Characterization of an Iron Ore Tailing Sample and the Evaluation of Its Representativeness

The massive annual tonnage of iron ore tailings, plus the more demanding environmental policies after the catastrophic collapses of Fundão Dam in 2015 and Feijão Dam I in 2019, have forced Brazilian mining companies to search for more sustainable and safer alternatives for tailings disposal. The Brazilian mining industry has been studying the dry stacking of filtered tailings. Most companies seek the development of great dry stacks, over 200 m high, to absorb the whole tailings generation. As tailings assume a structural role, it is vital to characterize the dry stacks and understand their behaviour. This study aimed to characterize index properties, evaluate the effect of compaction on the hydraulic behaviour, and evaluate field variability of iron ore tailings from a mine in Quadrilátero Ferrífero, Brazil. Also, a database of iron ore tailings properties from Quadrilátero Ferrífero was collected and used to evaluate the results. Specimens were compacted to 93%, 95%, 97%, and 100% of Proctor’s maximum dry unit weight, and the hydraulic conductivity was defined for several confining pressures (100, 200, 400, 800, 1600 and 1900 kPa) for each degree of compaction. Index properties of the studied tailing were similar to other iron ore tailings from the Quadrilátero Ferrífero, hydraulic conductivity was proportional to the void ratio, and the field tests indicated some variability of both material properties and degree of compaction of the stacked material. The database presented in this study supplies the mining industry with a reference point for future projects.


Introduction
Tailings represent an environmental liability to the mining industry and must be appropriately deposited in a tailings storage facility (TSF). The massive tonnage of iron ore production in Brazil requires large TSFs and, after the catastrophic collapses of Fundão Dam in 2015 (the biggest environmental disaster in the country) and Feijão Dam I in 2019 (which led to hundreds of deaths), both by flow liquefaction (Morgenstern et al. 2016;Robertson et al. 2019;CIMNE 2021), mining companies in Brazil have been searching for more sustainable and safer alternatives for tailings storage based on a dry iron ore processing production. Vale S.A., responsible for 71,69% of all iron ore commercial production in Brazil in 2021 (ANM 2023), has invested about R$ 66 billion in the development of technologies and facilities for dry iron ore processing production in the last decade (Vale 2020). Dry stacking of dewatered tailings is one of the technics that have been studied and gradually implemented by major mining companies in Brazil.
Dry stacking is a stable way of storing tailings in an unsaturated and dense solid state. Even though other dewatering technics have been studied (Fourie et al. 2007;Fourie and Jones 2010;Lees 2016;Sahi et al. 2019;Zhang et al. 2022;Shafaei et al. 2022), tailings dewatering is usually done in a large capacity vacuum or pressure belt filters, so they are also called filtered tailings (Davies & Rice 2001). When tailings are dewatered, there is a decrease in the water content and an increase in the effective solids content that implies higher effective stresses, consequently, higher internal resistance in the material (Davies et al. 2010). However, the Brazilian mining companies intend to develop great dry stacks (over 200 m high) to absorb tailings generation, and before building enormous dry stacks, it is vital to understand the intrinsic characteristics and the tailings behaviour under high confining stresses (Lupo and Hall 2011;Davies 2011;Hore and Luppnow 2014;Crystal et al. 2018;Furnell et al. 2022).
This study aimed to 1) understand the intrinsic properties of tailings that will be stored in a dry stack in the Quadrilátero Ferrífero (Minas Gerais, Brazil), 2) understand the material's hydraulic behaviour in different states when changing the degree of compaction and confining pressures, and 3) verify the material's variability in the field after quality-control tests.
Despite the singularities of each company's mineral beneficiation processes, similarities among tailings from different facilities in the Quadrilátero Ferrífero (QF) are expected because the mines are in the same geological formation. Therefore, a database of the tailings characteristics of iron ore mines in the QF has been collected and used to analyse the tailings studied in this research. Comparing the properties of different tailings generated in the QF helps identify patterns and ranges of variation, which can be used to analyse iron ore tailings from the same formation and with similar physical characteristics.

Materials and Methods
Disturbed samples of iron ore tailings were collected in airtight plastic bags directly from the pressure belt filter before being deposited and compacted in the TSF in a mining complex located in the QF, where the total tailings are currently being press filtered and deposited in the TSF. The experimental program described in this chapter was developed in the Laboratory of Soil Mechanics of the Federal University of Viçosa (LMS-UFV) (Viçosa, Brazil) and the Geotechnical Laboratory of the Faculty of Engineering of the University of Porto (LabGeo-FEUP) (Porto, Portugal). Additionally, a trial embankment was built by the mining company, in which field compaction quality-control tests were performed and made available for the research.

Sample Preparation
Tailing samples were dried at room temperature (~ 22 ℃), loosened, sieved, homogenized, and stored in airtight plastic bags. Before each test, samples were loosened, sieved, and homogenized again to avoid lumps.

Index Tests
Index tests were performed to determine the Atterberg limits, maximum and minimum void ratios, grain size distribution, and specific gravity of the iron ore tailings. Samples were prepared according to the Brazilian Association of Technical Standards (ABNT) Brazilian Regulatory Standard (NBR) 6457:2016.

Grain Size Distribution
Grain size analyses were conducted according to the International Standards Organization (ISO) 17,892-4:2016 and ABNT NBR 7181:2016. Twentyfour analyses were performed at LMS-UFV: one with the homogenized sample, and twenty-three with the material resulting from the hydraulic conductivity tests. Three analyses were performed at LabGeo-FEUP: two with the homogenized samples and one with the material resulting from the standard Proctor test.
The grain size analyses were performed to evaluate the uniformity of the samples sent to both laboratories. Additionally, it was evaluated the evolution or not of the grain size distribution due to the static (material resulting from the hydraulic conduction tests) and dynamic compaction (material resulting from the standard Proctor test).

Specific Gravity
Three tests were performed to define the specific gravity of the iron ore tailing samples: one test at LMS-UFV and two at LabGeo-FEUP to verify the similarity of the samples sent to both laboratories. The tests followed the ABNT NBR 6458:2016 and ISO 17892-3:2015 methods, respectively.

Atterberg Limits
Tests to determine plasticity and liquid limits were performed at LMS-UFV in conformity with the ABNT NBR 7180:2016 and NBR 6459:2016, respectively.

Maximum and Minimum Void Ratios
At LMS-UFV, the minimum and maximum void ratios were determined following test method A.1 of ABNT NBR 16,843:2020 and test method A of ABNT NBR 16,840:2020, respectively.

Proctor Compaction Test
Samples tested in the compaction tests were prepared according to the ABNT NBR 6457:2016 method. Standard Proctor tests were performed in both laboratories. At the LMS-UFV it was adopted the equipment and procedures according to the ABNT NBR 7182:2016: small rammer with 2,5 ± 0,01 kg and 50,0 ± 0,5 mm of diameter; small mould with an inside diameter of 100,0 ± 0,1 mm and height of 127,3 ± 0,3 mm; drop height of 305 ± 2 mm; 26 blows per layer and three layers. At LabGeo-FEUP it was adopted the equipment and procedures according to the National Laboratory for Civil Engineering (LNEC) E 197:1966: small rammer with 2,49 kg and 50 mm of diameter; small mould with an inside diameter of 102 mm and height of 117 mm; drop height of 305 mm; 25 blows per layer and three layers.

Field Test
To evaluate the quality of field compaction, the mining company determined the dry unit weight (γ d ) and water content (w) of 218 samples collected in different areas of a trial embankment. Each lift of the trial embankment was 0.5 m tick. Two samples were collected in some of the checked points to evaluate the compaction homogeneity of the layer: one on the top and another at the base of the compacted layer. Bulk unit weight (γ) was determined by the Drive-Cylinder Method, following the ABNT NBR 9813:2016; water content was determined by oven drying the sample at 110 °C ± 5 °C following the Brazilian National Highway Department (DNER) test method (ME) 213:1994.
The γ d and w values of each sample were compared to the maximum dry unit weight (γ d,max ) and optimum water content (w opt ) defined after standard Proctor tests performed every 1000 m 3 of compacted tailings. A total of thirty-three standard Proctor tests were performed by the mining company technical staff, and the results were made available for interpretation. The standard Proctor tests were performed according to the ABNT NBR 7182:201 (with similar equipment and procedures as presented in subsection 2.3).

Hydraulic Conductivity Test
The hydraulic conductivity tests were performed in a flexible wall permeameter adapted from the triaxial system following the ASM D5084:2016, Method A (constant head). It used triaxial-type cells with all the components. Sixteen hydraulic conductivity tests were performed in the different confining stresses with gradients equal to 4, 8 and 12, and no significant difference was observed in the results. Therefore, the other eighteen tests were realized after saturation and consolidation with a constant hydraulic gradient equal to 8 until at least four values of constant hydraulic conductivity were obtained. The data acquisition was done after every 20 s with electronic pressure transducers.
Four tests (one for each degree of compaction) were conducted after the saturation stage, at 10 kPa of effective confining pressure, and thirty tests were performed after the isotropic consolidation at different effective confining stresses (100 kPa, 200 kPa, 400 kPa, 800 kPa, 1600 kPa, and 1900 kPa): eight specimens with the degree of compaction, DC = 93%; eight with DC = 95%; eight with DC = 97% and six with DC = 100%.
At LMS-UFV it was adopted specimens circa (ca.) 50 mm in diameter × 100 mm long. At LabGeo-FEUP it was used sixteen specimens ca. 72 mm in diameter × 144 mm long, and five specimens ca. 60 cm in diameter × 120 mm long. The specimens were prepared through static compaction using a split mould and a press. At LMS-UFV, samples were compacted in four layers. For each layer, it was determined the exact mass of tailings to compensate for the effect of over-compaction. After the compaction of each layer, the material was scarified to avoid smooth planar surfaces between the compacted layers and improve the adherence between them. Specimens tested at LabGeo-FEUP followed the same preparation procedures, but they were compacted into six layers.

Iron Ore Tailings from the Quadrilátero Ferrífero, Brazil
Based on a comprehensive literature collection of data, it was possible to provide an overview of the index (physical), compaction and permeability characteristics of the QF iron ore tailings.

Grain Size Distribution
The grain size distribution of the tailings is affected by different factors, such as the mineral composition, the efficiency of the processing, and the disposal methods (Vick 1990). Robertson et al. (2019) subdivided the grading curves of Feijão Dam I into coarse (maximum fines content = 50%), fine (50% to 96% fines), and slime (~ 100% fines) tailings. Besides grain size distribution, the main difference among materials was the material's response to CPTu loading. According to the authors, coarse tailings presented a drained behaviour during CPTu; fine tailings presented an undrained behaviour with rapid pore pressure dissipation (measured time for 50% dissipation, t 50 < 400 s), and slimes presented an undrained behaviour with longer slow pore pressure dissipation (t 50 > 1000 s). Oathes et al. (2022) used the same delineation of materials to model the Feijão Dam I failure.
A gradual change in the mechanical and hydraulic behaviour of the tailings with the fines content increase is expected. However, the approach presented by Robertson et al. (2019) allows us to search for tendencies in the tailings' behaviour based on their index properties. Thus, in Table 1, the limits determined by Robertson et al. (2019) were adopted to classify the grain size distributions collected in the literature (Fig. 1), and the subsequent index properties are given in Sects. 3.2 to 3.6 From 72 grading curves, 25 were coarse tailings, and 40 were fine tailings. Based on the high values of the coefficient of uniformity (C u ) and the coefficient of curvature (C c ), coarse and fine tailings would be classified as well-graded or gap-graded soils, respectively. However, most grade curves presented in Fig. 1 are reasonably vertical, followed by a tail of finer particles that confers the high values of C u and C c . The same characteristic was identified by Li et al. (2018) for gold tailings. According to the authors, these materials are relatively poorly graded instead of well graded.
Therefore, nearly all coarse and fine tailings presented poorly graded curves composed mainly of silt and fine sands. The average grain size distribution curve of the coarse tailings indicates a silty sand material with 49.9% of fine sand and 23.3% of silt, while the average grain size distribution curve of the fine tailings indicates a sandy silt material with 53.9% of silt and 33.0% of fine sand. The slimes showed more well-graded curves composed mainly of silt size material (average = 72.2%) and clay size material (average = 25.9%). Nonetheless, the standard deviation values of all categories indicate considerable variability of the tailings' grading curves.
Most tailings grade curves are between the lower boundary of the fine tailings samples presented by Robertson et al. (2019) and the upper boundary of the samples tested by Silva (2010). Figure 2 presents the contours in which the grain size distribution curves of slimes (in light pink), and fine and coarse tailings (in navy blue) from the Quadrilátero Ferrífero are most likely to be. In white, it is also presented the average curve that separates the fine tailings from the coarse tailings.
It is worth mentioning that Fundão Dam and Feijão Dam I collapsed through flow liquefaction. As the defined contours are inside the grain size distribution range initially presented by Robertson et al. (2019) for Feijão Dam I, all the tailings of the Quadrilátero Ferrífero located inside these contours may be susceptible to liquefaction. Yet based on Robertson et al. (2019), the tailings below the average curve (with more than 50% of fines) are prone to have undrained behaviour under saturated conditions. The grain size distribution alone is not enough to dictate if the material is susceptible to liquefaction or not, but it may be used as a preliminary reference to identify potentially liquefiable tailings, as presented by Ishihara et al. (1980) for cyclic liquefaction susceptibility.

Specific Gravity
The tailings' specific gravity (G s ) depends on the specific gravity and concentration of each constituent mineral. As iron minerals have high specific gravity, the higher their contents, the higher the specific gravity of the tailings. Table 2 shows the high variability of iron mineral contents among samples. Some tailings are composed essentially of quartz, such as QF-AM01 (Pires et al. 2019), and others are composed mainly of iron minerals, such as Feijão D. I-Coarse Tailings (Robertson et al. 2019). The hematite, magnetite, and goethite grades also vary from each sample, and the hematite concentration is usually the highest. Iron content is inversely proportional to the efficiency of iron ore processing. Old tailings deposits (e.g., Feijão Dam I) tend to have higher iron content than tailings generated nowadays (e.g., Fernandinho mine future storage). Moreover, slime samples tend to present higher grades of iron minerals than other tailings. Table 3 shows that the average specific gravity of all samples, coarse, fine tailings, and slimes do not diverge much: 3.66 g/cm 3 considering all samples, 3.63 g/cm 3 for coarse tailings, 3.59 g/cm 3 for fine Table 1 Statistic of the grain size distribution curves and percentage of grain sizes of the QF iron ore tailings tailings, and 3.97 g/cm 3 for slimes. However, they present significant standard deviation, and comparing Table 2 with Table 4, it is identified that tailings with higher iron content have higher specific gravity and tailings with lower iron content have lower specific gravity, as expected.

Atterberg Limits
Most of the analysed samples of the QF iron ore tailings, presented in Table 5, were non-plastic (69%). The number of plastic samples tends to increase in the finer tailings. Evaluating the type of tailings, 94% of the coarse tailings were non-plastic, 77% of the fine tailings were non-plastic, and 100% of the slimes were plastic. However, it is noteworthy that the average values of the plastic samples (Table 3) indicated low plasticity, even for slimes. Based on the Casagrande plasticity chart, high-plastic materials (LL > 50%) were identified only by Ferreira (2018), Miranda (2018), and Robertson et al. (2019).

Maximum and Minimum Void Ratios
The maximum and minimum void ratios (limiting void ratios) are affected by several factors, such as the grain size distribution, fines content, mean grain size, particles shape, and intrinsic characteristics of the fine and coarse fractions Ishihara 1999, 2002;Carraro and Prezzi 2008). The liming void ratios reflect not only the physical properties, but also the characteristics of soil behaviour, such as compressibility, contractiveness, and flow potential Ishihara 2000, 2002). Non-plastic soils tend to pass from a coarse-dominated structure to a fines-dominated structure when increasing the fines content. Until the transition zone, around 30% to 40% of fines content, e max and e min tend to vary at the same rate. However, for soils with fines content above 40%, e max tends to increase at a faster rate than e min , resulting in more compressible soils (Lade et al. 1998;Cubrinovski and Ishihara 2002;Cubrinovski et al. 2010;Mijic et al. 2021).
Different authors in the literature have determined the maximum and the minimum void ratios of tailings with fines content up to 40% to extend the use of relative density to tailings (Vick 1990). Torres-Cruz & Santamarina (2020) expand the determination of e max and e min to characterize the behaviour of non-plastic silts. According to the authors, determining the limiting void ratios may assist the elaboration of sampling programs. However, Carraro & Prezzi (2008) alert that e max and e min , specially e max , are affected by the method adopted to determine them, and the real e max of natural loose deposits and tailings dams' reservoirs may not be assessed by the conventional standard methods. Table 4 presents the maximum and minimum void ratios of the QF iron ore tailings found in the literature. As expected, the coarse tailings show average e min and e max slightly lower than the fine tailings (Table 3). The maximum void ratio is two times more dispersed than the minimum void ratio for both fine and coarse tailings. High values of e max were found for coarse and fine tailings, even though the increase of e max should be proportional to the fines content increase. The variability of e max may result from the physical properties of the tailings and the method adopted by each author to define it.

Maximum Dry Unit Weight and Optimum Water Content
Most w opt. values of the tailings generated in the Quadrilátero Ferrífero found in the literature and presented in Table 6 are between 10 and 17%. A small increase in the average w opt. is observed with the fines content increase: 12.6% for coarse tailings, 13.4% for fine tailings, and 16.5% for slimes ( , which is consistent with the tendency of slime tailings to present higher iron ore content (Vick 1990). Notice that the maximum dry unit weight is directly proportional to the specific gravity of the material. Therefore, it is expected to have higher values of γ d,max for the samples with higher G s -as for Forquilha III Dam samples (Dornas 2008)-and lower values of γ d,max for samples with lower G s -as for Germano Dam, Bay 3, coarse tailings samples (Ferreira 2016).

Hydraulic Conductivity
The hydraulic conductivity (k) of tailings is hard to generalize because it is affected by several characteristics, such as fines content, plasticity, confining stresses, void ratio, stratigraphy, material structure, sampling, and testing procedures. According to Vick (1990), hydraulic conductivity can vary from 10 -10 m/s in slimes to 10 -4 m/s in coarse tailings. Moreover, hydraulic conductivity tends to reduce up to five times the initial value by reducing the void ratio, while this reduction can be about ten times in slimes. Chapuis & Aubertin (2003) studied the applicability of the Kozeny-Carman (KC) equation to mine Fig. 2 Contours in which the grain size distribution of the QF iron ore tailings are most likely to be. In light pink, the slimes contours; in navy blue the fine and coarse tailings; and in white the average curve, which separates fine tailings from coarse tailings tailings. According to the authors, the KC equation may be used to predict the k-value of tailings if it accounts for the particle shape effect, as in Eq. 1. However, the KC equation does not estimate k-values of intact tailings samples due to their increased stratigraphy and high anisotropy.
where, k is the hydraulic conductivity, e void ratio, G s specific gravity, S p specific surface.
(1) log k 1m∕s = 1, 46 ⋅ 0, 5 + e 3 G 2 s S 2 p (1 + e) + 1, 99 Table 7 presents average values of the hydraulic conductivity obtained by different authors of tailings generated in facilities located in the Quadrilátero Ferrífero. The range of variation of the hydraulic conductivity is very similar to the one presented by Vick (1990): 1.15 × 10 −10 m/s (Morgenstern et al. 2016) to 1.45 × 10 −4 m/s (Santos 2004). However, the lowest k-value was obtained in an oedometer test in coarse tailings with e = 0.70, which does not match the expected values for such material. This value is even lower than the values obtained for slimes of the same TSF, analysed by the same authors.  The void ratios of most samples shown in Table 7 were between 0.60 and 0.90. For this range, taking the average values of each author and excluding the sample analysed by Morgenstern et al. (2016), considered as an outlier, the average value of the hydraulic conductivity of the tailings is 1.92 × 10 −5 m/s , with a standard deviation of 3.36 × 10 −5 m/s. In the tests where k was calculated to different confining stresses, there was a progressive reduction of its value with the increasing stresses and void ratio reduction. Overall, coarse tailings presented a variation of one-half to five times the initial value, following the usual reduction presented by Vick (1990). One exception was the sample analysed by Morgenstern et al. (2016), which reduced a hundred times the initial value. The fine tailings showed a variation of three times the initial value. The slimes studied by Morgenstern et al. (2016) reduced three times the initial value in the oedometer test; in the large strain consolidation test, the hydraulic conductivity at the final void ratio was sixty times lower than the initial value. The hydraulic conductivity is considerably affected by the test used to measure it. Furthermore, comparing the hydraulic conductivities of fine and coarse tailings measured by the same author and the same type of test, Table 7 shows that, on average, fine tailings tend to have hydraulic conductivity ten to one hundred times lower than coarse tailings.

Grain Size Distribution
As synthesized in Table 8, the grain size distributions presented 83% to 94% of fines passing the nº 200 sieve, from which the majority are silt size particles. The samples UFV and FEUP-Sample A (Fig. 3) showed high similarity. Nonetheless, FEUP-Sample B showed some divergence on the central portion of the curve. The FEUP-Sample B, likewise the FEUP-Sample A-Proctor, exhibited higher content of particles ranging from 0.02 mm to 0.04 mm than the other samples.
When comparing FEUP-Sample A with FEUP-Sample A-Proctor the fines content increases in the Table 3 Statistical parameters of the QF iron ore tailing index properties: mineral and elemental compositions, specific gravity, void ratios, maximum dry unit weight, optimum water content, and Atterberg limits Stat statistics, Hm hematite, Mg magnetite, Gh goethite, Fe iron content, Gs specific gravity, e min and e max minimum and maximum void ratios, w opt optimum water content. γ d,max maximum dry unit weight, LL liquid limit, PL plastic limit, PI plasticity index, Avg average, Stdev Standard deviation, Max maximum value, Min minimum value  central portion of the curve, indicating some particle breakage due to dynamic compaction. On the other hand, UFV-Post-hydraulic conductivity test samples did not show a significant difference from the UFV sample, indicating that samples compacted statically (with a press) did not significantly evolve. Additionally, the tailings studied in this research are within the limits defined in 0 for the QF iron ore tailings (Fig. 4). The grain size curves of the studied material are very similar to the upper limit of the samples of fine tailings of Feijão Dam I tested by Robertson et al. (2019). The values of C u , and C c are in the one standard deviation range of the QF fine tailings (Table 9) indicating that curves have similar shapes. The high values of C u and C c are a consequence of the tail of fines shown by the curves, likewise the other tailings from the Quadrilátero Ferrífero. Therefore, the material could also be classified as relatively poorly graded instead of well-graded.

Specific Gravity
Specific gravity was determined on three samples: UFV (G s = 3.218 g/cm 3 ), FEUP-Sample A (G s = 3.184 g/cm 3 ), and FEUP-Sample B (G s = 3.213 g/cm 3 ). These values were consistent with each other and the literature, with an average of 3.205 g/cm 3 . The specific gravity of the analysed tailings is considerably close to the average value of QF fine tailings (G s = 3.59 g/cm 3 ) presented in subsection 3.2 when compared to the range of variation (2.66-5.14). Moreover, this value is consistent with the samples in Table 4 with low iron content.

Atterberg Limits
As with most iron ore tailings displayed in Table 5, the plasticity and liquid limit tests indicated that the studied tailings are non-plastic. Following the Unified  (

Maximum and Minimum Void Ratios
With an average water content of 0.66%, the sample dried at room temperature presented e min = 0.60 and e max = 1.18, and respective γ d,max = 19.70 kN/m 3 and γ d,min = 14.46 kN/m 3 . The maximum dry unit weight is lower than γ d,max = 20.71 kN/m 3 and γ d,max = 21.05 kN/m 3 obtained after the standard Proctor tests presented in 0, but still compatible with them. The slight variation of γ d,max is partially explained by the fines content increase identified after the standard Proctor test, indicating some particle breakage (see 0), which results in the rearrangement of the material. The broken particles tend to fill the voids resulting in higher dry unit weights. The e min and e max are a little bit lower than the average values found in Table 3 for the fine tailings of the Quadrilátero Ferrífero (e min = 0.67 and e max = 1.31), but they are into the one standard deviation range of the sets.

Maximum Dry Unit Weight and Optimum Water Content
The dry unit weight-water content curves defined after the standard Proctor tests performed at LMS-UFV and LabGeo-FEUP are plotted in Fig. 4. In both curves, the optimum water content was w opt = 11.7%, and the maximum dry unit weight presented a small variation (γ d,max ) FEUP = 20.71 kN/m 3 and (γ d,max ) UFV = 21.05 kN/m 3 . The divergence of the values of dry unit weight found at UFV and FEUP may be associated with the variability of the samples or the intensity of particle breakage in each test.
Compared to the parameters presented in Table 3, the values of w opt. and γ d,max are into the one standard deviation range of the QF fine tailings, which means that the values are considerably close to the average Table 6 Maximum dry unit weight and optimum water content of the QF iron ore tailings * Value not used to calculate the averages on Table 3. w opt optimum water content, γ d,max maximum dry unit weight, D dam, C concentration, M mine, Crs coarse, F fine, Slm slime   Crystal et al. (2018), materials with S higher than 85% are likely to behave as saturated in a dry stack. However, with degrees of compaction higher than 90% of standard Proctor, no samples (FEUP and UFV) presented degrees of saturation (S) above this reference. Even though the material does not reach S = 85%, it is close to reaching saturation, and water infiltration must be prevented at all costs when compacting wet of optimum. Additionally, when stacking the tailings, if the deposited tailings are compressible, as they are gradually loaded, the material will tend to consolidate, extruding the air existing in the internal voids, reducing the void ratio, and increasing saturation (Lupo and Hall 2011).

Field Tests
The standard Proctor tests performed in the field exhibited average w opt and γ d,max similar to those found in this research (Table 9). The one standard deviation range indicated a high concentration of values between γ d,max = 20.33 kN/m 3 and γ d,max = 22.13 kN/m 3 , and w opt between 11% and 12.6%. However, the maximum and minimum values diverge significantly from the average. Two samples have even shown pairs of values (w opt , γ d,max ) above the zero air voids curve (S = 100%), just as the other 42 pairs of values determined in the field tests (Fig. 4). These results suggest

Fig. 3
Grain size distribution curves of the studied sample Fig. 4 Field tests results (γ d and w) plotted with dry unit weight-water content and saturation curves of the studied sample. The saturation curves were defined after the average specific gravity, G s = 3.205 kg/m 3 considerable variability in the characteristics of the tailings in some portions of the trial embankment.
Taking the FEUP and UFV curves as references, the average water content of the field samples in Table 9, w field = 13.3%, is equivalent to DC between 95 and 97%. The average dry unit weight of the field samples, γ d,field = 21.0 kN/m 3 is equivalent to DC = 98.9%, adopting the average γ d,max = 21.2 kN/ m 3 as reference. Moreover, 82% of the field samples indicate compaction wet of optimum. Of the 218 field tests, 78 show a degree of compaction higher than 100% of standard Proctor and, from these 78 tests, 64 show water content above the optimum. There is no significant divergence of the values of γ d and w when comparing the sampling positiontop or bottom of the layer-in Fig. 4. It suggests that the same degree of compaction was achieved at the top and the bottom of the layers indicating that the compaction energy was adequate.
The samples on the top of the chart (Fig. 4) have pairs of dry unit weight and water content higher than the maximum dry unit weight and optimum water content of the standard Proctors of reference. A homogeneous material could never achieve a state of DC > 100% compacted wet of optimum, even if compacted with higher energy. These results firmly indicate variability of the material in the field, and a unique sample is not enough to define the behaviour of the dry stack. The macro-scale behaviour of the tailings is strongly dependent on the microscale interparticle contact, which is affected by the morphology (shape, size and surface roughness and fabric) and elastic characteristics of the tailings' particles (Sandeep and Senetakis 2019; Nardelli and Coop 2019; Ren et al. 2021). Therefore, the variability of the tailings composition directly affects the mechanical behaviour of a dry stack.
The variability of the tailings requires not only the constant characterization of the material but also the adjustment of the specifications for stacking it. This dynamic may not be easily implemented in largescale production, so the impact of the tailings' variability in the dry stack behaviour should be minimized. Stabilization of mining tailings with different binding agents has been studied ( Barati et al. 2020;Bruschi et al. 2021;Pereira dos Santos et al. 2022;Servi et al. 2022) and should become an attractive alternative to overcome the impact of tailings variability in large scale dry stacks.

Hydraulic Conductivity
The void ratios of the specimens varied from 0.64 to 0.47 under different degrees of compaction and confining stresses. Figure 5 shows that permeability decreases with the reduction of the void ratio. The specimens compacted with higher void ratios-lower degrees of compaction-showed higher hydraulic conductivities, except those tested at 10 kPa of confining stress, which presented almost the same permeability. The KC equation curve in Fig. 5 was determined adopting the average specific gravity (G s = 3.205 g/ m 3 ) and the specific surface (S p = 402.44 m 2 /kg) determined from the average grain size distribution curve of the UFV-Post-hydraulic conductivity test samples displayed in 0 following the method of Chapuis & Légaré (1992). The trend defined by the KC equation (Eq. 1) is not adherent to the results of the tests. The measured hydraulic conductivities and reduction rate are higher than those determined by the KC equation. It indicates that the studied tailings, in the studied range of void ratios, are more sensitive to void ratio reduction than the materials used to formulate the KC equation.
Comparing the highest (4.10 × 10 −7 m/s) and the lowest (6.23 × 10 −8 m/s) values, the factor of reduction of the hydraulic conductivity with decreasing void ratio is about 6.6 times. This rate is higher than the maximum expected reduction factor stated by Vick (1990) for coarse tailings (5 times) and lower than the maximum reduction factor for slimes (10 times). Thus, the value is consistent with the material, since it is classified as fine tailing composed predominantly of silt size particles.
After constant-head tests in flexible wall permeameter, Ferreira (2018) found k = 7.85 × 10 −8 for a sample of non-plastic fine tailings with e = 0.60 (Table 7). Compared to the studied tailings, the hydraulic conductivity is 2.3 times lower than the value defined by the lower boundary in Fig. 5 (k = 1.83 × 10 −7 ) and 5.8 times lower than the value defined by the average trend line (k = 4.58 × 10 −7 ) for the same void ratio. The value determined by Ferreira (2018) is 3.4 times higher than the value predicted by the KC equation (k = 2.31 × 10 −8 ), so it is closer to the lower boundary defined in Fig. 5.

Conclusions
The studied tailings exhibited characteristics consistent with some other iron ore tailings of the Quadrilátero Ferrífero. The grain size distribution curves of the samples compacted statically did not show significant divergence from the initial sample. Nonetheless, the dynamically compacted specimen showed a fines content increase in the central portion of the curve. It indicates the evolution of the grain size distribution due to dynamic compaction, but not static. The standard Proctor tests done for quality control of the trial embankment showed that the field samples converged to average values of optimum water content and maximum dry unit weight close to those found at laboratory tests. However, some samples showed considerable variation. Variability of both material properties and degree of compaction was identified in the trial embankment. Therefore, mapping the tailings' variations is essential to stack the material in an adequate initial state. The main variations of the material must be characterized, and the proper degree of compaction and optimum water content must be specified for each of them. The higher the degree of compaction, the lower the hydraulic conductivity. The hydraulic conductivity determined after the constant head tests were proportional to the void ratio of each specimen. For the studied range of void ratios, k varied from 6.23 × 10 −8 m/s to 4.10 × 10 −7 m/s, a reduction factor of approximately 6,6 times. The k measured was higher than that predicted by the KC equation, and the material showed to be more sensitive to the void ratio variation for the studied range.
The data presented in section 3 is a literature review of results found by different authors for different iron ore tailings of the Quadrilátero Ferrífero. The data survey had the objective of compilating the physical properties and trying to identify trends. The properties showed variability, but samples with similar index properties were also identified. Some properties, such as maximum and minimum void ratios and hydraulic conductivity, are more sensitive to the methodology adopted for their measurement and showed higher variability. All properties presented must be carefully analysed together, and variability must be accounted for when using them as references. For a 200 m high dry stack with an average bulk unit weight of 23,5 kN/m 3 (based on w opt. and γ d,max defined in subsection 4.5), the confining stresses in the base of the stack exceed the maximum confining stress of 1.900 kPa adopted in the hydraulic conductivity tests. Higher confining stresses could not have been adopted due to the stress limitation of the equipment. As further research, it is proposed to evaluate the void ratio and hydraulic conductivity reduction with higher confining stresses. Additionally, it is proposed to evaluate the deposition, layering, and microstructure formation in TSFs; the impact of the confining stresses, and variability of degree of compaction in the mechanical behaviour of the tailings; the impact of stabilization techniques in the mechanical and hydraulic behaviour of the tailings, and the numerical modelling of the dry stacking with adequate constitutive models.