Active control of compressor noise in the machinery room of … · 2020-02-07 · Noise Control...

Active control of compressor noise in the machinery room of refrigerators

a) School ofBusan, KO

b) School ofBusan, KO

c) School ofKOREA;

350

J.M. Kua), W.B. Jeongb) and C. Hongc)

(Received: 26 February 2019; Revised: 19 August 2019; Accepted: 30 August 2019)

The low-frequency noise generated by the vibration of the compressor in the ma-chinery room of refrigerators is considered as annoying sound. Active noise controlis used to reduce this noisewithout any change in the design of the compressor in themachinery room. In configuring the control system, various signals are measuredand analyzed to select the reference signal that best represents the compressornoise. As the space inside the machinery room is small, the size of a speaker is lim-ited, and the magnitude of the controller transfer function is designed to be small atlow frequencies, the controller uses FIR filter structure converged by the FxLMSalgorithm using the pre-measured time signal. To manage the convergence speedfor each frequency, the frequency-weighting function is applied to FxLMS algo-rithm. A series of measurements is performed to design the controller and to eval-uate the control performance. After the control, the sound power transmitted by therefrigerator is reduced by 9 dB at thefirst dominant frequency (408 Hz in this case)and 3 dB at the second dominant frequency (459 Hz here), and the overall soundpower decreases by 2.6 dB. Through this study, an active control system for thenoise generated by refrigerator compressors is established. © 2019 Institute ofNoise Control Engineering.

Primary subject classification: 38.2; Secondary subject classification: 12.4.6.1

1 INTRODUCTION

As people's perception of living standards and noiseincreases, interest in sound comfort and reduced noisefrom home life is increasing. Many studies to reduce thenoise generated from household appliances have been con-ducted1,2. In particular, since the refrigerator operates allday, researchers are trying to reduce various noise gener-ated by the refrigerator. Kim et al.3,4 reduced the noise gen-erated by the refrigerant by changing the structure of therefrigerant pipe and created a noise map of the refrigerantto predict flow noise, and Heo et al.5,6 researched the bladepass frequency (BPF) noise of the centrifugal fan of the re-frigerator and changed the design to reduce the noise.

However, among noises generated by a refrigerator,the low-frequency compressor noise is frequently describedas an unpleasant noise7, and the noise reduction cannot beeffectively achieved by passive method, e.g., sound absorp-tion and/or damping. Therefore, studies were conducted toreduce the noise by changing the structure of themachineryroom, the shape of the compressor, and the mounts8–10.

Mechanical Engineering, Pusan National University,REA; email: [email protected] Engineering, Pusan National University,REA; email: [email protected] Engineering, Ulsan College, Ulsan,Corresponding author. Email: [email protected].

Noise Control Engr. J. 67 (5), September-October 2019

Kim et al.11 reduced the noise by modifying the stiffnessand mass of the shell of a reciprocating compressor, andLee et al.12 modified the shape of the joint of the dischargecase of the linear compressor to reduce the noise.

In addition to ineffective reduction performance ofpassive methods at low frequencies, the passive methodsredesigning the compressor and/or the machinery roomrequire high costs. Without the change of the current de-sign, the compressor noise can be controlled using activenoise control effectively at low frequencies.Many research-ers implemented noise attenuation using active control forvarious applications13–19, but it is hardly found that activecontrol is applied for refrigerator application.

In similar studies, active noise control was applied toenclosure structures20,21. Lee et al.22 attempted to controlthe noise propagated through the opening of enclosurestructures surrounding the sound source using feedforwardcontrol. Ji et al.23 also tried to control the noise propagatedfrom the enclosure structure using causal finite impulse re-sponse (FIR) filter with an additional time delay accordingto the harmonics fundamental frequency. In two studies onenclosure structures, the entire transfer function betweenthe disturbance and the reference signal within the fre-quency range of interest can be measured. Also, the loud-speaker size is not limited to satisfy the space of theenclosure, and the controller can be designed regardlessof the target frequency. Nevertheless, in the case of a refrig-erator machinery room, a compressor, pipe, and blower fan

Published by INCE/USA in conjunction with KSNVE

are disposed therein. Thus, space for a control loudspeakerhas to be small, and a method of measuring transfer func-tions between the reference and disturbance signals is noteasy. Under this constrained condition, Koo et al.24 mea-sured the transfer function when the refrigerant and themachinery room fan were not operating. However, in thecase of a real refrigerator, the transfer function of the noisesource cannot be secured by the method as in Ref. 24 be-cause the transfer function changes with time.

In this study, a reference signal and a loudspeaker areselected for the practical control system, and the con-troller is designed to reduce the acoustic power radiatedfrom the machinery room. A time domain controller de-sign method is utilized, and the design method of thecontroller is studied when the size of the loudspeakeris small because of the limited space inside the machineryroom. To design the controller, the transfer function of theloudspeaker is measured, and an FIR filter is designedusing the sound pressure and acceleration time signal.Experiments are carried out by installing the designedcontroller on the digital signal processor (DSP) equip-ment, and the control performance is verified.

2 CONTROL SYSTEM CONFIGURATION

2.1 Selection of the Reference Signal

Figure 1 is a schematic of an active control system forthe compressor noise generated in a refrigerator machineryroom. In the compressor located on the left side of the ma-chine room, noise is generated owing to vibrations, and thenoise is transmitted through the left window and the rightwindow. The light gray squares inside the machinery roomrepresent the window of each side. The compressor oper-ates at an operating frequency of 51 Hz, and the spectrumof the sound pressure when operating at that frequency isshown in Fig. 2. The sound pressure generated by the com-pressor is large at 408 Hz and 459 Hz. Therefore, the con-trol target frequencies are selected as 408 Hz and 459 Hz.

To reduce this noise component, a feedforward controlsystem is chosen; the control system consists of a reference

Machinery Room

Error SensorControl

SpeakerCompressor

Reference Sensor

DSP

Fig. 1—Schematic diagram of activecontrol system for machineryroom of refrigerators.


signal sensor, an actuator, an error sensor, and a controller.In this section, reference signal sensors, actuators, and er-ror sensor selection are described. The controller designis discussed in the next section.

To find an appropriate reference signal, we measuredvarious signals related to the refrigerator noise. As can-didates for the reference signals related to the noise, thedriving voltage signal to the compressor, signals measuredby an accelerometer attached to the compressor shell, andsignals of the accelerometer attached to the machine win-dow were selected.

The transfer function and the coherence of the distur-bance for the three reference signals were calculated, andtime-dependent changes of the transfer function at controlfrequencies of 408 Hz and 459 Hz are shown in Fig. 3 andFig. 4, respectively. When the voltage used to drive thecompressor is the reference signal, it can be seen inFig. 3 and Fig. 4 that the fluctuation width of the mag-nitude and phase of the transfer function is large and the co-herence is low. If the accelerometer signal attached to themachine room window is the reference signal, it can beseen that the transfer function also fluctuates and the coher-ence is not enough for the control. When the accelerometersignal attached to the compressor shell is the reference sig-nal, the variation of the transfer functionwith time is almostzero, and the coherence is always higher than 0.95. There-fore, the accelerometer attached to the compressor shellwasselected as the reference signal sensor to control the noisegenerated by the compressor.

Figure 5 shows the components of the machine room tobe controlled. A loudspeaker is installed on the left frontpart near the window, and a small loudspeaker is selectedbecause of the internal space problem. A reliable acceler-ometer in the frequency band of interest was selected withcalibration data. The coherence between the accelerationsignal and the sound pressure outside the window wascompared with each location, and the accelerometer wasattached to the upper right part of the compressor shellwith the highest coherence.

2.2 Selection of the Actuator

In order to perform active noise control, it is importantthat the sound pressure generated by the speakers is line-arly proportional to the applied voltage to achieve the con-trol performance. However, as large voltage is applied tothe loudspeaker relatively to the their size, the transferfunction of the loudspeaker can be nonlinear25. Therefore,the loudspeaker performance should be considered for de-signing the controller to exhibit the control performance.

There was not enough space in the machinery room, itcannot be possible to choose loudspeaker with appropri-ately designed cabinet. The global control performancewould be improved with a monopole-like secondary loud-speaker having the cabinet.However, ifweuse the loudspeaker

351Published by INCE/USA in conjunction with KSNVE

0 200 400 600 800 1000-10

0

10

20

30

40

50

Frequency(Hz)

Pre

ssur

e(dB

A)

Fig. 2—Sound pressure spectra of the compressor noise.

having the cabinet which should be smaller loudspeakerto fit the space, it would be difficult to satisfy that thedriver works in the 400 Hz band, which is the main tar-get frequency of this control. Thus, in this article, the 3-inch loudspeaker shown in Fig. 6(a) was selected. Thefrequency response of the loudspeaker with white noiseexcitation is shown in Fig. 6(b). Since the resonant fre-quency of the control source is 260 Hz, this device is notdesigned to work below this frequency. Peaks below260 Hz are noise amplification. This can be confirmedby the coherence of Fig. 6(c).

Generally, the frequency response of the loudspeaker ismeasured using swept sine method. Because this method

0 100 200 30

2

4

Mag

nitu

de

0 100 200 3-200

-100

0

100

Pha

se(D

egre

e)

0 100 200 30.8

0.9

1

Coh

eren

ce

Tim

Fig. 3—Time history of the value of transfer functiosignals to the error sensors: input voltage(green dashed), and machinery room wind

352 Noise Control Engr. J. 67 (5), September-October 2019

ensures a separation of linear and non-linear part of thesound generation, this allows a more accurate measure-ment of the frequency response function. Nevertheless,there are two reasons for measuring the frequency responseusing white noise in this article. First, this procedure wasconsidered for mass-produced refrigerators. White noisewas considered appropriate to reduce production timesince system identification would be required for tuning onthe production line. Second, as shown in Fig. 6(c), we canobtain frequency response with high coherence in the con-trol band even with white noise. This is because whitenoise with low RMS value is applied to the loudspeaker,resulting in a small non-linear part.

00 400 500 600

00 400 500 600

00 400 500 600e(s)

n and coherence at 408 Hz from the reference(blue solid), compressor shell accelerationow acceleration (black dotted).


Pha

se(D

egre

e)

00

2

4

Mag

nitu

de

0-200

0

200

00.8

0.9

1

Coh

eren

ce

100

100

100

200

200

200

300

300

300Time(s)

400

400

400

500

500

500

600

600

600

Fig. 4—Time history of the value of transfer function and coherence at 459 Hz from the referencesignals to the error sensors: input voltage (blue solid), compressor shell acceleration(green dashed), and machinery room window acceleration (black dotted).

Fig. 5—Component inside machinery room.

2.3 Selection of the Error Sensor

A microphone is used as an error sensor for evaluatingthe active control performance by the controller. Micro-phones are selected as instruments capable of measuringthe frequency range of interest according to calibrationdata. The position of the error sensor is selected as thecenter point of the left and right windows of the machineroom window. There are slits on the left, slits on the right,and other small holes in the noise path of the machineroom. However, except for the left slit and the right slit,others are excluded from the controller design becausethe cross-section of those holes is small and the contribu-tion to the noise transmission is low. The error sensor isplaced at the center of both windows because the transferfunction change according to the error sensor position inthe window of the machine room is not significant withinthe frequency of interest.

3 CONTROLLER DESIGN THEORY

3.1 FIR Filter Design Using theFxLMS Algorithm

In general, the filtered-x least mean squared (FxLMS)algorithm converges to a controller by minimizing an er-ror signal using a reference signal x(k) and an error sig-nal e(k) measured in real time. However, in this article,the controller used in the experiment was designed asoff-line using the signal measured in advance. That is, theFxLMS algorithm was only used to design the control-ler, and the use of microphones is eliminated in actual


experiments. Thus, the actual control experiments wereconducted in a simple feedforward system without a feed-back loop. The controller is converged by the FxLMS algo-rithm by using an acceleration signal and a sound pressuresignal, which are measured in advance.

The block diagram of the FxLMS algorithm is shownin Fig. 7. The reference signal x(k) generates a disturbanced(k) through the plant HP(z), and the control source y(k)is generated through the controller C(z) updated with anLMS filter and the secondary path transfer function Hs(z)of the control loudspeaker. The control source and the dis-turbance act in opposite phases. The LMS filter receivesthe filtered reference signal x ′ (k) and the error signale(k), which is the difference between the disturbance andthe control signal at the error microphone. Then, filter con-verges the controller C(z) in a direction to minimize the


c

ba

Fig. 6—Experimental setup (a), frequency response (b) and the coherence (c) of the soundpressure to the voltage applied to the loudspeaker.

Fig. 7—Block diagram of FxLMS algorithm.

square of the error signal. The convergence algorithm ofthe controller uses the gradient steepest descent method.26

The equation for updating the coefficient of the con-troller can be expressed as Eqn. (1).

C k þ 1ð Þ ¼ C kð Þ � m@J

@Ckð Þ ð1Þ

Here, k is sample index, and m is the convergence co-efficient of the controller, where the cost function J =e(k)2 and the error signal e(k) is expressed as:

e kð Þ ¼ d kð Þ � y kð Þ � Hs kð Þ ð2Þwhere * denotes linear convolution, y(k) = [y(k) y(k� 1)⋯y(k � L + 1)], y(k) is filter output and L is the filter lengthwhen Hs(z) is expressed in the form of an FIR filter.

When Eqn. (2) is substituted into Eqn. (1), the control-ler coefficients of the next discrete time converge as shownin Eqn. (3).

C k þ 1ð Þ ¼ C kð Þ þ 2mx′ kð Þe kð Þ ð3ÞThe larger the value of m, the faster the filter coefficient

converges, but the filter coefficient diverges when mis larger than a certain value. Therefore, the value ofm should be appropriately selected within the follow-ing range27:

0 < m <2

Sk k2 s2x Lþ 2Deq

� � ð4Þ

where s2x is the reference signal power, ‖S‖ denotesthe norm of the secondary path FIR coefficients, L is thefilter length and Deq is equivalent delay parameter of thesecondary path.


3.2 Frequency-weighted FxLMS Algorithmfor Controller Convergence Rate Control

Generally, the FxLMS algorithm can be used to designan FIR filter that minimizes the acoustic power measuredby an error sensor. However, when using the FxLMS algo-rithm, it converges from the frequency which have highpower of the disturbance signal and the reference signal. Ifthe response of the converged FIR filter is large at a low fre-quency and the loudspeaker performance at a low frequencyis not sufficient for generating a control source, harmonicdistortion may occur at the control frequency. In order toavoid this situation, an appropriate frequency-weighting fil-ter is applied to the reference signal and the disturbance sig-nal to slow down a convergence rate of the controller at alow frequency and to increase the convergence speed ofthe controller at the target frequency.

Figure 8 shows a block diagram of the FxLMS algo-rithm with a frequency-weighting filter. W(z) is the FIRfilter coefficient of the frequency-weighting function. Sincethe reference signal and the disturbance signal throughW(z) have a high magnitude in the target frequency band,the control coefficients converge faster in the target band.In addition, since the weighting function is applied to bothsignals, noise can be controlled without distortion of the


Fig. 8—Block diagram of frequency weightedFxLMS algorithm.

Table 1—Reference signal and pressure magnitudeof 150 Hz harmonics to verify effectof frequency weighting function.

Frequency (Hz) 150 Hz 300 Hz 450 Hz

Reference magnitude (V) 0.1 V 0.01 V 0.001 VPressure magnitude (Pa) 0.01 Pa 0.001 Pa 0.01 Pa

system.Moreover, by using the frequency-weighting func-tion filter, it is possible to increase the convergence speedof the controller at a frequency where the magnitude ofthe both signals are small.

Numerical simulation is conducted to verify the effect ofthe frequency-weighting function. The reference signal andthe disturbance signal are harmonic signals having a fun-damental frequency of 150 Hz. The magnitudes of eachfrequencies of the reference and disturbance signals forconvergence through the FxLMS algorithm are shown inTable 1. To set a different convergence speed at each fre-quency, the magnitude difference between 150 Hz and theother two frequency are set to be 10 times or more.

After convergence of the controller, themagnitude of theresponse of the controller without applying the frequency-weighting function is shown in Fig. 9(a). Also, the magni-tude of the reference signal is shown in Fig. 9(b), and thesound pressure before and after using this controller isshown in Fig. 9(c).

0 100 200 300 400 500

Frequency(Hz)

0

2

4

Magnitude

0 100 200 300 400 500

Frequency(Hz)

-20

0

20

40

Sig

nal M

agnitude(d

B r

ef 1)

b

a

Fig. 9—Before applying frequency weighting funct(b) Magnitude of the reference signal, and


In the case of the controller converged with the FxLMSalgorithm without the weighting filter, the controller con-verges only at 150 Hz, which is large in both signals. Evenat 300 Hz, the controller slightly converges, showing somecontrol performance when compared to that before and af-ter the control. However, in the case of the 450 Hz compo-nent, the controller hardly converges. In Fig. 9(b), themagnitude of the control signal at 150 Hz is too large.Therefore, when a small loudspeaker is used in an ac-tual control experiment, the control performance maybe adversely affected by the non-linear part of the lowfrequencies.

To overcome this phenomenon, the converged con-troller with the weighting filter of Fig. 10 has to be used.The frequency-weighting function consists of non-zerovalues at 300 Hz and 450 Hz in the frequency domain.The magnitude of each sinusoids are 0.5. The frequencyweighting was implemented as an FIR filter, and FIR filterwas designed by the frequency sampling method. Thesame procedure is repeated with the frequency-weightingfunction. The applied result is shown in Fig. 11. As a resultof weighting function, the filtered reference signal and the

0 100 200 300 400 500

Frequency(Hz)

-40

-20

0

20

40

60

80

So

un

d P

ressu

re(d

B r

ef

2.0

e-5

Pa

)

Before Control

After Control

c

ion: (a) Magnitude of the controller,(c) Sound pressure spectra.


0 50 100 150 200 250 300 350 400 450 500

Frequency(Hz)

0

0.2

0.4

0.6

Magnitude

Fig. 10—Frequency weighting function to adjust convergence rate at 150 Hz, 300 Hz, and 450 Hz.

disturbance signal are reduced at frequencies other than thetwo frequencies.

The control performance is high at target frequenciesof 300 Hz and 450 Hz, and the amplitude of the controlsignal at 150 Hz is small. Therefore, it is possible tocontrol the convergence speed of the controller for eachfrequency band by applying an appropriate frequency-weighting function. By this method, it is prevented thata high voltage at a low frequency is applied to the smallloudspeaker through the method of converging the con-troller above 250 Hz.

4 EXPERIMENT OF SOUND POWERCONTROL

4.1 Experiment Equipment andConfiguration

Figure 12 shows the experimental setup for the control-ler design and the acoustic power measurement before and

0 100 200 300 400 500

Frequency(Hz)

0

10

20

30

Magnitude

0 100 200 300 400 500

Frequency(Hz)

-20

0

20

40

Sig

nal M

agnitude(d

B r

ef 1)

b

a

Fig. 11—After applying frequency weighting funct(b) Magnitude of the reference signal, an


after the control. The internal configuration of the machin-ery room is shown in Fig. 5. The reference signal and theerror signal were measured in an anechoic chamber to de-sign the active controller. The position of the microphoneis shown in Fig. 12(a). It is installed at the center of the leftwindow and the right window outside the machine roomcover. A charge-type accelerometer is used to measurethe acceleration value as the reference signal for designingthe controller. B&K Nexus conditioning amplifier wasused to connect the acceleration signal measured by the ac-celerometer to the dSpace RTI 1103 in the form of voltage.To reduce aliasing occurring in the DSP, the accelerationsignal was passed through a low-pass filter with a cut-offfrequency of 1000 Hz28. Low-pass filter SR640 of theStandard Research System was used, and LMS SCADASIII was used to measure the reference signal and the soundpressure. The schematic diagram of this procedure isshown in Fig. 13. The sampling time of measured time sig-nal and applied signal to DSP was set to 1/5120 s, and this

0 100 200 300 400 500

Frequency(Hz)

-40

-20

0

20

40

60

80

So

un

d P

ressu

re(d

B r

ef

2.0

e-5

Pa

)

Before Control

After Control

c

ion: (a) Magnitude of the controller,d (c) Sound pressure spectra.


a Anechoic chamber b Reverberant chamber

Fig. 12—Experimental setup for designing controller (a) and for measuring sound powerdifference before/after control (b).

time step has a reduction of more than 30 dB at the fre-quency at which aliasing would occur.

The performance of the designed controller was evalu-ated in both anechoic chamber and reverberation cham-ber. First, the sound pressure before and after the controlwas measured on both windows of the machinery roomand on evaluation point of the refrigerator. After that, tomeasure the acoustic power control performance of thedesigned controller, the refrigerator was moved to the re-verberation chamber. The change in the acoustic powerbefore and after the control was measured by the rever-beration method. The installed refrigerator and micro-phone are shown in Fig. 12 (b).

4.2 Frequency Response FunctionMeasurement and Controller Design

The frequency response function (FRF) of loudspeakerHs(z) is the sound pressure at the error sensor position rel-ative to the voltage signal applied to the loudspeaker. Thesound pressure and voltage were measured to calculatethe FRF. The measurement point of sound pressure is thesame as the microphone position in Fig. 12(a). Using thefunction generator, white noise with 1 V of peak valuewas generated and applied to the speaker. During the mea-surement, both the fan and the compressor in the refriger-ator were stopped. The frequency response function wascalculated by dividing the sound pressure of microphone

Fig. 13—Schematic diagram to measuring acceler


by voltage applied to the loudspeaker. Averaging methodwas used to reduce measurement noise. The number ofaverages is 100, and the overlap percentage is 99%. Weset the sampling frequency to be 5120 Hz and the fre-quency resolution to be 2.5 Hz. The measurement resultsof the FRF are shown in Fig. 14. The magnitudes of theloudspeaker FRF Hs(z) are sufficient at above 250 Hz asin Sec. 2.2, and it is confirmed that the phase is not accu-rately measured owing to low performance at frequenciesbelow 250 Hz.

Because of this characteristic, the controller transferfunction of the corresponding frequency should be set tonearly 0, so that the control sound pressure is not generatedbelow 250 Hz. Also, the magnitude of controller transferfunction should not be large at a frequency of 250 Hz orgreater.

In order to control the noise of the compressor, thecontroller was designed by measuring the time signalusing the equipment described in Sec. 4.1. The refer-ence signal and the disturbance signal were measuredduring the operation of the refrigerator. The spectrumof the measured signal is shown in Fig. 15 along withthe spectrum of the signal using the frequency-weightingfilter. The weighting function consists of frequenciesof 408, 459, and 500–800 Hz, where the frequencies areharmonic components of 51 Hz. The weighting functionis applied in the form of FIR filter, and the filter lengthis 2048. Also, it is designed by frequency samplingmethod.

ation signal for using reference signal.


0 100 200 300 400 500 600 700 800 900 10000

0.1

0.2

0.3

0.4

Mag

nitu

de(P

a/V

)

0 100 200 300 400 500 600 700 800 900 1000-200

-100

0

100

200

Frequency(Hz)

Pha

se(D

egre

e)

Fig. 14—Magnitude and phase of the transfer function to the control speaker at left window(solid)and right window (dashed).

Time-domain and frequency-domain response of theweighting function are shown in Fig. 16. It can be seen thatboth signals are large only at the frequency to which theweighting function is applied.

The controller is converged using the FxLMS algorithm,which is described in Sec. 3, for time signals with andwithout the application of the weighting filter. The lengthof the FIR filter to be converged is set to 2048. The fre-quency response of the converged controller is shown inFig. 17. In the case without weighting function, all ofthe harmonic components of the operating frequencyof 51 Hz have magnitude values. However, the responseof the controller using the weighted time signal has ampli-tudes of 408 Hz and 459 Hz, which are the control target

0 200 400 600 800 1000-80

-60

-40

-20

0

20

40

Frequency(Hz)

Vol

tage

(dB

ref

1V

)

No weighting

Weighting

Sou

nd P

ress

ure(

dB r

ef 2

.0e-

5Pa)

a b

Fig. 15—Spectra of reference signal (a) and disturweighting function.


frequencies, and it is confirmed that the magnitude alsoappears between 600 Hz and 800 Hz, which is the targetof theweighting function. By applying theweighting func-tion, the amplitude of the controller can be made small atbelow 250 Hz, so it can prevent a situation where a highvoltage is applied to the loudspeaker at a low frequency.

4.3 Control Experiment and Result

The control performance of the refrigerator was mea-sured when the controller was turned on/off, and the mea-surement was conducted in the anechoic room and thereverberation room. The instrument setup during themeasurement is shown in Fig. 12.

0 200 400 600 800 1000-20

0

20

40

60

80

100

120

Frequency(Hz)

bance signal (b) with/without frequency


0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

Time(s)

-5

0

5

Filt

er

Co

eff

icie

nt

0 200 400 600 800 1000

Frequency(Hz)

0

0.5

1

We

igh

tin

g F

acto

r

a

b

Fig. 16—Time-domain (a) and frequency-domain response (b) of the FIR filter forfrequency weighting.

Figure 18 shows the experimental results of the controlusing the designed controller described in Sec. 4.2. Thesound pressure in the left window near the noise sourcecompressor is reduced by about 9 dB at 408 Hz and by17 dB at 459 Hz. The sound pressure measured in the rightwindow decreased by 6 dB at 408 Hz and increased by3 dB at 459 Hz. The peaks of the mid-frequency noiseare generated by the machine room fan. Because this fanis closer to the right window, the mid-frequency noisemeasured in the right window is louder than left window.Since this noise is relatively smaller than the compressornoise, there is little influence on the change in the overallnoise after the control.

0 100 200 300 400 50

0.5

1

Magnitude

0 100 200 300 400 5-200

0

200

Freque

Phase(D

egre

e)

Fig. 17—Frequency response of controller by usingwith/without frequency weighting functio


In addition, in general, the noise of the refrigerator isevaluated at a distance of 1 m from the back of the refrig-erator. The change in the sound pressure before and afterthe control is plotted in Fig. 19. At the rear at 1 m29, themaximum sound pressure is at 459 Hz, and the soundpressure decreases by 5 dB after the control. The con-trol performance of the 408 Hz component is measuredat 2 dB.

Finally, the control performance was verified by mea-suring the change of the sound power before and after thecontrol in the reverberation room. The change in the acous-tic power before and after the control is shown in Fig. 20for the 1/3 octave band. The control frequencies of 400 Hz

00 600 700 800 900 1000

00 600 700 800 900 1000

ncy(Hz)

No weighting

Weighting

measured time data through FxLMS algorithmn.


0 200 400 600 800 1000-10

0

10

20

30

40

50

Frequency(Hz)

Pre

ssur

e(dB

A r

ef. 2

0µP

a)

Left window

Before Control

After Control

0 200 400 600 800 1000-10

0

10

20

30

40

50

Frequency(Hz)

Sou

nd p

ress

ure(

dBA

ref

. 20µ

Pa)

Right window

Fig. 18—Control performance at the left and the right window before/after control.

and 500 Hz are reduced by 9 dB and 3 dB, respectively, andthe overall acoustic power is reduced by 2.6 dB.

5 CONCLUSION

In this study, a method for designing an active controllerfor mass-product, limited-space refrigerator is studied. Thenoise is generated mainly by the compressor in a refrigera-tor machinery room. The time domain method is used todesign the controller. To adjust the convergence speed,the time signal is filtered using the frequency-weightingfunction. The designed controller is implemented intoDSP equipment. A series of experiments is conducted toverify the performance. The following conclusions arethen obtained:

1) Acceleration of the compressor shell is the most

Fig.

360

suitable reference signal for control system of refrig-erator. Various signals are compared as a referencesignal that correlated with the compressor noise.The coherence and the transfer function are con-sidered for the most appropriate reference signal.The coherence between the acceleration and the

100

125

160

200

250

315

Centre frequency of th

5

10

15

20

25

30

35

Pre

ssure

(dB

A r

ef. 2

0P

a)

Before Control

After Control

19—Control performance at 1 m behind from re


sound pressure is always high regardless of time,and the transfer function is hardly changed overtime. As a result, the corresponding signal is selectedas the reference signal.

2) Frequency-weighting filter is suggested for adjustingconvergence speed at each frequency of FxLMS al-gorithm when using small loudspeaker. The spaceinside the refrigerator is narrow, only small loudspea-kers are available for the control. As a result, the con-trol frequency is limited to 200 Hz or greater. Toimplement these constraints in the controller design,the control target frequency is limited by applying afrequency-weighting filter to the reference signaland the disturbance signal. Simulation and exper-iment are conducted for verifying performance,and the result of these can prove effectiveness offrequency-weighting function.

3) The control performance can be confirmed throughthe experiment in an anechoic room and a reverber-ation room. In the case of the control performancemeasured in the anechoic chamber, the sound pres-sure at thewindow near the compressor is greatly re-duced. On the far side, the sound pressure decreases

400

500

630

800

100

0

e 1/3 octave band

frigerator in the 1/3 octave band.


100

125

160

200

250

315

400

500

630

800

100

0

125

0

160

0

200

0

250

0

315

0

400

0

500

0

630

0

800

0

1000

0to

tal

Centre frequency of the 1/3 octave band

10

20

30

40

So

un

d P

ow

er(

dB

re

f 1

0-1

2W

)

Before Control

After Control

Fig. 20—Control performance of sound power in the 1/3 octave band measured in thereverberation room.

Noise C

at 408 Hz but slightly increases at 459 Hz. However,the sound pressure measured at the rear 1 m point,which is the method of evaluating the refrigeratornoise in the industrial field, decreases at both fre-quencies. Moreover, the sound power in the reverber-ation chamber decreases in the 400-Hz and 500-Hzoctave bands. Talking about mid-range frequencynoise, this band is dominantly influenced by the fan,so there is no control effect. However, the compres-sor noise has a higher contribution to overall noise.Thus, after control, the sound power is reduced byabout 3 dB.

Through this study, a method to reduce the low-frequency noise generated by a refrigerator compressorby using a loudspeaker of a limited size is established.As a method of designing the controller, a method ofusing a frequency-weighting function filter is presented,and the convergence speed changes according to the pres-ence or absence of a filter and its effect are confirmed. Inaddition, the acoustic power reduction performance of themachinery room is verified through the control experimentusing an actual refrigerator. In the future, we will study thecontrol performance change according to the number oftaps of the filter required for refrigerator commercializationand control model establishment for refrigerator control atvarious operating frequencies.

6 ACKNOWLEDGMENTS

This work was supported by the 2018 Research Fundof Ulsan College.

7 REFERENCES

1. F. Gue, C. Cheong, and T. Kim, “Development of low-noise axialcooling fans in a household refrigerator”, Journal of MechanicalScience and Technology, 25(12), 2995–3004, (2012).

2. Y. Xia, Y. Liu, Y. Liu, Y. Ma, C. Xiao, and T. Wu, “Experimentalstudy on reducing the noise of horizontal household freezers”,Appl. Therm. Eng., 68(1–2), 107–114, (2014).

ontrol Engr. J. 67 (5), September-October 2019

3. M.S. Kim, W.B. Jeong, and H.S. Han, “Development of noisepattern map for predicting refrigerant-induced noise in refrig-erators”, Journal of Mechanical Science and Technology, 28(9),3499–3510, (2014).

4. M.-S. Kim, H.-S. Han, T.-H. Kim, and W.-B. Jeong, “Reductionof Refrigerant-induced Noise of the Refrigerator by Modificationof the Evaporator Inlet Pipe”, Transactions of the Korean Societyfor Noise and Vibration Engineering, 19(10), 1012–1020, (2009).

5. S. Heo, C.-U. Cheong, M.-Y. Seo, and S.-R. Kim, “Applicationof hybrid techniques for the prediction of internal BPF noise ofcentrifugal fan in refrigerators and its reduction”, in: Proceed-ings of the Korean Society for Noise and Vibration EngineeringConference, The Korean Society for Noise and Vibration Engi-neering, 2009.

6. S. Heo, C. Cheong, and T.-H. Kim, “Development of low-noisecentrifugal fans for a refrigerator using inclined S-shaped trailingedge”, International Journal of Refrigeration, 34(8), 2076–2091,(2011).

7. K. Kyung-Soo, J. Weui-Bong, K. Tae-Hoon, S. Dae-Sik, andA. Se-jin, “Psychoacoustical Study for Sound Quality Indexof Steady State Noise of Refrigerators”, Transactions of theKorean Society for Noise and Vibration Engineering, 26(5),536–545, (2016).

8. V. Done, V. Balide, B. Tamma, K. Soni, S. Dey, and S. Angadi,“Muffler Design for a Refrigerator Compressor”, (2014).

9. J.-l. Guo, J. Luo, Y. Guo, X.-y. Pan, X. Fang, and X. Wu, “Noisetest and control for household refrigerator compressor”, in: Infor-mation and Automation, 2015 IEEE International Conference on,IEEE, 2015, pp. 112-115.

10. H.C. Kim, M.G. Cho, J. Kim, J.H. Park, and J. Shim, “Coherencetechnique for noise reduction in rotary compressor”, Journal ofMechanical Science and Technology, 26(7), 2073–2076, (2012).

11. Y.-T. Kim, J.-W. Lee, and J.-M. Joo, “Reduction of RadiatedNoise in aReciprocatingCompressor”, in: Proceedings of theKoreanSociety for Noise and Vibration Engineering Conference, TheKorean Society for Noise and Vibration Engineering, 2005.

12. J.-H. Lee, W.-B. Jeong, and D.-J. Kim, “Flow-induced noiseanalysis and its reduction design of discharge system in linearcompressor”, in: Proceedings of the Korean Society for Noiseand Vibration Engineering Conference, The Korean Societyfor Noise and Vibration Engineering, 2009.

13. M. Misol, S. Algermissen, and H.P. Monner, “Experimental inves-tigation of different active noise control concepts applied to a pas-senger car equipped with an active windshield”, J. Sound Vib.,331(10), 2209–2219, (2012).

14. S. Hu, R. Rajamani, and X. Yu, “Invisible speakers in home win-dows for simultaneous auxiliary audio playback and active noisecancellation”, Mechatronics, 22(8), 1031–1042, (2012).

15. B. Kwon and Y. Park, “Interior noise control with an active win-dow system”, Appl. Acoust., 74(5), 647–652, (2013).


16. J. Cheer and S.J. Elliott, “Active noise control of a diesel generatorin a luxury yacht”, Appl. Acoust., 105, 209–214, (2016).

17. N. Lee, Y. Park, and G.W. Lee, “Frequency-domain active noisecontrol for magnetic resonance imaging acoustic noise”, Appl.Acoust., 118, 30–38, (2017).

18. I.-H. Yang, J.-E. Jeong, U.-C. Jeong, J.-S. Kim, and J.-E. Oh,“Improvement of noise reduction performance for a high-speedelevator using modified active noise control”, Appl. Acoust., 79,58–68, (2014).

19. L. Liu, K. Beemanpally, and S. M. Kuo, “Application of multi-channel hybrid active noise control systems for infant incubators”,Noise Control Engineering Journal, 61(2), 169–179, (2013).

20. M.R. Bai and S. Chang, “Active noise control of enclosed har-monic fields by using BEM-based optimization techniques”,Appl. Acoust., 48(1), 15–32 (1996).

21. Y. Sun, T. Yang, M. Zhu and J. Pan, “The use of an active con-trolled enclosure to attenuate sound radiation from a heavy ra-diator”, J. Sound Vib., 392, 1–17, (2017).

22. H. Lee, C. Hong, and W. Jeong, “Active Control of PropagatedNoise through Opening of Enclosures Surrounding a NoiseSource”, Transactions of the Korean Society for Noise andVibration Engineering, 25(4), 223–231, (2015).


23. S. Ji, C. Hong, W.-B. Jeong, and T.-H. Kim, “Active Controlof Noise Propagated through Ventilation Openings of Enclo-sures Using an FIR Filter”, Transactions of the Korean Societyfor Noise and Vibration Engineering, 25(3), 191–198, (2015).

24. J.M. Koo, W.B. Jeong, T.H. Kim, and C. Hong, “Active Con-trol of Noise Transmitted through Ventilation Openings of theMachinery Room of Refrigerator”, Transactions of the KoreanSociety for Noise and Vibration Engineering, 26(4), 475–482(2016).

25. W. Klippel, “Assessing large signal performance of transducers”,Klippel GmbH, (2003).

26. P.A. Nelson and S.J. Elliott, Active Control of Sound (AcademicPress, Cambridge, Massachusetts, 1991).

27. I. Tabatabaei Ardekani and W.H. Abdulla, “Theoretical conver-gence analysis of FxLMS algorithm”, Signal Processing, 90(12),3046–3055, (2010).

28. S.M. Kuo and D. Morgan, Active noise Control Systems:Algorithms and DSP Implementations (John Wiley & Sons,Inc., Hoboken, New Jersey, 1995).

29. J.Y. Jeon, J. You, and H.Y. Chang, “Sound radiation and sound qual-ity characteristics of refrigerator noise in real living environments”,Appl. Acoust., 68(10), 1118–1134 (2007).


Active control of compressor noise in the machinery room of … · 2020-02-07 · Noise Control...

Documents

Transcript of Active control of compressor noise in the machinery room of … · 2020-02-07 · Noise Control...