Data center noise (or "data centre noise" in the UK) is an ever-increasing problem across the planet as more are built to service our digital world. Crypto mining and energy center noise also fall into the same category of problem. The primary noise issues from data centers, energy centers and crypto mining operations are the cooling systems, hosts of chiller fans that generate both broadband noise and low-frequency tones (hums and drones). In both cases, conventional noise control methods typically reduce efficiency and increase power consumption.
In addition to the cooling systems, there will also be energy center power generation on-site that can be continuously online (e.g. CHP, gas turbines) or on standby (e.g. diesel generator sets that are run to a test schedule), all of which can cause noise problems.
There are innovative environmental noise reduction technologies that solve both problems. These don't just cut noise without reducing efficiency, they can actually increase fan cooling efficiency by up to c 20%, potentially making projects self-financing.
There are 2 primary types of noise from the dry air cooler or chiller systems, both generated by the fans.
The following is a detailed guide to data center noise attenuation best practices for these sources based on the latest technology that can even improve cooling efficiency to reduce running costs.
Data center cooling fan noise sample: the same low-frequency hum or drone recorded outside and then inside a building.
The difference illustrates the effect of the sound passing through glazing or building structure that attenuates higher frequencies, enhancing the low-frequency content of the interior sound. This low-frequency tonal noise content is a very common cause of complaints that is usually not addressed as it does not contribute to the headline overall dB(A) value (the "A" weighting progressively attenuates low frequencies).
Data center noise
Email us a video from your smartphone for an evaluation of best practice
This plot is an overlay of the noise signatures from the recording above, measured both outside and inside a building. The latter (red trace) is overlaid with the corresponding "A" weighted signature showing how the dominant 73Hz tone is attenuated by 25dB by the "A" filter as illustrated by the green trace.
Not understanding this fact and the role it plays in the cause of complaints is a very common mistake and the reason for many failed data center and crypto mining noise mitigation strategies where the main sources (fans and generators) produce low-frequency tones.
All the traditional noise reduction methods are based on adding noise control equipment to the plant. Our innovative approach is primarily based on modifying aerodynamics to render the cooling systems more efficient (in one case, 20% more efficient). As a result, fans can be run at lower speeds to cut broadband noise and save energy. The aero-mods also eliminate problem low-frequency tones. This approach is dramatically lower cost and more efficient than the traditional methods.
If necessary, residual noise can be reduced using a combination of other new technologies or much more modest conventional measures.
There are 2 separate noise issues that need to be addressed:
Air-cooling system fans in chillers and adiabatic coolers etc pull high volumes of air through heat exchangers that generate mid-high frequency sound that is directional.
Aerodynamics / close shields / meta-barriers
These provide greater attenuation and enhanced cooling efficiency than is possible using conventional methods and at dramatically lower costs.
Acoustic louvres and barriers
Acoustic louvres typically provide modest attenuation (virtually none at low frequencies) with significant backpressure that can affect free cooling and hence overall system efficiency. Often fitted along roof edges where structural strength and wind loading factors have to be taken into account. Where the noise is dominated by broadband dB(A) and relatively little attenuation is required, they can be an effective option but should be costed against alternatives.
Conventional acoustic barriers must be located as close to the source as possible to provide good attenuation. However, as this geometry blocks airflow, they are usually located as far as practical from air intakes to maintain cooling efficiency at the expense of compromised attenuation. Note wind loading and structural considerations.
Neither of these options can attenuate low-frequency sound.
One of the most common complaints about data center noise is a low-frequency humming sound that travels long distances.
There is an elegant, low-cost engineering solution to the fan noise problem (see below) that can increase cooling efficiency (self-financing). Add improved silencers for the engines where necessary or retrofit elements to turbine silencers to improve low frequency attenuation.
Air-cooling system fans very often generate high levels of low-frequency tonal noise (hum or drone) that is omnidirectional and travels large distances. Sadly, it is virtually impossible to get information re tonal noise from suppliers. We resort to analysing smartphone video clips from anywhere in the world to evaluate tonality.
This is based on a similar approach as that used to cut broadband noise (see above). We use Computational Fluid Dynamics (CFD) modeling to aid in the design of retrofit aerodynamic components that reduce the low-frequency tonal noise elements - typically by 90% (10dB) to 99% (20dB). As this cuts the sound at source, both intake and exhaust noise is simultaneously reduced. In most cases, the fan performance is significantly improved, increasing efficiency and cutting running costs which can make these projects self-financing.
This is best practice as it is the only practical and effective method that can be used to reduce low-frequency tonal noise from cooling systems without very seriously compromising efficiency. It is also a dramatically lower cost option.
Silencers and enclosures
Tonal fan noise is radiated equally from the intakes and the exhausts. Consequently, both must be treated to provide any significant attenuation. In practical terms, this means very high backpressure or very long outlet silencers (the minimum one wavelength silencer for an 80Hz tone is 4m) coupled with intake silencing (usually as part of an acoustic enclosure) that suffer similar problems.
As a result, there is no practical way to use conventional silencers and enclosures to attenuate low-frequency tones. Perhaps this is why there is such focus on reducing the overall dB(A) (which can be attenuated using these techniques) rather than the often much more problematic low-frequency noise features.
We have extensive experience in creating noise models for data centers to generate noise contour maps both for planning applications for new projects and to aid in meeting environmental noise criteria on existing sites. The factor that determines the accuracy and usefulness of noise models is the acoustic knowledge and expertise of the consultant. The key elements we consider are:
In addition to desktop noise modeling exercises (which can include checking or updating existing noise models), we can also provide full environmental noise measurement and assessment services supported by the provision of detailed mitigation recommendations based on current innovative best practices.