Engineering properties of fine-grained red mud

ABSTRACT

Basic engineering parameters of fine-grained red mud tailings were evaluated via laboratory tests; the coupled effect of mineralogy and microstructure was investigated by appraising the stress-deformation characteristics; and the static and seismic liquefaction susceptibilities were evaluated via critical state soil mechanics and empirical equations, respectively. Results indicated the presence of cementitious minerals without swollen clay minerals. The textural and compression properties were similar to those of clays, whereas the strength parameters were similar to those of sand. The stress-deformation curves show overall strain softening with a temporary strain-hardening curve at the beginning, which can be qualitatively explained by the coupled effect of the microstructure and mineral components. Hydroxysodalite precipitated at the contact points among the particles afforded microstructural cementation bonds, allowing the aggregates to resist external loads at the initial loading stage. Red mud presented strong resistance to static liquefaction but was potentially seismically liquefiable.

KEYWORDS:

• Tailings
• critical state
• liquefaction
• consolidation
• triaxial tests
• cementation
• mineralogy

1. Introduction

The production of valuable minerals through ore processing leads to a large amount of fine waste materials called tailings. Red mud (bauxite residue) is a type of metalliferous tailing that is generated during the extraction of alumina from bauxite. Typically, tailings are transported in slurry form by discharging pipes to surface impoundments enclosed by dams and embankments, called tailings dams. Statistical results [1,2] indicate that the failure rate of a tailings dam is almost two orders of magnitude higher than that of conventional water-retention dams, which might be partially caused by the poor mechanical stability and contractive nature of impounded tailings. These failure incidents negatively impact the environment and public safety [3,4]. Thus, extensive experimental investigations on compression behaviours for safety management [5–13] and mechanical properties for the stability assessment of tailings dams, including coal, copper, gold, and iron tailings have been conducted [5,9,14–20]. Owing to the significant dependence of tailing properties on site-specific conditions, including parent ores, mineralogy, and technical processes [6,9], the conclusions of previous experimental investigations vary from case to case [4,6,7]. Additionally, most existing investigations regarding red mud are limited to basic geotechnical parameters (e.g. internal friction angle and cohesion) [7,9,21–23]. Therefore, for a better stability assessment and safety management of the red mud tailings dam, a deeper understanding of its stress-deformation characteristics is necessary.

Microstructure (fabric) [8,15,24] and mineralogy [25] are among the major factors that affect the mechanical properties of cohesionless soils, especially tailings. With current technological advancements, mineralogy can be correctly identified [7,9,10,21,26] and microstructure can also be clearly captured within the tailings [8–10,15,24,27]. However, most previous studies have investigated the effects of mineralogy and microstructure individually, and the combined mechanism of mineralogy and microstructure on the stress-deformation behaviour of fine-grained tailings, which are particularly important in metalliferous tailings as cementitious precipitation among particles may improve their overall performance [10,26,27], remain elusive.

Liquefaction, which is characterised by a sudden loss of strength caused by excess pore water pressure induced during either seismic or static loading, is one of the most common triggers of tailings dam incidents [2]. A liquefaction susceptibility assessment is required for tailings or other granular materials, which are brittle and potentially contractive [28]. Many studies have investigated seismic [6,29–34] and static [5,6,9,14,18,35–37] phenomena. The liquefaction susceptibility of tailings is significantly influenced by particle grading [5]. However, most of the aforementioned research is limited to silt-sized metalliferous tailings with a D₅₀ value greater than 0.005 mm (0.005 mm is the boundary of the clay content in seismic analysis [38]). To the best of our knowledge, there have been no specific investigations into the liquefaction susceptibility of clay-sized metalliferous tailings. Given the differences in generating and building up excess pore pressure between silt- and clay-sized particles [39,40], more studies are necessary to provide further insight into the liquefaction susceptibility of clay-sized metalliferous tailings.

This study investigated these issues in detail. First, a comprehensive laboratory testing program (particle size distribution, index, standard compaction, triaxial compression, consolidation) was conducted on clay-sized red mud to ascertain its textural, mineralogical, and mechanical properties, and the results were analysed. Subsequently, the combined effects of the microstructure and cementitious minerals on the stress-deformation characteristics were qualitatively analysed using a three-stage compression model. Finally, the static liquefaction susceptibility was investigated under the framework of the critical state soil mechanism (CSSM), and the seismic liquefaction susceptibility was evaluated via empirical equations. The outcomes of this study provide fundamental inputs for both tailing management and dam stability assessments.

Furthermore, the novelty of this study lies in the investigation of the coupled effects of microstructure and mineralogy on the stress-deformation characteristics and liquefaction susceptibility of clay-sized metalliferous tailings.

2. Materials and experimental procedures

The fine-grained red mud investigated in this study was collected from a local tailing storage facility in northern Queensland, Australia. A comprehensive laboratory testing program was undertaken to characterise its engineering parameters, including mineralogy, specific gravity, particle size distribution, Atterberg indices, compression, and strength parameters. The specific gravity, particle size distribution, and Atterberg indices were measured using the pycnometer method, ASTM D5550–14 [41], sieving method and hydrometer tests, ASTM D6913/D6913M–7 [42] and ASTM D7928-21e1 [43], and fall cone test, ASTM D 2487–06 [44], respectively. The maximum dry density and corresponding optimum water content were determined through standard compaction tests following the ASTM standard D698–12(2021) [45]. One-dimensional consolidation tests and consolidated undrained triaxial tests were carried out to investigate the compression and strength-deformation characteristics. XRD and SEM were used to determine the mineralogical composition and microstructure, respectively.

A one-dimensional oedometer manufactured by WILLE was used to conduct the consolidation test complied to the ASTM standard D2435M–11 [46]. The sufficiency of this kind of apparatuses for the consolidation tests of tailings has been demonstrated in the literature [5–11,13,22]. Oven-dried red mud was first ground and sieved, then mixed with water and compacted into a ring with a diameter and height of 10 and 2 cm, respectively. Load steps of 100, 200, 400, 800, and 1000 kPa were applied to each sample. Statistics [3] show that most of the tailings dams are less than 50 m high which is equal to vertical load of 900 kPa; thus, the selected load steps could represent most site conditions.

Consolidated isotropically undrained (CIU) triaxial compression tests were conducted according to ASTM standard D4767–11 [47] in a triaxial testing system manufactured by GDS. A moist-tamping method was used to prepare the triaxial samples. The oven-dried red mud was mixed well with distilled water in a proportion of 4% by weight. After the initial dry density was determined and the total weight was calculated, the moistened red mud was placed in a split mould installed on the pedestal and compacted into six layers. Each layer was carefully compacted to the same height. Prior to compacting the next layer, the top surface was roughened to ensure that no horizontal seepage occurs. Following specimen preparation, a triaxial cell was assembled by mounting the chamber on a pedestal and filling it with water. After the triaxial cell and other system components were assembled, the degree of saturation was enhanced by slowly percolating de-aired water from the bottom with the top valve open to the atmosphere to fill the specimen voids, transducer, and drainage lines. This was followed by linearly increasing the cell and back pressures to 510 and 500 kPa, respectively, to dissolve air with the top valve closed. During this process, B-check tests were performed to determine the B-values [48]. A significant B-value of 0.95 or more was required to fully saturate the specimen. Upon full saturation, the samples were isotropically consolidated by increasing the cell pressure in steps while maintaining a constant back pressure to achieve the effective stress required for shearing. When the volume change ΔV of the specimen became insignificant and at least 95% of the excess pore pressure had dissipated, consolidation was considered to be admissible [49]. The samples were then sheared at a constant strain rate of 0.03 mm/min under undrained conditions, with the drainage valve closed.

Subsequently, the samples were pulverised in ethanol using a McCrone micronising mill. The resulting slurries were oven-dried at 60°C and thoroughly mixed using an agate mortar and a pestle. Finally, XRD analyses were performed on the dry specimens to qualitatively identify the mineral phases using PANalytical ‘X’Pert Pro-Multi-purpose X-ray Diffractometer with Fe-filtered CoKα radiation, and Rietveld refinement method was employed to quantitatively estimate the mineralogical composition.

3. Testing results

3.1. Basic geotechnical parameters

The measured specific gravity was 2.98, which is relatively greater than natural inorganic soil but still within the range of the red mud sorted out by Reddy et al. [4], (G_S = 2.7 ~ 3.7), reflecting the presence of some iron minerals discussed later. Particle size distribution of the red mud, shown in Figure 1, indicate the presence of large quantities of clay-sized grains (54.43%), followed by silt-sized grains (43.25%), and sand (2.32%). The coefficients of uniformity (C_U) and curvature (C_C) indicate that the investigated red mud is poorly graded. Note that the red mud is classified as lean clay (CL) according to the Unified Soil Classification System (ASTM standard D2487–06) [44], based on the testing results of plastic limit, liquid limit, and plasticity index. The preliminary geotechnical parameters of the red mud are listed in Table 1.

Figure 1. Particle size distribution of red mud.

Table 1. Preliminary geotechnical parameters of red mud.

Parameter	Value
Standard compaction tests
Maximum dry density, MDD (g/cm^3)	1.26
Optimum water content, OWC, (%)	34.5
Minimum dry density (g/cm^3)	0.803
Particle size distribution
Percentage of sand (0.02–2 mm) (%)	2.32
Percentage of silt (0.002–0.02 mm) (%)	43.25
Percentage of clay (<0.002 mm) (%)	54.43
D10 (mm)	0.001
D30 (mm)	0.001
D50 (mm)	0.0023
D60 (mm)	0.0045
Coefficient of uniformity Cu (D60/D10)	4.5
Coefficient of curvature Cc [(D30)^2/(D60×D10)]	0.222
Specific gravity (Gs)	2.98
Plastic limit, WP (%)	33.5
Liquid limit, WL (%)	47.9
Plasticity index, IP (%)	14.4
USCS soil classification	CL
Internal fraction angle, ϕ (°)	30.48

3.2. Mineralogical properties

The quantitative mineralogy (relative abundance) values are presented in Table 2. The constituent mineralogy of bentonite was also analysed and presented together with that of the ‘sand’ sample from Newson et al. [10] for reference. The XRD patterns of the micronised and Ca-saturated samples are shown in Figure 2.

Figure 2. XRD pattern of red mud, micronised - Co Kα radiation.

Table 2. Relative mineralogy by quantitative XRD analysis (wt.%) of bulk, micronised red mud, sand, and bentonite samples.

Mineral phase	Formula	Red mud (%)	Sand (%)	Bentonite (%)
Quartz	SiO2	2	1.8	10
Calcite	CaCO3	2	–	<1
Albite/Anorthite	CaAl2Si2O8	–	–	5
Orthoclase	KAlSi3O8	–	–	1
Smectite	A0.3D2–3[T4O10] Z2 · nH2O	–	–	84
Kaolinite	Al2Si2O5(OH)4	2	–	<1
Anatase	TiO2	2	–	<1
Hematite	Fe2O3	36	22.0	<1
Goethite	FeO(OH)	–	22.7	–
Boehmite	γ-AlO(OH)	2	–	–
Gibbsite	γ-Al (OH)3	<1	–	–
Nosean	Na8Al6Si6O24(SO4)	14	–	–
Sodalite	Na8Al6Si6O24(Cl2)	–	1.8	–
Rutile	TiO2	1	1.2	–
Zircon	ZrSiO4	1	–	–
Perovskite	CaTiO3	–	0.6	–
Gibbsite	Al (OH)3	–	15.7	–
Amorphous		38	34.2	–

Many geotechnical soil properties depend on their constituent mineralogy. Significant quantities of iron minerals were found in both red mud and sand, present as haematite (Fe₂O₃) and goethite (FeO(OH)), respectively, whereas the main mineral composition of bentonite was smectite, followed by quartz. Smectite often referred to as a ‘swelling’ or ‘expansive’ clay mineral contains interlayer spaces and exhibits a high expansion (swelling) capability in the presence of water [50]. Thus, the absence of smectite and other claylike minerals indicates that red mud is similar to ‘sand’ sample, yet it undergoes essentially infinite swelling as the limited expansion of composition occurs with hydration, which has been proven in related literature [10]. Apart from iron minerals, the red mud featured significant quantities of nosean (Na₈Al₆Si₆O₂₄(SO₄)), distinguishing it from the ‘sand’ sample wherein large quantities of gibbsite (Al(OH)₃) were present.

Nosean, belonging to the sodalite group, is Feldspathoid whose crystal structures are based on the tetrahedral framework (Al_XSi_1−XO₂)^x– with large cavities (β-cages) holding large cations (Na⁺, K⁺, Ca²⁺), additional anions (Cl⁻, F⁻, SO₄^2–, S^2–, S³⁻, etc.), and neutral molecules (H₂O, CO₂) [51]. The β-cages provide similar ion exchange and water retention properties to sodalite groups [52]. The extra-framework exchangeable cations loosely occupy the central cavities of the tetrahedral structure and are enveloped by water molecules, balancing the net negative charge of the anions [10]. During hydrolysis, anions accept the protons (H⁺) from the water in the reaction and leave hydroxide ions (OH^–). In nosean (Na₈Al₆Si₆O₂₄(SO₄)), the anion is sulphate (SO₄₂^–). When anions (SO₄₂^–) are absent, hydroxide ions are present in positions formerly occupied by the anions (SO₄₂^–), resulting in the compound hydroxysodalite (Na₈Al₆Si₆O₂₄(OH)₂). Iron oxide [53, 54] and hydroxysodalite [10] are considered as cementitious agents in soils. The latter has a close structural relationship with zeolites.

Based on the setting velocity of red mud, Li and Rutherford [55] indicated that amorphous and cryptocrystalline iron (Fe_d) may act as bonding agents for red mud particles and synthetic flocculants. Later, by comparing the mineralogical composition and shearing behaviour of the original and acid-washed red mud samples, Newson et al. [10] further revealed that hydroxysodalite is a cementing/bonding agent in red mud and the potential source of bonding from flocculants can be ignored. The solubility of hydroxysodalite significantly improves in an alkaline environment, which is the case for red mud, whereas it is slightly soluble in water. Thus, when in contact with water, the dissolved hydroxysodalite in red mud acts as a bonding/cementing agent when precipitating over grain surfaces and at contact points between hydroxysodalite crystals [10].

3.3. Compression properties

The measured oedometer test results as a function of the void ratio ($e$) and logarithm of the applied stress ($p$) are shown in Figure 3. The compression curve can be approximated using the classical logarithmic function:

(1) $e=e_0-C_c\times log\left(p/p_0\right)$(1)

Figure 3. One-dimensional compression curve of red mud.

where $p_0=1$kPa is the initial vertical stress; $e_0$is the corresponding void ratio and the compression index ($C_c$) is the gradient of the virgin compression line (VCL). After fitting, we obtained $=2.7216-0.4849\times log\left(p/p_0\right)$.

The value of $C_c$ for the red mud in this study ($C_c$ = 0.4849) was higher than those presented by Newson et al. [10] ($C_c$ = 0.41), Islam [7] ($C_c$ = 0.221), and Reddy and Rao [56] ($C_c$ = 0.0929), but was still within the range of those presented by Gore et al. [57] ($C_c$ = 0.28–0.56). The variability of $C_c$in red mud indicates that compressibility is strongly influenced by the parent materials of the ores used in the extraction and technological processes of the refinery [4]. According to the values of $C_c$, the investigated red mud can be characterised as highly compressible, similar to fine-grained soil.

3.4. Shearing behaviour

Red mud with different dry densities (0.95, 1.0, and 1.1) was tested, and a series of initial consolidation stresses of 50, 75, 125, and 175 kPa were used. The results of the triaxial tests are presented in Figure 5. The stress-strain curves and stress paths were initially characterised by a significant bedding curve, which was also found in previous experimental investigations on loose samples with fine grains [58–60] and might be attributed to the high initial compressibility of fine grains. The current study focused on the critical or final states of the triaxial tests; therefore, the initial bedding curves had not been addressed in this paper.

3.4.1. Stress-deformation data

The results of the undrained triaxial tests and brittleness index ($I_B$) are summarised in Table 3. The brittleness index ($I_B$), characterised by Bishop [61], is defined as

(2) $I_B=\left({\tau }_p-{\tau }_r\right)/{\tau }_p$(2)

Table 3. Summary of the undrained triaxial tests on prudent specimens.

Specimen identity	Initial dry density (g/cm^3)	Initial specific volume	Initial consolidation stress (kPa)	Dry density before shearing (g/cm^3)	Specific volume before shearing	Peak shear strength (kPa)	Residue shear strength (kPa)	Brittle-ness index (IB)
CIU-1	0.95	2.137	50	0.999	1.802	19.50	16.56	0.151
CIU-2	0.95	2.137	75	1.012	1.757	29.52	25.15	0.148
CIU-3	0.95	2.137	125	1.053	1.647	52.29	41.10	0.214
CIU-4	0.95	2.137	175	1.094	1.559	76.84	62.31	0.189
CIU-5	1.0	1.98	50	0.962	1.909	17.82	15.34	0.139
CIU-6	1.0	1.98	75	0.980	1.858	29.79	24.77	0.169
CIU-7	1.0	1.98	125	1.034	1.709	57.92	41.44	0.285
CIU-8	1.0	1.98	175	1.081	1.591	80.46	64.27	0.201
CIU-9	1.1	1.709	50	1.022	1.741	17.42	15.93	0.086
CIU-10	1.1	1.709	75	1.050	1.666	36.57	31.32	0.144
CIU-11	1.1	1.709	125	1.091	1.566	67.06	57.90	0.137
CIU-12	1.1	1.709	175	1.131	1.476	92.50	78.15	0.155

CIU isotropically combined undrained test.

Peak (${\tau }_p$) and residual (${\tau }_r$) strengths were defined under the same effective normal stress. The brittleness index ($I_B$) is a discriminator to quantify the amount of undrained shear strength reduction and characterises the susceptibility of the soil to flow failure. It ranges from 0 to 1, where $I_B$ = 1 indicates a complete strength loss with true flow failure, whereas $I_B$ = 0 indicates no strength loss or strain hardening.

Generally, the stress-strain curves of red mud (as shown in Figure 5) are strain-softening under undrained loading conditions but vary in detail depending on the specific dry density and initial consolidation stress. In all the loose samples (dry density = 0.95, 1.0), the deviator stress increased monotonically with the strain and quickly peaked at approximately 1–2% axial strain. This was followed by collapse until a quasi-steady state was reached, which lasted only for a short strain interval. Subsequently, a temporary strain-hardening characteristic emerged as the deviator stress increased. However, in contrast to the classical quasi-steady state reported in other studies [4,62,62,63,63], wherein the deviator stress continues to increase until it reaches the ultimate steady state, in this study, the increase in the deviator stress after a quasi-steady state was only maintained until an axial strain of approximately 10% was reached; subsequently, the deviator stress decreased with a further increase in the axial strain. Interestingly, the temporary quasi-steady state that developed in the red mud under undrained conditions was dependent on the initial effective consolidation stress. The drop and regain in the deviator stress were significant under high initial consolidation stresses (125 and 175 kPa). As the initial consolidation stress decreased, the volatility gradually decreased.

In all the dense samples (dry density = 1.1), the deviator stress rapidly plateaued to the yield strength at approximately 1% axial strain and then increased slowly until the final peak value was reached at approximately 8–10% axial strain. Subsequently, the deviator stress kept decreasing until the tests were completed without reaching a constant value. The changes in the deviator stress indicated that the red mud exhibited dilative behaviour from approximately 1% axial strain after the initial compression, but this dilative behaviour became less significant with the increase in axial strain and finally it tended to be contractive at approximately 8–10% axial strain, implying that red mud is ultimately contractive. No temporary quasi-steady states were observed in the dense samples. The strain-softening characteristic was more significant under high initial consolidation pressures (125 and 175 kPa) than under low confining stresses (50 and 75 kPa), implying that the contraction increased with the initial consolidation pressures at the same initial dry density.

3.4.2. Stress paths and critical state line

A stress path is defined as a track formed by a point representing the stress state in the stress plane [64]. For the consolidated undrained test, the effective stress path is usually depicted in $q-p{ }^{\mathrm{\prime }}$plane, where $q$ is the deviator stress and $p{ }^{\mathrm{\prime }}$ is the average effective principal stress. These are defined as follows:

(3) $p{ }^{\mathrm{\prime }}=\left({\sigma }_1+2{\sigma }_3\right)/3$(3)

(4) $q={\sigma }_1-{\sigma }_3$(4)

The stress paths of red mud with different dry densities are shown in Figure 5. The phase transition is defined as a temporary state of transition from contractive to dilative behaviour of the granular material, irrespective of whether it involves a temporary drop in the deviator stress [62]. The shapes of the undrained stress paths depend on both the dry density and initial consolidation stress. Under low initial consolidation stress (50 and 75 kPa), especially for loose samples (dry density = 0.95 g/cm³), the stress paths are more ‘C’ shaped with a well-defined phase transition point. However, as shown in Figure 5, under high initial consolidation stress, the paths become more ‘T’ shaped, in which a distinct reversal was observed at the end of the test, and the phase transition point was less well defined, as proven by the volatility in the stress-strain curves. Generally, the stress path showed only a small downward curve, indicating that the red mud exhibits only slight strain softening.

Regarding the strain-rate-controlled undrained tests for loose cohesion-less soils, the critical state, defined in the test as a state of deformation without an effective stress increment or decrement [65], was achieved at a certain axial strain level ranging from 5% to 60% depending on specific experimental conditions. In this set of triaxial results, constant deviator stress was not reached even at the measured maximum axial strains of about 30% when higher initial consolidation stresses (125 kPa, 175 kPa) were applied, which contradicts previous empirical observations on loose contractive tailings [8,10,11,18], where the critical state is reached at small (15–20%) axial strains under consolidated undrained condition. However, it complies to the observation of loose sands with high silt content [66] and silt-sized gold tailings [9], in which the critical state is usually achieved around 40% axial strains. The particle size distribution (PSD) analysis verified the large quantities of silty content in the red mud; this may highlight a close relationship between red mud and silt-rich cohesionless soils in terms of characteristics at the final stage of deformation.

By inspecting the stress path after phase transformation, it can be seen that despite the significant difference in the stress-strain curve over the entire loading range, the samples tended to have the same stress path at the end of the tests. All loci at the end of the stress path were located in a straight line through the origin, representing the approximated CSL in the $q-p{ }^{\mathrm{\prime }}$ stress plane. Therefore, the state at the end of the test can be used to represent the critical state [67]. The gradient of the CSL (M) in the $q-p{ }^{\mathrm{\prime }}$ plane was 1.22, corresponding to a critical state internal friction angle,${\phi }_{cs}^{\prime }$ = 30.46°. The high friction angle was similar to the shearing parameter of sandy soils, which has also been reported in the literature [4,10]. Moreover, the coincidence of effective stress paths before the critical state suggests that the initial dry density and consolidation stress did not significantly affect the critical state of the soil, probably because a common fabric developed at the stage of large deformation [62] derived from the same mineralogy, particle size distribution, and angular to subangular particle shapes of all particles [8].

4. Discussion

4.1. Effect of mineralogy and microstructure on mechanical behaviour

4.1.1. Cementation bonds within the microstructure

The mechanical behaviour of tailings is related to the fabric effects of the sample preparation method. Previous studies [68–73] using SEM and mercury intrusion porosimetry (MIP) techniques on the compacted fine-grained soils (e.g. silt, clay) have confirmed a well-displayed microstructure composed of voids and aggregates, characterised by a skeleton made of silty grains linked together by clayey particles, when compacted under low moisture content [69]. This as-compacted structure displays well-defined inter-aggregate pores (macrovoids between soil aggregates and unbonded grains) and intra-aggregate pores (microvoids within the soil aggregates).

In red mud and other fine tailings classified as silt (ML) or lean clay (CL), PSD analysis typically shows a high percentage of silty and clayey grains. These characteristics suggest that fine tailings may develop the same aggregates under compaction as cohesive soil. This hypothesis is proven by the SEM images shown in Figure 4, wherein the red mud samples compacted at a low moisture content display a microstructure composed of dissociative grains and integrated aggregates with well-defined large macrovoids. The aggregates are bonded by fine grains with small microvoids. No obvious particle orientation was observed in the aggregates, indicating that they formed a heterogeneous continuous matrix. Hydroxysodalite, produced through hydration reactions in an alkaline environment, lead to the formation of aggregates with a certain unconfined compressive strength [74]. Therefore, the aggregates are not loose clusters, like sandy materials, but can bear external loads as equivalent coarse particles. Similar observations have also been reported in the literature (red mud [10,75], cemented tailings backfill [27,74] and fly ash [76]. A common feature of the materials employed in this study was the presence of cementitious active minerals.

Figure 4. SEM images of compacted red mud sample at 100 nm.

The as-compacted structure may get erased during saturation, as reported for high-plasticity Maryland clay under oedometric conditions [68]. This is possible because of the swelling of clay minerals during observation, which leads to an increase in micropores and decrease in macropores [77]. PSD analysis usually shows large contents of fine particles in red mud [4,10,75] and other fine tailings [76]. The XRD patterns of red mud showed virtually no swelling clay minerals (e.g. Montmorillonite, Bentonite, and Vermiculite) present and yet present ‘claylike’ behaviour (swelling). Thus, the as-compacted structure of red mud can be sustained until consolidation during the triaxial tests.

From the foregoing mineralogical discussion and verification of the SEM images, because of the existence of hydroxysodalite in an alkaline environment, the aggregates were bonded together by the hydrosodalite crystals growing around them over the particle surfaces and at contact points with a skeleton made of silt-sized grains. This crystallisation forms cementation bonds within the aggregates. The load-bearing capacity of cementitious bonding in red mud is quickly re-established after remoulding [10] and is demonstrated to be approximately constant during compressive loading [78].

The three proposed phases of the skeleton in red mud after compaction and crystallisation, composed of aggregates and other dissociative grains, are shown in Figure 6 (A). The bonding force in red mud can be classified as intra-aggregated or inter-aggregated bonding. The total resistance to loads can be divided into resistance to frictional bonding from the unbonded soil skeleton and resistance to cementitious bonding in cemented soils [78]. Intra-aggregate bonding comprises frictional bonding between the bonded particles and cementitious bonding from the hydrosodalite crystals, whereas inter-aggregate bonding is frictional bonding between the aggregates and dissociative grains. Intra-aggregate bonding can enclose aggregates and allow them to resist external loads, until the aggregate is crushed.

Figure 5. Undrained shearing behaviour of red mud: (a) deviator stress-strain curves of effective confining stress = 175 kPa and 125 kPa; (b) deviator stress-strain curves of dry density = 1.1 g/cm³; (c) deviator stress-strain curves of dry density = 0.95 g/cm³; (d) critical state line; (e) effective stress paths of dry density = 1.1 g/cm³; (f) effective stress paths of dry density = 0.95 g/cm³.

Figure 6. Schematic illustration of the stress-deformation process of compacted red mud with soft cemented aggregates with proposed three different stages: A) loads taken by bonded aggregates and dissociative grains; B) bonded aggregates start breaking down under external loads; C) bonded aggregates fully destructured and loads taken by grains; D) possible structure of soft cemented aggregates; E) possible structure of aggregates start breaking down; F) aggregates completely destructured; G) schematic stress-strain curves; H) schematic stress path.

4.1.2. Coupled impacts of mineralogy and microstructure

For properties associated with a wide range of axial strains of red mud in triaxial tests, the stress-deformation characteristics can be described in three distinct stages based on internal resistance, which is related to the transformation of aggregates owing to an increase in stress, under undrained conditions (no volume change). These aggregates were formed in the compaction and sustained during the consolidation. The stress deformation characteristics are schematically illustrated in Figure 6.

As for the stress-deformation characteristics, as illustrated in Figure 6 (A), when the loads are applied to the specimen, the aggregates and dissociative unbonded grains resist the load through inter-aggregate frictional bonding. Intra-aggregate bonding resists a portion of the load transferred to the aggregates. In soft cemented soils, within an aggregate, inter-aggregate bonding resulted from the resistance of skeletons sliding over each other through friction or interlocking, with the cementitious bonds providing additional constant resistance. In cemented, saturated, and fine-grained soils, frictional and brittle cementitious bonding operate simultaneously [78]. For loads within the yield strength, the cementitious bonds could withstand additional loads, with negligible deformation, as illustrated in Figure 6 (A). Once the load applied to the aggregates was greater than their yield strength, as illustrated in Figure 6 (B), the aggregates started to break up, causing the brittle cementitious bonding to disappear. As the development of axial strain continued, an increasing number of aggregates collapsed (Figure 6 (C)), resulting in yielding and transferring of the load carried by them to the frictional bonding among the disintegrated grains. The corresponding frictional bonding could not withstand these additional loads under undrained conditions (no volume change), and a deviation in the stress-strain curve occurred (Figure 6 (G)). In this study, the cementation bonds broke down at an axial strain of approximately 8–10%.

From the above discussion, the steep strain-stress curve (Figure 6 (G)) under a small axial strain (1–2%) can be attributed to changes in the components of the stresses on the aggregates. Because the aggregates were larger than the red mud grains, the stress-strain curve behaved as sand. As the strain increased, the curve exhibited strain-softening or a quasi-steady state depending on the confining stresses, which is also the typical undrained shear behaviour of dense sand under large deformation (Yoshimine and Ishihara [65]). With a further increase in strain, the stress-strain curve becomes strain-softening after the peak deviator stress, implying that the aggregates are breaking down and friction bonds cannot withstand the extra loads among the fine particles. In other words, before the peak deviator stress, the red mud behaves like coarse-grained soil (sand), whereas it behaves like fine-grained soil (clay and silt) after passing the peak. This observation is also consistent with the high friction angles and high residual strength characterised by $I_B$ values of red mud ranging from 0.08 to 0.29.

The stress and strain curves (Figure 5) conform to the hypothesis that red mud is soft-cemented, leading to a high axial strain at the critical state. As illustrated in Figure 6 (G), under low stress levels, the cementation bonds enclose the aggregates and the deformation derived from the rearrangement of the aggregates and macrovoids dominates. Once the yield strength is reached, where the aggregates begin to break down with decreasing cementitious bonding, the existing soil structure built through the initial rearrangement, mainly from the aggregates, collapses. The intra-aggregate microvoids are released into the inter-aggregate macrovoids, the soil contracts significantly, and excessive pore pressures generate. With the disassembly of aggregates, comes a transformation in the size and shape of the load-bearing materials. External loads are applied by smaller disintegrated grains rather than the original aggregates. As the loading continue, a new rearrangement occurs between the disintegrated grains and changed voids which suggest that further axial strain is required to rebuild the pore pressure and reach the critical state.

The evolution of the stress paths (Figure 5) can also be explained from the perspective of the collapse of the cemented microstructure. As illustrated in Figure 6 (H), after phase transformation, the stress paths initially turn right to approach the critical state, which coincides with the strain-hardening stress path of dense sand [65], indicating that the resistance against loads mainly arises from inter-aggregate bonding. As the intra-aggregate cementitious bonds began to disappear, the stress path deviated to the left, implying the collapse of the aggregate structure and an attempt at rearrangement of the disintegrated grains. In comparison, the stress path is consistent with loose sand with high non-plastic fines under a low initial consolidation stress [66,79], suggesting that the load-bearing capacity derived from cementation bonding disappears, and the resistance against loads mainly arises from the friction bonds among disintegrated grains at this stage (Figure 6(C)).

4.2. Liquefaction susceptibility

4.2.1. Static liquefaction potential under CSSM

Liquefaction is frequently associated with seismic events. However, mine tailing impoundments have demonstrated more static liquefaction events than seismic-induced events [80]. Static liquefaction, also known as flow failure, is defined as the rapid reduction of shear strength in saturated contractive granular materials in response to applied static loads owing to increased pore pressure, which triggers the development of large strains and strain-softening behaviour [81,82].

The susceptibility of tailings to static liquefaction is typically evaluated based on CSSM. The critical state conditions for red mud are shown in Figure 7, where the void ratio $e$ is plotted against the logarithm of the effective mean stress $p{ }^{\mathrm{\prime }}$. The normal compression line (NCL) is presented for reference. The red mud exhibited a nonlinear CSL which could not be approximated by a single straight line in the $e-ln{p}^{\prime }$ plane. However, a curved CSL can fit within the entire range of mean stress.

Figure 7. Void ratio (e) plotted against effective stress (p′) under the concept of Carrera et al. [5].

When the samples were sheared from NCL to CSL, the susceptibility to static liquefaction was controlled by the corresponding locations on the CSL and the mean stress. At higher mean stresses, the gradient of the CSL was approximately parallel to that of the NCL, similar to the literature on other tailings [5,9,14]. As the mean stress decreased, CSL tended to flatten. Samples under shearing exhibit strain softening to a critical state associated with large deformations and high pore pressures, causing the shear stress resistance of the material to decrease from the initial peak to its residual value at the critical state. The samples achieved a stable state in the CSL, although they underwent significant strength loss. The significance of the strain-softening effects can be estimated using the distance from the NCL to the CSL. This strain-softening behaviour may eventually lead to flow failure when the static loads exceed the residual shear strength under undrained conditions [82]. As the mean stresses decreased further, the initial state shifted beyond the horizontal asymptote of the CSL, but flow liquefaction must occur because the critical state can be defined in the $e-ln{p}^{\prime }$ plane without a corresponding position on the CSL. For red mud, under undrained compression, a reduction in shear strength occurred; however, flow liquefaction, which had previously been reported for other silt tailings [5,14,18,36,37], did not occur.

In problems associated with a wide range of stresses, the deviation of the CSL from the straight line defined by the slope in the $e-ln{p}^{\prime }$ plane at mean stresses ranging from high to low under undrained conditions highlights the necessity of describing static liquefaction susceptibility in three distinct state conditions, conceived in most cases of static liquefaction assessment under the framework of CSSM critical state soil mechanics [5,14,18], as follows:

1. Stable condition. At higher mean stress, the sample exhibits compressive and general strain hardening under shearing. The stable critical states are located in the straight part of the CSL, where the slope is essentially parallel to that of the NCL.

2. Flow instability. Moderate to low mean stresses. The sample suffers from a certain strain softening but can still reach a distinct position on the curved part of the critical state line. Static liquefaction occurs when permanent loading exceeds the residual shear stress under undrained conditions.

3. Flow liquefaction. At a very low mean stress level, the sample displays pronounced strain softening, which ultimately leads to static liquefaction characterised by a complete loss in effective shear strength and the development of excessive shear strains [14]. The critical state lies beyond the horizontal asymptote of the CSL.

The triaxial test results indicate that the critical states were located under the horizontal asymptote of the CSL, ranging from the upper bound of the flow instability to stable conditions, implying a strong resistance to static liquefaction.

4.2.2. Assessment seismic liquefaction through empirical criteria

The liquefaction potential of fine-grained soils under cyclic/seismic loading can be evaluated using either geotechnical tests (e.g. cyclic triaxial (CTX), cyclic direct simple shear (CDSS), and cone penetration tests (CPT)) or empirical criteria [83–85]. However, the testing approaches require sophisticated apparatus and specially trained staff. Alternatively, evaluating the cyclic/seismic liquefaction potential of fine-grained soils via empirical criteria [83–85] has been extensively applied in current engineering practices, which rely on the Atterberg indices, percentage of clayey particles, and water content.

Seed and Idriss [84] proposed the standard known as ‘Chinese criteria’ based on the in-situ data collected by Wang [86], where the fine-grained soils with a percentage of fine particles (smaller than 0.005 mm) less than 15%, liquid limit (W_L) less than 35%, and water contents to liquid limit (W_C/W_L) ratio greater than 0.9, are potentially liquefiable under cyclic/seismic loading. As listed in Table 1, the liquid limit (W_L) and plastic index (I_P) of the red mud were 35.5 and 14.4, respectively. Therefore, red mud is safe under cyclic loading according to the Chinese criteria.

However, with advances in evaluating the liquefaction potential of fine-grained soils, the criterion utilising the fine content as a discriminator has been proven to be unreliable in engineering practices [83]. Plasticity index (I_P) was recently recommended as a better discriminator. Seed et al. [85] reported that soil is potentially liquefiable under cyclic loading conditions if the liquid limit (W_L) is less than 37 and the plastic index (I_P) is less than 12. The soil with a liquid limit (W_L) between 37 and 47 and a plastic index (I_P) between 12 and 20 require further laboratory testing. More recently, Bray and Sancio [83] suggested soils with a plastic index (I_P) less than 12 and a water content to liquid limit (W_C/W_L) ratio greater than 0.85 are liquefiable and that with a plastic index (I_P) between 12 and 18 and W_C/W_L ratio between 0.8 and 0.85 have moderate susceptibility. The liquefaction evaluation criteria are summarised in Table 4.

Table 4. Summary of liquefaction evaluation criteria.

Criterion	Plastic Index (IP)	Fine particle contents (%)	Liquid limit (%)	Water content to liquid limit ratio WC/WL	Liquefaction susceptibility
Seed and Idriss [84]	–	<15	<35	>0.9	Potentially liquefiable
Seed et al. [85]	<12	–	<37	–	Potentially liquefiable
	12–20	–	37–47	–	Require further identification
Bray and Sancio [83]	<12	–	–	>0.85	Potentially liquefiable
	12–18	–	–	0.8–0.85	Moderate susceptibility

Despite some controversy over the effectiveness of using the plasticity index (I_P) as an indicator for estimating the liquefaction potential of clayey soils with certain pore water chemistry under cyclic loading [87], the criteria developed by Seed et al. [85] and Bray and Sancio [83] are commonly used. According to the criteria indicated in Figure 8 (a and b), red mud may have the potential to liquefy under cyclic loading and requires advanced tests such as CTX or CDSS tests to further identify its liquefaction susceptibility. [88–90];

Figure 8. Liquefaction susceptibility of red mud: (a) criterion proposed by Seed et al. [85]; (b) criterion proposed by Bray et al. [83].

5. Conclusions

In this study, the engineering properties of fine-grained red mud from TSF in Australia were thoroughly investigated, and the following conclusions were drawn:

1. The most significant components in the investigated red mud were clay-sized grains, followed by silt-sized grains. The median particle size (D₅₀) was 0.0023 mm. The dominant mineralogical constituents are haematite and nosean, with small amount of kaolinite. This implies that red mud can be classified as a non-plastic clay-sized granular material.

2. The red mud had distinctive mechanical properties compared with common observations in natural soils, that is, a higher frictional angle with higher compressibility and higher axial strains to the critical state. The microstructure compacted at a low moisture content was the same as that of fine-grained soils. This microstructure comprises macrovoids, well-defined aggregates, and microvoids inside the aggregates.

3. The samples displayed an overall strain-softening behaviour under undrained loading conditions, as verified by the Brittleness Index. This strain-softening trend is coupled with a temporary strain-hardening trend during the initial loading phase. This unique feature can be explained by the coupled effect of microstructure and mineralogy. The absence of clay minerals allows the aforementioned microstructure to survive during the saturation process. Meanwhile, the dissolved hydroxysodalite produced by nosean through the hydrolysis reaction induced cementation bonds among the grains within the aggregates. Therefore, the aggregates could bear external loads up to a threshold value as equivalent assemblages. Subsequently, further increased loads caused the breaking up of cementation bonds and consequent disassembly of aggregates. Thus, the external loads were borne by the disintegrated grains, causing a reversal in the stress path and a drop in the strain-stress curve.

4. A nonlinear CSL was identified in the volumetric plane, subdividing the susceptibility to static liquefaction into three distinct conditions: flow failure, flow instability, and stable condition. The critical states in this set of triaxial tests were located at the upper bound of the flow instability to the stable condition zones, indicating strong resistance to static liquefaction.

5. The assessment using empirical criteria implied that red mud might liquefy under cyclic loads. Advanced experiments, including cyclic triaxial tests and direct simple shear tests, are required to confirm and quantify the cyclic/dynamic liquefaction susceptibility.

This study focused on a laboratory approach and did not consider field-scale information (e.g. CPT) when investigating the response to static loading. The static liquefaction potential in the clay-sized red mud contradicts existing research on other silt-sized tailings, which might be attributed to the liquefaction resistance from the high percentage of clay-sized fines. Further studies are required to understand the influence of non-plastic clay-sized fines content on the static liquefaction of metalliferous tailings.

Disclosure statement

No potential conflict of interest was reported by the authors.