the rate of 100mL of DI water per gram of dry solid (giving a final wash water total soda concentration of 2gL-1, as Na2CO3), followed by simple drying in air.

Point (c) is an important one deserving further explanation. Early experiments revealed important differences between predesilication and digestion DSPs formed from identical liquors. There was also evidence that DSPs formed in predesilication survive, essentially unaffected, through subsequent exposure to digestion conditions (see Section 4.6 for further details on this point).

2.2 Choice of Test Liquor Compositions

Once the decisions described in Section 2.1 had been made, two major campaigns of DSP preparation were carried out. These provided predesilication and digestion DSPs for a total of 91 liquors, working in a set order through two liquor composition matrices.

Table 1

Target Concentrations Covered by the Two Liquor Matrices

  Matrix 1 (excluding run 55) Matrix 2 (high TC, low impurity portion)
Na2SO4 g/L 0 – 2 – 4 – 4 – 7 – 11 – 18 – 30 – 30 0.1 – 0.20.51 – 2
NaCl g/L 2 – 51125 – 56 0.5 – 12.25 – 10
NaF g/L 0.5 – 124 – 8 0.5 (fixed)
Na2CO3 g/L 13 – 203250 – 79 14 – 202840 – 56
TC g/L 180 – 200220240 – 260 260 – 260280300 – 300
A/TCexit dig. 0.62 – 0.650.680.71 – 0.74 0.7 – 0.70.720.74 – 0.74
A/TCexit predesil. 0.3 – 0.33 – 0.365 – 0.40 – 0.43 0.4 – 0.40.440.48 – 0.48

Footnote: In the above, bold values are "cube point" values (used in a fraction of all possible combinations ( in Matrix 1, in Matrix 2), designed to allow checks on interactions between effects of different liquor variables). Underlined values are "centre point" values (forming a single, central, liquor combination, repeated once a week as a form of control). Others are "star point" values (explored one factor at a time, with all other variables held at their centre point values).

Matrix 1, comprising 54 liquor compositions, was designed to cover the liquor compositions met in the majority of Alcoa refineries, including liquors with a wide range of impurity concentration ratios. The compositions were chosen using a 6 factor central composite experimental design (Box, Hunter and Hunter, 1978), arranged into 6 randomised blocks of 9 runs, each containing a repeat run of a centre point liquor. Matrix 2 also contained a similar central composite selection of liquor compositions, but with a focus on high TC, low impurity liquors as found in the Alcoa refineries not covered by Matrix 1. It also contained a repeat of one block of 9 liquor compositions from Matrix 1, but with the digestion and predesilication runs both performed at predesilication values of A/TC (to check whether the differences previously seen between predesilication and digestion DSPs were due to temperature and time or alumina content differences).

 

2.3 Producing the Test DSP Batches

Liquors were prepared in batches of 9, at the start of each experimental week, by mixing portions of a single caustic-aluminate concentrate with inorganic salt solutions. In the remainder of the week, predesilication and digestion DSPs were produced from each liquor in the week’s batch. Gibbsite was added as necessary to target each run’s A/TC ratio, with the silica charge at the start of the run.

3.0 analytical methods

3.1 Techniques Used for DSP Solids Analysis

The total Na2O, Al2O3, SiO2 and sulphate contents of the DSP solids were measured by ICP analysis of solutions prepared by fusion with lithium metaborate and dissolution in nitric acid. The chloride contents of the solids were measured by silver nitrate titrations of the same solutions.

Considerable effort was expended using XRD and other techniques to identify non DSP phases which were sometimes found in the solid samples (e.g. phases co-precipitated with DSP or from unreacted impurities in the kaolin). Consequently, the primary method of determination of DSP Na2O, Al2O3 and SiO2 was ICP analysis of solutions formed by leaching the DSP in weak sulphuric acid (0.05M). This method assumes that, of all the possible phases, only DSP is soluble in weak acid.

Inorganic carbon contents were measured by sparging oxygen through acid digests of the DSPs, and analysing the resultant gas stream through a Dohrman DC190 carbon analyser. This replaced an earlier method based on thermal evolution of CO2, which was found to give an incomplete measure of CO32- incorporation.

Liquors were diluted 100 times with nitric acid and analysed for Na, Al, S, and Si by ICP. Chloride (and where appropriate fluoride) was analysed by capillary electrophoresis using a silica column. Standard liquor titrations (for "A", "TC" and "TA") were performed in duplicate using the Alcoa ALIAN procedure (Rosenberg, 1990).

3.2 Error Control

Analytical method choices, and precision and accuracy in execution were optimised as much as possible. Accuracy was validated by checking that: (a) mass balances existed between liquor and solids analyses, and (b) solids analyses were reasonably consistent with the charge balance required by the DSP structure described in Section 1.2.

Precautions to exclude the effects of batch to batch analytical variation included: (a) keeping the DSPs produced from the weekly blocks of 9 liquors together in any batching of preparation and analysis procedures (ensuring batch to batch analytical variations could be kept out of the final assessments of important liquor composition effects), (b) use of the centre point DSPs included in every block as controls (on preparation and analysis), and (c) use of analytical controls.

4.0 Results & discussion

In the following, concentrations of all components of DSP solids are presented as mole ratios relative to 6SiO2, the theoretical silica content of one DSP formula unit (see Section 1.2). These mole ratios are signified by prefixing the component’s formula with an X. e.g. XNa2O = the DSP Na2O:6SiO2 mole ratio. Concentrations of liquor species in gL-1 are signified by prefixing the species formula with a C. e.g. CNa2SO4 denotes the number of grams of sodium sulphate in 1L of liquor.

4.1 Soda

XNa2O, the total soda mole ratio, was found to be independent of changes in liquor composition over the experimental range. The only significant effect on total soda incorporation was the difference between formation during predesilication (average experimental value 3.90) and digestion (average experimental value 4.03). The 2 sigma random variation around these values was less than 5% in each case. Within experimental error, the value for digestion equals the theoretical value corresponding to all DSP cages being full (4Na2O:6SiO2), while that for predesilication corresponds to 90% of cages being filled.

4.2 Sulphate

Using the prefix "P" to denote model prediction, the classical form of a Langmuir isotherm to predict XSO4 from CNa2SO4, with other variables held fixed would be:

PXSO4 = CNa2SO4/( a + b.CNa2SO4)

A simple modification to this was found to do an excellent job of predicting measured values of sulphate inclusion. This was achieved by adding linear terms in CNa2CO3 and CNaCl to the denominator:

PXSO4 = CNa2SO4/( a + b.CNa2SO4 + c.CNa2CO3 + d.CNaCl)

Fitted values for the parameters in the sulphate model differed significantly between digestion and predesilication, reflecting the fact that sulphate incorporation was clearly higher under digestion conditions.

Figure 2

Sulphate Model Goodness of Fit Graph – Matrix 1 Data

Although goodness of fit of the modified Langmuir model for PXSO4 was excellent within each data set (Matrix 1 and Matrix 2), significantly different parameter values were required for the two major liquor types. This is a warning that the model should not be extrapolated beyond the range of the data to which it has been fitted (Table 1).

Of all the liquor composition variables, CNa2SO4 has the greatest effect on DSP composition. This finding is illustrated by the following Figure (showing a subset of the Matrix 1 data in which CNa2SO4 was the only manipulated liquor variable, with model curves as described above (PXSO4) and in Section 4.3 (PXCO3 and PXCl)).

Figure 3

Typical Effects of CNa2SO4 on Predesilication DSPs

4.3 Carbonate & Chloride

Carbonate and chloride inclusion ratios were found to be more sensitive to variations of sulphate in liquor than to concentrations of their own anions (compare Figure 4 below with Figure 3 in Section 4.2). There was a lower chloride inclusion ratio under predesilication conditions. Carbonate inclusion ratios were more or less independent of DSP formation conditions (predesilication vs digestion).

After trying numerous possibilities, an adjustment proportional to PXSO4 (the predicted fraction of the DSP cages occupied by sulphate) was deemed the best practical way of modelling the effect of CNa2SO4 on XCO3 and XCl. Once this choice was made, very similar models could be used for carbonate and chloride inclusions:

PXCO3 = a + b.PXSO4 + c.CNa2CO3 + d.PXSO4.CNa2CO3

PXCl = a + b.PXSO4 + c.CNaCl + d.PXSO4.CNaCl + e.CNaCl2

Figure 4

Predicted impurity inclusion ratios

The horizontal axes in Figure 4 show percentages across the "cube point" concentration ranges chosen for Matrix 1. Each curve assumes fixed concentrations of all but the nominated anion.

 

Figure 5

Carbonate and Chloride Model Goodness of Fit Graphs – Matrix 1 Data

4.4 Fluoride

Fluoride inclusion in DSP was below detection (weight% NaF <0.1) by flux/acid digest/ICP. Fluoride levels in liquor had no detectable impact at all on DSP composition.

4.5 Charge Balance, Hydroxide and Alumina

The Na2O:6SiO2 ratio needed to balance the charge associated with alumina, sulphate, carbonate and chloride (but not hydroxide or fluoride) can be calculated as

XNa2Oc = XAl2O3 + XSO4 + XCO3 + 0.5XCl

where XAl2O3 is the measured total alumina mole ratio (structural + included as aluminate) relative to 6SiO2. Figure 6(a) shows the relationship of measured XNa2O to XNa2Oc. The predesilication vs digestion difference noted in Section 4.1 is clear.

Assuming that XF=0 (see 4.4), hydroxide inclusion can be estimated by charge balance: XOH = 2(XNa2O - XNa2Oc). From Figure 6(a), these differences can be seen to be somewhat scattered (on the chosen scale), with a slight negative bias. These features are presumed due to the accumulation of error inherent in differences involving so many separate analyses. Overall, the conclusion is that hydroxide inclusions were too low to be detected, under our experimental conditions (including washing to 2g/L TS in wash water).

Although there were some correlations between liquor composition variables (particularly CAl2O3) and XAl2O3 (the total alumina inclusion), these were not strong. As a result, the following predictive model, based on charge balance, assuming XOH and XF are both zero, was deemed as good as any other, for practical purposes:

PXAl2O3 = PXNa2O – (PXSO4 + PXCO3 + PXCl/2)

Figure 6

Charge Balance vs Measured Values of (a) Total Soda & (b) Total Alumina

Figure 6(b), above, compares this model with the weak acid leach/ICP analyses for XAl2O3. Points to note are the small bias between measured and model values for digestion DSPs (average measured XAl2O3 > average model XAl2O3), and that the variation within each group is small relative to the mean (r.s.d. 3%). Reasons for the bias have been investigated, but are still unclear. Choosing the stated model for PXAl2O3 ignores this problem in favour of creating a set of prediction equations respecting charge balance. Other models have also been developed with fit to data as the guiding principle. The simplest of these is to predict XAl2O3 by the experimental average for the relevant DSP formation conditions (3.26 for digestion, 3.12 for predesilication (Matrix 1 experimental range)). Most experimental DSPs were within 3% of these averages (Matrix 1 two sigma limits). Whichever method is chosen to calculate XAl2O3 , Lowenstein’s Rule (structural Al:Si ratio = 1:1) can be used to calculate predictions of aluminate inclusions: PXAl(OH)4 = 2.(PXAl2O3 – 3).

4.6 The Nature of the Predesilication, Digestion Differences

Figure 4 (Section 4.3) illustrates some of the differences found between DSPs formed under digestion and predesilication conditions (in particular that sulphate inclusion was higher under digestion conditions). In general, to match plant concentrations, the digestion runs were conducted at higher A/TC ratios than the predesilication runs (see Table 1). To check whether this was the main cause for the differences between the resulting impurity inclusion rates, one block of 9 liquors from Matrix 1 was rerun in Matrix 2 with the digestion A/TC ratios set to match that block’s predesilication values. The differences between the resulting DSPs were not altered, leading to the conclusion that this difference is due to temperature and/or reaction time rather than A/TC.

The preliminary work prior to starting liquor matrix 1 included comparisons between DSPs formed under 3 sets of conditions: predesilication only, predesilication followed by digestion and digestion only (from identical starting liquors, but with extra gibbsite being added for digestion runs). With respect to impurity inclusion (but not the soda silica ratio), the results were that predesilication predesilication+digestion digestion alone. Impurity inclusion ratios in DSPs formed during predesilication appear to survive through digestion.

5.0 Conclusions

The inclusion of sulphate into DSP dominates other inclusions. Sulphate inclusion is best modelled by a Langmuir type dependency on sulphate, chloride and carbonate in solution. The prediction is unaffected by fluoride, alumina or free caustic in solution.

The inclusions of chloride and carbonate can be predicted by regression models involving terms in their own solution variable, a linear dependence on the predicted sulphate inclusion and cross terms. Fluoride is not significantly included into DSP under the conditions covered by liquor matrices 1 and 2 (1-4 g/L NaF). Neither does fluoride in solution significantly affect the inclusion of other salts.

DSPs formed under predesilication (slurry storage) conditions or predesilication followed by 150 C digestion contain higher levels of chloride and lower sulphate inclusions than those formed under digestion only conditions. This is due to the different formation conditions (temperature and holding time) rather than differences in alumina concentration. Aluminate inclusion is also higher under digestion conditions but in this case is likely to be due to the alumina content of the liquors. Carbonate inclusion is relatively unaffected by formation conditions or alumina in solution.

The DSPs produced under digestion conditions during this project have soda to silica mole ratios within  5% of theoretical predictions assuming full occupancy of the sodalite cages. Soda to silica mole ratios were independent of liquor composition. There was however, a significant difference in the averaged soda to silica ratio for digestion DSPs (0.67) compared to the predesilication DSPs (0.65). The reason for this difference is not known.

Measured alumina to silica mole ratios in excess of those expected from charge balance could indicate small analytical biases, a larger structural alumina component of the DSP or the presence of another alumina containing phase coprecipitated with the DSP (the last possibility was not always supported by XRD examination of selected DSPs).

Accurate impurity incorporation models have been worked out for DSPs prepared from a wide range of liquors under laboratory conditions. These are now being tested against analyses of plant samples, and have been programmed into impurity balance models giving predictions of steady state impurity levels in plant liquors.

Acknowledgments

The project underlying this paper was a team effort between Alcoa and CSIRO Minerals, with most of the analytical and all of the experimental work being carried out by CSIRO. We gratefully acknowledge the outstanding work of the CSIRO Minerals analytical services unit (Waterford) in analysing the samples in this report. In particular, we thank Milan Chovancek (solids and liquid analyses), Ian Davies (XRD and alumina titrations) and Peter Choo (carbon analyses). On the analytical side at Alcoa, we also thank Tony White, Margaret McIntyre and co-workers (liquor analyses), and Nick Pearson (DSP TGA). Thank you also to Barry Whittington and Ben Fletcher of CSIRO, who contributed substantially to scientific and experimental aspects of the first half of the project. Finally, we wish to thank Gerald Roach and Laurie Stonehouse of Alcoa for their technical and project management guidance.

References

Box, G.E.P., Hunter, W.G. and Hunter, J.S., 1978, Statistics for Experimenters, J.Wiley & Sons.

Catlow, C.R.A.; George, A.R.; Freeman, C.M., 1996, Ab initio and molecular-mechanics studies of aluminosilicate fragments and the origin of Lowenstein's rule, Journal of the Chemical Society, Chemical Communications, 1311-1312.

Hassan, I., 1983, The crystal structure and crystal chemistry of the cancrinite and sodalite groups of minerals, PhD Thesis, McMaster Universtiy, Canada.

Lectard, A., Nicolas, F., 1983, Influence of Mineral Impurities on the Alumina Trihydrate Precipitation Yield in the Bayer Process, Travaux, 13 (18) pp345-351.

Rosenberg, S.P., 1990, The Alcoa liquor analyser (ALIAN), Second International Alumina Quality Workshop Proceedings, Perth, Western Australia pp 355-365.

Smith., P.G., Pennifold, R.M., Whittington, B.I. and Fletcher, B.L., 1998, Measurement of inorganic impurity inclusions into desilication product: final report, CSIRO Minerals Report DMR-705, 107 pp. CONFIDENTIAL

Seimiya, S., 1963, Some Properties of Sodalite in Red Mud, The Extractive metallurgy of Aluminium Vol 1, Interscience

Stroud, C.E., Stencel, J.M. and Todd, L.T. Jnr., 1979, Infrared spectra of cathodochromic sodalite, J.Phys.Chem., 83, pp2378-2382.

Teas, E.B., Kotte, J.J., 1980, Effect of Impurities on Process Efficiency and Methods of Impurity Control and Removal, Proc Bauxite Symposium, Geol Soc Jamaica, pp100-129

Whittington, B.I., Smith., P.G. and Fletcher, B.L., 1997a, Measurement of inorganic impurity inclusions into desilication product, CSIRO Minerals Report DMR-580, 51 pp. CONFIDENTIAL

Whittington, B.I., Fletcher, B.L. and Talbot, C., 1997b, DSP Formation in the Bayer process: the effect of reaction conditions on DSP composition and impurity content, CSIRO Minerals Report DMR-429, 64 pp. CONFIDENTIAL