1-s2.0-S0892687515001417-main.pdf

7/24/2019 1-s2.0-S0892687515001417-main.pdf

1/17

Development of a geometallurgical framework to quantify mineral

textures for process prediction

Cecilia Lund a,, Pertti Lamberg a, Therese Lindberg b

a MiMeR Minerals and Metallurgical Research Laboratory, Lule University of Technology, 971 87 Lule, Swedenb LKAB, Research & Development, 983 81 Malmberget, Sweden

a r t i c l e i n f o

Article history:

Received 31 December 2014Revised 26 March 2015Accepted 4 April 2015Available online 5 May 2015

Keywords:

Geometallurgical frameworkIron oreMineral texturesParticle trackingTextural archetypes

a b s t r a c t

A geometallurgical framework was developed in three steps using the Malmberget iron ore deposit,northern Sweden, as a case study. It is based on a mineralogical-particle approach which means thatthe mineralogical information is the main focus. Firstly, the geological model describes quantitativelythe variation in modal composition and mineral textures within the ore body. Traditional geological tex-tural descriptions are qualitative and therefore a quantitative method that distinguishes different mineraltextures that can be categorised into textural archetypes was developed.

The second step of the geometallurgical framework is a particle breakage model which forecasts howore will break in comminution and which kind of particles will be generated. A simple algorithm wasdeveloped to estimate the liberation distribution for the progenies of each textural archetype. The modelenables numerical prediction of the liberation spectrum as modal mineralogy varies. The third stepincludes a process model describing quantitatively how particles with varying particle size and com-position behave in each unit process stage. As a whole the geometallurgical framework considers the geo-logical model in terms of modal composition and textural type. The particle breakage model forecasts theliberation distribution of the corresponding feed to the concentration process and the process modelreturns the metallurgical response in terms of product quality (grade) and efficacy (recovery).

2015 Elsevier Ltd. All rights reserved.

1. Introduction

Geometallurgy embraces geological and metallurgical informa-tion to create spatially-based predictive models (3D) of ore bodiesthat supply all relevant information for mineral processes(Lamberg, 2011). The industrial application of geometallurgy is astructured effort to bridge all the relevant knowledge of theresource for production planning and management, also calledgeometallurgical program.

Geometallurgical programs are needed for better resource man-

agement and to lower the risk in the process operation related togeological variations within the ore body. It is a vital part of theprofitability of the operation. The mine needs to have the capabil-ity to adjust the concentration process and the product qualities tomeet the requirements of a changeable global market e.g. by amore effective utilisation of the ore resources or the ability to han-dle larger volumes of lower grade ore. Today there exist differentkinds of geometallurgical models depending on the ore, its quality

and the mineral processing circuit (e.g., Alruiz et al., 2009; Suazoet al., 2010; Hunt et al., 2012).

Most of the geometallurgical programs are established by usingcertain steps and rely on metallurgical and geometallurgical test-ing (Dobby et al., 2004; Bulled and McInnes, 2005; David, 2007;Lamberg, 2011). Commonly a series of representative ore samplesis collected and are then tested to measure the metallurgicalresponse directly with a standard methodology (e.g. standard flota-tion test). There are fundamental reliance on the representative-ness of the samples and tests since they link the ore with the

metallurgical response. As the sample set should include all vari-ability in the ore this is often called a variability test. Based onthe test results, a mathematical model is created to explain themetallurgical response based on the sample characteristics.

Iron mines are big volume operations and the production is dri-venby throughput. Most iron ore companies produce high volumesof iron ore products with a Fe grade between 62% and 64%.Examples of such production are direct shipping of hematite oresin Australia and Brazil (Poveromo, 1999). The Swedish iron ore pro-ducer, LKAB, represents another type of production strategy. Theyproduce custom high grade iron ore pellets (>67% Fe) and fines forblast furnaces and direct reduction (LKAB, 2011). A good

http://dx.doi.org/10.1016/j.mineng.2015.04.004

0892-6875/2015 Elsevier Ltd. All rights reserved.

Corresponding author. Tel.: +46 920 492354.

E-mail address:[email protected](C. Lund).

Minerals Engineering 82 (2015) 6177

Contents lists available at ScienceDirect

Minerals Engineering

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / m i n e n g

7/24/2019 1-s2.0-S0892687515001417-main.pdf

2/17

understanding of the tangible properties of the raw material andits variability is essential in both the strategies. In the direct ship-ping ores the production chain is very short; normally it includesonly crushing and screening, and therefore if the ore is not suitablefor the product requirements the tools available in mineral pro-cessing to address this are very few. In the high grade productswhere the tolerable grade of impurities like silica, aluminium,phosphorous and sulphur, are very low, the success of processingis very dependent on the identification of ore quality before itcomes to the processing plant.

In the literature there is very little information on existing geo-metallurgical programs of iron ores, but a few can be found, e.g.from Kiruna, Sweden (Niiranen and Bhm, 2012) and NWAustralia (Paine et al., 2011). The technique that is frequently usedin evaluating the metallurgical variation in magnetite-bearing ironores is the Davis tube (Farrell et al., 2011; Niiranen and Bhm,2012). It is a small-scale magnetic separation test and the ore sam-ples are normally ground to the liberation size before testing. Thecorresponding concentrate and tail are chemically analysed andthe distribution of elements is then calculated. The iron dis-tribution (recovery) and the concentrate quality are used in pre-dicting the metallurgical response in the full scale operation.

The problem with using only the chemical components is that itis not always the typically primary reason for the metallurgicalresponse. As chemical components are bonded in minerals we sug-gest it is more appropriate to use mineralogy for building themetallurgical functions and the geometallurgical domains.However, minerals do not occur independently in the processesthey occur in particles which vary in size, shape and composition.

Lund et al. (2013)developed a practical, fast and inexpensivemethod to derive modal composition from routine chemical assaysusing an element to mineral conversion technique which formedthe first part of a geometallurgical framework of the Malmbergetdeposit. However, the modal mineralogy alone is not sufficient todescribe the ore behaviour when processing the ore. The mineraltextures play a significant role and need to be considered when a

geometallurgical model is developed.The texture characterisations are typically subjective (e.g.

Bonnici et al., 2008) and traditionally more related to orecharacterisation than process mineralogy (Perez-Barnuevo et al.,2012). During ore processing, the effects of mineral textures andthe liberation are closely associated. The textures in an ore areone important family of parameters that can limit the ability toupgrade the ore (e.g.Butcher, 2010). The purpose of the comminu-tion stage is to liberate ore minerals appropriately for the concen-tration process to enable reaching required concentrate qualitywith adequate recovery.

In mineral processing the relationship between the mineral(micro) textures and liberation has been a separate research sub-ject for a long time. Basically the aim has been to forecast the

liberation distribution from a two-dimensional picture of an ore.

This is generally called the liberation model. The principle wasintroduced already byGaudin (1939).Andrews and Mika (1975)developed a graphical presentation and this was further developedby King (1979) and King and Schneider (1998). These modelsassume random breakage which is unfortunately rarely true espe-cially in grindingVizcarra et al., 2010).King and Schneider (1998)developed the model further and included a kernel function whichovercomes the problem of random breakage.Hunt et al. (2011a,b)used chess-board pattern for crushed samples and reduced theeffect of the random breakage assumption. All these methodsrequire a two dimensional microphotographs of an ore sample tobe investigated. In a geometallurgical context this means prepara-tion of thin sections or grain mounts, their photographing andimage processing for a large number of samples. This is not verypractical and an alternative way is needed.

This case study aims to find a solution to allow the incor-poration of mineral texture information into a particle-basedapproach (Fig. 1) modified from a concept by Lamberg (2011).This is done through a case study of Malmberget iron ore deposit,in Northern Sweden. It focuses on mineral parameters, such asmodal mineralogy, mineral textures, mineral association, mineralgrain sizes and their relationships to liberation characteristics.

The final purpose is to deliver a geological model which canoffer quantitative rather than descriptive data to be used in a pro-cess model. Firstly, the geological model is complemented withtextural information. Secondly, it is demonstrated how such a geo-logical model can be linked with a process model capable of fore-casting the metallurgical response such as grade and recovery forany given geological unit (sample, block, or domain).

2. Sampling, experiment and analytical work

Two different ore bodies in Malmberget deposit were selectedin this study. Fabian ore body that is proved to be one of the largerore bodies in the deposit and the Printzskld ore body that wasconsidered for validating the results of Fabian ore but also to iden-

tify the differences of the ore bodies in the deposit. In the first data-set were over 100 mineralogical samples selected for polished thinsections aiming to characterise the mineralogy and the texturalproperties of the ore. The second dataset consist of the metallurgi-cal samples, sampled from five drill cores of Fabian (Fa) andPrintzskld (Pz) ore body of both massive ore and semi-massiveore, referred as ore type inTable 1. The drill cores were carefullylogged, and three dominate ore types were identified from themineralogical study, generating a total of >100 kg in five differentcomposite samples, namely (i) Feldspar (albite and orthoclase) richore (Fsp GEM-type) of Fa and Pz, (ii) Apatite rich or of Fa and Pz (ApGEM-type) and (iii) Amphibole rich ore of Fa (Amph GEM-type),referred as sub-ore type inTable 1. Sampling, sample preparationand analytical methods used for the mineral analyses (optical

microscopy, electron microprobe (EPMA)) and the element to

Fig. 1. The particle-based geometallurgical concept modified from Lamberg (2011). Modal mineralogy and texture links the geological model and process model. In theprocess model minerals are treated as particles. From the geological information the particle population is generated through the particle breakage model.

62 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

3/17

mineral conversion (X-ray fluorescence (XRF)) are described in

detailed byLund et al. (2013), Lund (2013).In an earlier paper ofLund et al. (2013)a system for ore type

classification was developed based on the Fe-grade, Fe-mineralogy,gangue mineralogy and presence of sulphides in the drill cores andore which were named geometallurgical ore type (GEM-type). Inaddition to the modal mineralogy the optical microscopy examina-tion in this study identified a variation of mineral textures for eachGEM-type and an attempt to develop the oretype classification sys-tem, now also including the mineral texture was made (Table 1).

The (metallurgical samples) composite samples were processedand analysed at the LKAB, Malmberget laboratories. They were firstcrushed to a particle size below 3 mm. This was followed by drymagnetic separation tests with Sala Mrtsell magnetic separationusing standard test procedures by LKAB producing two outputs:

concentrate and tailings (Lund et al., 2010). The test products wereweighed and sized into ten size fractions using sieves of 38, 75,150, 300, 425, 600, 1190, 1680 and 3000 lm. The size fractionswere analysed for their chemical composition by X-ray fluores-cence and SATMAGAN magnetic balance (Lund et al., 2013).Polished resin mounts were prepared for the size fractions 3875, 75150, 150300, 300450 and 450600lm and analysedwith QEMSCAN E430 Pro at LKAB, Lule. The QEMSCAN data wasexported into Excel files and read into HSC Chemistry for dataprocessing.

The purpose of the processing of the mineral liberation data wasto quantitatively track how different kinds of multiphase particleswere derived from different GEM-types of different size fractionsduring processing. For that purpose a particle tracking technique

developed byLamberg and Vianna (2007)was used. The ParticleTracking was done with HSC Chemistry software developed byOutotec (Lamberg et al., 1997). The algorithms of the ParticleTracking were also used in textural classification of progeny parti-cles, as described below.

Particle tracking uses mass balanced mineral by size data.Instead of using mineral grades such as those determined byQEMSCAN (Gottlieb et al., 2000; Pirrie et al., 2004), the modalanalysis was done with an element to mineral conversion tech-nique and calculation procedure developed byLund et al. (2013).Using HSC Chemistry software the raw mineral grades with solidsflow rates data were first balanced on an un-sized basis and thenon a size by mineral basis with the additional constraint that theun-sized mass balance is conserved.

The QEMSCAN analysis identified a total of 16 minerals but tomake this match with the mineralogy determined from element

to mineral conversion some of the trace minerals were combined

with more abundant minerals. This gave a total of five most impor-tant minerals (the average value of ore sample; magnetite(55 wt.%), feldspar (25 wt.%), amphibole (9 wt.%), biotite (5 wt.%)and apatite (4 wt.%)) (Lund et al., 2013) to be used in textural clas-sification and in the particle tracking. After combining the lessabundant minerals, the next step of the Particle Tracking is toadjust the mass proportions of the particles to meet the mass bal-anced modal composition on sized level. This stage adjusts themass proportion of measured particles but does not change theircomposition. The adjustment for the liberation data is very smallbut is needed to make the particle data fully consistent with thesized level chemical mass balance. This means that if the modaland chemical composition of the size fractions and streams is backcalculated from the adjusted particles the result is exactly the same

as received from the sized level mass balance.To mass balance particles they must be grouped somehow andin

the particle tracking this is done based on composition. The basicbinningstage groups particlesfor exactlydefinedbinaryandternarycombinations. In this study the system of five minerals plus others,(tot. six minerals) and five size fractions the total number of parti-cles after this step, in each stream, end up to be 2166. Some of theparticle groups may have very few particles if any. Therefore, theadvanced binning is a stage which combines established groups toensure that each group has enoughparticles for sound mass balanc-ingof particle groups. The liberation measurementis challenging invery fine (

7/24/2019 1-s2.0-S0892687515001417-main.pdf

4/17

3. Description of textures

Textures are critical mineralogical characteristics in governingore behaviour in mineral processes. This is commonly acceptedand has been widely known for decades (e.g.Butcher, 2010). Butthe question remain how to describe textures and how to use thetextural information e.g. in geometallurgy (e.g.Bonnici, 2008).

The geological definition fortexturesis: The relative size, shape,and spatial interrelationship between grains and internal featuresof grains in a rock (e.g.Spry, 1969; Fettes and Desmons, 2007).Texture, refers thus to a general description on the small scalephysical properties. These attributes are usually described in qual-itative terms, e.g. pegmatitic texture, fined grained granoblastictexture. Such descriptive information, such this is insufficient ingeometallurgy as it only gives possibilities to use the texture infor-mation for classification of different ore types. In geometallurgythis may lead to a great number of domains and for each of themseparate sample sets need to be collected for geometallurgical/metallurgical testing. If, on the other hand, texture could bedescribed numerically, and even with additive parameters, thisinformation could be processed with geostatistical methods andbe used in ways similar to those for metal grades that are currentlyused in resource estimation (Glacken and Snowden, 2001).

3.1. The Malmberget iron ore deposit

The Malmberget iron ore deposit consists of several ore bodiesof massive magnetite and hematite (Geijer, 1930) that have beenexposed to extensive regional metamorphism, deformation andintruded by several generations of felsic and mafic dykes(Martinsson and Virkkunen, 2004). As a result magnetite andhematite ores are metamorphically overprinted evidenced byrecrystallized coarse grained ore minerals, metamorphic textures,different oxidation textures and a chemical redistribution of ele-ments between the magnetite and hematite (Lund, 2013).

Compared to the well-known Kiirunavaara ore body, theMalmberget deposit is lower in tonnages, slightly lower in Fe gradeand has a moderate but varying P content (Bergman et al., 2001).Mineralogically the iron ores are rather simple. Magnetite andhematite are the main ore minerals, and typical gangue mineralsare apatite, amphibole and pyroxene.

Massively-textured ore represent the bulk of the Malmbergetiron ore bodies. These massive orea broad variationof mineral-tex-tures can be identified. The magnetite has textural variationsmainly in the magnetite grain size, grain shape and association.The texture is mostly granoblastic with distinct triple junctions atthe grain boundaries. Coarse grained magnetite occurs as porphy-roblasts or as veins in a finer grained matrix of magnetite. Largermagnetite grains can have an elongated shape following the min-

eral lineation in the host rock. In the western part of the deposit,hematite is the main mineral together with magnetite. Typicalmetamorphic textures are exsolutions patches of spinel in mag-netite or oxidation surfaces of hematite. The magnetitegrains showalso intergrowth of ilmenite as lamella or rutile needles.

As contrast to the massive ore, semi-massive ore surrounds theore bodies. It abounds in particularly in the eastern part of theMalmberget deposit. It is lower in Fe and higher in SiO2 and canextend several tens of meters with a decreasing iron grade. Themain non-iron minerals are the silicates e.g. amphibole, feldspar(albite and orthoclase), quartz and biotite in various proportionsthat generate several different mineral assemblages with complextextures.

The distinct chemical discrepancies of the massive ore and the

semi massive ore are important to consider as the ore resourceestimation uses traditional 3D block models which solely are based

on the chemical analyses of the ore samples. The mass proportionamounts of semi-massive ore that is present in the process is con-sequently very significant and will not only influence the orereserve estimation but also contribute to elemental variations inbeneficiation. Therefore, a classification between the massive oreand the semi-massive ore can clarify the true distribution of thechemical elements from the valuable and the gangue mineralsand give more precise information for defining the boundaries ofthe ore bodies.

3.2. Textural classification

Using drill core logging (e.g. macroscopic observations) andoptical microscopy an attempt was made to further develop theclassification systemof the GEM-types which now also will includemineral textures. The classification is based on the mineralogicaland textural composition of both the massive and the semi-mas-sive ore (Table 2).

The textures were described and classified according to mag-netite grain size, shape and the main mineral associated with mag-netite. For that purposes more than 4500 grains of magnetite andassociating minerals were measured using the EquivalentProjection Area of a Circlewhich included 60 polished thin sections.For the magnetite grain size the relative standard deviation (RSD)was between 7% and 11% (Fig. 2). The following eight texture types,arranged in the order of increasing magnetite grade, were identi-fied: disseminated (1), banded (2), veiny (3), patchy (2) granules(5), clustered (6), speckled (8) (semi-massive ore) (Table 3) andmassive texture (massive ore) (Table 2).

In Fig. 2 the different texture type of the semi-massive ore givesa clear correlation between the magnetite grain size and the mag-netite grade (e.g. Fe wt.%). The low (Fe) grade texture type calleddisseminated texture (1)(Table 2) have an average magnetite grainsize of about 40 lm, and the magnetite grain size increases foreach texture type through the semi-massive ore up to 140 lm, toa higher (Fe) grade texture type, calledspeckled texture (8). In the

massive ore, the grain size increases up to 1400lm (Table 2).Even this can be regarded as a general rule, the massive ore andsome texture types of the semi-massive ore show also variationin the magnetite grain size and can be further classified as fine(100400lm), medium (400800 lm) and coarse grained (8001400 lm).

The feldspar (Fsp) GEM-type in the semi-massive ore shows thelargest variation in textural features. It basically consists of twodifferent materials identified by their colour. Melanocratic mag-netite-rich material brecciates the leucocratic feldspar-rich mag-netite-poor matrix. The melanocratic parts consist of magnetite,apatite and amphibole veins, bands or patches that locally developinto networks in the leucocratic rock. The mineralogy of the leuco-cratic rocks consists of a varying proportion of albite, orthoclase,

quartz and biotite with minor magnetite (Table 3).Thegrainsizeofthemagnetiteintheleucocraticpartsaresmaller

in the samples for alltexture types than in the melanocratic parts asshown inFig. 2. The difference is about 4060 lm for each texturetype. In bothleucocratic and melanocratic parts the magnetite grainsize increases with magnetite grade, and in addition the overallmagnetite grain size increases from texture type 18 (Fig. 2). Thetexture type 2 (banded) is deviating from the linear correlationand thus shows an uncertainty when the Fe-grade is low.

The Bt-(Amph-Ap) GEM-type is structurally controlled and islocated in the footwall of the massive ore. Texturally, theBt-(Amph-Ap) GEM-type is unique showing a schistose appearancedue to the orientationof biotite. Tabular biotite grains, mainly (10400lm), sometimes as large as 2000lm, occur disseminated or as

a cluster that follows the lineation in the host rock either intersti-tially to a magnetite- or a silicate granoblastic matrix (Table 4).

64 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

5/17

The Py bearing Amph-(Ap-Bt) GEM-type contains a significantamount of sulphides. Single euhedral grains of pyrite andchalcopy-

rite are in general 5200lm in size, occasionally up to 4800 lmand occur as dissemination and in veinlets, at the grain boundariesor interstitially to magnetite. This texture occurs both in the mas-sive ore but more commonly in the semi-massive ore. The com-position of this GEM-type regarding the mineral texture and themodal mineralogy is closest to the Amph-(Ap-Bt) GEM-type.

Two different GEM-types exist in the massive ore: Amph-(Ap-Bt) and Ap-(Amph). The Amph-(Ap-Bt) type shows wide variationin the amphibole (-pyroxene) grain sizes and grain shapes, andtwo end members can be identified. The coarser grained variantconsists of large recrystallised amphibole (-pyroxene) grainsoccurring as a cluster (0.83.2 mm) enclosing smaller magnetitegrains (Table 4). In the finer grain variant, amphibole grains (0.21 mm) occur interstitially to magnetite grains, either with or with-

out the inclusion of small magnetite grains. In both types the mag-netite grains are similar in size; the biggest difference between

them is that in the coarse grained variant the magnetite grains alsooccur as inclusions in amphibole, whereas in the finer grained vari-ant, the magnetite occurs along the grain boundaries. These twotextures can easily be identified during drill core logging due tothe significant difference in amphibole grain size (Table 4). The

Ap-(Amph) type is texturally uniform and is marked by equallysized apatite and magnetite that are closely associated to form ina granoblastic or porphyroblastic texture (Table 4).

The classification visualise that the combinations of the differ-ent mineral textures are unique for almost each GEM-type, butthey do have in common that each of them has both a fine and acoarse grained variant (Table 1).

The textural study shows that when the mineral textures areconsidered the modal GEM-types divide into numerous types, andthe closer one looks the more complicated and numerous theclasses become (Table 1). In a geometallurgical context the use ofclasses is problematic since the treating of non-numeric data inblock modelling is challenging. Therefore, to change the geologicalmodel from descriptive to practical mineral textures must be chan-

ged from qualitative to quantitative. Therefore, the mineral tex-tures are now considered from the mineral processing viewpoint.In mineral processing, ore is comminuted to liberate ore from

gangue minerals and to make the particle size suitable for down-stream processes. Comminution is an energy intensive stage, andtherefore a good balance is needed between mineral liberationand throughput. Full liberation is not a feasible target since besideshigh energy required the separation efficiency of downstream pro-cesses tends to decrease towards very fine particle sizes (

7/24/2019 1-s2.0-S0892687515001417-main.pdf

6/17

In comminution, the ore is broken into particles using multiple

stages of size reduction, such as blasting, crushing and grindingcombined with classification processes such as screening and

hydrocycloning. The behaviour of an ore during comminution is

dependent on machine parameters, such as nature and magnitudeof the applied comminution energy (unit operation properties and

Table 3

Eight representative and different mineral textures identified from Fsp GEM-type of Fabian and Printzskld ore bodies.

66 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

7/17

operational parameters), and on the materials physical properties,such as elasticity, hardness and strength. These together will definein a given process the relationship between specific energy

(energy/mass) and overall size distribution of the material. Theterm grindability describes this, and it is themeasure of the specificenergy consumption required to reduce a certain mass of material

Table 4

Different ore textures that are identified and defined in massive and semi-massive ore from Fabian and Printzskld ore bodies.

(continued on next page)

C. Lund et al. / Minerals Engineering 82 (2015) 6177 67

7/24/2019 1-s2.0-S0892687515001417-main.pdf

8/17

from a given fresh and initial size down to a defined product size by

grinding. Similarly, the term crushability is used for crushing.Comminution circuits are designed and operated to provide a

targeted product size, and almost without exception this figure isfixed or changed very seldom (Alruiz et al., 2009). However, it iscommon that the grain size and association of the ore minerals

vary within the ore body and therefore in plant feed on a dailybasis. A good geometallurgical model should therefore not onlyforecast the metallurgical response but also give the best opera-tional parameters, e.g. the target liberation degree and accordinglythe target grinding size for any given rock unit or plant feed(blend).

Table 4 (Continued)

68 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

9/17

The liberation distribution of a comminuted ore is dependentnot only on (1) nature and magnitude of applied comminutionspecific energy and (2) type of comminution equipment (3) crusha-bility and grindability, but also on (4) modal mineralogy and (5)mineral textures. The first factor is independent of material. Thefour other factors cannot be separated fully from each other sincethe physical properties controlling the comminution behaviour arehighly dependent on modal mineralogy and mineral texture (Gay,2004; Ozcan and Benzer, 2013; Mwanga, 2014). Since there areestablished techniques to experimentally determine the crushabil-ity and grindability (Bond, 1961; Toraman et al., 2010) and to usethem in process simulations (King, 2001), the third factor is takenas an isolated parameter, and it determines together with the firstparameter the overall particle size distribution of the material aftercomminution.

How does one recognize respective effects of the modal min-eralogy and texture on the liberation distribution? Firstly, the min-eral grain size is basically a pure textural property, but as shownabove the magnetite grain size has a positive correlation with themagnetite grade (Fig. 2). Similar interdependence is noted for theeffects of the associated minerals that are controlled both by themodal mineralogy and textural parameters. In the Fsp GEM-typethe association of magnetite in the melanocratic and leucocraticparts is different (Table 2). These examples show that the modalmineralogy and the mineral textures are strongly interrelatedand their separation using a traditional textural definition is prac-tically impossible. Therefore, a new definition for mineral texturesis proposed:

Two samples are geometallurgical different in texture if the libera-

tion distribution by size (compensated against modal mineralogy)

show significantly different metallurgical result when processed

in a given circuit.

This is an effort to separate the comminution properties and themodal mineralogy from the textural properties as well as establisha way to do the textural classification.

3.3. Textural archetypes

By using the liberation distribution of the different GEM-typesan attempt was made to quantify the mineral textures using sam-ples including different textural types (Table 1). All the samplesshow a variation in the modal mineralogy by size fraction as illus-trated by the magnetite grade by size in which also identifies thegrade distribution varies by ore type (Fig. 3a). In general, the FspGEM-types show an increase in the magnetite content by particlesize. A similar pattern is found in the Ap GEM-types with an excep-tion that the finest size fraction has the highest magnetite grade.Differing from the previous the Amph GEM-type shows the highestmagnetite content in the middle size fractions, e.g. 75150 and

150300 lm. The variation in the mineral grade by size gives thefirst challenge when comparing the liberation distribution(Fig. 3b).

The degree of liberation of magnetite decreases in all samples asparticle size increases (Fig. 3b). However, the overall liberationdegree (of magnetite) can be different in the bulk sample even ifthe liberation degree by size is similar, if the overall particle sizedistribution if different. Therefore, the degree of liberation fromthe textural point of view must be studied by size. In (Fig. 3b)the Fsp GEM-types from Fabian (Fa Fsp) and Printzskld (Pz Fsp)

show clearly that they have similar liberation degrees for mag-netite in four size fractions, but in the coarsest size fraction 425600lm. The Fabian sample shows much better liberation and this

causes the bulk liberation degree to be different (Fig. 3b).In order to compare the liberation distribution of five different

samples, a characteristic size fraction, 150300 lm, was selected.This size fraction was selected because the magnetite grade inthe 150300 lm size fraction is close to the bulk sample; the massproportion in crushed sample was high enough and the number ofparticles measured gives sound statistics. In this size fraction thedegree of liberation of magnetite has a positive correlation withthe magnetite grade, but the both Fsp GEM-types of Fa and Pz dif-fers significantly from the Amph GEM-type and Ap GEM-type. Inthe Fsp GEM-types the magnetite liberation is better than thegrade would suggest (Fig. 4a).

The comparison of mineral association for non-liberated parti-cles is challenging. If just comparing the mass proportion of mag-

netite with, for example, albite the value is affected by themagnetite liberation and the albite grade. If the liberation degreeis high, the association of magnetite with albite must be low, andsimilarly if the albite grade is low the mass proportion of the bin-ary magnetite-albite grains will also be low. Therefore, a newAssociation Index (AI) was developed. It aims to describe howcommon it is to find the target mineral associated with the otherminerals regardless of the liberation degree and the modal com-position. The association index for each mineral pair is calculatedusing the following formula:

0

10

20

3040

50

60

70

80

90

100

Bulk 38-75 um 75-150 um 150-300 um 300-425 um 425-600 um

Magneteli

beraondegree(%)

Size fracon

Fa Fsp GEM-type Pz Fsp GEM-type Fa Amph GEM-type

Fa Ap GEM-type Pz Ap GEM-type

(a)

(b)

Fig. 3. (a) The magnetite grade plotted against the particle size in five differentGEM samples from Fabian (Fa) and Printzskld (Pz). (b) The degree of liberation formagnetite by size fraction in the five different GEM samples.

AIAB Association of mineral A with mineral Bwhen fully liberated grains are excludedwt%

Mineral B grade in a fraction excluding mineral Awt%

C. Lund et al. / Minerals Engineering 82 (2015) 6177 69

7/24/2019 1-s2.0-S0892687515001417-main.pdf

10/17

If the association index of a mineral pair A B is 1, then theassociation of mineral A with B is as common as the modal com-position would suggest. If the association index is greater than 1,then the association is more common than expected, and if it islower than 1 then it is rarer than expected. An index value of zeroshows that there is no association between minerals A and B.

Hypothetical examples of the calculations of the associationindex are given inTable 5. The fictitious samples 13 inTable 5show a similar association index with different modal com-positions. Sample 4 inTable 5shows that even if the deportmentpercentages are equally high for the pair AB and AC, the associa-tion index shows that A can be found relatively more often with

phase C than with B. Sample 5 shows that the association indexcannot be calculated for binary systems (C = 0).

The association indices of magnetite with feldspars(albite + orthoclase), biotite, amphibole and apatite are shown forthe selected size fraction (e.g. 150300lm) of five differentGEM-types types from the Malmberget iron ore inFig. 4b. TheFsp GEM-ore types of Fa and Pz show quite similar associationindices for feldspar, amphibole and apatite, but not for biotite.The associationindex of magnetite with feldspar (AIMgt-Fsp) is smal-ler than 1in Fa and Pz Fsp GEM-type, which means that magnetite,is associated with feldspar less often than the modal mineralogywould suggest. On the other hand the association index of mag-netite with amphibole (AIMgt-Amph) is higher than 1 in Fa and PzFsp GEM-type showing that the magnetite is more often associatedwith amphibole than the modal mineralogy would suggest. Theassociation index of magnetite with apatite shows a significant dif-ference between the Fsp and Amph (and Ap) GEM-types. In the FspGEM-types the magnetite is rarely occurring with apatite, but inthe Amph and Ap GEM-types the association is more common thanthe modal composition would suggest.

Because the association index is calculated for particulate mate-rial, it carries information on both the mineral textures (grain size,shape, associating mineral) and the ore breakage. This is clearlyillustrated inFig. 4c which shows the variation of the associationindex by size for theFa FspGEM-type samples. In principleit meansthat if the texture is fully homogenous and the breakage fully ran-dom, the association indices should be 1 for all minerals in all sizes.Deviation from 1 can be due to heterogeneous texture or non-ran-dombreakage.The associationindicesapproach1 as theparticlesizegets coarser, as shown inFig. 4c. This is because in coarse particlesizes the particles start to be similar in their modal compositionand liberation distribution. This point is not reached yet at the425600 lm particle size for Fsp GEM-type sample of Fabianbecause the liberation degree of magnetite is as low as 74% (Fig. 3).

The decoupling of the texture and breakage is not possible withthe association index. Some more informationis needed, and there-fore the association index values are compared to the texturaldescription of original uncrushed ore samples. The Fsp GEM-type

consists of leucocratic and melanocratic parts and the magnetiteand amphibole content is higher in the latter and feldspar grade ishigher in the former. Association index values for magnetite withfeldspar (AIMgt-Fsp) lower than 1 and higher values with amphibole(AIMgt-Amph) can therefore be explained based on the mesoscopictexture (Fig. 4b). Additionally, the variation in the association indexof magnetite with apatite between the different GEM-types can beexplained fromtextural differences. In the Fa Amph GEM-type, apa-tite preferentially occurs in contact with magnetite, but this is notthe case in the Fsp GEM-type. The association index of magnetitewith biotite shows strong variation by particle size: in fine particlesizes the value is very high and decreases by size (Fig. 4c).

Coming to the definition of texture, the Fsp GEM-types ofFabian and Printzskld can represent similar textures since the

liberation degree of magnetite and association indexes do not dif-fer from each other strongly. The Fabian Amph and Ap GEM-typesare also potentially similar in texture, and the Printzskld Ap GEM-type is a textural type of its own. The question remains how smallor big the difference between the liberation distributions and therelated key figures used here (degree of liberation and associationindex) should be to justify saying that two samples are similar ordifferent in their textures In addition how this information willbe gathered and used in a geometallurgical context.

To extend the modal compensation also into liberation degree,the calculation algorithms of the particle tracking technique wereapplied (Lamberg and Vianna, 2007). The particle tracking tech-nique is for mass balancing the liberation data, but for doing thatit includes steps where the particles are classified in a systematic

way (basic binning and advanced binning), and the modal com-position of the liberation measurement is refined to match with

0

0.5

1

1.5

2

2.5

3

3.5

4

AI (Mgt-Fsp) AI (Mgt-Bt) AI (Mgt-Amph) AI (Mgt-Ap)

Associaonindex(AI)150-300m

Fa Fsp GEM-type Pz Fsp GEM-t ype

Fa Amph GEM-type Fa Ap GEM-type

Pz Ap GEM-type

(a)

(b)

(c)

Fig. 4. (a) The degree of liberation for magnetite versus the magnetite grade ofthree different GEM-type textures (Fsp, Amph and Ap) representing the sizefraction, 150300lm. (b) The association index (AI) for magnetite in size fraction150300 lm in the same GEM-type texture. (c) The association index of magnetitein sample by size from Fa Fsp GEM-type fractions.

70 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

11/17

the modal composition calculated by the element to mineral con-version. Therefore, the liberation data of five GEM-type sampleswas classified in such a way (e.g. basic binning) as to produce anidentical particle population in each sample. After this, the libera-tion distribution of each sample was forecasted using the texturalinformation from another sample. For this an algorithm given in

the textbox below was used (Lamberg and Lund, 2012). For exam-ple, to forecast the liberation distribution of sample A compared tosample B (called archetype), the liberation spectrum of the arche-type (sample B) was taken, and it was refined by using the modalcomposition of sample A.

The particle population for a given sample is calculated by

taking the particle population of the corresponding arche-

type. This is now iteratively adjusted, and for the adjust-

ment, a correction factor, k, is calculated for each

mineral (i) in a size fraction before each iteration round:

ki;fraction MifractionPnj1

pjfractionxip

Basically the formula above is a ratio of mineral grade

from the geological model (M(i)) and the mineral grade

back calculated from the liberation data of the archetype

(denominator). p refers to mass proportion of particle in

a size class and x(i) is the mass proportion of mineral in

a particle.

The mass proportion of particle j (pj) is recalculated on

each iteration round using the correction factor and an

equation:

pj;fraction pj;fraction PL

j1xij ki;fraction

Using equations given above, the particle mass propor-

tion is iteratively adjusted until the difference between

mineral grades of the geological model and archetype

has reached the required tolerance defined case by case.

The results of the magnetite deportment of one pair Fa and PzFsp GEM-types are given inTable 6. The algorithm forecasts theliberation distribution well in all size fractions except in the coars-est 425650 lm. This is most probably due to the small number ofparticles measured (about 400).

Fig. 5shows the error done in forecasting the liberation dis-

tribution with all of the combinations of the five GEM-types sam-ples. For example when Fa Fsp GEM-type liberation distribution isused to forecast the mode of occurrence of magnetite in Pz FspGEM-type the difference (e.g. error) is less than 1% but for the FaAp GEM-type it is almost 10%. Putting a limit that the average errormust be lower than 1%, and then the five GEM-types can be simpli-fied into two textural archetypes: Fsp and Amph/Ap. One samplefrom each group is selected to represent the group, and this sampleis called a textural archetype.

4. Framework for a metallurgical model

The process model takes the information of the geological

model and transfers it to forecast on the metallurgical perfor-mance. When developing a metallurgical model, one needs toanswer questions, such as: (1) what is the purpose of the model?(2) What is the level of the complexity of how material (ore) isdescribed? (3) And how detailed is the information that the modelcan provide on a metallurgical performance?

Typically in the processing plants this would include informa-tion like throughput, concentrate quality, recovery of the com-modities and the tailing quality. In the case of Malmberget, themetallurgical parameters considered by the process model are:

(1) Throughput, recovery of iron and concentrate grade in termsof iron, phosphorous and silica. The tailing quality is not alimiting factor in Malmberget, and there is no need toinclude that in the model.

Table 5

Hypothetical calculation examples for the association index. For the calculation modal [(1)(3)] and deportment [(4)(7)], the

analysis result is needed.

C. Lund et al. / Minerals Engineering 82 (2015) 6177 71

7/24/2019 1-s2.0-S0892687515001417-main.pdf

12/17

In this study the geological model describes the resource on amineralogical basis, and the process model logically uses the sameapproach. Three different levels can be used

(2) The un-sized level uses minerals on an un-sized basis, whichmeans that there is no information on the particle sizesavailable. The sized level uses the mineral by size for thematerial description and for the process models. Thisenables including particle size and the metallurgical func-tions can be size dependent. The liberation level uses libera-tion information. Therefore, the process must describe thebehaviour of particles by their composition and size.

(3) The level of processing details means how detailed the pro-cess is described in terms of unit operations and operational

parameters. In the lowest level, one black box represents thewhole process and no operational parameters are included.In the other end of the complexity lies a model where allunit operations are described and all effective operationalparameters are included.

In this study the feasibility of the mineralogical approach istested using a simple process: the one stage dry magnetic sep-aration (cobbing) test for the Fsp GEM-type of Fa and Pz ore. Theprocess model is developed here includes all three levels.

To reach the sized level, the calculated modal mineralogy needsto be mass balanced on a size fraction level. The readers arereferred toLund et al. (2013)for the calculation rules and detailedresults. The liberation distribution level was reach by the liberation

analysis of the concentrate and tail and was then mass balancedusing the Particle Tracking technique.

4.1. Comminution-particle breakage model

The comminution unit models (grinding mills, crushers), is itpossible to use the particle breakage model. Therefore, in themodel the forecasting of the liberation distribution and the totalsize distribution can be decoupled. In the latter the traditional pop-ulations balance breakage models can be used (Weller et al., 1996:Alruiz et al., 2009; Vogel and Peukert, 2003).

The particle breakage model gives the liberation distribution ofa sample when the information on corresponding textural arche-type and modal composition is given (Figs. 6 and 7). The modelconverts the un-sized modal composition, to the liberation dis-tribution using a textural archetype as a basis and a simple algo-

rithm in adjusting the liberation data of the archetype to matchwith the given modal mineralogy. The textural archetype alsoincludes the information on how the modal composition variesby size. The overall size distribution model developed by Koch(2013)was used.

4.2. Concentration model dry magnetic separation

For the Malmberget concentration process, the first unit modelwas developed for the dry magnetic separation stage, e.g. cobbing.Fig. 8 shows the distribution of minerals on an un-sized basisbetween the concentrate and tailing for the Fsp GEM-type fromFabian. In a particle size of P80 = 1 mm, about 30% of the material

is rejected into the tail (MagTail) with about 6% magnetite losses,e.g. the recovery of magnetite into the magnetic concentrate is 94%.

Fig. 5. The average error showing the magnetite deportment in the samples.

0

10

20

30

40

50

60

70

8090

100

Liberaon(%)

Size fracon (m)

Mgt=80

Mgt=70

Mgt=60

Mgt=50

Mgt=40

Mgt=30

Fig. 6. The mass proportion of fullyliberatedmagnetite in different size fractionsasthe magnetite grade in the ore varies between 30 and 80 wt%. The bulk refers tocombined size fractions.

Table 6

The magnetite deportment measured and estimated using the Fsp GEM-type of Fabian and Printzskld and the difference

between the estimated and measured.

72 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

13/17

Studying the behaviour of the minerals by particle size it can beseen that for magnetite the recovery is quite constant between 92to 96% in particle size range from 1.68 mm (Fig. 9). Allof the gangue minerals show a similar pattern having the recoveryminimum between 38 and 106 lm (Fig. 9). Noteworthy is that the

biotite shows higher recoveries than the other gangue minerals,especially in the finest particle size fraction

7/24/2019 1-s2.0-S0892687515001417-main.pdf

14/17

and for recovery 1.3% (FaFa). The Fa model succeeds to forecastthe metallurgical performance for Pz when the liberation informa-tion is readily available: the standard deviation of the difference forgrades is 0.93% and for recovery 2.7% (PzFa). When the liberationinformation is not available and the modelled feed sample is gen-erated from the textural archetype, the forecast is still reasonablygood; the standard deviation of the difference in grades is 0.92%and the recovery 5.5%.

As the process model is based on the particle properties, themodel can be used more widely than traditional unit models whichneed to be calibrated if there is a change in size distribution, modalmineralogy or liberation.

5. Discussion

This is the first effort to develop a geometallurgical frameworkrelying on two important mineralogical characteristics: modalcomposition (Lund et al., 2013) and mineral textures. The frame-work demonstrates how the mineralogical information can be usedfor metallurgical predictions. The method relies on qualitative dataprovided from drill cores and ore samples. The requirements forthis framework to be useful in modelling and simulation necessi-tate a capacity to handle large amounts of data fast, be inexpensiveand also to have a practical applicability in the process.

To get the full benefit, the geometallurgical model should be

established already in a feasibility stage. For ore deposits in pro-duction, like Malmberget, the geometallurgical model can poten-tially be used to recognise the process limitations and re-evaluate the mining plan. It facilitates also the possibility to makerealistic process predictions which potentially improve the perfor-mance by giving on daily basis useful and achievable productiontargets.

To use modal mineralogy in geometallurgical classification forsome weathered iron ores and bauxite ore has shown to be usefulbefore (Paine et al., 2011; Neumann and Avelar, 2012) but a

0

10

20

30

40

50

60

70

80

90

100

10 100 1000 10000

Recovery,%

Size fracon (m)

Mgt

Ab

Bt

Amph

Ap

0

20

40

60

80

100

0 10 20 30 40 50 60 70 80 90 100

MineralReco

very%

Mineral association with magnetite %

0

20

40

60

80

100

Mineral

Ab

Amp

Ap

Bt

Mgt

(a)

(b)

Fig. 9. (a) The cobbing test with a sample from the Fsp GEM-type of Fabian showsthe recovery of minerals by size. (b) The mass proportion of minerals associatedwith magnetite vs. mineral recovery where five size fractions between 38 and600lm are shown. The fully liberated grains are shown for magnetite.

Table 7

QEMSCAN pictures of typical particles from the magnetic concentrate and tail, Fabian Fsp GEM-type samples. Brown is

magnetite, orange is feldspar.

0

10

20

30

40

50

60

70

80

90

100

Recovery

intoconcentrate(%)

The mass proporon of magnete in parcle (wt%)

0-38 um

38-75 um

75-150 um

150-300 um

300-425 um

425-600 um

Fig. 10. The recovery of the magnetite-feldspar binary particles into the magneticconcentrate by size fraction as a function of magnetite grade in the particle.

74 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

15/17

question still remains about how accurate the analysis needs to bein a geometallurgical context. The error in the element to mineralconversion for Malmberget (Lund et al., 2013) is too high to beused directly in resource estimation (Pitard, 1989), but can it beenough for geometallurgical modelling? A study byKoch (2013)on Malmberget samples shows better accuracy; the relative stan-dard deviation for magnetite grade estimate is 45%. One way tofind this out would be Monte Carlo simulations and identify howmuch the error of the modal analysis affects the estimate formetallurgical performance, such as iron recovery and silica gradein the magnetite concentrate.

The quantification and usage of mineral textures is still a chal-

lenge and so far quite undeveloped and qualitative. There are somenew methods to produce the mineralogical and textural data,rapidly and inexpensively (Hunt et al., 2009, 2011; Bonnici et al.,2008, 2009; Perez-Barnuevo et al., 2012) and even use the texturalinformation in modelling (Hunt et al., 2012). These methods arebased on optical image system which also addresses a moreinexpensive alternative than the use of automated mineralogy sys-tems. The two textural archetypes developed in this study aims toestablish a library of different textural archetypes for an oredeposit or even to be generic for all ores. This method rely onthe assumption that textures can be classified in drill core loggingwithout the need of optical or electron microscopy for a large num-ber of samples.

However, these tested textural archetypes do not answer how

to recognise whether two samples are texturally different and ifa new item in the textural archetype library should be added.

Here, the comparison and identification was based on the degreeof liberation by size and the association index (AI). Another ques-tion for textural archetypes is how large the variation in the modalmineralogy one textural archetype can cover and whether eachseparate iron ore deposit needs textural archetypes of its own(e.g. variation in grain size). This will require more experimentalwork to be conducted and also use several different ore depositsto be tested.

The developed metallurgical framework adopted here relies onassumption that similar particles, in terms of composition and size,will behave similarly regardless where they coming within the ore.Instead of extensive variability testing only few samples were

tested and the behaviour of different particles was studied indetails. Once being capable to forecast the behaviour of individualparticles accurately in the process the challenge is shifted to beable to forecast what kind of particles are generated when an oreis comminuted. An approach which uses textural archetypes andsimple compensation algorithm to correct the effect of mineralgrades was developed. The results from the tests of this approachshow error level below 5% which fulfils the criteria for technicalsampling and estimations (Gy, 1982; Pitard, 1989). Even if thisframework is based on a one stage magnetic separation it givessome indications that the approach could be used for producingmetallurgical parameters in a future for block model. Question is,if it is possible to track some textural impact in the final concen-trates remains open and parameters that are not included in the

model, needs to be validated and further developed. In addition aunit operation models of concentration and comminution capable

Table 8

A model for the dry magnetic separation. Split values for minerals by size when occurring fully liberated.

Table 9

The observed grades and recoveries in the magnetic concentrate in the cobbing test. The results are the sum of the size fractions

from 38 to 600 size fractions. Diff: Difference (Sim-Meas), R.Diff: relative difference = 100 (Sim-Meas)/Meas.

C. Lund et al. / Minerals Engineering 82 (2015) 6177 75

7/24/2019 1-s2.0-S0892687515001417-main.pdf

16/17

to handle the variation in hardness and throughput need to beconsidered.

6. Conclusion

In this paper a geometallurgical framework using theMalmberget iron ore deposit as a case study was developed. The

results of this work provided an indication of the potential of usingmodal mineralogy and mineral texture to forecast the productgrade and recovery.

A classification scheme was established to identify the mineraltextures that are present in the ore and is used to verify theaccurateness of the developed textural archetypes. Two differenttextural archetypes are defined in terms of the degree of liberationby size and the association index (AI). This developed techniquedescribes and distinguishes quantitatively different mineral tex-tures by processing the liberation data of the included particleswithout the influence of the modal mineralogy to be used in a geo-logical model. Based on the modal composition and the mineraltextures, three different geometallurgical ore types (GEM-type)were established for the Malmberget ore body. Each of these

GEM-types describes quantitatively: the present mineral and itschemical composition, rules how to calculate the modal com-position from routine chemical assays (Lund et al., 2013) anda tex-tural archetype in a library of archetypes.

Finally, the geometallurgical framework and the geologicalmodel were tested comparing the forecasted metallurgical key fig-ures against the ones received by metallurgical test. The differencein terms of product quality (Fe, Si and P grade) and in iron recoverywere reasonable small validating the developed methodology.

Acknowledgements

This project is financial supported by the Hjalmar LundbohmResearch Centre (HLRC), which we are thankful for. The authors

want to express their gratitude to people at the LKAB for their con-tribution of data and knowledge. ke Sundvall, Magnus Stafstedt,LKAB are acknowledged for comments and support.

References

Alruiz, O.M., Morrell, S., Suazo, C.J., Naranjo, A., 2009. A novel approach to thegeometallurgical modelling of the Collahuasi grinding circuit. Miner. Eng. 22,10601067.

Andrews, J.R.G., Mika, T.S., 1975. Comminution of a heterogeneous material:development of a model for liberation phenomena. In: Proceedings of 11thInternational Mineral Processing Congress, Cagliari, Italy, 2026 April,Proceedings, pp. 5988.

Bergman, S., Kbler, L., Martinsson, O., 2001. Description of regional geological andgeophysical maps of northern Norrbotten County (east of Caledonian orogen):SGU Geological Survey of Sweden, v. Ba 56, p. 110.

Bond, F.C., 1961. Crushing and grinding calculations. Brit. Chem. Eng. 6, 378385.Bonnici, N., Hunt, J., Walters, S., Berry, R., Collett, D., 2008. Relating TexturalAttributes to Mineral Processing Developing a MoreEffective Approachfor theCadia East CuAu Porphyry Deposit: Ninth International Congress for AppliedMineralogy, Brisbane, Australia, 810 September, Proceedings, pp. 415418.

Bonnici, N., Hunt, J., Berry, R., Walters, S., McMahon, C., 2009. Quantified mineralogyand texture: informed sample selection for communition and metallurgicaltesting: The Tenth Biennial SGA Meeting, Townsville, Australia, 17th20thAugust, Proceedings, pp. 679681.

Bulled, D., McInnes, C., 2005. Flotation plant design and production planningthrough geometallurgical modeling Publication Series, Australasian Institute ofMining and Metallurgy, v. Publication Series, pp. 809814.

Butcher, A., 2010. A practical guide to some aspects of mineralogy that affectflotation, flotation plant optimisation. Spectrum Ser. 16, 8393 (Chapter 4).

David, D., 2007. The importance of geometallurgical analysis in plant study, designand operational phases., v. Australasian Institute of Mining and MetallurgyPublication Series, pp. 241247.

Dobby, G., Bennett, C., Bulled, D., Kosick, X., 2004. Geometallurgial modeling thenew approach to plant design and production forecasting/planning, and mine/

mill optimization. In: Proceedings of 36th Annual Meeting of the CanadianMineral Processors, Ottawa, Canada, 2022 January, Proceedings, p. paper 15.

Farrell, J., Miller, A., Gaze, R., 2011. Geometallurgical Sampling and ResourceEstimation for Magnetite Deposits: First AusIMM International GeometallurgyConference (GeoMet), Brisbane, Australia, 57 September, Proceedings, pp.311319.

Fettes, D., Desmons, J., 2007. Metamorphic Rocks A Classification and Glossary ofTerms: Cambridge. Cambridge University Press, U.K., p. 244.

Gaudin, A.M., 1939. Principles of Mineral Dressing. McGraw-Hill Companies, NewYork, USA, p. 554.

Gay, S.L., 2004. A liberation model for comminution based on probability theory.Miner. Eng. 17, 525534.

Geijer, P., 1930. Geology of the Gllivare iron ore field: geological Survey of Sweden,v. Ca 22, 1115.

Glacken, I.M., Snowden, D.V., 2001. Mineral resource estimation. In: MineralResource and Ore Reserve Estimation The AusIMM Guide to Good Practice.The Australasian Institute of Mining and Metallurgy, Melbourne, Australia, pp.189198.

Gottlieb, P., Wilkie, G., Sutherland, D., Ho-Tun, E., Suthers, S., Perera, K., Jenkins, B.,Spencer, S., Butcher, A., Rayner, J., 2000. Using quantitative electron microscopyfor process mineral applications. JOM J. Metals, 2425.

Gy, P., 1982. Sampling of Particulate Materials - Theory and Practise. Elsevier, NewYork, US, p. 431.

Hunt, J., Berry, R., Bonnici, N., Walters, S., Kamenetsky, M., McMahon, C., 2009. FromDrill Core to Processing A Geometallurgical Approach to Mineralogy andTexture From Meso- to Micro-Scale: The Tenth Biennial SGA Meeting,Townsville, Australia, 17th20th August, Proceedings, pp. 685687.

Hunt, J., Berry, R., Bradshaw, D., 2011a. Characterising Liberation and FlotationPotential Using Image Analysis, Simulated Fragmentation and Small-ScaleFlotation: The first AUSIMM international geometallurgy conference, Brisbane,Australia, 57 September, Proceedings, pp. 331333.

Hunt, J., Berry, R., Bradshaw, D., 2011b. Characterising chalcopyrite liberationand flotation potential: examples from an IOCG deposit. Miner. Eng. 24, 12711276.

Hunt, J., Berry, R., Bradshaw, D., Triffett, B., Walters, S., 2012. Development ofliberation/recovery domains: examples from the Prominent Hill IOCG deposit,Australia: Process Mineralogy12, Cape Town, South Africa, 79 November,Proceedings, pp. 116.

King, R.P., 1979. A model for the quantitative estimation of mineral liberation bygrinding. Int. J. Miner. Process. 6, 207220.

King, R.P., 2001. Modelling and Simulation of Mineral Processing System.Butterworth-Heinemann, Oxford, Great Britain, p. 399.

King, R.P., Schneider, C.L., 1998. Mineral liberation and the batch communitionequation. Miner. Eng. 11, 11431160.

Koch, P., 2013. Textural variants of iron ore from Malmberget: characterization,comminution and mineral liberation: Master thesis, Lule, Sweden, SustainableProcess Engineering, Lule University of Technology, 101p.

Lamberg, P., 2011. Particles the bridge between geology and metallurgy:conference in mineral engineering, Lule, Sweden, 89 February, Proceedings,

pp. 116.Lamberg, P., Lund, C., 2012. Taking liberation information into a geometallurgical

model-case study, Malmberget, Northern Sweden: Process Mineralogy12, CapeTown, South Africa, 79 November, Proceedings, pp. 113.

Lamberg, P., Vianna, S., 2007. A Technique for Tracking Multiphase Mineral Particlesin Flotation Circuits: 7th Meeting of the Southern Hemisphere on the MineralTechnology, Ouro Preto, Brazil, Proceedings, pp. 195241.

Lamberg, P., Hautala, P., Sotka, S. and Saavalainen, S., 1997. Mineralogical balancesby dissolution methodology:Short Course on Crystal Growth in EarthSciences,S. Mamede de Infesta, Portugal, September 810, Proceedings, pp. 129.

LKAB, 2011. Annual Report: Lule, Sweden, Lule Grafiska, 122p.Lund, C., 2013. Mineralogical, chemical and textural characterisation of the

Malmberget iron ore deposit for a geometallurgical model, Department ofCivil, Environmental and Natural Resources Engineering, Lule, SwedenDoctoral thesis, p. 190.

Lund, C., Lindberg, T. and Martinsson, O., 2010. Mineralogical-texturalcharacterisation of different apatite-iron ore bodies, Malmberget deposit,Sweden, treated in a sorting process in laboratory scale: ProcessMineralogy10, Cape Town, South Africa, Proceedings, pp. 19.

Lund, C., Lamberg, P., Lindberg, T., 2013. Practical way to quantify minerals fromchemical assays at Malmberget iron ore operations an important tool for thegeometallurgical program. Miner. Eng. 49, 716.

Martinsson, O., Virkkunen, R., 2004. Apatite Iron Ore in the Gllivare, Svappavaara,and Jukkasjrvi Areas: Society of Economics Geologists. Guidebooks Series 33,167172.

Mwanga, A., 2014. Test Methods for Characterising Ore Comminution Behaviour inGeometallurgy: Licentiate thesis, Lule, Sweden, Division of Minerals andMetallurgical Engineering, Lule University of Technology, 112p.

Neumann, R., Avelar, A.N., 2012. Refinement of the isomorphic substitutions ingoethite and hematite by the Rietveld method, and relevance to bauxitecharacterisation and processing: Process Mineralogy 12, Vineyard Hotel, CapeTown, South Africa, 79 November, Proceedings, pp. 110.

Niiranen, K., Bhm, A., 2012. A systematic characterization of the ore body formineral processing at Kiirunavaara iron ore mine operated by LKAB, NorthernSweden.: XXVI International Mineral Processing Congress (IMPC), New Delhi,India, Proceedings, p. 1039.

Ozcan, O., Benzer, H., 2013. Comparison of different breakage mechanisms in terms

of product particle size distributionand mineral liberation. Miner. Eng.49, 103108.

76 C. Lund et al. / Minerals Engineering 82 (2015) 6177

7/24/2019 1-s2.0-S0892687515001417-main.pdf

17/17

Paine, M., Knig, U., Staples, E., 2011. Application of rapid X-ray diffraction (XRD)and cluster analysis to grade control of iron ores, In: Proceedings, 10thInternational Congressfor Applied Mineralogy (ICAM), Trondheim, Norway, 15August, Proceedings, pp. 495501.

Prez-Barnuevo, L., Pirard, E., Castroviejo, R., 2012. Automated textural analysis ofthe Kansanshi Copper ore: Process Mineralogy12, Cape Town, South Africa, 79November, Proceedings, pp. 116.

Pirrie, D., Butcher, A.R., Power, M.R., Gottlieb, P., Miller, G.L., 2004. Rapidquantitative mineral and phase analysis using automated scanning electronmicroscopy (QemSCAN); potential applications in forensic geoscience:

Geological Society, London, Special Publications, vol. 232, pp. 123136.Pitard, F.F., 1989. Sampling theory and sampling practice. Pierre Gys sampling

theory and sampling practice, vol. 2. CRC Press, Boca Raton FL, US, p. 247 .Poveromo, J.J., 1999. Iron ores. Chapter 8 in The Making, Shaping and Treating of

Steel. In: Wakelin, D.H. (Ed.), 11th ed. Warrendale, PA: Association for Iron andSteel Technology, pp. 547642.

Spry, A., 1969. Metamorphic Textures. Pergamon Press, p. 350.Suazo, C.J., Kracht, W., Alruiz, O.M., 2010. Geometallurgical modelling of the

Collahuasi flotation circuit. Miner. Eng. 23, 137142.Toraman, O.Y., Kahraman, S., Cayirli, S., 2010. Predicting the crushability of rocks

from the impact strength index. Miner. Eng. 23, 752754.Vizcarra, T.G., Wightman, E.M., Johnson, N.W., Manlapig, E.V., 2010a. The effect of

breakage mechanism on the mineral liberation properties of sulphide ores.Miner. Eng. 23, 374382.

Vizcarra,T., Wightman, E.,Johnson, N., Manlapig,E., 2010. The physical basisof non-random breakage in an iron-oxide hosted copper gold ore. XXV International

Mineral Processing Congress. Brisbane, QLD. pp. 13431350.Vogel, L., Peukert, W., 2003. Breakage behaviour of differentmaterialsconstruction

of a mastercurve for the breakage probability. Powder Technol. 129, 101110.Weller, K.R., Morrell, S., Gottlieb, P., 1996. Use of grinding and liberation models to

simulate tower mill circuit performance in a lead/zinc concentrator to increaseflotation recovery. Int. J. Miner. Process. 4445, 683702.

C. Lund et al. / Minerals Engineering 82 (2015) 6177 77