Prediction of the noise from mechanical ventilation systems in dwell- ings: case studies Arne Dijckmans 1 Belgian Building Research Institute Rue de Lombard 42, B-1000 Brussels, Belgium

ABSTRACT Noise caused by mechanical ventilation systems in dwellings remains a common problem in practice. To help designers and installers, a simplified acoustic calculation tool has been developed which can be used without detailed acoustic knowledge. The tool is based on the prediction models in EN 12354- 5 and VDI 2081 and takes into account the airborne duct sound. The model has been validated by multiple case studies in dwellings. For each case study, the noise levels and airflow rates of the mechanical ventilation system have been measured in each room. Generally, the global A-weighted noise level is predicted with an accuracy of ± 3 dB, on the condition that the airborne duct sound is dominant and sufficiently accurate acoustical input data are available for key components like fans, silencers and air openings. However, some measurement results cannot be predicted by the simplified calculation framework, e.g. when turbulence in two closely placed fittings leads to additional flow noise. The uncertainty also increases when the fan sound is dominated by tonal components.

1. INTRODUCTION

Noise is one of the key elements that determines the occupants satisfaction of mechanical ventilation systems in dwellings. Too noisy ventilation systems are often turned off by the users, which has a direct impact on the indoor air quality. Therefore, national standards and regulations impose maxi- mum noise levels on ventilation systems in buildings. For small projects, the acoustical requirements are typically only accounted for by means of simple design guide lines and rules of thumbs (e.g. placing a silencer and limiting the air velocity). However, this does not necessarily guarantee that the ventilation system satisfies the acoustic requirements. To help designers and installers, we want to develop an acoustic calculation tool which can predict the acoustical performance with sufficient accuracy, but doesn’t require detailed acoustic knowledge from the user.

Different acoustical design methods, software tools and guidelines exist. EN 12354-5 gives a gen- eral framework to calculate installation noise. Guidelines are given in various handbooks (often com- plementary to the standard and each other), like the German VDI [1], the American ASHRAE [2] and the Dutch ISSO [3].

Several questions arise when applying these calculation schemes [4]. The guidelines use simplified tables or empirical relations but their field and limits of application are not well known. Specific acoustic input data are often lacking for components used in ventilation systems (e.g. networks with manifolds). There is also a lack of confidence in the calculation process, as e.g. interaction between components is not considered. Previous and current research mainly focusses on improving the acous- tic calculation methods for ventilation systems by addressing these issues. The ACOUREVE project

1 arne.dijckmans@bbri.be

worm 2022

[4] focused on improving the quality of the acoustical input data for components like junctions, bends, dampers and manifolds. Similarly, numerical analysis and laboratory measurements are being used to investigate the interaction effects between components. On system level, more advanced models have been developed based on a transfer matrix method and a two-port [5] or multi-port [6] charac- terization of individual components. While this approach can give valuable insight, it is not readily applicable for real-life situations as input data covering the entire frequency range of interest are still lacking for many ductwork elements.

The aim of this study was therefore to investigate the accuracy and limitations of the standardized prediction framework of EN 12354-5 when applied to mechanical ventilation systems typically en- countered in dwellings in Belgium.

2. PREDICTION MODEL

2.1. General description The acoustic design tool for mechanical ventilation systems, developed at BBRI, is based on the prediction models in EN 12354-5 and VDI 2081 [1] and takes into account the airborne duct sound. This includes both the inlet or outlet fan sound power radiated into the ducts and the flow noise generated in each component. The calculation is performed in octave bands from 63 Hz to 8000 Hz.

The sound power level radiated into a room by the air terminal device (ATD) is determined by a sequential calculation, starting at the fan. For each element, the outgoing sound power level is calcu- lated from the incoming sound power level, the sound power level reduction of the element ( Δ L w ), and the flow noise generated in the element.

The standardized sound pressure level L Aeq,n T in the room is calculated at 1.5 m from the ATD (i.e. the minimum distance to the noise source in the measurement procedure according to ISO 10052), accounting for both the direct sound and the reverberant field. When more than one ATD is present in a room, the L Aeq,n T is determined at 1.5 m from the noisiest ATD, taking into account only the reverberant field contribution of the other ATD(s).

2.2. Input data for elements The following type of elements are implemented in the acoustic calculation model: fans, straight ducts, bends, area changes, junctions, valves, manifolds, silencers and air terminal devices. For all the elements, the generic formulae of VDI 2081 [1] are implemented for flow noise and sound level reduction. Some of the generic formulae are simplified to minimize the required input from the user.

 Fans : because the fan speed n and optimum efficiency are generally unknown, tonal compo-

nents and efficiency terms are neglected. A standard relative octave sound power spectrum is used (based on n = 2100 / 3000 / 250 min -1 for fan type AM / RR / T, respectively).  Flow noise in bends and junctions : a relative rounding radius of r/d a = 0.15 is assumed.  Bends : the sound power reduction of a bend of 90° with a bend radius and without a lining is

used (which is the most conservative and generally most relevant for bends used in dwellings). No sound power reduction is taken into account for bends smaller than 90°. For junctions, the general formula based on the area ratio is adjusted as follows [3,4].

 The sound level reduction in the straight branch is reduced by 1 dB above the limit (or cut-

off) frequency f G of the straight branch.  For the additional level reduction in a bend, the spectrum of a 90° sharp-edged bend without

lining is used. Only half the bend reduction is accounted for when the section of the straight branch changes.

worm 2022

Manifolds are modeled as a junction, but only accounting for the area ratio level reduction, thus ne- glecting any additional reduction due to bend or plenum effects. Finally, for fans, silencers, valves and ATDs, measured acoustic data from technical sheets can be given as input. 3. CASE STUDIES

worm 2022

3.1. Description

In total, 12 dwellings with recently installed mechanical ventilation systems were selected for the study. Table 1 gives an overview of the main characteristics of each case study. The nominal flow rates varied between 200 m³/h for the smallest dwelling and 588 m³/h for the largest dwelling. In three dwellings, a mechanical ventilation system C was installed with natural supply through venti- lation grilles above the windows in the living room and bedrooms and mechanical extraction in the wet rooms (kitchen, bathroom, toilet) and bedrooms. For the other eight dwellings, a system D with balanced mechanical supply (in living room, bedrooms, playrooms and/or studies) and mechanical extraction (in kitchen, bathroom, toilets, storage rooms and/or technical rooms) was installed. The ventilation system of case 3 was optimized in four steps [7]. The dwelling of case 5 was divided in two sections with two separate ventilation systems.

Three different types of network lay-out and ductwork were encountered. For the dwellings with system C, the fans have multiple connections (Figure 1a). The airflow rate in each connected branch is determined by a single-leaf damper valve which is controlled by one or more sensors in the corre- sponding room(s) (CO 2 , humidity and/or VOC). For the dwellings with system D, five have a branched network of rigid metal ducts (Figure 1b) and three make use of a manifold system to which multiple, smaller round plastic ducts are connected (Figure 1c). In dwelling 9, a combination of both systems has been used (manifold for ground floor and branched network for upper floor).

In most dwellings, primary silencers (either rigid or flexible circular silencers with an absorptive material of 25 mm or 50 mm thickness) are installed in the technical room close to the fan. Only for case 1 and case 9, no silencers were present. Here, non-acoustic flexible ducts were used to connect the ducts to the fan. In three cases, additional secondary silencers were installed in the branched network (Figure 1b).

(a) (b) (c)

Figure 1: (a) Fan of case 8, (b) ductwork and secondary silencer of case 7, (c) manifold of case 5.

3.2. Measurements Measurements have been done for the fan setting corresponding to the nominal flow rate (which is generally the highest setting of the fan). The flow rates in the rooms have been measured at the air terminal devices (ATD) using a zero pressure compensation measurement device. The electrical power absorbed by the ventilation unit has been measured using a power meter placed directly into

the electrical plug. The standardized service equipment sound pressure levels L Aeq,n T in the rooms have been measured following ISO 10052. The equivalent sound pressure levels were measured over a time interval of 30 s in three fixed positions (two in the reverberant field of the room and one in the corner with the most reflective surfaces) using a handheld sound level meter. The reverberation time in each room was measured in octave bands using an impulsive noise source (hand clap). The octave- band service equipment sound pressure levels were standardized with the measured reverberation time in each octave band and corrected for background noise following the procedure of ISO 16032.

Table 1: Main characteristics of the mechanical ventilation systems of the case studies

Nr. System Nominal flow rate

Network lay-out and ductwork

(1) Primary silencers (2) Secondary silencers

1 C 310 m³/h Multiple connections at fan. Flat, rectangular ducts in floor.

(1) None (2) None

2 D 225 m³/h Manifold with multiple round, semi-flexible ducts in floor.

(1) Rigid 90 cm (2) None

3 D 300 m³/h Branched. Rigid, round metal ducts.

(1) Flexible 50 cm + rigid 90 cm (2) Rigid 90 cm (bedroom 1)

4 D 468 m³/h Branched. Rigid, round metal ducts

(1) Rigid 90 cm (2) Rigid 60 cm

5 D 266 m³/h + 164 m³/h

Manifold with multiple round, semi-flexible ducts.

(1) Flexible 100 cm (2) None

6 C 410 m³/h Multiple connections at fan. Flat, rectangular ducts in floor.

(1) Flexible 30 cm / 50 cm (2) None

7 D 588 m³/h Branched. Rigid, round metal ducts.

(1) Rigid 120 cm (2) Rigid 120 cm / rigid 60 cm

8 C 276 m³/h Multiple connections at fan. Flat, rectangular ducts in floor.

(1) Flexible 50 cm (bedrooms) (2) None

9 D 224 m³/h Combination of manifold (round, semi-flexible ducts) and branched network (rigid, round metal ducts).

(1) None (2) None

10 D 324 m³/h Manifold with multiple round, semi-flexible ducts in floor.

(1) Flexible 100 cm (2) None

11 D 230 m³/h Branched. Rigid, flat, rectangu- lar ducts in floor

(1) Flexible 50 cm / 75 cm (2) None (?)

12 D 200 m³/h Branched. Rigid, flat, rectangu- lar ducts in floor

(1) Flexible 50 cm (2) None (?)

worm 2022

3.3. Results and comparison with prediction For the prediction of the case studies, first the fan sound power was estimated at the working point (airflow rate – pressure drop) of the fan. The total airflow rate was determined from the measured airflow rates in each room. The pressure drop was estimated from a detailed aeraulic model of the ventilation system and/or the measured electrical power consumption of the fan. If detailed acoustic data were available at or around the working point of the fan, the measured sound power spectrum was used as input. Sometimes, only global values were available in the technical sheets, in which case a standard relative spectrum was applied. Generally, an interpolation was needed between dif- ferent working points stated in the technical sheets.

Afterwards, an acoustic calculation was performed for each branch. The flow noise of the ATDs was determined at the working point of the ATD (measured airflow rate and estimated pressure loss after balancing of the system). For most ATDs, no detailed spectrum of the flow noise is available and thus the relative frequency spectrum of VDI was used. The global A-weighted sound power level was determined by an interpolation of the available acoustic data in the technical sheets.

The noise levels in the technical rooms (where the fans are installed) were calculated based on the sound power levels of the housing radiation of the fan at its working point.

Figure 2 compares the global measured and predicted L Aeq,n T - values in each room of the 12 dwell- ings. For the measured values (in red), an error of ± 1 dB is indicated on the figure, corresponding to the measurement margin in the new acoustic Belgian standard for dwellings (NBN S 01-400-1:2022) which is based on the precision indicated in ISO 16032. For the predicted values (in blue), an error of ± 3 dB is indicated.

For more than 95% of the rooms, the measured L Aeq,n T was determined by the airborne duct sound. For the other cases, either the background noise was higher than the ventilation noise (bedroom 1 of case 3 step 4, studio case 4) or the ventilation noise was caused by other transmission paths, like airborne transmission from the technical room via the building structure or duct breakout (atelier and bedroom 1 case 7, laundry case 10). For the cases where the airborne duct sound was dominant, the prediction error is smaller than 2 dB in 54% of the cases, smaller than 3 dB in 65% of the cases and smaller than 4 dB in 75% of the cases. Larger prediction errors can e.g. be caused by an inaccurate model of the actual ventilation network (which often differs from the design), inaccurate input data for the ductwork elements, the presence of tonal components or the interaction between components creating additional turbulence and flow noise (see section 4).

4. DISCUSSION OF RESULTS

Generally, the L Aeq,n T at low frequencies (63 – 250 Hz) is determined by the fan sound, while at high frequencies (2 – 8 kHz), the flow noise of the ATD is the most important noise source. As a result, the accuracy at low frequencies depends on the prediction of the fan sound power level and the sound power level reduction of the network, which is mainly determined by the silencers, junctions, mani- folds and the end reflection of the ATD at low frequencies. The accuracy at high frequencies depends on the prediction of the ATD flow noise and thus on the prediction of the pressure losses.

In dwellings, the airflow rate in each room is generally controlled by turning the ATD in the correct setting point. When the system is well balanced and the ATD of the branch with the highest pressure loss is put completely open, all the ATDs will be relatively open and produce little flow noise. In this case, the measurements at high frequencies were often determined by the background noise. How- ever, when the network is not well balanced or the ATDs are not well adjusted during installation, the flow noise is often clearly noticeable and even determining the global L Aeq,n T .

worm 2022

L Aeq,n T [dB]

60

55

50

45

40

35

30

25

20

15

10

Wellness

Toilet 1

Toilet 2

Atelier

Study room

Storage room 1

Storage room 2

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 4

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 4

Bedroom 5

Study

Bathroom 1

Bathroom 2

Living room

Living room

Living room

Living room

Living room

Living room

Living room

Studio

Bathroom

Bathroom

Bathroom

Bathroom

Bathroom

Bathroom

Bathroom

Technical room

Technical room

Technical room

Open kitchen

Open kitchen

Open kitchen

Laundry room

Laundry room

Toilet

case 1 case 2 case 3 case 4 case 5 step 0 step 1 step 3 step 4

60

55

50

45

40

35

30

25

20

15

10

Laundry

Playroom 1

Playroom 2

Toilet 1

Toilet 2

Toilet 1

Toilet 2

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 4

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 2

Bedroom 1

Bedroom 2

Bedroom 1

Bedroom 2

Bedroom 3

Bedroom 1

Bedroom 1

Bedroom 2

Atelier

Bedroom 1

Bedroom 2

Living room

Living room

Living room

Bathroom 1

Bathroom 2

Technical room

Storage room

Toilet

Toilet

Toilet

Toilet

Technical room

Technical room

Technical room

Technical room

Technical room

Technical room

Open kitchen

Open kitchen

Open kitchen

Open kitchen

Open kitchen

Bathroom

Bathroom

Bathroom

Bathroom

Bathroom

Bathroom

Living room

Living room

case 6 case 7 case 8 case 9 case 10 case 11 case 12

Figure 2: Comparison between measured L Aeq,n T ( ● ) and predicted L Aeq,n T ( ♦ )

worm 2022

4.1. Accuracy of network model and input data In some cases, it was clear that the actual ventilation network differed from the design, e.g. because the location of the ATDs was different. When pictures were taken during installation, the exact layout of the duct network could be modeled. Otherwise, it was assumed that the actual network corresponds to the design, with estimated adjustments to account for the actual location of the ATDs. Generally, small differences between the actual and modeled geometry of the network don’t have a large influ- ence on the accuracy, because straight ducts and bends do not considerably contribute to the overall sound power level reduction of the network at low frequencies.

The presence and type of secondary silencers was not always indicated in the design and verifiable in situ . For case 11 and 12 (situated in the same building), for example, the measured ventilation noise levels in the bedrooms are much lower than in the other rooms and also much lower than the predicted values. Possibly, additional silencers were installed before the ATDs of the bedrooms.

Sometimes, critical input data are missing for the model, for example for bedroom 4 of case 2 where the noise was clearly dominated by the ATD flow noise. As no laboratory measurements were available for the ATD, the general formula of VDI was used, leading to an overestimation. Generally, VDI is on the safe side by overestimating the flow noise of ATDs used in dwellings. When the L Aeq,n T is dominated by the flow noise, the use of laboratory data of the ATD flow noise is therefore essential.

Another problem encountered, is the estimation of the Δ L w of non-acoustic flexible ducts. While the high-frequent insertion loss is negligible, there is a significant amount of damping at low frequen- cies due to duct breakout. As no measurement data are available, this was estimated based on the average difference between the insertion loss of flexible and rigid acoustic silencers. This gave good results for case 1 and case 9, but for the bathroom and toilet of case 8, the damping is strongly under- estimated.

As a rule of thumb, the laboratory insertion loss values of silencers at the lowest frequency bands seem also less reliable and less relevant for the actual insertion loss of the silencers as installed.

4.2. Tonal components For case 10, the measured L Aeq,n T -values at the nominal flow rate were determined by a tonal compo- nent in the one-third octave band of 315 Hz. Measurements were also made for a higher fan airflow rate of 400 m³/h in some of the rooms (living room, 3 bedrooms and bathroom 1). In this case, the tonal component was less pronounced in the measured L Aeq,n T - spectra and the prediction errors on the overall L Aeq,n T are significantly lower (average prediction error of 2.4 dB for 400 m³/h compared to 4.5 dB for nominal flow rate).

It can be expected that the prediction is less accurate for tonal components. First of all, the tonal component of the fan is strongly sensitive to fan speed. Because no fan sound power level data were available at the nominal flow rate, an interpolation was needed. While such an interpolation generally gives a good estimation of the broadband fan noise, the shift of tonal components in the fan power spectrum is impossible to account for. Secondly, the acoustic properties of the elements are only available in octave bands. These average values are appropriate for noise levels which are relatively constant in that frequency band. However, this is not the case for tonal components. Furthermore, the insertion loss of specific elements like silencers and manifolds can vary significantly with frequency, especially at lower frequencies where modal effects can be important (e.g. the wavelength at 315 Hz corresponds to the length of the silencers and the dimensions of the manifold used in the dwelling).

4.3. Turbulence When two fittings (like bends, junctions, valves or ATDs) are placed close to each other, the airflow will become more turbulent and noise generated in the fitting can be substantially higher. For exam- ple, when an ATD is placed immediately after a bend or a junction, the flow noise can increase by as

worm 2022

much as 12 dB according to ASHRAE [2]. Furthermore, there can be a strong interaction between the elements, influencing the Δ L w -value [5]. This additional flow noise and interaction is not taken into account in the acoustic model and may explain some of the larger prediction errors. For example, the high airflow rate in the junction placed immediately before the ATD of the study room of case 4 probably leads to strong turbulence and a significant increase of the ATD flow noise (estimated in- crease of 12 dB). The foam sound attenuators initially placed behind the ATDs in the living room of dwelling 3 (step 0) also led to a significant increase of the flow noise generated by the ATDs [7]. 5. CONCLUSIONS

The accuracy of a simplified prediction tool for mechanical ventilation noise based on EN 12354-5 and VDI 2081 has been investigated by comparing measurements and predictions for 12 dwellings. While more complex modeling techniques exist, the simplified model is generally sufficient to check the acoustic requirements of a mechanical ventilation system to be installed in a dwelling. One of the problems is that the exact lay-out of the ductwork is often not known in the design phase. Further- more, acoustic input data are not always available for every element in the ductwork. Therefore, a simplified prediction with generic prediction formula is usually the best option. For 75% of the rooms where the airborne duct sound was dominant, the global A-weighted noise level in the room can be estimated with an accuracy of ± 3 dB, incorporating the fact that the measurement error is ± 1 dB. Sometimes (< 5%), the noise is determined by alternative transmission paths (e.g. airborne transmis- sion from technical room, duct breakout or structure-borne transmission). While these contributions cannot be easily estimated without detailed knowledge of the building design, they can be avoided by following the general rules of thumb given in design guides like VDI, ASHRAE and ISSO. 6. ACKNOWLEDGEMENTS

This work has been supported by the Flemish Agency for Innovation and Entrepreneurship (VLAIO) in the Flux50 project Smart Ventilation (HBC.2020.2520). I would like to thank C. Delmez for the help with the aeraulic measurements. 7. REFERENCES

1. VDI 2081 Noise generation and noise reduction in air-conditioning systems (2001) 2. ASHRAE Handbook – HVAC Applications – Chapter 48: Noise and Vibration Control (2015) 3. ISSO-publicatie 24. Installatiegeluid – Ontwerpaanbevelingen en theoretische grondslagen t.b.v.

kleine utiliteit en onderwijsgebouwen (2019) 4. Bessac, F., Guigou-Carter, C., Bailhache, S. & Lefebvre, C. Ductwork noise calculations: main

outputs of AcouReVe project. Proceedings of 39 th AIVC conference . Juan-les-Pins, France, 18- 19 September 2018. 5. De Roeck, W. & Desmet, W. Indirect acoustic impedance determination in flow ducts using a

two-port formulation. Proceedings of 15 th AIAA/CEAS Aeroacoustics Conference , pp. 2831-2840, Miami, Florida, USA, 11-13 May 2009 6. Denayer, H. Flow-acoustic characterization of duct components using multi-port techniques. PhD

Thesis, KU Leuven (2017) 7. Caillou, S. & Dijckmans, A. Improvement of the acoustical performance of mechanical ventila-

tion systems in dwellings: a case study. Proceedings of 39 th AIVC conference . Juan-les-Pins, France, 18-19 September 2018.

worm 2022