The Use Of Duplicates.pdf

The use of duplicates, Thompson and Howarth method, on evaluation of sampling protocol, Cuiabá mine, Brazil Andrade, V. R.1; Villanova, F. L. S. P.2; Canela, J. H. C.2 1 2

Federal University of Minas Gerais, Brazil AngloGold Ashanti, Sabará MG, Brazil

Duplicate samples are a requirement from resource definition to mining operation, for any code reporting investment. The duplicates usually consist of assayed pulp which is re-labelled and resubmitted to the laboratory. The use of duplicates on quality assurance and quality control is well known mostly to test the assay repeatability of each sample type, i.e. rig, crushed and pulp. Consequently, duplicates can monitor the variance in the sub-sampling processes and identify sample swops and carry-over contamination. However they can also be used to qualitatively evaluate sampling protocols. Thompson and Howarth (1976) presented a straightforward alternative to evaluate imprecision of sampling protocol using duplicates. It is a simple method that estimates graphically and quantitatively, the coefficient of variation or the imprecision of paired data. In January of 2018, a change on the protocol occurred in Cuiaba gold mine, Brazil, due to a crusher upgrade, which reduced the crushed material to be pulverized. To evaluate this change, the duplicate analysis suggested by Thompson and Howarth (1976) is used in this study, to both crushed and pulp duplicates. From this, based on the last two years historical data, it is possible to determine whether or not the change of protocol is suitable, according to the lithology, mineralization style (gold associated with sulfides or gold free in quartz) and gold grain size. When the imprecisions of crushed e pulp duplicates are compared, is expected the later to be superior compared to crushed imprecision. The opposite suggested the presence of coarse gold grains, which increases the variability. The mass of the actual protocol is adequate for lithologies with gold associated with sulfides, but as expected, the mass proved to be insufficient when gold was free in quartz.

1. Introduction Random sampling errors result on erroneous decisions regarding to the selection of ore and waste. This, in its turn, can take to reconciliation errors and error on production calculation. These errors must be identified and reduced to an acceptable and controlled level. One of the ways to do this is by regular duplicate sampling on each sampling stage. Duplicate samples are a requirement from resource definition to mining operation, for any code reporting investment. The duplicates usually consist of assayed pulp which is relabelled and re-submitted to the laboratory. The use of duplicates on quality assurance and quality control is well known mostly to test the assay repeatability of each sample type, i.e. rig, crushed and pulp. Consequently, duplicates can monitor the variance in the sub-sampling processes and identify sample swops and carry-over contamination. However they can also be used to qualitatively evaluate sampling protocols. The Australasian Joint Ore Reserves Committee (JORC), the National Instrument (NI43/101) from Canada and the South African Code (SAMREC ) are the most known codes that regulate the public reports of mining companies, Mineral Exploration Results, Mineral Resources and Ore Reserves; assuring a level of confidence. JORC in terms of “quality of assay data and laboratory tests” criteria determines that “the QA/QC program includes CRMs, blanks, preparation duplicates and field duplicates and is acceptable according to industry standards.”. SAMREC code states that for QA/QC it is necessary to “Demonstrate that

adequate field sampling process verification techniques (QA/QC) have been applied, e.g. the level of duplicates, blanks, reference material standards, process audits, analysis, etc.” AngloGold Ashanti has an internal document providing QAQC guidelines that declares the general Quality Assurance (QA) and Quality Control (QC) requirements for all assay practices and subsequent processes used for Metal Accounting purposes. This document has the SAMREC CODE as one of its references. The following table informs the percentage of samples that must be analyzed according to the guideline for mine geology samples (grade control and Measured resources) (Table 1). Table 1 - Minimum of QAQC samples that must be put in sample batchs. The red rectangle highlights phases of Mine Geology.

The sampling protocol applied to core drill samples and channel samples on the AGA’s laboratory, until the moment of acquisition of Boyd crusher machine (December 2017), can be summarized according to the following stages: 1) drying the sample at 110 ° C; 2) material crushing - 90% less than 2.0 mm; 3) reduction of the mass of the primary sample to 500 g with the use of a Jones splitter; 4) pulverization in a pulverizing mill - 80% less than 75 μm; 5) selection of a 30 g aliquot (using a spatula) for fire assay analysis, with gold determination by atomic absorption spectroscopy (AAS) or gravimetric finish. The actual protocol is: 1) drying the sample at 110 ° C; 2) material crushing - 90% less than 3.175 mm; 3) sample mass split to 300 g, by a rotatory splitter coupled on the crusher; 4) pulverization in a pulverizing mill - 90% less than 75 μm;

5) selection of a 30 g aliquot (using a spatula) for fire assay analysis, with gold determination by atomic absorption spectroscopy (AAS) or gravimetric finish. In both protocols, the materials that are not selected on stages 3 and 5 of mass reduction are named crusher residue and pulp residue, respectively. These samples can be archived for further analysis. This paper aims to understand the variability of each lithotype in sampling stages, as well as to verify the possible impact of sampling protocol change, mainly of the most variable litologies. Geological setting: Iron Quadrangle (IQ) is a worldwide known metalogenetic province for its gold and iron deposits, located in Minas Gerais, Brazil. Cuiabá mine is a suphide-hosted gold deposit very known for its grade and ton, located on the north border of IQ. Cuiaba mine rocks comprises rocks of Nova Lima Group, Rio das Velhas Supergroup, an Archean greenstone belt sequence (Dorr, 1969). It is within the volcanoclastic association of Baltazar and Zucchetti (2007), part of a matavolcano-sedimentary sequence. The deposit is a lithological succession of metavolcanic, metavolcanoclastic and metasedimentary rocks. The basal unit is metandesite with intercalated phyllites, overlaid by chemical-sedimentary rocks (banded iron formation and ferruginous chert). Above this unit there are carbonaceous phyllite and metabasalt with intercalated micaceous phyllite, overlaid by a unit composed by micaceous phyllite and metavolcanoclastics rocks (Vial, 1980). The mineralization is structurally controlled and the major orebodies are hosted on banded-iron formation (BIF) and ferruginous chert. The mineralization is directly related to hydrothermal alteration and gold occurs associated and included in sulphides (pyrite, pyrrothite and arsenopyrite) in the BIF and in venulated sulphide schists (XS). Gold also occurs free in quartz veins (QZ) that cross-cut metabasalts and metassediments, and can be fine grained (10 micrometers) to coarse (1 millimeter). Table 2 summarizes the lithologies that host the mineralization and were analyzed in this study. Table 2 – Logged lithotypes in Cuiabá mine.

Lithotype

Acronyms

Banded-iron formation

BIF

Massive sulphide

S

Sulphide schist

XS

Quartz veins

QZ

Description Rock with alternating layers of dark quartzcarbonate-carbonaceous matter and translucent to orange quartz-carbonate-chert bands. Visually estimation of a maximum of 5% of sulphides along the bands. Banded-iron formation with more than 5% of sulphides (pyrite or pyrrhotite). Beige rocks intensely veined with quartz and carbonate, showing a penetrative mylonitic foliation. Resulted from hydrothermal alteration of metavolcanoclastic rocks, mostly on the proximal zone (sericite). Disseminated pyrite and/or pyrrhotite are common. Veins composed by smoky and/or white quartz,

sometimes with associated carbonates and sericite and chlorite-rich films. Sulphides can occur as accessory minerals and free gold can be observed macroscopically if it is coarse-grained.

The main orebodies in production are: Fonte Grande, Serrotinho, Balancão and Galinheiro. Figure 1 is a geological map of level 11 of the Cuiabá mine, showing the main orebodies and mineralized zones.

Figure 1 - Geological map of Level 11 of the Cuiabá mine showing the main orebodies associated with BIF (FGS – Fonte Grande Sul, SER – Serrotinho, BAL – Balancão, GAL – Galinheiro, GAL EXT – Galinheiro extensão, DDO – Dom Domingos, CGA – Canta Galo, SUR – Surucu), and mineralized zones related to quartz-carbonate veins (VQZ, SER FW – Serrotinho Footwall, FGS FW – Fonte Grande Footwall) (Vitorino, 2017).

2. Methodology In this paper, for the application of Thompson and Howarth (1976) method, two years historical data available is grouped with regards to the orebody and the lithology, according to Table 3. First of all, to use this method it is necessary a minimum of 50 pairs of data (original sample and duplicate) and the groups with less than this were ignored. The Thompson & Howarth (1976) method applied in this study can be summarized as follows:   

From de duplicate list, calculate the mean e absolute difference of the pairs (original sample and duplicate); Organize the data in increasing order considering the mean values; From the first 11 results of the list, calculate the concentration mean value and the median of the absolute difference values;

 

Repeat the previous procedure to each successive group of 11 samples, ignoring those that left and are less than 11 in total; Complete the linear regression of median values on mean values.

The variation coefficient of all batch used for the regression is approximately the graph inclination. The graph intercept is the detection limit, once it should be related to the standard deviation (in this case de median) of the system measured at or near zero concentration value (Thompson & Howarth, 1976). According to Carsweel et al (2009), the imprecision is twice the variation coefficient.

3. T-H Plot Thompson and Howarth method is described in Methodology. This is method was chosen for this study because it is a straightforward alternative to evaluate imprecision of sampling protocol using duplicates. It is a simple method that estimates graphically and quantitatively, the coefficient of variation or the imprecision of paired data. Any sampling stage, which means, mass reduction, results on errors (sample collection, in case of channel samples, quartering of crusher or pulp fragments and selection of the reduced sample). A duplicate collected in one of these stages will show the errors of post stages, however it is not possible to determine in which of this phases this error occurred. In this way, the comparison of two stages duplicates, theses stages consecutive, allows possible errors to be identified. For example, the crusher duplicate contemplates an error that could have occurred on crushing phase, associated to ideal top size and/or the mass, or that could have occurred on the following stages of subsampling. Pulp duplicates, on the other hand, indicates errors that occurred during pulverization, also associated to ideal top size or mass and the following stage, and during the selection of the final analytical aliquot. Comparisons between error values associated to crushed and pulp duplicates can bring understanding from protocol adequacy to mineralization types. It is expected an imprecision reduction all throughout sampling stages, from collection to analysis, and any inconsistency on this reduction is intrinsically related to gold particles size and to top size and/or masses considered on the protocol. However, even using these comparisons, it is not viable to be certain if the imprecision is due to the sample mass or top size. Pierre Gy was a pioneer creating the Theory of Sampling (TOS) to address sampling problems related to the size of the samples. His well know formula (Gy’s equation) is applied to predict, estimate or minimize the variance of the “Fundamental Sampling Error” (FSE). This variance, VarGy(FSE), is related to a sampling error resulted from representing a lot by an aliquot (Gy, 1979, 1988), and is defined as: (

)

⁄

( ) is the prediction made by Gy’s TOS for Var(FSE), is particle shape factor, g where is size range factor of the particles in the population, is liberation factor of the particles in the population, is mineralogical composition factor of the particles in the population, is material top-size, and is the mass (or weight) of a sample (Gy, 1979). The understanding of Gy’s formula is of the utmost importance for suggesting changes based on the T-H plot method. As noted on this equation, any change on the top-size produces

significant changes on the variance of the error. However, this is often limited to the equipment available. Therefore, another way of reducing the error is by increasing the mass of the sample, which could also be a bottle neck, but could be more viable.

4. Results Data from Balancão, Fonte Grande, Serrotinho, Galinheiro e VQZ orebodies of Cuiaba mine was analyzed. The information is divided according to the lithotypes: iron banded formation (BIF), massive sulfide (S), sulphide schist (XS) and quartz veins (QZ). For each of these lithotypes, crushed (DCR1) and pulp (DP1) duplicates data are available. Table 3 summarizes available dada. Table 3 – Duplicate data available for this study. The data with 5 or more groups, in which T-H method was applied, is highlighted in green. Orebody

Lithology BIF S

BALANCÃO XS QZ BIF S FONTE GRANDE XS QZ BIF S GALINHEIRO XS QZ BIF S SERROTINHO XS QZ BIF S VQZ XS QZ

Top size

Number of duplicates

Number of groups (11 paired data in each)

DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1 DP1 DCR1

4 2 26 1 13 11 57 14 134 78 56 10 30 11 65 20 116 44 49 12 46 24 124 60 69 38 36 18 43 11 82 30 n.a. n.a. n.a. n.a. 9 6 53 29

0 0 2 0 1 1 5 1 12 7 5 0 2 1 5 1 10 4 4 1 4 2 11 5 6 3 3 1 3 1 7 2 0 0 4 2

According to the method proposed, only the groups that comprise 55 or more pairs of duplicates should be considered. Linear regressions of median values over mean values for data that fits this requirement are presented on Figure 2.

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

T-H Plot Fonte Grande_BIF_DP1 DH Grouped data

N = 66 Imp. = 15%

y = 0.0757x + 0.0209 R² = 0.9904

0

5

10

15

Group Median Absolute Difference (g/t)

Group Median Absolute Difference (g/t)

T-H Plot Fonte Grande_BIF_DCR1 DH Grouped data

1.2

N = 132 Imp. = 24,5%

1 0.8 0.6

y = 0.123x + 0.0399 R² = 0.923

0.4 0.2 0 0

Group Mean Gold Grade (g/t)

2 1.5

y = 0.0538x + 0.7293 R² = 0.6886

0.5 0 0

10

20

30

40

50

Group Mean Gold Grade (g/t)

Group Median Absolute Difference (g/t)

Group Median Absolute Difference (g/t)

2.5

1

16

y = 0.0425x + 0.0465 R² = 0.3916

0.05 0 0

1

2

3

4

Group Mean Gold Grade (g/t)

5

Group Median Absolute Difference (g/t)

Group Median Absolute Difference (g/t)

0.15 0.1

10

12 10 8 6

y = 0.5384x - 0.1252 R² = 0.9993

4 2 0 -2 0

10

20

30

Group Mean Gold Grade (g/t)

T-H Plot Galinheiro_QZ_DCR1 DH Grouped data

N = 110 Imp. = 8.5%

0.2

8

N = 55 Imp. = 107.7%

14

T-H Plot Galinheiro_BIF_DP1 DH Grouped data

0.25

6

T-H Plot Fonte Grande_QZ_DP1 DH Grouped data

N = 55 Imp. = 10.8%

3

4

Group Mean Gold Grade (g/t)

T-H Plot Fonte Grande_S_DP1 DH Grouped data

3.5

2

0.7

N = 55 Imp. = 38%

0.6 0.5 0.4

0.3

y = 0.1886x + 0.0422 R² = 0.981

0.2 0.1 0 0

1

2

3

Group Mean Gold Grade (g/t)

4

1.4

T-H Plot Serrotinho_BIF_DP1 DH Grouped data

N = 176 Imp. = 42%

1.2 1 0.8 0.6

y = 0.2106x + 0.0346 R² = 0.976

0.4 0.2 0 0

2

4

6

8

Group Mean Gold Grade (g/t)

Group Median Absolute Difference (g/t)

Group Median Absolute Difference (g/t)

T-H Plot Galinheiro_QZ_DP1 DH Grouped data

0.2

N = 66 Imp. = 2.4%

0.15 0.1 y = 0.012x + 0.0643 R² = 0.2429

0.05 0 0

2

4

6

8

10

Group Mean Gold Grade (g/t)

Group Median Absolute Difference (g/t)

T-H Plot Serrotinho_QZ_DP1 DH Grouped data

1.4

N = 77 Imp. = 32%

1.2 1 0.8 0.6 0.4

y = 0.1602x - 0.0346 R² = 0.9465

0.2 0 -0.2

0

2

4

6

8

Group Mean Gold Grade (g/t)

Figure 2 – T-H plots of lithotypes of Fonte Grande, Galinheiro and Serrotinho orebodies.

5. Discussions The TH plots can be discussed in terms of imprecision: for the same lithotype and orebody, observe the difference in imprecision for pulp and crushed duplicates; difference in imprecision of different lithotypes; comparision of the imprecision of a lithotype in different orebodies. The limitation of available data for this study causes some of these correlations not possible to be addressed. In terms of comparison between imprecision of crushed and pulp duplicates, it is expected an inferior value of the latter, compared to the imprecision of crushed duplicates of the same lithology, and these values can bring discussions in terms of gold grain size. Another discussion that can be brought observing T-H plots is with respect to the detection limit, as the graph intersection is the detection limit. Moreover, throughout the distribution profile of the samples, is possible to argue with relation to mineralization continuity. Balancão and VQZ orebodies do not have sufficient duplicate data to be analyzed according the T-H plot method, used in this paper. Therefore, only plots of some lithotypes of Fonte Grande, Galinheiro and VQZ were further discussed. Comparing the imprecision on BIF pulp duplicates, samples from Fonte Grande have the higher imprecision of 24.5%, whereas Galinheiro and Serrotinho have much lower values

of 8.5% and 2.4% respectively. A noteworthy point for Fonte Grande is that the orebody is cut by a fault with divides it in half. The South side, being associated with sulphides and the North side associated with a high degree of BIF silicification. The gold bearing quartz alteration on BIF is related to an increase of gravimetric recovery seen at the plant, when this ore body is mined. The presence of free gold is marked by an increase on variability of the duplicates. For QZ lithotype, pulp duplicates of Fonte Grande oreboby pointed an imprecision of 107.7% and 42% for Galinheiro. The high value of Fonte Grande QZ, can also be explained by the peculiarity of the mineralization on each side of the fault. The only sufficient duplicate data available for S is of Fonte Grande oreboby, and a value of 10.8% of imprecision was found. For this orebody, because of structural aspects pointed above, the meaning of this value is not explicit. Even though the imprecision value can be considered acceptable, the value of 0.7 of intercept, is noticeable greater than the detection limit (0.05). However, it is important to observe that there is a dispersion of the data along the regression line (R2=0.68), and this can result on a bad inference of the detection limit, which means that certainly it is not near 0.05, but also not 0.7. Since the lithotype is massive sulfide, that is generally easy to be sampled, a high value of detection limit can indicate coarse gold grains (associated to sulphides), which are easily released during the milling process. For BIF lithotype of Fonte Grande orebody, it is possible to compare the imprecision of crushed and pulp duplicates. The former is 15% and the latter is 24.4%. As it is expected an inferior value of imprecision on pulp duplicates, the opposite suggests the presence of coarse gold grains, which increases variability when gold is released. Less expressively, this fact was also observed in Galinheiro quartz vein (QZ) lithotype, for which T-H plot showed an imprecision of 38% for crushed duplicates and of 42% for pulp duplicates. T-H plots can be discussed in terms of the distribution profile of the samples. The discontinuous nature of the plotted points indicates an erratic mineralization, most commonly associated with free gold. This was strongly observed for QZ duplicates of Fonte Grande and Serrotinho. T-H plots with a more continuous distribution along the line (pulp duplicates of Galinheiro QZ and Fonte Grande BIF) indicate a more homogeneous mineralization, in which a higher grade is attributed to a higher concentration of sulfides. According to the plant requirements, fit-for-purpose imprecision values suggested are of a maximum of 15% for pulp duplicates and 30% for crushed. For all the lithotypes in which a high value of imprecision was observed the easier manner of suiting the protocol is by increasing the mass to be pulverized. If, from this, the value of imprecision is still high, the analytical mass could be increased, and two or three analysis could be done. The value considered would be the mean value.

6. Conclusions Mass reduction for lithotypes that have shown high values of imprecision leads to an increase on variance, resulting on an even greater imprecision. Mainly for lithotypes with coarse gold grains, prone to generate a nugget effect, a mass reduction on the protocol is not suitable. Instead, an increase on the mass would be suggested.

For lithotypes with fine gold grains or associated with sulphides, the mass reduction would not impact that much on the value of imprecision. However, this might be experimentally checked. It is suggested the sampling protocol to be changed, focusing on samples with grade at least 10 times the detection limit, because high grades defines the ore and consequently, should define the sampling protocol. Low grades tend to be less erratic, possibly due to fine gold (free gold sulphide associated or microscopic) that is easily sampled. For the results to be analyzed again, a minimum of 55 duplicates of each lithotype and orebody are requested, and this will take a while in order to use data exclusively of the actual protocol. When a sufficient number of duplicates are available, will be possible to compare the results of both protocols and come up with proposals of suiting it.

7. References Baltazar, O.F., Zucchetti, M., 2007. Lithofacies associations and structural evolution of the Archean Rio das Velhas greenstone belt, Quadrilátero Ferrífero, Brazil: a review of the setting of gold deposits. Ore Geol. Rev. 32, 471–499. Carswell, J. T., Yulia, K., Lesmana, D., & Steamy, K., 2009. Grade control sampling quality assurance/quality control in a high-grade gold mine—Gosowong, Indonesia. In Proceedings of the 7th International Mining Geology Conference (pp. 283-290). Melbourne. The Australasian Institute of Mining and Metallurgy. Dorr, J.V.N., 1969. Physiographic, Stratigraphic and Structural Development of the Quadrilátero Ferrífero, Minas Gerais, Brazil. Regional Geology of the Quadrilátero Ferrífero, Minas Gerais, Brazil. Gy, P., 1979. Sampling of Particulate Materials. Theory and Practice. Elsevier, Amsterdam. 431 pp. Thompson, M and Howarth, R J, 1976. Duplicate analysis in geochemical practice part 1: Theoretical approach and estimation of analytical reproducibility, Analyst, 101:690698. Vial, D.S., 1980. Mapeamento Geologico do Nivel 3 da mina de Cuiabá, Mineração Morro Velho as: Internal Report. Vitorino, A.L.A., 2017. Mineralização aurífera associada aos veios quartzo-carbonáticos hospedados na unidade máfica basal da jazida Cuiabá, greenstone belt Rio das Velhas, Quadrilátero Ferrífero, Minas Gerais, Brasil. Vanessa Resende de Andrade Geologist Vanessa is a Geologist graduated in Federal University of Minas Gerais, Brazil, in July 2018, when she also finish, after 18 months, an internship in AngloGold Ashanti company, at Lamego and Cuiaba mines. With the experience at AngloGold Ashanti and an international experience of 15 months of exchange in Australia, she is currently looking for a position as junior geologist or trainee.