Frank M. Kimmerle, Frank Feret, and Stéphane Paré

Alcan International Limited

Arvida Research and Development Centre

P.Q. Box 1250, Jonquière, Québec, Canada G7S 4K8

Barbara Feret

BF Simulation

3099 St. Patrick St., Jonquière, Québec, Canada QC, G7S 2P7


Preliminary analysis of mine exploration samples has traditionally been confined to the determination of elemental composition complemented by limited processibility studies. Detailed evaluation of the mineralogy through laboratory digestion simulations (and thus an estimate of the true value of the deposit) is usually postponed until the development of a mining plan. However, the initial exploration data, including older data limited to the four major oxides, can be exploited to determine the mineralogy of the deposit. Using non-linear programming, the BQuant software provides quantitative phase determination based on an elemental mass balance without having to rely on time-consuming wet chemical digestion nor deposit-specific X-ray diffraction or regression methods. It can even deal with overburden, clay floor and is independent of the exact geological nature of the deposit.

The BQuant approach is being used to characterise the Ely deposit (Northern Queensland), Jamaican and Boké (Guinea) bauxite mine sites. This communication describes the quality control practices implemented, the methodologies used and the data treatment involved. It describes the sample preparation steps followed by automated loss of mass determination, elemental analysis using ICP or robotised XRF fusion. It also compares the BQuant predictions with actual laboratory digestion results for the various geological deposits involved and provides estimates of the technician-contact time for the manual and the robotised analysis and data treatment.


mineralogy, quantification, bauxite, software, mass balance



Frank M. Kimmerle, Frank Feret, Stéphane Paré and Barbara Feret


Bauxite deposits and aluminium bearing minerals, like any other mineral ores, can be described from a multitude of points of view. The geologist seeks to identify and quantify mineral phases or amorphous material that he knows from experience to be amenable to alumina extraction. He wants to be able to measure the extent of a mineral deposit and perhaps attach to it an approximate monetary value. The analytical chemist seeks to characterise a series of samples using standard analytical techniques. He wants to convert these results into data, which the geologist and the process engineer can use. The latter wants to be able to blend mined bauxite and be able to predict its behaviour in a given alumina extraction process. He will also want to measure the efficiency of his plant against defined benchmarks. These different needs give rise to different descriptors, which moreover differ somewhat from one aluminium company to another. We have developed a Windows-based software, called BQuant 98, which will guide the non-expert user from elemental composition and loss of mass on ignition (LOI) to an estimate of the mineralogy and to progress from the mineralogy to laboratory and plant digest parameters.

Unlike empirical XRF regressions that tend to be specific to a given mineralogical deposit (Feret, 1991), the present approach is thought to be universal and does not require site-specific calibration, (although independent identification of some of the phases would make the approach even more rigorous). XRD calibration methods (Boski, 1988; Andrews (1980) are also limited to the concentration ranges of variables for which the calibration coefficients were calculated. Moreover, the minerals that are the most critical to this analysis, such as gibbsite, goethite, or kaolinite, show wide variations in X-ray diffraction response due to natural diversification in crystallinity, poor crystallinity, and preferred orientation. The Rietveld refinements (Young, 1993) are already represented by a number of commercial programs, but unfortunately, the Rietveld approach is also hampered by the amorphous phases and gibbsite-preferred orientation effect. Sajó attempted to overcome the limitations using the XDB approach (Sajó, 1994; Feret, 1997), which involves deconvoluting X-ray diffractograms along with a complete mass balance. Both Rietveld and the XDB methods are too slow for any but research purposes.

Since the number of unknowns, (mineral phases), is larger than the number of known parameters, (oxides, T.O.C. and LOI), the algorithm used in BQuant employs Monte Carlo techniques to quantify the mineral composition from a given set of phases. Details of the computation have been described previously (Kimmerle, 1997) and will only be summarised here. Under given Bayer process conditions, alumina, silica, as well as minor constituents including phosphates, vanadates and zinc will be extracted into the caustic liquor. While the geologist may be satisfied with a description of the mineralogy of the deposit, it is nevertheless also necessary to express the phases in terms of the traditional process-related parameters. This constitutes the last phase of the BQuant development and the correlation will be limited here to standard laboratory rather than variable plant conditions.



In order to accurately estimate concentrations of the major extractable phases: gibbsite, boehmite, goethite and kaolinite, BQuant 98 requires accurate analysis of LOI and the major oxides, Al2O3, SiO2, Fe2O3, TiO2 to better than 0.1% and preferably of the minor oxides and total organic carbon to 0.02%. The expected input data and the acceptable composition range are indicated in Table 1.

Table 1

BQuant Input Data: Elemental Composition (Expressed as Oxides)

Data Input

Input Range

Desired Accuracy (1s)





> 11 and <34



> 30 and < 72

t.Al2O3/LOI < 4.0



< 20



t.Fe2O3+ t.SiO2 <45



< 10


TOC (total organic carbon)

default 0.1



< 5



< 5



< 2



< 2



< 2



< 5



< 2



< 5



< 0.2



< 1



< 2



< 1



where < 97.5% < S LOI, element oxides < 102.5%

preferably, where < 99.5% < S LOI, element oxides < 100.2%

The analytical equipment currently used is indicated in Table 2. X-ray fluorescence (XRF) and inductively coupled plasma spectrometry (ICP) allow accurate analysis of elements > Na in bauxites at the 0.01% level and below. However, for highest accuracy, in order to avoid particle size and specific orientation effects, it is necessary to prepare XRF samples as glass pearls and not powder briquettes. During fusion, care must be taken to avoid loss of sulphur and the same 12:22 lithium borate/metaborate flux can also be used to prepare solutions for ICP analysis avoiding tri-acid or NaOH digestion. Any method based on mass balance needs to divide water of crystallisation between the various phases, principally aluminium mono and tri-hydrate, goethite and hematite and LOI measurements become the primary criteria. At best, manual gravimetric techniques can attain 0.1% precision and we now use automatic thermogravimetric analysis throughout our operations. LOI needs to be corrected for losses due to the decomposition of carbonates (implicitly by BQuant) and explicitly for the pyrolysis of organic carbon. The total weight loss after calcination involves about 2.2 times the TOC measured (Kuhnhard, 1986). Should TOC vary considerably within a deposit, it then becomes necessary to measure it in order to deduce the H2O from LOI. Either pyrolysis (LECO SC444DR) or wet oxidation (OIC-700) can be used, with the latter having the drawback of using an aliquot of only about 10 mg and requiring grinding to - 45µm to assure sample representativity.

We use semi-automatic Claisse fluxers to produce the XRF pearls in our mines and alumina refineries but have introduced a completely automated HERZOG fluxer in our Jamaican operation in 1998. The resulting 30 mm pearls are analysed using wavelength dispersive X-ray spectrometers in most of our plants and using rugged simultaneous units in the Guinea, (Africa) operations. Samples for the Australian Ely deposit were analysed using ICP in a contract laboratory that recently acquired a Phoenix semi-automatic fluxer.

Table 2

Instrumentation used to obtain Input Data



Australia (Ely)

Jamaica (Kirkvine)

Guinea (CBG)

LOI 1996/7












Oxides 1996/7



Phoenix Fluxer/ICP

Claisse fluxer BIS



Claisse fluxer BIS

WLD simul XRF

2.1 Quality Control Samples

In many instances, exploration-mining samples are analysed in the form of powder briquettes with more emphasis being placed on cost/sample than on accuracy. Since we wanted to integrate the exploration data into a future mining plan, a more stringent analytical quality control was adopted for the Ely deposit and similar controls are now being copied by us elsewhere. The reduced analytical costs involved in BQuant allow us to analyse drill holes as a function of depth (25-cm intervals) and restrict confirmatory wet chemical digestion analysis to a limited number of composite samples.

2.1.1 Sample Preparation

As indicated in the simplified sample scheme, Figure 1, drilling core samples are split in the field, identified with bar codes and 98% single and 2% duplicate splits (typically 2 kg, 0.6 mm) shipped to the contract laboratory. If beneficiation will be necessary, the moisture is determined, the samples washed (on 28 and/or 48 Tyler Mesh) sieves and the % recovery calculated. The entire sieved fraction is ground, riffled and a sub-sample of about 50 g further ground to - 150 M. A split of about 25 g is set aside for analysis with 3% of the samples to be analysed in duplicate. For analyses requiring 1 g sample or more, the aliquots may be taken directly, if smaller quantities are required, grinding in a ring and puck mill to -325 M (-45 µm) is recommended to assure representative sampling.

2.1.2 Calibration Samples

Alcan certified reference bauxites are used to establish or correct calibration curves for LOI, XRF or ICP and TOC analysis. In principle LOI, relying as it does only on strict temperature control and an analytical balance, is an absolute measurement requiring no calibration. In practice we have found that it is necessary to identify and correct any sources of experimental bias in the operating procedures.

2.1.3 Quality Control Standards (working reference material)

During the preliminary exploration of the deposit, about 25 kg of a "typical" sample is set aside to produce a quality control standard. For deposits containing both tri-hydrate and monohydrate bauxites, one sample of each is required. The sample is dried, ground to - 150 Mesh and bottled in 100 g lots. 10% of the bottles are tested for homogeneity and carefully analysed in a reference laboratory. This leaves about 200 units available for an analytical quality assurance program and any analytical method development. The quality control sample is introduced after every 50 unknown samples and follows the same analytical sample procedures as the unknowns.

Figure 1

Sample Preparation Scheme

2.2 Analytical Quality Control

The internal analytical quality control involves a number of steps. Duplicate or triplicate readings for the same sample or prepared solution (for about 3% of the load) will give the laboratory technician an immediate indication of any instrument instability. The results of the repetitive analysis of the quality control samples are recorded on SPC charts, see Figure 2, and indicate both the reproducibility and any bias with respect to the certifying laboratory. Preparation and analysis of 2% of both halves of the core splits are used to judge the homogeneity of the original sample and the reproducibility of the entire procedure including beneficiation. Finally, a small fraction, usually a series of composites, is reanalysed by a second laboratory both for elemental composition and by wet chemical digestion.

2.3 Analytical Costs

Using automatic instrumentation for LOI, a semi-automatic fluxer and modern XRF spectrometer with automatic sampler changer for elemental determination, the technician contact time per sample will range between 15-20 minutes. Analytical costs varied little with the degree of automation and were relatively constant from one geographical region to the next and even between in-house and contract laboratories when amortisation of equipment was taken into account. Excluding site specific preparation and grinding, we estimate the cost of the elemental and loss of mass analysis between $15 to $25 US per sample with a similar amount for TOC. Unless organic carbon was known to vary considerably, we therefore assigned default values obtained from a limited number of samples. BQuant calculations include electronic data transfers, correcting the occasional data transcription errors, file manipulations and reporting. Using a Pentium 200 MHz PC operating under Office 95 or NT and Windows 95 or 97, BQuant can handle about 100 bauxite samples per hour and thus add less than $1 per sample to the actual analytical costs.


Figure 2

Statistical Process Control Chart: Total Alumina in Ely Bauxite Standard


The input data consists of LOI and TOC and up to 16 elements expressed as oxides, of which only the first five items are absolutely required (see Table 1). The list of "default" mineral phases most likely to occur is given in Table 3 and the operator has an opportunity to customise the list if additional information is available. BQuant begins by assigning the secondary elements according to an expert system that takes into account the mineralogy of the deposit. The remaining amounts of water, alumina, silica and iron oxide may then be distributed among the eight remaining unknowns (gibbsite, boehmite, kaolinite, quartz, goethite, hematite, and the degree of substitution of Al2O3 and H2O in goethite). Since the number of unknowns exceeds the number of variables, no analytical solution is possible and BQuant seeks an iterative solution using two independent Monte Carlo generators (Binder, 1987).

3.1 Data Manipulation

The input data for up 1000 samples at time may be transferred from a laboratory information management system or entered manually into an EXCEL work-sheet. Only the system administrator may select different combinations of phases; the casual operator is limited to correcting data entry. BQuant will indicate if the input range or conditions are not respected and proceed to estimate the phase composition. The output supplies both the percentage composition of the various minerals and the alumina and silica present in the major phases. In order to correlate this information with the traditional wet laboratory digestion protocols, we need to estimate the fraction of mineral phases attacked under the standardised conditions, conditions that vary somewhat throughout the industry:


For low temperature extractable alumina, presuming precipitation of kaolinite as sodalite:

Al203150 = a1* gibbsite + 30% * a2* crandallite [1]

For low temperature extractable silica:

SiO2150 = b1* kaolinite [2]

For high temperature total extractable alumina, presuming partial extraction of alumina from aluminium goethite and no quartz attack:

TEA = c1* gibbsite + c2* boehmite + 60% * c3* crandallite

+ 40% * c4* lithiophorite + 40% * c5* Al-goethite [3]

For high temperature extractable alumina, presuming partial extraction of alumina from aluminium goethite but complete quartz attack with precipitation as sodalite:

Al203225 = c1 * gibbsite + c2 * boehmite + 60% * c3* crandallite

+ 40% * c4 * lithiophorite + 40% * c5* Al-goethite - c6* quartz [4]

For high temperature extractable silica:

SiO2225 = b1* kaolinite + quartz [5]

where the stoichiometric coefficients are: a1 = 0.654, a2= 0.37, b1= 0.466, c1= 0.654, c2= 0.850,

c3 = 0.37, c4 = 0.185, 0 < c5 < 0.15 varies with the degree of substitution and c6 = 0.85.

Table 3

BQuant Default Mineral Phases





Gibbsite/Nordstrandite 0.1% <X1< 95% Al(OH)3 *
Boehmite/Diaspore 0.01% <X2< 25% AlOOH **
Kaolinite 0.05% <X3< 25% Al2O3.2SiO2.2H2O  
Goethite 0.05% <X4< 50% (Fe1-nAln)O(OH) 0 < n < 0.25
Hematite 0.1% <X5< 30% Fe2O3  
Quartz 0.05% <X6< 20% SiO2  
Anatase/Rutile 0.05% <X7 TiO2 ***
Crandallite 0.05% <X8 CaAl3(PO4)(PO3)(OH)7  
Calcite 0.05% <X9 CaCO3  
Apatite 0.05% <X10 Ca5(PO4)3.Ca(F,Cl,OH)  
Wavellite 0.05% <X11 Al3(PO4)2(OH)3 .5H2O  
Illite 0.05% <X12 K2O.3Al2O3.6SiO2.5H2O  
Zircon 0.01% <X13 ZrO2.SiO2  
Lithiophorite 0.05% <X14 Al2O3.Li2O.MnO2.MnO.H2O  
Chromite 0.01% <X15 FeO.Cr2O3  
Dolomite 0.1% <X16 CaMg(CO3)2  
Magnesite 0.05% <X17 MgCO3  
Dawsonite 0.1% <X18 Na2OAl2O3(CO2)2.2H2O  
Schubnelite 0.05% <X19 Fe2O3. V2O5.2H2O  
Gahnite 0.01% <X20 ZnO.Al2O3  
Celestite 0.05% <X21 SrSO4  
Alunite 0.05% <X22 K2SO4.Al2(SO4)3.4Al(OH)3  
+ Xi = % Weight of phase "i" in bauxite, dry basis

* BQUANT cannot distinguish between gibbsite and nordstrandite

** BQUANT cannot distinguish between boehmite and diaspore

*** BQUANT cannot distinguish between anatase and rutile

3.2 Comparison with Laboratory Digestion Data

For a number of Ely exploration samples, results obtained by laboratory digestions in one of Alcan’s R & D Centres were compared with the BQuant estimates based on the complete elemental analysis carried out in our laboratory.

Figure 3

Ely Deposit: Tri-hydrate Digestion Results versus BQuant Estimates

Figure 3 indicates the excellent agreement between the experimental and calculated results for low temperature extractable alumina (Equation [1]), where the digestions for low trihydrate bauxites were carried out such as to avoid attacking the boehmite mineral. The difference between equations [3] and [1] can be considered as the potentially extractable alumina lost to the mud residue in low temperature digestion plants, sometimes erroneously referred to as "monohydrate" alumina. Figure 4 indicates the good agreement and incidentally confirms that much of Ely, like the adjoining Andoom and Weipa bauxites are best used in alumina refineries operating under high temperature digestion conditions.

Figure 4

Ely Deposit: "Monohydrate" Digestion Results versus BQuant Estimates

The Jamaican bauxites exploited by Alcan contain mainly aluminium tri-hydrate but also considerable minor constituents and often a high degree of aluminium substitution in the high surface area goethite.

Comparison of routine wet digestion data and estimates of gibbsitic alumina by BQuant indicate a relative standard deviation of almost 0.9%. However, given the 0.4 to 0.5% interlaboratory reproducibility of the digestion data and a reproducibility of about 0.25% for the XRF analysis of total alumina, the agreement indicated in Figure 5 is considered adequate for exploration data. Over the next few months, we expect to fine-tune the phase choices and improve the agreement with the full implementation of the automatic fluxer and upgrading from the existing PW 1400 spectrometer. Close agreement was reported between experimental and calculated values of kaolinitic silica over the range 0.5 to 5.0% with BQuant reporting average values 0.15% higher with a standard deviation of 0.3%.


Figure 5

Jamaican Bauxite: Tri-hydrate Digestion Results versus BQuant Estimates


Figure 6

Bidikoum Bauxite: Tri-hydrate Digestion Results versus BQuant Estimates

Five Boké (Guinea) deposits are presently being explored by CBG and all can be treated using the same phase selection, assuming Russian (or hydro) goethite. LOM and the four major oxides, (Al2O3, Fe2O3, SiO2, TiO2) together with 0.1% organic carbon, describes about 99.3% of the composition. Using only these data and average trace constituents for the remaining 0.7%, BQuant was able to estimate the mineralogy with surprising accuracy. Taking into account the efficiency of the laboratory digestion procedure, the average standard deviation for the difference between the experimental and estimated extractable alumina at 143oC averaged about 0.3%. About 10% of the data examined are illustrated in Figure 6 for the Bidikoum deposit. Even closer agreement is observed for the extractable alumina under high temperature conditions.


BQuant 98 is presently being used to deduct the mineralogy of three major bauxite deposits, to estimate their commercial potential and to help produce a preliminary mining plan. It is able to quantify the composition of alumina-bearing minerals, employing a «balance of mass» approach, which incorporates an expert system and an ingenious iteration procedure. Using reliable elemental composition data, it estimates a complete list of the most likely mineral phases, which can then be related to extractable alumina and silica under standard laboratory conditions. Existing data, even if limited to LOI and the major oxides, can be integrated with more recent drilling results. For exploration of new deposits, the reduction of the analytical cost compared to conventional bomb digests or XRD regression procedures is such that it allows us to establish complete mineralogical depth profiles and should have a major influence on future mine exploitation.



We would like to acknowledge the advice of Neil Bliss and Monique Authier-Martin in interpreting the bauxite mineralogy and chemistry, the contribution of Stéphane Tremblay in adapting BQuant to a Windows environment and the dedication of Lori Wickert and Daniel Roy in crunching large data banks.



Andrews, W. H. and Crisp, A.J.,C (1980), Correlation of Bauxite Analysis with Mineralogy, Light Metals, pp 3- 18.

Binder, K., (1987), Application of the Monte Carlo Methods in Statistical Physics, Springer Verlag, Berlin.

Boski, T., Herbillon, A.J., (1988), Quantitative determination of hematite and goethite in latteritic bauxites by thermodifferential X-ray powder diffraction, Clays and Clay Minerals, Vol. 36, No. 2, pp. 176-180.

Feret, F. and Giasson, G.F., (1991), Quantitative Phase Analysis of Sangaredi Bauxites (Guinea) Based on their Chemical Composition, Light Metals, pp. 187-191.

Feret, F, Authier-Martin, M.and Sajó, I. (1997) Quantitative Phase Analysis of Bidikoum Bauxites (Guinea), Clays and Clay Minerals, Vol. 45, No. 3 pp. 418-427.

Kimmerle, F. M., Feret, F., (1997), BQuant: Cost Effective Calculations of Bauxite Mineralogy, Light Metals, pp. 1-8.

Kuhnhard, C., (1986), Bestimmung und Charakterisierung von Huminstoffen in Kreislauflaugen des Bayer-Prozesses, Aluminum,2, pp. 101-106.

Sajó, I., (1994), Powder Diffraction Phase Analytical System, Version 1.7, Users Guide, Aluterv-FKI, Ltd., Budapest.

Young, R.A., (1993), The Rietveld Method, International Union of Crystallography, Oxford Science Publications.