Why Metal Louver Facades Whistle and Work Loose: Back-Calculating Critical Wind Speed from Vortex-Induced Vibration

Six months after handover a louver screen starts humming in the wind, and a row of bolts is found loose. They are re-torqued with 20% more effort; three months later they are loose again and a blade has cracked at the end. The usual verdict on site is bad workmanship.

It is not workmanship. It is cross-wind vortex-induced resonance, and it was never calculated. The along-wind load procedure that everyone does run - wk = beta_gz * mu_s * mu_z * w0 in GB 50009, or the equivalent in EN 1991-1-4 - answers the question how hard does the wind push the screen inward. It says nothing about the vortices shedding alternately from the two faces of each blade and dragging it sideways, back and forth, all day. That second mechanism is what makes a facade sing and unbolt itself, and it can be predicted with three formulas at drawing-board stage.

What follows is a full worked example on one very ordinary blade: a 6063-T5 aluminium flat bar, 100 mm x 3 mm, 1.2 m span, bolted to mullions at both ends with no intermediate support.

The reference case

Step 1: How fast does it want to move - first natural frequency

Treating the blade as a simply supported beam (bolted ends sit between pinned and fixed; pinned is the conservative choice):

Counter-intuitive result no.1: switching to 304 stainless changes almost nothing

What actually enters the formula is the specific stiffness E/rho, not E or rho on their own. Raising E by a factor of 2.8 raises rho by 2.9 at the same time, and the two cancel. Every common structural metal sits in the band 2.43 to 2.62 x 10^7 - a spread of under 8%.

Stainless will fix it is one of the most expensive misconceptions in louver work.304 gives f1 = 4.66 Hz, 2.4% lower than the aluminium blade. The money buys corrosion resistance, not stiffness. If you need both, specify stainless plus thicker section or an intermediate support - do not expect the alloy to do the job.

The two knobs that do work: thickness to the first power, span to the second

The gain from a mid rail comes entirely from the shorter span, not from continuity.The first mode of a two-equal-span continuous beam is antisymmetric with a node at the central support, so its frequency equals the simply supported fundamental of one span. Do not count on tying everything together to stiffen it - the 1/L^2 term is all you are buying.

Step 2: When does the wind pick it up - vortex shedding and critical speed

Flow past a bluff section sheds vortices alternately from each side. The shedding frequency scales with speed and inversely with the reference depth:

Lock-in occurs when f_s meets f1. Solving for the wind speed:

What 4.0 m/s actually means

That is the whole problem: the critical speed is 4.0 m/s, sitting on top of the speed band the site sees most often.Foshan's annual mean wind is about 2.0 to 2.6 m/s and 3 to 6 m/s is the everyday working range, so this blade crosses its own resonance several times a day rather than once in a while. A practical screening rule follows directly: compute v_cr and compare it with 8 m/s - below that it will almost certainly sing, above 15 m/s it is a rare event.No wind tunnel required.

Why the whole facade appears to vibrate at once

The strict lock-in band is only about plus/minus 10% (3.6 to 4.4 m/s here), yet entire elevations buzz. Two reasons:

• Wind speed drifts continuously - it climbs and falls through the lock-in band repeatedly, dwelling there for tens of minutes to hours at a time.
• f1 is not identical across the array - length tolerances, bolt tightness and mullion alignment scatter f1 by plus/minus 10% to 20%, so at any moment some subset of blades is locked in.

Step 3: How far does it move - the Scruton number

Amplitude is governed by the mass-damping parameter, the Scruton number. It is the highest-leverage quantity in the whole problem:

Bare metal structures typically run zeta = 0.002 to 0.006; a bolted or welded louver screen is around zeta = 0.004:

(Orders of magnitude read from the Sc-dependent curves in EN 1991-1-4 Annex E, for option screening. Use the code curves or an aeroelastic test for final design.)

Damping is the cheapest fix on the list

Put a 3 mm neoprene pad (60 to 70 Shore A) between blade and mullion, or bond a constrained damping layer along the blade, and zeta goes from 0.004 to 0.04:

Retrofit priority by effect per dollar: (1) damping pads at the blade ends - Sc x10 for almost no money, and retrofittable without dismantling; (2) a mid rail to halve the span - f1 x4 for the price of one rail; (3) thicker blade - f1 x1.33 for 33% more metal.Do 1 and 2 first; you will rarely need 3.

Step 4: What breaks first - bolt loosening, long before fatigue

Stress range

With mode shape w = y*sin(pi*x/L), curvature kappa = y*(pi/L)^2, moment M = EI*kappa, and edge stress:

Cycle count

Two years puts it near 1e8 cycles - deep in the high-cycle fatigue regime.

But the first casualty is the bolt hole, not the parent metal

What actually happens first: self-loosening (the Junker mechanism)

Once transverse displacement relative to the joint reaches 0.1 to 0.3 mm, the faying surfaces micro-slip, nut and bolt head rotate relative to each other, and preload drops 30% to 50% within 1e2 to 1e3 cycles, fully loosening within a few thousand (DIN 65151 transverse vibration test).

Our amplitude is 15 mm - 50 to 150 times the trigger. At 4.78 Hz:

The failure sequence is therefore fixed: low-frequency large-amplitude vortex response → bolt self-loosening (minutes to hours, heard as rattling, seen as lost torque) → support conditions soften, f1 drops, amplitude grows (a vicious circle) → hole-edge fatigue cracking (years) → fastener fracture and blade drop."Re-torque and it comes back" is stage two. More torque only raises the starting point; transverse slip grinds preload away regardless.

Anti-loosening: what works

Where the whistling comes from: three sources, three frequency bands

The arithmetic behind quiet by day, noisy at night:night-time wind often parks in one band, and louver whistling scales as v^6. Going from 5 to 10 m/s is +18 dB(A). Against the GB 3096 class-2 night limit of 50 dB(A), a facade that reads 45 dB unnoticed in the day can cross the line once the evening breeze picks up. Treat the slot, not the sound - break the hard gap at the blade end with a gasketed floating connection and vary slot widths across the elevation.

Five ranked fixes

1. End damping pads - highest return, lowest cost.3 mm neoprene or a bonded constrained layer between blade and mullion lifts zeta from 0.004 to 0.04 and Sc from 3.25 to 32.5, cutting amplitude by an order of magnitude. Retrofittable on existing screens.
2. Mid rail, halve the span.f1 scales as 1/L^2, so 1.2 m to 0.6 m is a factor of 4 and v_cr goes from 4.0 to 16 m/s - clear of the everyday range. The gain is from span, not continuity.
3. Break span-wise coherence of the vortex street.Equal pitch makes vortices shed in phase across the whole elevation and their energy adds. Switch to a varying pitch (180/200/220 mm repeating), slot the blades, bevel or serrate the ends - near-zero cost, and the step most often skipped.
4. Upgrade the anti-loosening detail.Drop the spring washer; use wedge-locking washers plus medium-strength anaerobic threadlocker, and mark a torque witness line across bolt, nut and substrate for inspection. On new work, weld or blind-rivet and delete the failure mode.
5. Manage the gaps.Replace the hard blade-end-to-mullion gap with an EPDM-gasketed floating joint to kill local cavity resonance, and stagger gap widths along the elevation so the whole facade cannot whistle on one note.

Three-step field check, no instruments needed

Step 1 - measure the frequency

Push the blade at mid-span and release. Record the decay with a phone slow-motion video (240 fps) or any accelerometer app, count ten full cycles t10, then f1 = 10 / t10. For D about 100 mm, f1 below 8 Hz means v_cr below 6.7 m/s - it will sing in everyday wind; above 18 Hz it is a rare event.

Step 2 - measure the damping

zeta below 0.01 means Sc is too low and amplitude will be large - fit damping pads.

Step 3 - confirm self-loosening

Draw a continuous witness line across bolt, nut and substrate on every fastener. Re-check after a month. A broken line proves transverse slip is occurring, at which point more torque is pointless and the anti-loosening detail must change.

Five common mistakes

1. Expecting stainless to stop the vibration.f1 depends on E/rho: aluminium 2.556e7, 304 stainless 2.434e7, Q235 2.624e7 - a spread under 8%. In this example 304 gives 4.66 Hz against 4.78 Hz for aluminium. Material change buys corrosion resistance, not stiffness.
2. Thickening without shortening the span.f1 is proportional to t, so 3 to 4 mm buys 33% for 33% more metal; f1 is proportional to 1/L^2, so 1.2 to 0.6 m buys 300% for one extra rail. Spend on the rail first.
3. Specifying spring washers against transverse vibration.In the DIN 65151 transverse test they are largely ineffective. Use wedge-locking washers, anaerobic threadlocker, or weld.
4. Uniform pitch.Equal spacing makes the Karman street fully coherent across the elevation so the whole facade sheds in phase. Varying the pitch or slotting the blades costs almost nothing and is the fix most often omitted.
5. Checking only along-wind load.wk = beta_gz * mu_s * mu_z * w0 covers along-wind static and gust response and says nothing about cross-wind vortex resonance. Louver blades, tubes and slender members need a separate Strouhal-number vortex check - otherwise the screen passes the strength calculation and still shakes itself apart.

In one line:a singing, self-loosening louver is three numbers that were never calculated - f1 = (pi/2L^2)*sqrt(EI/m) (4.78 Hz here, and stainless does not help), v_cr = f1*D/St (4.0 m/s here, right inside the 3 to 6 m/s band the site sees most), and Sc = 2*m*delta/(rho_air*D^2) (3.25 bare, 32.5 with a neoprene pad). Bolts loosen first, in minutes to hours, and spring washers will not stop it; hole-edge fatigue arrives years later. Fix in this order: damping pads, mid rail, break coherence, upgrade the locking detail, manage the gaps - the first two are usually enough.