Propeller CFD — Zixian Zhang

1.0

Objective

A candidate thruster for a drone built to both fly and run underwater needed thrust, torque and cavitation estimates before anyone drew the vehicle. The supplied propeller would not mesh. I traced why, rebuilt the blade from published experimental cross-sections, and ran a Fluent sweep from 300 to 2000 rpm that sized the drive and bounded where the propeller can run without cavitating.

2.0

Hardware

InflowPropeller: three blades, D 390 mm, pitch/diameter about 0.3, drawn from the rebuilt mesh-ready geometryConsole dials, left to right: thrust −0.2 to 2 kN, torque 0 to 50 N·m, open-water efficiency 0 to 0.6Bubbles: suction side below the 2.3 kPa vapour pressureThrust, torque and efficiency follow from the advance ratio J = V/(nD): at a fixed inflow a faster shaft lowers J and raises the thrust and torque coefficients, and both inflow speeds fall on one curve. Every value is a CFD result (Ansys Fluent, steady MRF, k-ω SST, seawater at 10 m depth, 1.0 and 1.5 m/s inflow), interpolated between the computed speeds of 300 to 2,000 rpm. Bubbles appear only where the computed minimum absolute pressure falls below vapour pressure, from about 1,110 rpm, and cover the computed share of the blade area. The rig is illustrative: the CFD modelled the propeller alone in open water. A strobe makes the blades appear to turn slowly.

3.0

Work performed

3.1Traced why every meshing attempt failed. The two-blade original carried sliver edges as short as 0.0017 mm and a zero-thickness trailing edge; on the three-blade model, the rotating and stationary fluid bodies each carried a copy of their shared interface (31,701 cross-intersections, about 1 mm clearance).
3.2Reconstructed the blade from published experimental cross-sections (the T-Motor 15×5 carbon blade measured by Russell & Sekula at NASA): 46 radial sections rebuilt as cubic B-spline surfaces, largest deviation 0.64 mm and most stations within 0.04 mm. Then delivered a mesh-ready three-blade solid with shared topology and a resized rotating zone (R = 215 mm, ±30 mm).
3.3Ran the Fluent sweep at 1.0 and 1.5 m/s. Both inflows collapse onto one K_T(J), K_Q(J) curve: zero thrust at J ≈ 0.48, peak efficiency about 0.5 near J ≈ 0.3, and 1,898 N for 9.2 kW at 2000 rpm.
3.4Mapped cavitation risk: none up to 800 rpm, small patches from 1200 rpm, and 18% of the suction side below vapour pressure at 2000 rpm (σ = 0.23 on tip speed). Recommended staying under about 900 rpm at 10 m depth.
3.5Matched the propeller to the candidate motor: direct drive reaches only about 19 N at 1 m/s, while an 8–10:1 reduction gives 220–300 N inside the cavitation-free range. These are the pre-CAD load estimates for the vehicle's propulsion sizing.

4.0

Specification

Propeller	3 blades · D = 390 mm · P/D ≈ 0.3
Blade source	Published sections of the T-Motor 15×5 carbon blade (Russell & Sekula, NASA Ames/Langley, 2017)
Solver	Ansys Fluent 2026 R1 · steady MRF (frozen rotor) · k-ω SST · pressure-based coupled
Mesh	950,526 poly-hexcore cells · 1.5 mm blade faces · 4 prism layers
Conditions	Seawater at 10 m depth (201.9 kPa abs) · 1.0 and 1.5 m/s inflow · 300–2000 rpm
Cavitation	Single-phase criterion: area below 2.3 kPa vapour pressure · σ on tip speed
Outputs	Thrust · torque · K_T, K_Q, η vs J · blade pressure · slipstream · cavitation risk
Use	Motor matching (T-Motor MN505-S KV380 model) · gear-ratio and propulsion sizing before CAD

5.0

Plates

Pl. 1Fluent meshing evidence for the original three-blade model. Left: the propeller solid itself is valid, with no self-interference; the three blade-root cuts are marked. Right: the Workbench rotating and stationary fluid domains shared a duplicated interface (31,701 cross-intersections) with only about 1.00 mm axial and 2.00 mm radial clearance; the revised MRF zone is R = 215 mm, Y = ±30 mm.

Pl. 2Original versus rebuilt blade sections at z = 17.5, 27.5, 100 and 192.75 mm. The blade was resampled at 46 radial sections and rebuilt as cubic B-spline surfaces; the largest sampled deviation is 0.64 mm near the root trailing edge at z = 27.5 mm, and most stations are within 0.04 mm.

Pl. 3Volume mesh cut through the rotation axis (poly-hexcore, refined rotating zone and wake) and the blade surface mesh near a tip: 950,526 cells in all, 1.5 mm blade surface size.

Pl. 4Blade surface pressure relative to ambient (201.9 kPa absolute at 10 m) at 2000 rpm and 1.0 m/s, suction side facing upstream and pressure side facing downstream. Scale clipped at ±300 kPa; the computed range is −722 to +697 kPa.

Pl. 5Cavitation indicator from the single-phase solution at 2000 rpm: in red, blade area where the computed absolute pressure is below the 2.3 kPa vapour pressure, 18 % of the suction side and 0 % of the pressure side.

Pl. 6Velocity on the plane through the rotation axis at 2000 rpm with 1 m/s inflow: the slipstream contracts to about 0.16 m radius and reaches 6 to 7 m/s.

Pl. 7Thrust and torque coefficients against advance ratio J = V/(nD) from the CFD sweep at 1.0 and 1.5 m/s inflow; both speeds fall on one curve, with zero thrust near J = 0.48.

Pl. 8Propeller thrust against shaft speed at both inflow speeds; dots are CFD points, curves are drawn through the shared K_T(J) curve (negative thrust means the propeller is braking).

6.0

Anomaly report

Closed

Observed: Every Fluent meshing attempt on the supplied propeller failed.
Root cause: Mostly not the blade. The rotating and stationary fluid domains each carried a copy of their interface, which crossed itself 31,701 times with about 1 mm of clearance; sub-millimetre sliver faces at the blade roots and tips made it worse.
Corrective action: Shared topology between the domains, a re-partitioned and re-sewn mesh-ready solid (volume change −0.0001%), and a rebuilt rotating zone. The volume mesh then built with zero errors.
Lesson: Most CFD failures are geometry failures, and not always in the part you suspect. Check the fluid domains, not just the solid.

7.0

Limits & next

Russell & Sekula, NASA 2017 (blade geometry) →