Filter Flow Analysis

Category: Fluid Analysis (CFD) | Integrated 2026-04-06
CAE visualization for filter cfd theory - technical simulation diagram
Filter Flow Analysis

Filter Flow: Theoretical Foundations

Overview

๐Ÿง‘โ€๐ŸŽ“

Teacher! In what situations is CFD analysis of flow inside filters used?


๐ŸŽ“

It's used for predicting pressure loss and evaluating particle collection efficiency for flows passing through porous media, such as air filters, oil filters, DPF (Diesel Particulate Filter) for exhaust gas catalysts, water purification filters, etc.


Governing Equations

๐Ÿง‘โ€๐ŸŽ“

Filters are porous, right? Do you use the Navier-Stokes equations as-is?


๐ŸŽ“

There are two approaches: directly resolving the filter pores (Pore-Scale Simulation) and using a volume-averaged porous media model. In practice, the latter is used almost exclusively.


๐ŸŽ“

In the porous media model, flow resistance is expressed by the Darcy-Forchheimer equation.


$$ -\frac{\partial p}{\partial x_i} = \frac{\mu}{\alpha} v_i + C_2 \frac{\rho}{2} |v| v_i $$

๐Ÿง‘โ€๐ŸŽ“

The first term is Darcy's viscous resistance, and the second term is Forchheimer's inertial resistance, right?


๐ŸŽ“

Exactly. $\alpha$ [mยฒ] is the permeability, and $C_2$ [1/m] is the inertial resistance coefficient. For low-speed flow (Re < 1), the Darcy term dominates, while for high-speed flow, the inertial term becomes important.


๐ŸŽ“

For packed beds, the Ergun equation is used.


$$ \frac{\Delta p}{L} = \frac{150 \mu (1-\phi)^2}{\phi^3 d_p^2} V + \frac{1.75 \rho (1-\phi)}{\phi^3 d_p} V^2 $$

๐Ÿง‘โ€๐ŸŽ“

$\phi$ is the porosity and $d_p$ is the packed particle diameter, right? What is the relationship between the Ergun equation and Darcy-Forchheimer?


๐ŸŽ“

Porous parameters can be back-calculated from the Ergun equation.


$$ \alpha = \frac{\phi^3 d_p^2}{150(1-\phi)^2}, \quad C_2 = \frac{3.5(1-\phi)}{\phi^3 d_p} $$

Modeling Particle Collection

๐Ÿง‘โ€๐ŸŽ“

How is the particle collection efficiency of a filter modeled?


๐ŸŽ“

The basic theory is Single Fiber Theory. The efficiencies of various collection mechanisms for a single fiber are summed.


Collection MechanismDominant Particle SizeEfficiency Formula (Outline)
Interception> 0.5 um$E_R = \frac{R^2}{(1+R) Ku}$
Impaction> 1 um$E_I \propto Stk^2$
Brownian Diffusion< 0.3 um$E_D \propto Pe^{-2/3}$
Gravitational Settling> 5 um$E_G = G/(Ku \cdot Re_f)$
ElectrostaticAll sizesDepends on charge amount
๐Ÿง‘โ€๐ŸŽ“

The MPPS (Most Penetrating Particle Size) is around 0.1~0.3 um because of the gap between diffusion and inertia, right?


๐ŸŽ“

Exactly. That's why the collection efficiency curve for HEPA filters is V-shaped. In CFD, methods using DPM to track particle trajectories and determine arrival at the fiber layer are used.


Practical Considerations

๐Ÿง‘โ€๐ŸŽ“

What should be especially careful about in filter CFD?


๐ŸŽ“
  • Porous parameters ($\alpha$, $C_2$) should be fitted from the filter manufacturer's pressure loss data.
  • Since actual filter pressure loss characteristics are nonlinear, it is desirable to obtain data at multiple face velocities.
  • Pressure loss increase due to filter clogging (dust accumulation) is a time-varying problem, so introducing temporal changes via UDFs, etc., may be necessary.

Coffee Break Casual Talk

The "Three Mechanisms" of Filter Collection โ€” Caught Whether Too Big or Too Small

An interesting point in filter flow theory is how the particle collection mechanism switches depending on particle size. Large particles (several ฮผm and above) are caught by "inertial impaction" โ€” deviating from streamlines and colliding with fibers. Mid-sized particles (0.5~2ฮผm) are caught by "interception" โ€” contacting fibers while following streamlines. Ultrafine particles (below 0.1ฮผm) are caught by "diffusion collection" โ€” moving zigzag due to Brownian motion and accidentally adhering to fibers. This means the hardest particles to collect are those in the transitional zone of 0.1~0.3ฮผm, which have "weak inertia and weak Brownian motion." The Single Fiber Theory, which combines these three mechanisms, forms the theoretical basis for filter design.

Computational Methods for Filter Flow

Details of Numerical Methods

๐Ÿง‘โ€๐ŸŽ“

Please teach me the specific implementation methods for filter CFD.


๐ŸŽ“

There are two implementation methods for the porous media model: Volumetric Porous Zone (distributing resistance over the entire volume) and Porous Jump (setting a pressure jump on a thin surface).


Porous Zone vs. Porous Jump

MethodApplicationAdvantagesDisadvantages
Porous ZoneThick filters (packed beds, catalyst layers)Obtains velocity distribution inside the filterRequires mesh
Porous JumpThin filters (pleated filters)No mesh needed, computationally lightNo internal flow information
๐Ÿง‘โ€๐ŸŽ“

So thin filters like HEPA filters can be handled with Porous Jump, right?


๐ŸŽ“

Yes. If the filter thickness is on the order of a few cm and detailed flow in the thickness direction is unnecessary, Porous Jump is efficient. For cases like DPF where thickness is 100mm or more, use Porous Zone.


Procedure for Setting Up Porous Media in Fluent

๐Ÿง‘โ€๐ŸŽ“

Please teach me the specific steps to set up a Porous Zone in Fluent.


๐ŸŽ“

1. Select the filter region in Cell Zone and enable Porous Zone.

2. Set Direction-1 Vector to the filter normal direction.

3. Input Viscous Resistance (1/ฮฑ [1/mยฒ]) and Inertial Resistance (Cโ‚‚ [1/m]).

4. Input Porosity (default 1.0 means the entire space is flow path; change to the physical porosity).


๐ŸŽ“

Here's an example calculation of resistance values. For a HEPA filter with face velocity 0.5 m/s, pressure loss 250 Pa, and thickness 65 mm.


$$ \Delta p = \left(\frac{\mu}{\alpha} V + C_2 \frac{\rho}{2} V^2\right) L $$

๐ŸŽ“

Assuming the Darcy term dominates in the low-speed region.


$$ \frac{1}{\alpha} = \frac{\Delta p}{\mu V L} = \frac{250}{1.8 \times 10^{-5} \times 0.5 \times 0.065} \approx 4.3 \times 10^{8} \text{ [1/mยฒ]} $$

๐Ÿง‘โ€๐ŸŽ“

So Viscous Resistance is about 4.3e8. Is Inertial Resistance determined from data at another face velocity?


๐ŸŽ“

Yes. If you have pressure loss data at two or more face velocities, you can separate the Darcy and Forchheimer terms using simultaneous equations. Use data like pressure loss 140 Pa at face velocity 0.3 m/s and 380 Pa at 0.7 m/s.


Coupling with DPM

๐Ÿง‘โ€๐ŸŽ“

How is particle collection in filters modeled with DPM?


๐ŸŽ“

There is a method where the wall condition for DPM particles inside a Porous Zone is set to Trap, treating collisions with the filter fiber surface as collection. However, this is a simplified model; for more accuracy, the collection probability is implemented as a function of particle size via a UDF.


๐ŸŽ“

The overall collection efficiency of a filter is expressed by the following equation.


$$ E_{total} = 1 - \exp\left(-\frac{4 \alpha_f E_f L}{\pi (1-\alpha_f) d_f}\right) $$

๐Ÿง‘โ€๐ŸŽ“

$\alpha_f$ is the filter packing fraction, $E_f$ is the single fiber efficiency, and $d_f$ is the fiber diameter, right?


๐ŸŽ“

Yes. In CFD, the local single fiber efficiency is calculated from the local face velocity in each cell and applied as the collection probability for DPM particles.


Mesh Considerations

๐Ÿง‘โ€๐ŸŽ“

What about the mesh for the filter region?


๐ŸŽ“
  • Porous Zone: Ensure at least 5~10 cell layers in the filter thickness direction.
  • Provide sufficient space upstream and downstream of the filter (each 3~5 times the filter thickness).
  • For pleated filters, simplify the pleat shape or calculate using a periodic model of one pleat.

๐Ÿง‘โ€๐ŸŽ“

Pleated filters have complex pleat shapes, but do you model them all?


๐ŸŽ“

Modeling all pleats in 3D is often not realistic. A practical approach is to determine representative characteristics using a periodic boundary model of 1~3 pleats and treat the entire model as an equivalent porous surface.


Coffee Break Casual Talk

Inside the "Porous Media Model" for Modeling Filters in CFD

The "Porous Media Model" often used in filter flow analysis is a method that ignores the detailed fiber structure inside the filter layer and treats it as a pressure loss source distributed over the entire volume. Both Fluent and OpenFOAM have implementations based on Darcy's law, where pressure loss coefficients (viscous resistance coefficient $\alpha$ and inertial resistance coefficient $C_2$) are fitted from measured ฮ”P-velocity curves. A point of caution is that it ignores "the actual non-uniformity in the filter thickness direction" โ€” when clogging progresses locally in actual use, the model's accuracy decreases. Therefore, for analysis including temporal degradation, it's necessary to couple it with particle tracking (DPM).

Related Simulators

Experience the theory firsthand with the interactive simulator for this field

All Simulators

Related fields

Thermal AnalysisV&V ยท Quality AssuranceStructural Analysis
Rate this article
Thank you for your feedback!
Helpful
More details
Report error
Helpful
0
More details
0
Report error
0
Written by NovaSolver Contributors
Anonymous Engineers & AI โ€” Sitemap
About the Authors