Why Fan Noise Is a Selection Problem Before a Treatment Problem

Why this matters

The cheapest decibel on any air system is bought before anything is built, by putting the fan's duty near the peak of its own efficiency curve. The most expensive decibel is bought afterwards, in the 63 Hz octave, with mass and length and shaft space. A fan selected well off peak does not just move less efficiently, it generates additional low-frequency sound power, and that is the one part of the spectrum where every downstream treatment is at its weakest and the A-weighted number on the meter is at its least honest. This card follows one fit-out where every treatment installed performed as published, and the complaint only ended when the wheel was changed.

Before any panel comes off

A fan wheel coasts long after the disconnect opens, and the stored rotational energy in a large wheel will take a hand. Lock and tag the unit under 29 CFR 1910.147, then look through the access port and confirm the wheel has actually stopped before reaching in, rather than assuming a timer. Where the work is at the drive, de-energize, lock, tag and prove dead under 29 CFR 1910.333(b)(2) and wait the drive manual's stated bus discharge time before opening the enclosure. Every measurement taken with the fan running is taken with all guards in place as 29 CFR 1910.212 requires, from outside the rotating plane, and where your exposure at the measurement position reaches the 29 CFR 1910.95 action level for the time you will be standing there, wear what the program specifies for that position.

The complaint, and the two numbers that disagreed

An open-plan office fit-out, occupied three weeks. Steady rumble, worse when the floor was quiet, no tonal quality reported. The specification for the space named an A-weighted limit and the commissioning report showed the space passing it.

At the worst-reported position, microphone 1.2 m above floor, 5-minute equivalent-continuous, referenced to 20 micropascals:

The A-weighted figure met the specification. The 63 Hz band did not meet the criterion curve the same specification also named, and it is the band an A-weighting network discards most of before the overall number exists at all. A sibling card owns why that filter behaves this way; the operational consequence is that a compliant overall number and a genuine low-frequency complaint sit together comfortably, and the spectrum is the only thing that separates them.

Background margin in the 63 Hz band is 18 dB, at or above 10, so the running reading is the system's own band level within a few tenths and the correction was declined and recorded as declined.

What got eliminated, and on what evidence

Isolation. Free height minus installed height was measured at every mount and matched the computed deflection, and the frequency ratio at running speed cleared the amplification gate comfortably. Nothing at shaft rate stood above the background at the complaint position. Cleared.

Breakout. Two positions and a blanked terminal, run as the breakout card specifies. The mid-room level under the main did exceed the level under a diffuser, so breakout was real and it was treated: the crossing main was changed to round spiral. That work bought 5 dB in the 63 Hz band at the complaint position, taking it from 63 dB to 58 dB. Real, measurable, and not enough.

The silencer. Measured static pressure across the unit matched the published figure at the computed free-area velocity, and the effective insertion loss, once its self-noise was combined in, matched what the published data predicted. The unit was performing to specification. Cleared as a fault, though its published insertion loss in the 63 Hz octave was modest, which is normal for a dissipative silencer in that band.

Three correct answers, and the room was still 18 dB over background at 63 Hz.

Where the fan actually sat on its own curve

The fan's published sound power data is given at specific points of rating, tested to an AMCA method in the edition the certified rating names. That data is only descriptive of your fan if your duty is near the point it was published at. So the duty got plotted.

The selection software reported the installed duty at 62 percent of the fan's peak static efficiency, well to the left of peak, which for a centrifugal fan means low flow relative to the wheel at high pressure. That region is where flow over the blades becomes unstable and separates, and the additional sound power that instability produces is concentrated at low frequency. The standard estimation procedure in the handbook chapter treats this explicitly, adding an efficiency penalty in dB that is zero near peak and rises as the operating point moves away from it; the table of values belongs to that procedure and to the fan manufacturer's own data rather than to a rule of thumb.

Name the other end of the range, or you have asserted a direction rather than established it. Far to the right of peak, at high flow and low pressure, the same fan also gets louder, but the character is different: more broadband and higher-frequency content from velocity through the wheel and the discharge, plus a motor loading problem on some fan types. Low-frequency rumble specifically is the left-of-peak symptom, and that is what matched the measured spectrum here.

Why the low bands are the expensive ones

Three independent things go wrong at once below about 125 Hz, and each one is owned by another card rather than re-derived here.

A dissipative silencer's insertion loss falls off in the low octaves, so buying more attenuation there means length and shaft space you generally do not have. A duct or partition wall gains roughly 5 to 6 dB per doubling of surface mass at a stated frequency and incidence, which means a serious low-frequency improvement is a serious weight of material. And the A-weighting network removes most of the low-frequency content before the overall figure exists, so the number that would justify the spend to a client is the number least able to show the problem.

Against all three of those, moving the fan's operating point costs one component.

The reselection and what it bought

The wheel was changed for a larger one turning slower at the same duty, placing the operating point at 91 percent of peak static efficiency. From the manufacturer's own selection output, published sound power in the 63 Hz octave fell from 96 to 85 dB re 1 picowatt, a reduction of 11 dB, and the 125 Hz octave fell 8 dB.

Path attenuation between the fan and the room is a fixed per-band subtraction at constant duct geometry, constant flow and the same terminal, none of which changed. So an 11 dB reduction in source sound power should appear as an 11 dB reduction in sound pressure at the complaint position in that band.

Measured after, same position, same 5-minute equivalent-continuous basis, re 20 micropascals: 63 Hz octave, 47 dB running against the same 40 dB background. That margin is 7 dB, which sits in the zone where a correction applies, so 1.0 dB comes off and the corrected figure is 46.0 dB.

The comparator gets the same treatment or the comparison is void. The before figure had an 18 dB margin, where the correction is a few tenths and was declined, so it stands at 58 dB. Corrected before against corrected after is 58 down to 46.0, a measured reduction of 12.0 dB against a predicted 11 dB. The 1.0 dB difference is comfortably inside the uncertainty of a background-corrected reading at a 7 dB margin, so the prediction is confirmed rather than beaten.

Sibling-rule check. Every level carries its quantity, weighting, bandwidth, time basis, reference and position, and sound power levels re 1 picowatt are never mixed with sound pressure levels re 20 micropascals in the same arithmetic. The background margin gate is applied per reported figure using the sibling procedure's three zones, and both the before and after figures are corrected or declined on their own margins, so no corrected figure is benchmarked against an uncorrected one. The efficiency penalty and the mass-law relationship are each stated with the condition they hold under and routed to their owning data. The blade-pass and shaft-rate content is left to the isolation and spectrum cards rather than re-derived.

The comparison the client actually needed. The treatment work already installed bought 5 dB in the governing band. The reselection bought 11 dB, better than twice as much, and the wheel change consumed well under half the labour hours the duct modification had already consumed. That ratio is not a property of this site; it is what happens whenever a low-frequency complaint is treated downstream of a fan that is generating avoidable low-frequency power.

When treatment is the right answer after all

When the fan is already near peak. If the duty plots at or near peak static efficiency, there is no selection penalty to recover and the remaining sound power is the fan's honest output. Treatment is then the only path and the silencer and duct work are correct spending.

When the fan cannot be changed. An existing unit still under warranty, a wheel with no larger option in the housing, a duty fixed by a process, or a lead time the schedule cannot absorb. Then the decision is which treatment buys the most in the governing band, and that comparison is made per band and not on the overall.

When the governing bands are mid-frequency. A hiss or a whine at a terminal is regenerated close to the room and no fan reselection touches it, because the energy is not coming from the fan at all.

When the complaint is a tone rather than a rumble. A tone is judged by different rules and can be worth several times its measured contribution; that is a sibling card's subject and it changes the priority order entirely.

References

• ASHRAE Handbook chapters on sound and vibration control, for the fan sound power estimation procedure and the efficiency penalty it applies
• AMCA test standards for fan sound power rating, in the edition the certified rating names, which own the conditions the published data was measured under
• Fan manufacturer's selection output and published octave-band sound power data for the specific wheel and duty
• 29 CFR 1910.147 and 29 CFR 1910.333(b)(2), for isolating the mechanical and electrical energy before a cabinet or drive enclosure is opened
• See related: What a Duct Silencer Costs You in Pressure; Why Duct-Borne Noise and Duct Breakout Are Different Faults; What a Weighting Network Is Doing to Your Reading