RESOURCES · SOLUTIONS BLOG

The Vortex Street — and a Little Plate That Tames It

RS Rand Simulation · Applications Engineering AI · June 2026 · 11 min read

Wind past a flagpole, water past a bridge piling, flow past a heat-exchanger tube — behind every bluff body in a steady stream, the wake peels off into a tidy, alternating procession of vortices. That von Kármán vortex street doesn't just look beautiful; it pushes sideways on the body at a precise frequency, and when that frequency finds a structural resonance, things shake apart. We reproduced the textbook shedding in Ansys Fluent, validated it against the literature, and then added one of the oldest tricks for killing it: a splitter plate.

The von Kármán vortex street behind a circular cylinder at Re = 150. The wake sheds alternating vortices at a single dominant frequency — the source of the sideways (lift) oscillation.

First, get the benchmark right

The bare cylinder is the control, and it's run at Re = 150 — squarely in the clean, two-dimensional laminar shedding regime, so a 2D simulation is honest. The number that defines the wake is the Strouhal number, the dimensionless shedding frequency. Our solve lands at St = 0.187, about 1.5% off the published reference of 0.184 (Qu et al., 2013), with the mean drag (C_d ≈ 1.36) and the lift fluctuation in the same close company. The spectral peak is sharp and single — the signature of clean, periodic shedding — which is what lets the control comparison mean something.

The lift-force spectrum, measured by Fourier-transforming the *computed* C_L(t) over a long developed-shedding window (~8 cycles). The bare cylinder (red) has a **single, sharp peak at St = 0.187**, landing essentially on the published reference of 0.184 (gold dashed). One peak, no harmonics worth speaking of — that is what a clean, periodic von Kármán street looks like in frequency space, and it is the benchmark we had to hit before the splitter comparison could mean anything. The splitter peak (teal) is previewed here too; the next sections unpack it.

A subtle but important methodology note hides in that peak. Our first attempt produced a wildly wrong St ≈ 0.04 at a verified Re = 150 — not a physics error but a numerical one: too few inner iterations per time step left each step under-converged, which low-pass-filters the dynamics and artificially slows the shedding. Tightening to 20 inner iterations at a 10⁻⁵ residual and 160 time steps per cycle restored the true St = 0.187. We flag this because it is exactly the kind of quiet under-resolution that produces a confident, wrong answer — the validation against a known St is what caught it.

Then break the feedback loop

A vortex street is a feedback loop: the two shear layers off the cylinder's shoulders roll up and interact across the wake centerline, each one triggering the other. A splitter plate — a thin fin mounted on the rear of the cylinder, along the centerline — physically gets between them. It delays that cross-talk, lengthens the recirculation bubble, and weakens the roll-up. The result is a quieter wake: the shedding doesn't stop entirely, but the sideways forcing drops sharply.

The result: the rear splitter plate cuts the lift-fluctuation amplitude (C_l,rms) by roughly 53% — from ~0.376 on the bare cylinder to ~0.176 — and drops the mean drag (~1.36 → ~1.08) and shedding frequency too. This is strong attenuation, not full suppression: the wake is still periodic, just far gentler on the structure.

Before and after, on one frame

The numbers are honest, but a 53% drop in a coefficient is abstract. Here is the same physics you can see — both cases frozen at a fully developed shedding state, the bare cylinder above and the splitter below, in the field that actually generates the vortex street: the signed z-vorticity (red = counter-clockwise swirl, blue = clockwise).

Instantaneous z-vorticity, built directly from the solved velocity gradients (ω_z = ∂v/∂x − ∂u/∂y). **Top — bare cylinder:** the two shear layers roll up almost immediately behind the body into a tight, vigorous alternating street. **Bottom — splitter plate:** the thin fin physically separates the two shear layers along its length, so the roll-up is *pushed downstream*, the recirculation bubble is visibly *lengthened*, and the vortices that finally form are weaker and more diffuse. Same flow, same Reynolds number, one small plate — a completely different wake.

The same Re = 150 flow with the rear splitter plate in place (vorticity-magnitude field). The thin fin sits on the wake centerline and holds the two shear layers apart, so the roll-up is pushed downstream and the near-wake is visibly calmer than the bare cylinder above — the motion behind the 53% drop in lift fluctuation.

That picture is the mechanism. A vortex street is a feedback loop between the two shear layers; the plate gets between them and delays the moment they can start triggering each other. It does not abolish the loop — the layers still eventually roll up past the plate tip — it just makes the loop weaker and slower. That is why the shedding frequency drops too: with the formation region stretched out, the wake takes longer to complete each cycle, and the Strouhal number falls from 0.187 to 0.164.

The pressure field is the lift

Why does any of this change the force on the body? Because the sideways (lift) force is nothing more than the surface-pressure imbalance integrated around the cylinder — and every vortex that rolls up carries a low-pressure core. As vortices shed alternately off the top and bottom, that low-pressure zone flips from one side to the other, and the body feels a force that swings with it. Look at the pressure field and you can read the lift straight off it.

Instantaneous static pressure (≈ C_p). The blue low-pressure cores are the vortices; in the bare wake (top) they sit on *alternating* sides of the centerline and march downstream, and that side-to-side asymmetry, pressing on the cylinder surface, is exactly the fluctuating lift the spectrum measured. In the splitter case (bottom) the near-wake suction is milder and the alternating cores are weaker and pushed back — less imbalance on the surface, less lift. This is the physical link between a flow picture and the C_L,rms number.

So the chain is complete: the splitter lengthens the formation region (vorticity figure) → the alternating low-pressure cores are weaker and further from the body (pressure figure) → the surface pressure imbalance is smaller → the lift fluctuation collapses. The last link is the one we can put a hard number on, by watching the lift coefficient itself over time.

The computed lift coefficient C_L(t) over the developed-shedding window, bare (red) versus splitter (teal), aligned to a common start. Both are clean, near-sinusoidal limit cycles — the wake is *still shedding* in both cases — but the splitter's amplitude is roughly half. The root-mean-square lift drops from **C_L,rms = 0.376 to 0.176, a 53% reduction**. The mean drag falls too (C_d 1.36 → 1.08, ~20% lower), and in the spectrum the shedding peak loses ~78% of its power. None of these are asserted — they are read off the same force history the solver wrote at every step.

Notice what the splitter does not do: it does not flatten the trace to zero. The lift still oscillates, just gently. That is the honest headline — strong attenuation, not suppression. At L/D = 1.5 and Re = 150 the plate is sub-critical: it cannot fully kill the shedding, because the shear layers still find each other past the plate tip. Driving the wake completely steady would need a longer plate (the critical length grows with Reynolds number), and that is a different, longer run we did not make — so we report the attenuation we measured, not a suppression we did not.

Why Re = 150 keeps the 2D story honest

One more word on the choice of Reynolds number, because it is what makes every number above defensible. At Re = 150 the cylinder wake is in the laminar, parallel-shedding regime: the vortices come off as clean two-dimensional tubes, with no spanwise wiggle. The first three-dimensional instability (the “mode A” transition) does not set in until Re ≈ 190. Below that threshold a 2D simulation is not a convenient simplification — it is the physically correct model, and that is precisely why the 2D benchmark literature (Qu et al., 2013) exists to validate against. Push the same 2D mesh to Re = 300 and it would quietly lie to you, because the real wake there is no longer flat. We stay where 2D is the truth, hit the benchmark, and only then trust the control comparison.

Why this one matters

Vortex-induced vibration (VIV) is a real failure mode — it's why chimney stacks wear helical strakes, why marine risers carry fairings, why heat-exchanger tube bundles are spaced the way they are. The engineering question is never just “does it shed?” but “how hard does it push, and what knocks that down?” A validated shedding model answers the first; an honest before/after on a control device answers the second. Match the benchmark first, then quantify the fix — and don't oversell “reduced” as “eliminated.”

Honest scope. 2D laminar CFD at Re = 150, the regime where 2D is physically appropriate; results are reported as 2D and compared against 2D reference data (Qu et al., 2013). The splitter result is framed honestly as partial suppression / strong attenuation — the C_l,rms reduction is the headline, not a claim that shedding is eliminated. Each case was advanced to a converged, statistically stationary shedding window before the statistics were taken. The vorticity and pressure fields above are instantaneous snapshots of the developed wake (not time-averaged), each at its own representative phase; the force statistics (C_L,rms, C_d, St) are the true averages over a long developed window (~8 shedding cycles, >1,200 samples). All numbers come directly from the solver's own force history and field output — none are illustrative.

Is vortex shedding rattling something you’re responsible for — a stack, a riser, a heat-exchanger tube bundle? An Ansys Fluent model that had to hit the published benchmark before it was believed — St = 0.187 against the 0.184 of Qu et al. (2013), a 1.5% match that caught an under-converged first attempt in the act — and then put a hard number on the fix, a splitter plate worth a 53% cut in lift fluctuation read straight off the solver’s force history over ~8 developed shedding cycles — is how simulation tells you how hard the wake pushes and what knocks that down, before resonance finds your structure. That's innovation through insight.