# Best Practice Advice AC3-03

**Cyclone separator**

**Application Challenge 3-03** © copyright ERCOFTAC 2004

**Best Practice Advice**

*The Cyclone separator is perhaps the most widely used separation device to be found in industry. There are no moving parts in the device itself, and it can be easily manufactured from a range of materials. Combined with moderate pressure drop and a range of throughputs and efficiencies, these advantages have made the cyclone the most attractive solution to separation in both gas-solid and liquid-solid systems. The best practice advice presented in this document is applicable to cyclone flows modelled using finite volume based CFD solvers*.

**Best Practice Advice for the AC**

**Key Fluid Physics**

Cyclones are used to separate or classify secondary phases. The cyclone works by inducing spiral rotation in the primary phase and therefore imposes an enhanced radial acceleration on a particulate suspension. In conventional cylindrical cyclone devices there are two outlets both on the axis of symmetry. The underflow is situated at the apex of the cone and the overflow is an inner tube, which descends from the top of the cyclone. The density of the suspended particulate phase is normally greater than the primary phase. Due to the imposed swirl, larger particles migrate radially to the outer wall and then spiral down to the underflow. Smaller particles migrate more slowly and are captured in an upward spiral in the center of the cyclone and leave through the top via the vortex finder.

1.2. The prediction of flow behavior inside cyclones is a challenging one. Experimental work has shown that the tangential velocity increases sharply with radius in the central core region under the vortex finder and that thereafter it decreases with radius. This typical radial transition between free and forced vortex needs to be captured, as it is fundamental for predicting the secondary phase separation and the pressure field. The correct swirl profile is also implicitly coupled to the flow reversal which creates the two streams that transport the classified product from the cyclone.

1.3. The constrained free vortex flow in a cyclone means that turbulent fluctuations are restrained in the tangential and axial direction but less so in the radial direction. To accommodate this anisotropy of the turbulence it is necessary to chose a suitable turbulence model.

1.4. Due to the low pressure along the cyclone axis, back-flow can occur, in liquid cyclones open to atmosphere this can result in a gas core. It is critical to capture the low-pressure back flow into the cyclone correctly as it throttles the system and controls the flow split between the under and overflow.

1.5. Under certain operating conditions cyclones may have a number of unstable flow features particularly in the low-pressure central core. These unstable features can have an influence on the the operational performance of the cyclone, resulting in pressure fluctuations and particulate short circuiting.

1.6. By virtue of its nature the suspended second phase in a cyclone will be concentrated within the device. As the cone narrows the concentration of material will build up on the wall of the cyclone. For high particle loadings towards the underflow build up of material may effectively change the geometry and therefore constrict the apex which will in turn influence the flow split.

1.7. The UFR4-06 (swirl diffuser) is associated weakly with the cyclone application challenge. The swirling diffuser can exibit center line instabilities and flow reversals due to the high radial pressure gradients. In contrast to the cyclone the swirl diffuser decelerates the swirling flow. The tangential velocity profile observed in the UFR data is a low shear rotational vortex of moderate swirl number and it therefore has much more weakly anisotropic turbulence.

**Application Uncertainties**

Although simple, Geometric descriptions of cyclones are notoriously incomplete:

2.1.1. Cyclones are normally designed using empirical models and are then tweaked on plant to achieve the performance requirements, by changing spigot diameters or adjusting the length of the vortex finder.

2.1.2. In operation cyclones are prone to wear which may change the underflow and the inlet shape of the cyclone you are modelling.

2.1.3. Ramped helical inlets are difficult to describe in two-dimensional drawings. Always obtain 3 dimensional CAD from the manufacture if available. Ensure you know what you are modelling!

2.2. Single point static pressure measurements made at the outflows of an operating cyclone may not represent realistic boundary conditions for a CFD model. The pressure at the outlet of a cyclone has a strong radial pressure distribution due to the swirl.

2.3. In normal operation cyclones are typically down stream of another process. Consequently the inlet flow to the cyclone may not be constant.

2.4. Upstream piping introducing the particle laden flow to the cyclone is likely to have bends in it which can result in an uneven velocity profile to the cyclone and possibly cause pre-separation of the particulates prior to the inlet.

2.5. In many cyclones the flow field is inherently unstable, the resulting transient instabilities can affect the overall separation performance of the cyclone. It may as a result be difficult to represent the flow field as an averaged condition.

2.6. Particulate related uncertainties

2.6.1. The location and distribution of particulates in the inlet of the model directly impacts the separation efficiency. It is therefore important to understand what is happening in the real system.

2.6.2. Measured particle sizes and distributions are susceptible to error. Care should be taken to consider these errors when predicting separation performance.

**Computational Domain and Boundary Conditions**

The flow in a conventional cyclone with tangential inlet is a 3 dimensional problem. 2D idealisation can not be made without a priori knowledge of swirl and axial velocity distribution in the top of the cyclone.

3.2. Any pipe bends in the vicinity of the inlet to the cyclone should be included in the computational model.

3.3. If only a single outflow is being modelled which assumes the underflow is closed or discharging into a pot, it is sufficient to represent it by a single mass flow outlet. The mass flow outlet boundary condition constrains neither the velocity nor pressure. The outlet boundary condition will also resolve the radial pressure distribution correctly and therefore allow back flow into the cyclone.

3.4. If both the underflow and overflow are modelled the mass flow split must not be prescribed, this is over constraining the analysis. To calculate the flow split between the underflow and overflow apply static pressure boundaries at the under and overflow. When using a pressure boundary condition for a swirling flow ensure that a radial pressure distribution is applied as the fluid exiting the cyclone will normally be experiencing a high degree of swirl. Applying a constant pressure will artificially suppress swirl and influence the flow inside the cyclone. In the swirl diffuser underlying flow regime a constant pressure outlet has been applied the results consequently showed insufficient reduction in centerline velocity.

3.5. For an incompressible fluid at the inlet to the cyclone either a fixed velocity, or total inlet pressure, may be used.

3.6. The results of the application challenge were not assessed for sensitivity to the turbulent intensity at the inlet. In the application challenge, wall functions were used to represent the viscosity affected inner region near the walls. The results indicate that once the fluid enters the cyclone it enters a region in which the turbulence is principally determined by the vortex structure in the bulk flow, and is relatively insensitive to the inlet boundary conditions or the viscosity affected inner region near the wall.

**Discretisation and Grid Resolution**

Discretisation schemes play an important part in cyclone simulations. When the flow does not align with the grid low order schemes can cause the distribution of transported properties to become smeared. The best type of grid to model swirling flows is a hexahedral type with the mesh elements aligned with the circumference of the cyclone. In the application challenge to avoid skew cells at the axis a square core has been used, this type of meshing strategy shown in Figure 1 has been termed a butterfly mesh.

4.2. Use a high order discretisation scheme such as Quadratic upwind differencing scheme (QUICK). In addition, due to the rotating flow there are high radial pressure gradients which need to be accommodated. It is therefore advantageous to use PRESTO (PRESsure STaggering Option) for the continuity equations. PRESTO is a pressure interpolation scheme that is similar to staggered grid schemes used on structured meshes and uses the discrete continuity balance for a staggered control volume about the face to compute the face pressures.

4.3. In the cyclone application challenge a relatively course mesh of 41000 cells was found to be sufficient. However the diameter to length ratio of this cyclone is only 1:4. Cyclones can have diameter to length ratios of 1:20, therefore the most important factor in deciding on the exact mesh resolution is ensuring that the cell aspect ratio is not excessive, ideally the cell aspect ratio should not exceed 1:5. High cell aspect ratios in swirling flows cause convergence difficulties.

4.4. Boundary layer resolution is not critical as the turbulence is generated in the main flow. At the underflow and overflow a central back flow of fluid can occur. The backflow may occupy a significant proportion of the cross-sectional area of the outlet boundary as shown in Figure 2. It is therefore important to have sufficient mesh to resolve the narrow annular gap next to the wall through which the fluid leaving the cyclone will flow.

4.5. Fine mesh along the axis of the cyclone will begin to resolve any unstable flow features associated with the low-pressure central core, these have been shown to exist both experimentally and by CFD analysis. Even in steady state simulations a transient pattern may occur, this can be detected in a cycling in the residuals when this occurs it is very difficult to obtain a converged solution. If residual cycling happens it is necessary to change to a transient (URANS) calculation using a time step less than 100th of the residence time of the cyclone. In this situation it has been found that running the simulation in a transient manner improves the definition of the central low-pressure core.

**Physical Modelling**

An anisotropic turbulence model is required to correctly capture the free to forced vortex transition that occurs in cyclonic flows. Standard k-ε models and other models based on assumptions of isotropic turbulence are not suitable as they tend to over predict the turbulent viscosity and exaggerate the forced vortex. For the application challenge a Reynolds Stress Turbulence model has been used successfully to calculate tangential and axial velocity profiles. LES also gave good results but has a high computational overhead.

5.2. The underlying flow regime swirl diffuser exhibits only moderate swirl and conclusions about the suitability of turbulence models in the URF are not applicable to the cyclone.

**Recommendations for Future Work**

The best practice advice provided in this document is focused on the correct prediction of single-phase cyclonic flows. The flow problem is challenging but can be modelled following the advice outlined in this document. The next cyclone challenge to be evaluated is the separation and the prediction of secondary phases in the cyclone. Potential multiphase best practices that could be developed include:

6.1. Evaluate the suitability of different multiphase modelling and turbulent coupling for cyclone separation.

6.2. Evaluate the appropriateness of boundary conditions for modelling multiphase systems with reverse swirling flows.

6.3. Grid sensitivity and accuracy of discrete phase tracking schemes.

6.4. Find and consider experimental data suitable for modelling and testing the appropriateness of the techniques for different classes of cyclone operation.

Figure 1. Butterfly mesh used in cyclone simulation

Figure 2. Back flow at the outlet of a cyclone is shown by the blue region

© copyright ERCOFTAC 2004

Contributors: Chris Carey - Fluent Europe Ltd

Site Design and Implementation: Atkins and UniS