Centrifugal Pump (Centrifugal Pump) — CAE Glossary
What is a Centrifugal Pump?
So a centrifugal pump is basically a machine that spins an impeller to push water, right? What does CFD analyze?
Essentially, the impeller (runner) rotates to impart kinetic energy to the fluid, which is then converted to pressure energy in the volute (spiral chamber) or diffuser. In CFD, the primary objective is to predict the head-flow relationship — the H-Q curve (head-flow characteristics) — i.e., how much head (elevation gain) is produced at a given flow rate.
Pump CFD sounds complex. How is it different from typical external flow analysis?
The biggest difference is the presence of rotating components. The impeller rotates at hundreds to thousands of rpm, so we must handle both the fixed casing and the rotating blades simultaneously. To address this, we use MRF (Multiple Reference Frame) or sliding mesh methods. Also, the turbulence is intense, making turbulence model selection critical.
Definition and Basic Principles
How is the head in a centrifugal pump physically determined? If you increase the rotation speed, can you get unlimited head?
Theoretically, it's determined by Euler's equation for turbomachinery. The change in tangential velocity component between impeller inlet and outlet generates the head:
Here, $u$ is impeller peripheral speed, $c_\theta$ is the tangential component of absolute fluid velocity, and subscripts 1 and 2 denote inlet and outlet. Higher rotation speed increases $u$, thus increasing head, but losses and cavitation risk also increase. In practice, the similarity law approximates that head increases proportionally to the square of rotational speed.
So $u_2 c_{\theta 2}$ is fundamentally determined by the blade outlet angle, right?
Exactly. A larger outlet blade angle $\beta_2$ (backward swept) reduces $c_{\theta 2}$, lowering head but providing a steeper H-Q curve slope and better operating stability. Conversely, forward swept blades ($\beta_2$ smaller) produce higher head but are prone to surge. In practice, backward-swept blades ($\beta_2 = 20^\circ \sim 40^\circ$) are standard. CFD is invaluable for optimizing this blade angle.
H-Q Characteristic Curves and CFD
How do you generate H-Q curves using CFD? Do you run multiple calculations at different flow rates?
Exactly. Typically, you set 5–8 flow conditions (ranging from 0% at shutoff to ~150% overload) and run steady MRF analyses for each. You specify inlet mass flow, then compute head from the total pressure difference:
For example, a typical 200 mm diameter pump at 1450 rpm might produce a flow of $Q = 0.03\,\mathrm{m^3/s}$ and head $H = 25\,\mathrm{m}$ at the design point. Matching CFD H-Q curves to experimental data within ±3% error at the design point is considered good agreement in the industry.
Can you also predict efficiency? Pump efficiency is pretty important.
Absolutely. Hydraulic power is $P_w = \rho g Q H$, and dividing by shaft power $P_s = T\omega$ (where $T$ is torque and $\omega$ is angular velocity) gives hydraulic efficiency:
However, CFD efficiency typically runs 2–5 percentage points higher than experimental values because CFD cannot fully capture disk friction (impeller back-face losses) and leakage flows. So treat CFD efficiency as a trend indicator rather than absolute truth. Use it for parametric comparisons, not direct prediction.
Impeller CFD Analysis: Rotational Modeling
How does MRF work if the mesh doesn't move? How can it still capture rotation?
In MRF, the domain surrounding the impeller is defined as a rotating reference frame, and Coriolis and centrifugal terms are added to the Navier-Stokes equation:
Here, $\mathbf{v}_r$ is relative velocity, the term $-2\boldsymbol{\omega} \times \mathbf{v}_r$ represents Coriolis force, and the final term is centrifugal force. The mesh stays fixed; we're solving the flow in a rotating frame. This allows steady-state analysis, making it computationally cheap — ideal for exploring many design variants in early stages.
When do you actually need sliding mesh instead? If MRF is enough, why use it at all?
MRF is a steady approximation, so it cannot resolve impeller-volute interaction (blade passing frequency pressure pulsations) or unsteady phenomena near the tongue (cutwater). Sliding mesh becomes mandatory for:
- Pressure pulsations and vibration — When you need to evaluate pressure variations at blade passing frequency (BPF = blade count × rotational speed)
- Noise analysis — Time-dependent pressure fluctuations are essential for aeroacoustic sources
- Off-design operation — Rotating stall can occur at low flows, requiring transient analysis
- Transient response — Water hammer or surge during startup/shutdown
Sliding mesh costs 10–100× more, requiring at least 360 timesteps per impeller revolution (ideally 1-degree increments), and you must simulate 3–5 complete revolutions to reach periodic steady state. You typically use the last revolution's data.
How does data get transferred across the MRF/sliding mesh interface?
Interpolation occurs at the interface. For MRF, this is called the "frozen rotor" method — rotating and stationary sides exchange velocity and pressure across the boundary with coordinate transformation. For sliding mesh, mesh offset occurs every timestep, so non-conformal mesh interpolation (GGI: General Grid Interface) is used. In ANSYS Fluent, a "mixing plane" averages circumferentially for stage interfaces. The key is ensuring equal mesh resolution on both sides of the interface; mismatched densities can introduce interpolation errors causing several percent head error.
Cavitation Prediction
Cavitation is a recurring problem in pumps — bubbles forming and collapsing on impeller suction sides, causing erosion. Can CFD predict it?
Yes. Cavitation analysis uses multiphase flow models to compute coexistence of liquid and vapor. A common approach is the homogeneous mixture model with bubble growth/collapse rates from the Rayleigh-Plesset equation. Popular implementations include Schnerr-Sauer and Zwart-Gerber-Belamri models:
Here, $R_B$ is bubble radius, $p_v$ is saturation vapor pressure, and $p$ is local pressure. When $p < p_v$, bubbles grow; when pressure recovers, they collapse.
I hear about NPSH all the time. Can CFD predict that too?
Absolutely. Determining NPSHr (required NPSH) via CFD is very practical. You systematically reduce inlet pressure until head drops 3%, then define that NPSH margin as NPSHr:
For example, at 25°C water, $p_v \approx 3.17\,\mathrm{kPa}$. Reduce inlet pressure from 130 kPa to 120 kPa, 110 kPa, etc., and track head degradation. Critical point: cavitation analysis requires fine meshes — miss the bubble inception location otherwise. Mesh at least 20 layers across blade height, especially near leading edges.
Does cavitation mess with CFD convergence?
Severely. Rapid phase interface motion prevents steady convergence; you're often forced into transient analysis (sliding mesh + cavitation model). Vapor volume fraction oscillates sharply, causing pressure swings. Remedies include finer timesteps (≤0.5° impeller rotation per step) and reduced underrelaxation factors. OpenFOAM's interPhaseChangeFoam solver handles this. The takeaway: cavitation is computationally demanding and requires careful numerics.
Practical Considerations
How do you mesh a centrifugal pump? Impeller geometry looks complicated.
Pump meshing is indeed challenging. Here are best practices:
- Structured mesh in blade passages — Tools like TurboGrid or ANSYS BladeModeler create structured (H-O-H topology) meshes between blades, superior to unstructured for $y^+$ control
- Wall meshing — For SST k-ω, aim for $y^+ \approx 1$ (wall-resolved). Wall functions allow $30 < y^+ < 300$ but sacrifice separation accuracy
- Impeller-volute interface mesh — Match mesh density across interface. A 360-division circumferential target is reasonable
- Total mesh size — Single-stage pumps typically need 2–5 million cells. Allocate ≥300k cells per impeller blade
Which turbulence model do you recommend? k-ε or SST k-ω?
SST k-ω is the de facto standard for pump CFD. It excels at predicting boundary-layer separation under adverse pressure gradients on blade surfaces — critical for low-flow (partial-load) conditions. Standard k-ε is acceptable for near-design-point H-Q curves, but SST wins for off-design. For pulsation and noise, consider DES or SAS (Scale-Adaptive Simulation).
Any final critical warnings for pump CFD?
Three absolute must-dos:
- Inlet and outlet extensions — Avoid boundary conditions right at impeller. Include ≥5 diameters upstream of suction and ≥10 diameters downstream of discharge; otherwise, unphysical flow develops
- Tip clearance modeling — Leakage flows through impeller-casing gaps significantly reduce head. Neglecting them overpredicts head by 5–10%. Resolve clearances with ≥5 layers of mesh
- Convergence judgment — Never trust residuals alone. Monitor probe points (inlet pressure, outlet pressure, torque). If residuals decay but these quantities oscillate, convergence is incomplete
Following these three points alone dramatically improves pump CFD accuracy.
Related Terms
- MRF (Multiple Reference Frame) — Steady-state approximation for rotating machinery using rotating coordinate system
- Sliding Mesh — Transient method where mesh in rotating domains actually rotates, capturing dynamic interactions
- Cavitation — Bubble formation when local pressure drops below saturation vapor pressure
- H-Q Curve — Relationship between pump head (H) and flow rate (Q), defining performance characteristics
- NPSH — Net Positive Suction Head. Margin against cavitation on pump suction side
- BPF (Blade Passing Frequency) — Pressure pulsation frequency = blade count × rotational speed
- Specific speed $n_s$ — Dimensionless parameter for pump classification: $n_s = n\sqrt{Q}/H^{3/4}$
Precise understanding of CAE terminology is foundational for team communication. — Project NovaSolver supports learning for practicing engineers.
Share Your Centrifugal Pump CFD Challenges
Project NovaSolver is committed to solving real CAE bottlenecks — setup complexity, computational cost, result interpretation. Your hands-on experience drives better tools.
Contact (Coming Soon)Related Topics
helpful
detail
error