Effects of Ferrosilicon Fineness on the Multiphase Flow Field and Separation Performance of Dense Medium Cyclones for Low-Grade Ores Authors: Shuli Wang, Jun Wang, Guanzhou Qiu, Li Shen Published: 2024 ABSTRACT As natural resources continue to be exploited, dense medium cyclones (DMCs) are increasingly utilized for the preconcentration of low-grade ores to meet the demands for higher feed grade, increased processing capacity, and reduced energy consumption. However, determining the optimal fineness of ferrosilicon remains ambiguous for different types of ores and is often described as more of an art than a science. This paper investigates the subtle effects of ferrosilicon fineness on flow field characteristics, medium classification, and the ore separation process using a validated numerical approach, which integrates a two-fluid model, a turbulence dispersion model, and a discrete phase model. The results reveal that, under consistent operating conditions, increasing the ferrosilicon fineness enhances the tangential velocity and pressure drop while having minimal impact on axial velocity and water split. The ferrosilicon fineness also influences the distribution of the turbulence field, and in combination with the changes in the particle size distribution (PSD) itself, it modifies the characteristics of the density field. This results in a reduction in the density difference between the overflow and underflow with finer ferrosilicon particles, thereby lowering the cut density but improving the separation accuracy. Moreover, the residence time of particles with different densities in the DMC also varies significantly with changes in separation density. The study underscores that even minor changes in ferrosilicon fineness can impact separation performance, highlighting the need for careful consideration of this factor in process design and medium recovery. Introduction As mining resources are increasingly developed, high-quality and easily accessible deposits are becoming increasingly scarce, leading to more challenging and costly extraction processes. For example, China stands as one of the world’s largest producers and consumers of mineral resources, with a rich reserve of strategic metal minerals that are, however, characterized by being “low grade, fine grain, and complex symbiosis”.1 Traditional beneficiation methods, such as full grinding and flotation, are energy-intensive and generate abundant fine tailings, posing significant environmental2 and economic challenges.3,4 Hence, the preselection (or preconcentration) of ores before grinding has emerged as a critical strategy to address these issues by reducing the load on subsequent processes and enhancing overall mineral processing efficiency.5 This approach also allows for the utilization of discarded coarse gangue minerals as construction materials, promoting resource efficiency and sustainability.6 Among various preselection methods7−9 used in mineral processing, dense medium separation (DMS) has been widely employed in industries like coal,10 phosphate,11 diamond,12,13 lithium,14 and iron15 ores, due to its high adaptability and efficiency. In DMS, the ore-medium mixture is fed into separation vessels where separation occurs based on the relative densities of medium and ores. These vessels can be categorized as either static, relying on gravity as the primary driving force, or dynamic, utilizing centrifugal forces. The dense medium cyclone (DMC), patented in 1942 by the Dutch State Mines, is a typical piece of dynamic separation vessel. Its geometric structure is just like a hydrocyclone but involves a more complex separation process due to the heterogeneous suspension of water and solids used as the dense medium.16 Subject to the centrifugal force, the DMC involves both the classification of dense medium particles and the separation of ore particles within a nonuniform viscosity-density field, with interactive effects between these processes.17 Therefore, effective DMC operation highly depends on matching the medium properties with ore characteristics. Over the years, heterogeneous suspensions have been created using materials such as sand, galena, magnetite, and milled/atomized ferrosilicon, with the latter two being most widely used for low-density and high-density separation, respectively.18 Research by institutions such as JKMRC19,20 and PESCO16 has shown that these suspensions are unstable and exhibit non-Newtonian fluid characteristics, including yield stress and shear-thinning behaviors. The stability and rheology of these suspensions are oppositely affected by the concentration, fineness, density, and shape of suspended solids, as well as contaminant content and shear rate. For instance, increasing the density or fineness of DMS powder can significantly improve suspension stability but also increase apparent viscosity and flow resistance. Compared to milled ferrosilicon, atomized ferrosilicon has higher sphericity, and lower viscosity but poorer stability.21 In practice, an unstable or overly viscous medium due to improper PSD can lead to severe efficiency losses. Therefore, maintaining a balance between the stability and viscosity of the dense medium is crucial, especially for high-density separation.22 International suppliers offer various types and fineness of DMS powders, like 75D, 270D, Cyclone 40, etc., to meet the requirements of different operational requirements. However, this does not imply a fixed recommended fineness for a specific ore, and the selection remains a complex blend of art and science.23 Physical experiments can help determine the optimal type of ferrosilicon, but setting up and calibrating the test apparatus, as well as recycling the medium, is extremely labor- and resource-intensive. These experiments yield macroscopic results highly dependent on the properties of the ore, lacking insights into the internal flow field, making the conclusions often lack general applicability. Consequently, those empirical models developed by the Dutch State Mines,24 Collins,25 Davis,26 and Wood27 are often limited to coal or specific ores. In recent years, numerical modeling has become an essential tool for enhancing our fundamental understanding of fluid–solid flows in recent years. For DMCs, simulation accuracy predominantly hinges on the selection of the turbulence model and multiphase flow model.28 Due to the inherent anisotropy of turbulence within a DMC, the choice often falls between the Reynolds stress model (RSM) and the large eddy simulation (LES), with the latter demonstrating superior predictive capabilities for fluctuating velocities and turbulence dispersion effects.29 For the complex water–air–ferrosilicon–ore multiphase flow, many modeling approaches are available and usually involve several steps. For the simulation of dense medium flow, the two-fluid model (TFM) stands out as the most comprehensive Eulerian-Eulerian model suitable for varying solid concentrations,30 but its computational demands become quite pronounced when dealing with a broad PSD. Hence, a more prevalent alternative in the literature is a simplified TFM, the algebraic slip mixture (ASM) model.31,32 While the ASM model33 is generally used to simulate the dense medium flow field, few studies34 have used it to simulate the separation of ore particles, as they are typically in millimeter scale and the basic assumption of slip velocity in ASM is not suitable. It is important to note that in DMCs, the concentration of solid particles is very high, especially for metallic ores, and the effects of collision and friction between particles are significant. Using the original ASM combined with a simplified kinetic theory for granular flow (KTGF) makes it challenging to produce meaningful results.35 Therefore, previous studies31,35−38 often use an empirical viscosity model to correct medium viscosity. For example, Wang et al.31 used a modified Ishii model with a fixed correction factor (3.8), which worked well for coal beneficiation. However, by checking the original study39 involving viscosity measurements, ferrosilicon powder was used to make suspensions, whereas coal beneficiation typically uses magnetite as the dense medium, making the model’s applicability questionable. Narasimha et al.40,41 also used an empirical viscosity equation calibrated by fine fractions and the Saffman-Mei42 lift force model to successfully simulate the density field in a DMC, achieving good agreement with Gamma-Ray test results.43 Nevertheless, the above method assumes that the dense medium is a Newtonian fluid, which is also debatable.35 As ore particles are typically on a millimeter scale, the discrete element method (DEM) is more appropriate than TFM. It simulates the separation process by tracking the motion of individual particles according to Newton’s laws of motion, which combines the strengths of CFD and DEM to provide a comprehensive understanding of complex multiphase systems. To enhance model robustness and reduce computational load, coarse-grained DEM (CG-DEM)32 or Lagrangian particle tracking (LPT)44 models are more widely used. Scholars have employed these models to investigate the impact of various design parameters,45 operating conditions,46 and feed properties.47,48 Regarding the influence of ore PSD, studies by Chen et al.49,50 and Chu et al.47,51 explored the performance of DMCs when handling coal particles of varying sizes and densities. Their findings indicated that both particle and medium flows exhibit sensitivity to the ore PSD, as particles of different sizes follow distinct trajectories, significantly altering bulk particle-fluid interactions. However, the effect of ferrosilicon PSD was not considered in the above studies. This paper aims to address these limitations by exploring the intricate relationship between ferrosilicon fineness, flow field characteristics, and separation performance in DMCs. First, experimental trials were conducted on a carefully designed DMC setup using milled ferrosilicon. The resulting data serve as a basis for validating and refining numerical models. The Eulerian-Eulerian framework enhanced by a turbulent dispersion model (TDM) was utilized to accommodate a wider range of ferrosilicon PSD variations. The LPT model was used to track the trajectories of different ore particles. This approach provides valuable insights into flow patterns, ferrosilicon segregation, and underlying mechanisms within the DMC. Methodology Experimental Work This section provides a detailed description of the laboratory test rig’s construction, material properties, and operating conditions used in this study. Test Rig The physical experiments were designed and carried out in the semi-industrial laboratory of Weihai Haiwang Hydrocyclone Co., Ltd. Figure 1 shows the schematic diagram and implementation of the laboratory test rig, which consists of a measurement and control system, a feed system, and a separation system. The feed flow rate is controlled by a KFJ-35 centrifugal pump and measured by an MGG-50 magnetic flowmeter produced by Azbil Corporation. The feed pressure is measured by a Y-100 pressure transmitter. The slurry is prepared and maintained in a homemade 1200L mechanical stirring tank equipped with a frequency converter. The signals from all these electronic devices are brought together into an electric cabinet and controlled by an embedded micro-PC. A bypass pipe was mounted on the main pipe to ensure pulp circulation and feed sampling. The test platform contains a set of two-product Φ250 mm DMC and a set of three-product DMC, while only the former is used in the current study to investigate the performance of water–air two-phase flow and medium segregation. Schematic diagram and implementation of the laboratory test rig on the Φ250 mm DMC: 1 - Electric cabinet; 2 - Mixing tank; 3 - Centrifugal pump; 4 - Magnetic flowmeter; 5 - Pressure transmitter; 6 - Φ250 mm DMC; 7 - Platform. Properties of Dense Medium The separation of low-grade ores usually requires ferrosilicon with a higher density as the heavy medium. Similar to hydrocyclones or cyclones,52,53 particle shape and particle size play an important role in the stability/viscosity of ferrosilicon suspensions. Physical tests used 270D ferrosilicon with a silicon content of 14–16%, which has proven to provide the best corrosion protection without too great a reduction in particle density.19 The density and particle size of ferrosilicon were measured by using an Anton Paar DMA 4200 M powder density meter and a Malvern 3000 laser particle size analyzer, respectively. The results indicate that the average density of ferrosilicon is 6900 kg/m3, with an error of 15 kg/m3. Figure 2 compares the actual PSD of ferrosilicon used in the physical tests and varied PSDs considered in numerical simulations. Practically, the percentage passing 325 mesh (45 μm) sieve is often used to differentiate fineness. The actual ferrosilicon fineness is about 88%, with most ferrosilicon particles falling within the 20–30 μm size range. Numerical experiments involve five sets of ferrosilicon samples, each with varying −45 μm content levels: 75%, 80%, 85%, 90%, and 95%. For the convenience of expression, these samples are assigned to F75, F80, F85, F90, and F95, respectively. Their specific grain compositions were determined using the Johnson SB function,54 as shown in eq 1, which provides smooth variations in PSDs1where erf(x) is the error function, σj is the distribution coefficient, and dmin, dmax, and d50f represent the minimum, maximum, and median sizes, respectively. Based on the PSD shown in Figure 2(b), it can be observed that the primary differences between samples lie in the content of particles smaller than 20 μm and those larger than 40 μm. (a) Actual PSD of ferrosilicon used in physical tests and (b) varied PSDs of ferrosilicon in simulations. Test Procedure and Data Collection The laboratory tests were designed to assess the effects of feed pressure on water–air flow and medium segregation results. A typical test consists of three key steps: material preparation, sampling, and postprocessing. Material Preparation Initially, 400 L of tap water was injected into the stirring tank. To examine the medium segregation performance, 560 kg of 270D ferrosilicon powder was added to prepare the dense medium. It is important to note that the computed slurry density may not exactly match the actual feed medium density (ρf); in fact, it tended to be slightly higher. After thorough stirring for 10 min, the centrifugal pump was initiated at a frequency of 30 Hz, and the bypass pipeline was opened to allow for feed sampling. Continuous recirculation ensured that all slurry streams returned to the stirring tank, maintaining consistent feed qualities. Throughout this process, readings from both the pressure transmitter and the electromagnetic flowmeter were transmitted to a remote-control panel. Sampling Gradually increase the pump frequency to achieve desired feed pressure readings. Once a quasi-steady state is reached, samples from the underflow, overflow, and feed streams are gathered, with minimal temporal discrepancies. Note that the sampling locations are marked in Figure 1. The sampling mass is approximately 1.5 kg to ensure the representativeness of samples. For each operating condition, three samples were taken in parallel Postprocessing The wet mass of each sample was measured and then subjected to filtration, drying, and weighing to establish mass flux ratios. Note that in the following analyses, the medium or particle density is expressed as the relative density (RD), which is defined as the density ratio of medium (ρm) or particle (ρp) to water (ρw). A subset (5–10g) of samples was used to determine the PSD. After computing the mass balance and analyzing the PSDs in the input and output products, the partition curve can be drawn. The specific experimental scheme and results are shown in Table 1. For each test, three parallel runs were conducted to obtain an average result, which served as validation data for numerical experiments. Trail tests have shown that the measurements error was less than 3%. Since the pressure transmitter was installed on the horizontal pipe rather than the actual DSM input location, and there was a bend between them, it was necessary to perform a pressure correction based on a practical fluid Bernoulli equation. Table 1Laboratory Test Scheme and Results Feed pressure (kPa)Flow rate (m3/h)RDTest No.MediumSetCalibratedQfQuρfρuρo1 10081.3760.3510.37 2 12097.6466.1211.22 3water140113.9171.4211.99 1.00 4 160130.1876.3512.41 5 180146.4580.9812.71 6water and ferrosilicon14098.4763.7115.351.942.991.61 Laboratory Test Scheme and Results Numerical Simulation Model Description The internal flow field of the DMC is highly complex, and direct solutions for gas–liquid–solid multiphase flow often encounter convergence difficulties. To address this, the study employs a modeling strategy, as illustrated in Figure 3, to systematically explore the flow field characteristics, ferrosilicon classification, and motion behavior of ore particles in three steps. Modeling strategy used in this study. In the first step, the RSM-TFM model is used to obtain a preliminary water–air two-phase flow field, focusing on forming a stable air core. In the second step, ferrosilicon particles of all sizes are introduced, and the RSM-TFM coupled with the TDM is employed to examine the spatial distribution of ferrosilicon particles, thereby characterizing the density field within the DMC. In the final step, based on the time-averaged gas–liquid–solid multiphase flow field, the DPM is used to track the motion behavior of individual ore particles in the slurry, with an emphasis on analyzing their separation performance. Further details regarding these models are provided below. Multiphase Model The TFM treats both fluids and solids as interpenetrating continua through the implementation of the volume fraction concept. The equations of continuity and momentum are satisfied individually by each phase, which can be expressed as23where αq, ρq, and symbolize the volume fraction, density, and velocity of the phase q, respectively. In general, q = 1 pertains to water, q = 2 relates to air, and q ≥ 3 refers to ferrosilicon particles. The prediction of medium sedimentation and segregation greatly depends on the accuracy of rheological properties. Considering that different ferrosilicon PSDs are used in this study, the fluid shear environment and size composition in the DMC are such complex that using a simple Newtonian fluid slurry viscosity correction model may not be appropriate. Hence, the kinetic theory for granular flows (KTGF) is applied in this process. For granular flows in the compressible regime, a solid pressure pq is added in eq 3 to account for the particle interactions consisting of kinetic and collisional terms4where Θq is the granular temperature and proportional to the kinetic energy of particles’ random motion, is the average diameter, and g0,pq is the radial distribution function. The last term adjusts the probability of collisions among granular phases. If the total volume fraction of particles surpasses a certain threshold, additional pressure, attributable to particle friction, ought to be introduced. Another important term is the stress–strain tensor, , which can be described by5where μq and λq represent the shear viscosity and bulk viscosity, respectively. In the KTGF, the shear viscosity is highly dependent on the shear rate and is composed of collisional, kinetic, and frictional components6where μq,col and μq,kin are modeled by the Gidaspow et al.55 model and μq,fr is modeled by Schaeffer’s expression.56 The bulk viscosity λq is calculated according to the Lun et al.57 model. These equations are given below.78910 Turbulence Model It is well recognized that both RSM and LES can predict the turbulent flow field of cyclones, such as air cyclones, hydrocyclones, and DMC. Despite that the LES gives more accurate predictions on fluctuating velocities and pressure, the RSM shows advantages in memory requirements and computation efficiency as well as good compatibility with other models such as the TFM.52 Hence, the RSM is used to model turbulent flows. model turbulent flows. The equations have been well documented in the literature58,59 and are not repeated here. It should be noted that although the secondary phase volume fraction is comparable to the primary phase in some positions, their velocity differences are not that significant. Hence, a more stable option called the RSM mixture turbulence option is used which shares the same turbulence field for all phases. Turbulent Dispersion Model The turbulent dispersion force essentially acts as turbulent diffusion in particle motions, which is crucial for accurately predicting the spatial distribution of ferrosilicon particles, i.e. the density field. In numerical simulations, the immediate momentum transfer between dispersed and continuous phases can be expressed as the sum of the mean drag force and the resulting turbulent dispersion force from the averaging process11where and are the mean phase velocity vectors and Kpq is the interphase exchange coefficient. The second term, Kpq, accounts for the dispersion of the secondary phases due to transport by turbulent fluid motion. DMCs involve three phases of air, water, and solid, where a distinct interface exists between the air and water phases. As in previous studies33,60 concerning hydrocyclones, the coefficients Kpq used for fluid–fluid and fluid–solid momentum exchanges are determined by the Schiller–Naumann model61 and the Gidaspow model,55 respectively.1213For the turbulent dispersion force, this study applied a model62 based on Favre averaging of the drag term, and the final expression is like14where CTD is a user constant, Dq is the dispersion scalar estimated by turbulent viscosity of the continuous phase, σpq is the Prandtl number, and is the gradient of dispersed phase volume fraction. Discrete Phase Model Ore particles are relatively coarse, typically a few millimeters—comparable to the grid size. Meanwhile, their sphericity has a significant impact on particle movement. Hence, it seems inappropriate to continue using the TFM to simulate ore separation. In this study, the LPT model is used for ore particle tracking in a fixed dense medium flow field. According to differential equations for Newton’s Second Law, a general force balance on a particle can be written as15where the terms on the right-hand side denote the drag force, buoyancy force, and additional forces on per unit mass of particles, respectively. The symbol ρp is the particle density, and ρm is the medium density. The drag function is given by16where dp is the particle diameter, CD is the drag coefficient, and Rep is the relative Reynolds number of the equivalent volume sphere. The drag model for nonspherical particles derived by Haider and Levenspiel63 is used.17where ϕ is the particle sphericity. Assuming the shape factor of ore particles has a constant value of 0.8, the b1, b2, b3, and b4 are 0.28, 0.54, 1.39, and 463.86, respectively. The Reynolds number of a particle is defined by18where μm is the viscosity of dense medium. Note that, when using the TFM, the discrete phase can only interact with the primary phase (here the water phase) rather than the mixture phase. The water density and viscosity in eq 18 must be replaced with mixture properties (density and viscosity) before conducting particle tracking. These mixture properties can be calculated via volume averaging, like the algebraic slip mixture model. Additional forces include the pressure gradient force, virtual mass force, Saffman lift force, and others. Due to the large pressure gradient along the radial direction, the pressure gradient force should be considered, while other forces are negligible for DMCs. To account for the effect of turbulence dispersion, the discrete random walk model is used to predict the instantaneous velocity along the particle path. The particle is assumed to interact with the fluid phase eddy over the smaller of the eddy lifetime and the eddy crossing time. When this time is reached, a new value for the instantaneous velocity is obtained. Simulation Conditions The geometric parameters and computational mesh of Φ250 mm DSM are shown in Figure 4. It assembled a standard 20° cone and an involute feed body. The inclination angle of the axis is 15°. The 3D geometry was created on the Ansys platform using SCDM. The type and quantity of grids determine the calculation speed and precision of numerical tests. In this study, the ICEM meshing software is employed to generate high-quality hexahedral grids using a multiple O-shaped meshing method. This can enhance the orthogonality between the fluid flow direction and mesh surface, thus improving the convergence speed. The grids at the bottom and wall of the vortex finder are locally refined. This ensures that the first layer of grids is in the turbulent logarithmic zone, meaning the Y+ value is between 30 and 300. A growth rate parameter of 1.2 was utilized, indicating that the maximum increase in the size of a neighboring element cannot exceed 20%.64 Grid independence tests were conducted with grid numbers of 100k, 193k, 410k, and 830k, respectively. The tangential and axial velocities at the Z = 0 mm line were used to determine the most appropriate grid scheme, as shown in Figure 5. When the grid size exceeds 400k, further reducing the grid size has a minimal impact on the calculation results. Therefore, the appropriate number of grids for the Φ250 mm DMC is determined to be 400k. (a) Geometric parameters and (b) computational mesh of Φ250 mm DMC. Grid independence tests on the Φ250 mm DMC: (a) tangential velocity and (b) axial velocity. Similar to the laboratory tests, the feed density is set to 1.94 by injecting a specific volume fraction of ferrosilicon particles. The continuous PSD needs to be divided into several size fractions. Taking the actual ferrosilicon sample as an example, it was divided into six different fractions as shown in Table 2. Each fraction was allocated to new phases with median particle sizes and respective volume fractions within each interval. Table 2Original Size Distribution of Ferrosilicon Used in Numerical SimulationSize fraction (μm)Mean size (μm)Yield (%)Cumulative yield (%)Volume fraction (%)–12.756.4015.1315.132.37–21.25 + 12.7517.0024.3239.453.81–29.75 + 21.2525.5022.6462.093.55–38.25 + 29.7534.0017.0179.102.67–46.75 + 38.2542.5011.0790.171.74–85 + 46.7565.909.83100.001.54 Original Size Distribution of Ferrosilicon Used in Numerical Simulation A “velocity inlet” boundary condition is used at the inlet, while “pressure outlet” conditions are applied at the underflow and overflow boundaries. The inlet velocity is set according to the flow rates specified in Table 1. The static pressure at both outlets is maintained at 1 atm, representing ambient atmospheric pressure. The backflow coefficient of air is set to 1 to facilitate the formation of an air core. The turbulence intensity for both outlets is generally set to 5%. A “no-slip” boundary condition is applied to all walls. These boundary conditions are extended to solid particle phases. Simulations are conducted using a transient solver, with pressure–velocity coupling achieved through the phase-coupled SIMPLE scheme. The QUICK scheme is employed to discretize the equations related to momentum, turbulence kinetic energy, turbulence dissipation rate, and Reynolds stress. The Modified-HIRC interface tracking scheme is used to discretize the volume fraction in the TFM. To enhance the robustness of the RSM-TFM-TDM, a preliminary water–air flow field with a fully developed air core is established before injecting ferrosilicon particles. A fixed time step of 5 × 10–4 s, determined based on the Courant number and calculation speed, is used for all simulations, yielding converged results within 20–30 s of physical time. Unless otherwise specified, the results presented here are averaged over the final 3 s of the simulation. Model Applicability The TFM model used to simulate the flow field and particle classification in a hydrocyclone has been extensively validated in the literature.59 Here, this paper simply compares the pressure-flow rate relationship, water split ratio to underflow (Rf), and product density obtained from simulations and physical experiments. The results are shown in Figure 6 and Table 3. Water–air system validation: (a) pressure drop and (b) Rf. Table 3Comparison of Measured and Predicted Separation Performance FeedOverflowUnderflowIndexExp.Exp.CFDRelative errorExp.CFDRelative errorFlow rate (t/h)123.7677.8581.755.01%45.9142.03–8.45%Medium concentration (wt %)56.7544.3046.033.91%77.8577.61–0.31%RD1.941.611.652.48%2.992.96–1.00%Ferrosilicon yield (%)100.0049.1153.559.04%50.8946.45–8.72% Comparison of Measured and Predicted Separation Performance As shown in the figure, the pressure-flow relationship is well predicted for the water–air system, but the water split ratio is generally lower. This discrepancy may be due to the extension pipe with an elbow present in the laboratory DMC’s overflow pipe, which was not considered in the simulation. Additionally, wall wear and roughness were not accounted for in the model. Trail tests have shown increasing wall roughness helps increase water split ratio, but it is not crucial for studying the effect of ferrosilicon fineness. Table 3 shows that, for different computational indices, the errors between numerical and experimental results are controlled within 10%. This indicates that the numerical model employed can accurately describe the segregation process of ferrosilicon in a DMC. The LPT model used has also been validated by numerous studies,35,44,65,66 and the exclusion of volume fraction effects and particle interactions is considered acceptable for the scope of this study. Results and Discussion The effects of ferrosilicon fineness are analyzed from three perspectives, including the flow field, the classification of different-sized ferrosilicon particles, and the partitioning of ore particles. Effects of Ferrosilicon Fineness on Flow Field Velocity Profile and Strain Rate Although the density field is the most critical characteristic of the flow field in a DMC, it is fundamentally the velocity profile that dictates the distribution of ferrosilicon particles. Therefore, this section examines the impact of ferrosilicon fineness on tangential and axial velocities and compares them with the flow field in the water–air system, as shown in Figure 7. Two characteristic lines at Z = 0 and Z = −200 mm were selected for comparison. The addition of ferrosilicon significantly reduces the tangential velocity, with variations observed at different axial positions. At the cylinder-cone junction, i.e. line 1, the velocity change outside the quasi-free vortex is minimal, while in other regions, tangential velocity decreases as ferrosilicon fineness reduces. This is because larger particles exhibit poor fluid followability and lower momentum exchange efficiency with the fluid, leading to an overall reduction in tangential velocity. The PSD of ferrosilicon has a negligible effect on axial velocity, with the location of the zero-velocity remaining almost unchanged. The ferrosilicon particle size distribution (PSD) has a similar impact on the velocity profile at line 2. However, throughout the entire radial scope, the tangential velocity decreases as the ferrosilicon gets coarser. This trend is further illustrated in Figure 8, where decreasing ferrosilicon fineness results in a drop in feed pressure, but the water split ratio of the underflow remains relatively stable, aligning with the changes in tangential and axial velocities shown in Figure 7. Effects of ferrosilicon fineness on tangential and axial velocities. Effects of ferrosilicon fineness on pressure drop and water split. For incompressible non-Newtonian fluids, the strain rate or shear rate is crucial for determining apparent viscosity. It has been well-recognized that the dense medium made by ferrosilicon suspension is a typical non-Newtonian fluid. However, when the shear rate exceeds 200 s–1, the medium approximates Newtonian behavior, meaning the apparent viscosity remains constant with the shear rate. Figure 9 presents the distribution of strain rate on the Y–Z plane in DMCs with different ferrosilicon PSDs. The results demonstrate that the strain rate is generally larger than 200 s–1 near the vortex finder, along the whole chamber wall, and in the lower cone section. However, in the majority of the separation space, it mainly lies in the range of 80–120 s–1, which can lead to shear-thinning depending on the local volume fraction and PSD of ferrosilicon in each cell. This phenomenon underscores the superiority of the KTGF oversimpler Newton viscosity models. As the ferrosilicon fineness increases, the shear rate inside the cone increases, corresponding to the variations of the tangential velocity shown in Figure 7. Effects of ferrosilicon fineness on strain rate. Turbulence Intensity The competition between fluid radial drag and centrifugal force influences the settling direction and distance of ferrosilicon particles. For fine particles, the fluctuations in drag caused by turbulence are the primary driving force for radial dispersion, significantly affecting the density differences within the separation space of the DMC. Additionally, eq 14 shows that the turbulent diffusion coefficient is directly proportional to both turbulent viscosity (μt) and the concentration gradient. Hence, Figures 10 and 11 illustrate the effects of ferrosilicon fineness on turbulent intensity (I) and eddy viscosity ratio (μt/μw), which are defined as follows.1920where is root-mean-square of turbulent fluctuating velocity, U is local average velocity, Cμ = 0.09, k is the turbulence kinetic energy, and ε is the turbulence dissipation rate. One thing to note is that the average velocity in Fluent software is not taken from the local velocity. Instead, it is a reference velocity that can be set manually. This is done to prevent a very large number and resulting floating-point overflow in case the local average velocity is zero. In this study, the inlet velocity (6.7 m/s) is set as the reference. Effects of ferrosilicon fineness on turbulent intensity. Effects of ferrosilicon fineness on eddy viscosity ratio. The results demonstrate that the addition of ferrosilicon significantly alters the turbulence characteristics of the flow field. Maximum values shift from the lower part of the vortex finder in the cylinder section to the lower cone section. This shift suggests that ferrosilicon particles in the lower cone are less likely to segregate by size and instead exhibit improved dispersion. As the fineness of ferrosilicon increases, both the turbulent intensity and turbulent viscosity of the entire flow field rise correspondingly. It is foreseeable that the degree of dispersion of fine particles will also increase. Density Field The motion behavior of ore particles depends on the density field characteristics of the DMC, particularly its density and uniformity. Theoretically, a uniform dense medium is beneficial for improving separation accuracy.67 However, different sizes of ferrosilicon particles introduce classification effects within a DMC, resulting in axial and radial density differences. Figure 12 illustrates the density field changes with variations in the ferrosilicon fineness. As the axial position decreases or the radial position increases, the density of the medium gradually rises, peaking at the cone apex. Although increasing ferrosilicon fineness leads to an overall decrease in the density of the separation space, the distribution pattern remains largely unchanged. Notably, as the proportion of fine particles increases, the uniformity of the density field improves, which may enhance ore separation efficiency. Effects of ferrosilicon fineness on density field. Figure 13 shows the effect of ferrosilicon particle size on the radial variation of the dense medium density. At the cylinder-cone interface (line 1), the medium-density distribution is largely unaffected by the ferrosilicon fineness. It increases gradually from the axis to the wall, maintaining a constant value of 2000 kg/m3 between radial positions of 40 mm and 110 mm. In contrast, in the cone section, although there are regions with relatively constant density, the overall density of the dense medium decreases with increasing ferrosilicon fineness. This phenomenon will be further analyzed and explained in the following sections. Radial distribution of medium density with different ferrosilicon fineness. Effects of Ferrosilicon Fineness on Medium Segregation Medium Partition Curve Subject to the centrifugal force, ferrosilicon particles will be classified, with the coarser particles reporting to the underflow and the finer particles to the overflow. Figure 14 shows the impact of ferrosilicon fineness on the partition curves of ferrosilicon. As ferrosilicon fineness increases, the partition curve shifts to the left, which means the recovery to underflow of each size of ferrosilicon particles increases. This shift can be attributed to two main factors. First, finer ferrosilicon particles result in larger pressure drops and tangential velocities, which increase the centrifugal force acting on the particles. Second, a reduction in the overall content of coarse particles with finer ferrosilicon diminishes the congestion effect near the spigot, facilitating easier entry of coarse particles into the underflow. However, this does not necessarily translate to a higher overall yield of ferrosilicon in the underflow. As it depends not only on the grade efficiency of each size but also on the ferrosilicon PSD, expressed as21where Gi and γf,i represent the grade efficiency of each size ferrosilicon and its yield in the feed. Effects of ferrosilicon fineness on dense medium partition curves. Figure 15 illustrates how the densities of underflow (ρu) and overflow (ρo) as well as their difference (ρu – ρo), vary with ferrosilicon fineness. The underflow RD is about 3.0, while the overflow RD is around 1.5. As Rf remains roughly unchanged, finer ferrosilicon leads to higher overflow density and lower underflow density. As a result, the RD differential decreases from 1.5 to 1.2, implying an improved uniformity in the density field. Effects of ferrosilicon fineness on density differential. Spatial Distribution of Medium To better understand the seemingly contradictory changes in the density field, Figure 16 contrasts the spatial distribution of ferrosilicon particles of different sizes. The concept of enrichment ratio is introduced here to facilitate comparison of their spatial distribution patterns, given their varying percentages in the feed. Different sizes of ferrosilicon particles exhibit distinct motion behaviors in the DMC. Finer particles are more easily carried by the fluid and evenly distributed, while coarser particles tend to concentrate in specific areas, affecting the uniformity of the density field. As the particle size increases to 17 μm, particles start to concentrate in the inner vortex and near the cyclone wall. With a further increase in particle size, there is a notable rise in particle concentration near the wall, while the concentration in the inner vortex decreases. This observed spatial distribution pattern, in conjunction with the variations in ferrosilicon PSDs, explains the changes in the density field shown in Figure 12. Effects of ferrosilicon fineness on the enrichment ratio of different particles. Figure 17 illustrates the radial variations in the volume fraction of ferrosilicon particles at different axial positions, considering various ferrosilicon fineness levels. The comparison reveals that the radial distribution patterns for each particle size are similar. Specifically, the volume fraction of fine particles remains relatively constant across the radial positions. In contrast, the volume fraction of coarser particles shows a general increasing trend from the axis to the wall, followed by a gradual decrease and then a rapid increase. Local maxima and minima occur at radial positions of r = 50 and 100 mm, respectively. When comparing lines 1 and 2, the distribution of fine particles remains unchanged, while other particle sizes (except 65.90 μm particles) display a monotonically increasing trend radially, although changes in the central region are minimal. Notably, along line 2, the peak volume fraction of coarse particles is observed slightly away from the wall, aligning with the findings reported in the literature.68 Effects of ferrosilicon fineness on the radial distribution law of different particles. However, it is important to note that even if a point in the separation space has the same density, the medium properties at that point can still be completely different. Figure 18 illustrates the impact of different feed ferrosilicon fineness on the volume fraction and composition of different ferrosilicon particles at the equivalent density point 1 (0, 100, 0). The results indicate that under different operating conditions, the PSD at this point is slightly coarser than that of the feed. Effects of ferrosilicon fineness on (a) volume fraction and (2) PSD of ferrosilicon particles at an equivalent density point. Effects of Ferrosilicon Fineness on Ore Separation Ore Partition Curve Low-grade metallic ores and gangue minerals are typically in the range of 2000 and 3000 kg/m3 in density and between 0.5 and 10 mm in size after crushing. LPT simulations were conducted where particles in this size range were injected at the inlet and tracked. Each simulation was repeated 3 times and 800 particles were injected per run. The particles (parcels) reported to the underflow were recorded and used to construct partition curves as a function of particle densities and particle sizes. Figure 19 shows the partition curves so generated using a TFM simulation with a feed RD of 1.94 and the original ferrosilicon PSD. Ore particle partition curve as a function of ore size. As shown in Figure 19, for ore particles of different sizes, the partition number shows a gradual increase with the increase in ore particle density. The cut density significantly decreases as the mineral particle size increases. This implies that a larger particle size is beneficial for improving the separation accuracy. The pivot phenomenon, where partition curves for different sizes of ore intersect at a pivot point, has been successfully modeled. The predicted pivot parameters slightly deviate from the feed density, which is a common phenomenon for DMC with high densities.69 Taking 3 mm particles as an example, Figure 20 shows the impact of ferrosilicon particle size composition on the ore partition curves. As the fineness of ferrosilicon increases, the cut density initially decreases slowly and then more rapidly, while the Ecart probable (EP) first decreases and then increases when ferrosilicon fineness exceeds 85%. By comparing Figures 12, 15, and 20, it is evident that the cut density is closely related to the distribution of the density field, particularly aligning with the trend of the density difference between the overflow and underflow. This indicates that a uniform density field formed by finer ferrosilicon promotes high-precision separation, whereas a high-density difference field formed by coarser ferrosilicon can achieve high-density separation even under lower feed densities. It should be noted that the effect of the latter gradually weakens and significantly impacts the separation efficiency of fine ore particles. Additionally, when the ferrosilicon is too fine, the apparent viscosity of the density medium increases, the flow resistance becomes greater, and EP will also increase inevitably. Therefore, optimizing the fineness of ferrosilicon is crucial for improving separation performance, and must be tailored to specific ore and separation conditions. Effects of ferrosilicon fineness on partition curves of ore particles. Particle Trajectory Figures 21 and 22 compare the movement trajectories and residence time distributions of 3 mm ore particles with a RD = 2.5 in different density fields. From Figure 21, it is noticed that with the increase in ferrosilicon fineness, the residence time of particles in the cyclone gradually increases, reaching a maximum under the F90 condition. Examining the trajectories reveals that near-density particles undergo multiple separations in the conical section, which are difficult to separate.44 When the fineness of ferrosilicon is further increased, particles quickly enter the underflow, with residence times approximately equal to the system’s response time, which is simply the quotient of volume and flow rate (1.46 s). Effects of ferrosilicon fineness on trajectories of ore particles. Effects of ferrosilicon fineness on the residence time of ore particles. From Figure 22, it can be observed that in both the overflow (light product) and underflow (heavy product), the residence time of particles initially increases and then decreases with the increase in ore density, reaching a maximum at near-density, approximately 2–3 times the system’s response time. As the fineness of ferrosilicon increases, the residence time of low-density particles in the DMC gradually increases, and the peak residence times of both light and heavy products shift toward lower densities. Conversely, the residence time of high-density particles gradually decreases and approaches the system’s response time. These results are mainly related to the properties of the density field, which corresponds well with the observations in Figure 20. Conclusions This study investigates the impact of ferrosilicon fineness on multiphase flow characteristics, ferrosilicon classification, and the overall separation performance of dense medium cyclones (DMCs) through laboratory experiments and numerical simulations based on a TFM-TDM-DPM model. Five different ferrosilicon samples were considered, with their −45 μm contents being 75%, 80%, 85%, 80%, and 95% respectively. The key findings are as follows. 1The TFM-TDM model demonstrates good predictive capabilities for ferrosilicon segregation, with relative errors less than 10% in pressure drop and RD. As ferrosilicon fineness increases, the tangential velocity gradually increases but is still smaller than in the absence of ferrosilicon. The axial velocity is barely affected by the existence of ferrosilicon. Under a constant flow rate, finer ferrosilicon particles cause a higher pressure drop (from 78 to 106 kPa), while having a negligible impact on the Rf, which remains around 15%. In most separation space, the strain rate lies in the range of 80–120 s–1, which results in the slurry retaining non-Newtonian fluid characteristics. The interaction between ferrosilicon particles and flow field causes changes in turbulence characteristics and significantly reduces the medium density in cone section as ferrosilicon fineness increases. The TFM-TDM model demonstrates good predictive capabilities for ferrosilicon segregation, with relative errors less than 10% in pressure drop and RD. As ferrosilicon fineness increases, the tangential velocity gradually increases but is still smaller than in the absence of ferrosilicon. The axial velocity is barely affected by the existence of ferrosilicon. Under a constant flow rate, finer ferrosilicon particles cause a higher pressure drop (from 78 to 106 kPa), while having a negligible impact on the Rf, which remains around 15%. In most separation space, the strain rate lies in the range of 80–120 s–1, which results in the slurry retaining non-Newtonian fluid characteristics. The interaction between ferrosilicon particles and flow field causes changes in turbulence characteristics and significantly reduces the medium density in cone section as ferrosilicon fineness increases. 2At a fixed feed RD of 1.94, the underflow RD is approximately 3.0, and the overflow RD is around 1.5. As ferrosilicon fineness increases, the partition curve shifts to the left, indicating an increase in the recovery of each size of ferrosilicon particles to the underflow. As the impact is caused by the change in ferrosilicon fineness, finer ferrosilicon leads to a higher overflow density and a lower underflow density, reducing the RD differential from 1.5 to 1.2. At a fixed feed RD of 1.94, the underflow RD is approximately 3.0, and the overflow RD is around 1.5. As ferrosilicon fineness increases, the partition curve shifts to the left, indicating an increase in the recovery of each size of ferrosilicon particles to the underflow. As the impact is caused by the change in ferrosilicon fineness, finer ferrosilicon leads to a higher overflow density and a lower underflow density, reducing the RD differential from 1.5 to 1.2. 3In the whole separation space, finer particles are evenly distributed, while coarser particles tend to concentrate near the wall and in lower cone areas with enrichment ratios higher than 4. The feed PSD of ferrosilicon directly determines its local PSD at each point in the separation space, with a negligible effect on the radial distribution law. In the whole separation space, finer particles are evenly distributed, while coarser particles tend to concentrate near the wall and in lower cone areas with enrichment ratios higher than 4. The feed PSD of ferrosilicon directly determines its local PSD at each point in the separation space, with a negligible effect on the radial distribution law. 4For ore particles of different sizes, the partition number increases with ore particle density, so the cut density decreases. The pivot phenomenon was observed at a point (RD = 2.51) higher than the feed RD (1.94). As ferrosilicon fineness increases, the cut density gradually decreases from 2.73 to 2.40, while EP first decreases and then increases when ferrosilicon fineness exceeds 85%. A uniform density field formed by finer ferrosilicon promotes high-precision separation, while a high-density difference field formed by coarser ferrosilicon can achieve high-density separation even under lower feed densities, although its effect on fine ore particles gradually weakens. For ore particles of different sizes, the partition number increases with ore particle density, so the cut density decreases. The pivot phenomenon was observed at a point (RD = 2.51) higher than the feed RD (1.94). As ferrosilicon fineness increases, the cut density gradually decreases from 2.73 to 2.40, while EP first decreases and then increases when ferrosilicon fineness exceeds 85%. A uniform density field formed by finer ferrosilicon promotes high-precision separation, while a high-density difference field formed by coarser ferrosilicon can achieve high-density separation even under lower feed densities, although its effect on fine ore particles gradually weakens. 5The residence time of near-density particles is about 2–3 times the system response time and increases with the increase in ferrosilicon fineness. In different density fields, the movement trajectories and residence time distributions of ore particles vary. As ferrosilicon fineness increases, the residence time of low-density particles in the DMC gradually increases, and the peak residence times of both light and heavy products shift toward lower densities. Conversely, the residence time of high-density particles gradually decreases and approaches the system’s response time. The residence time of near-density particles is about 2–3 times the system response time and increases with the increase in ferrosilicon fineness. In different density fields, the movement trajectories and residence time distributions of ore particles vary. As ferrosilicon fineness increases, the residence time of low-density particles in the DMC gradually increases, and the peak residence times of both light and heavy products shift toward lower densities. Conversely, the residence time of high-density particles gradually decreases and approaches the system’s response time.