Measurement of Field Complex Noise Using a Novel Acoustic Detection System

Transcription

1 Southern Illinois University Carbondale OenSIUC Conference Proceedings Deartment of Electrical and Comuter Engineering Fall 04 Measurement of Field Comlex Noise Using a Novel Acoustic Detection System Jun Qin Southern Illinois University Carbondale, jqin@siu.edu Pengfei Sun Southern Illinois University Carbondale, sunengfei@siu.edu Jacob Walker Southern Illinois University Carbondale, walkerjacob393@yahoo.com Follow this and additional works at: htt://oensiuc.lib.siu.edu/ece_confs Recommended Citation Qin, Jun, Sun, Pengfei and Walker, Jacob. "Measurement of Field Comlex Noise Using a Novel Acoustic Detection System." (Fall 04). This Article is brought to you for free and oen access by the Deartment of Electrical and Comuter Engineering at OenSIUC. It has been acceted for inclusion in Conference Proceedings by an authorized administrator of OenSIUC. For more information, lease contact oensiuc@lib.siu.edu.

2 Measurement of Field Comlex Noise Using a Novel Acoustic Detection System Jun Qin, PengFei Sun, and Jacob Walker Deartment of Electrical and Comuter Engineering Southern Illinois University Carbondale Illinois, USA Abstract This aer reresents our recent exerimental measurement study of the comlex noise in industrial fields, using a novel acoustic detection system and wavelet transform algorithms. Noise induced hearing loss (NIHL) continues to be one of the most revalent occuational hazards in the United States. Number of research on NIHL showed a comlex noise could roduce more hearing loss than an energy-equivalent continuous or imulsive noise alone. Many worklaces in varied industries are subjected to the high level comlex noise (i.e., high-level imulsive noise mixed with continuous Gaussian noise). The current noise measurement guidelines and devices (e.g., conventional sound level meters) are based on the equal energy hyothesis (EEH), which states that loss of hearing by exosure to noise is roortional to the total acoustic energy of the exosure. However, the EEH does not accurately rate the imulsive noise and the comlex noise. Theore, the conventional sound level meter may not be able to accurately assess the comlex noise in industrial fields. In this roject, a new waveform rofile based noise measurement system has been develoed for evaluation of the high level comlex noise in industrial fields. The system consists of four ½ condenser microhones, and it can simultaneously detect and record four waveforms of the comlex noise with high samling rate (5 KHz). In addition, a wavelet transform based signal analysis algorithm has been modified and imlemented to characterize the comlex noise. Pilot field measurements have been conducted in selected local coal mining fields (e.g., wet coal rearation lant and dry coal handling lant) using the develoed system. The reliminary results showed that the system successfully detected and recorded waveforms of comlex noise in industrial fields. The modified algorithm can decomosed the comlex noise signals and dislay the detailed features in the time-frequency joint domain. The key arameters of comlex noise can be determined, and the hazardous comlex noise in industrial fields can be identified. In addition, when measuring the equivalent A-weighted averaged sound ressure level, the develoed system is comarable to a conventional sound level meter. Keywords noise induced hearing loss; comlex noise, noise measurement, wavelet transform; time-frequency domain characterization. I. INTRODUCTION Noise induced hearing loss (NIHL) continues to be one of the most revalent occuational hazards in the United States. Tyical worklaces are often subjected to a comlex noise environment in which high-level imulsive noise mixed with continuous Gaussian noises are embedded []. A number of animal studies showed that interaction effect between imulsive and broadband noises may actually exacerbate the NIHL []. For examle, an exosure to a comlex noise was observed to roduce a much greater ermanent threshold shift (PTS) and more extensive hair cell losses than an exosure to only an energy equivalent continuous or imulsive noise alone would have caused. The noise measurement guidelines and standards, such as ISO and ANSI S , are designed based on the equal energy hyothesis (EEH). This theory regards that NIHL deends on the total acoustic energy of the noise exosure []. In addition, conventional sound level meters (SLMs) are develoed based on the EEH. These SLMs commonly use condenser microhones to detect the noise signals, and take A- weighted equivalent sound ressure level (SPL), L Aeq, as criterion to evaluate the noise exosure levels. However, a number of studies showed that the imulsive noise and the comlex noise could not be accurately rated by the EEH [, 3]. Furthermore, L Aeq could not lect that frequency discreancy in hearing loss evaluation. Other studies also suggested that new noise metrics were needed for evaluation of the risk of high level comlex noises [4-]. Theore, new noise measurement systems and noise metrics (e.g. the time-frequency domain characteristics, Kurtosis, etc.) were needed for accurate assessment of comlex noises in industrial worklaces. In this study, we develoed a novel acoustic measurement system to collect the field comlex noise data. Unlike the convenient SLMs, the develoed system could record the ressure waveforms of noise signals with high samling rate u to 5 khz, and it included an acoustic test fixture to emulate the structure of human ear canals. A wavelet transform (WT) based algorithm was used to characterize the noise signals measured in fields. In addition, new noise metrics, including WT based equivalent SPL L WTeq (ω) and Kurtosis kurt(ω), were roosed for the accurate assessment of the risks of comlex noises in industry worklaces.

3 II. A. Noise measurement system METHODS AND MATERIALS A waveform rofile based noise measurement system has been develoed in this study. As shown in Figure, the system consists of an acoustic test fixture (GRAS 45CA) with two ½- inch ressure condenser microhones assembled, two ½-inch free-field condenser microhones, a 4-channel microhone signal conditioning amlifier, a 6-bits data acquisition device (NI DAQ USB654), and a lato. The test fixture is used to emulate human ear canal structure. Two assembled ½-inch ressure microhones are used to record acoustical ressure waveforms in the ear cannel, while the other two ½-inch freefield microhones are used to record the ressure waveforms in free field. A user interface is develoed using LabVIEW software to control the system, and record the waveforms of comlex noise signals. After develoment, the system was validated using varied noise signals, including ure tone noise, Gaussian noise, and comlex noise. Gaussian and comlex noises were simulated and generated by a digital noise exosure system. The SPL of varied noise signals were simultaneously measured by the develo system and a conventional SLM. The validation results indicate that the develoed system is comarable with the SLM. In addition, the develoed system has higher sensitivity and recision than the SLM. B. Field measurement The develoed noise measurement system has been used to conduct field measurements in a coal rearation lant. Figure B shows the setu of the develoed system in a field measurement. In each field measurement, a conventional SLM has been used for a quick noise level measurement at different locations in the lant. Three locations were selected for the comrehensive measurement using the develoed system, including Level nearby the entrance of the lants as background noise measurement, Level 3½ nearby the screen bowl as the highest averaged SPL oint, and Level 4 nearby the DR screen at the highest eak SPL oint. At each selected location, five noise signals (5 minutes time duration) were measured and recorded by the develoed system for further signal analysis. A conventional SLM was also used to measure SPLs of noise in the field, and it was comared with the develoed system. C. Wavelet transform technology for noise analysis. In the conventional noise evaluation methods, the noise signals are characterized in either the time domain or the frequency domain. However, time and frequency are relevant to each and change simultaneously. Theore a transient signal, such as imulsive or comlex noise, needs to be better characterized in the time-frequency (T-F) domain. The wavelets transform, which uses various kernel functions to decomose signals, is suitable for analysis of transient signals. FIG.: (A) Schematic diagram of the develoed noise measurement system, and (B) a hotograh of system setu in a field measurement at a coal rearation lant. Continuous wavelet transform (CWT) which decomoses a signal f(t) in the T-F domain can be defined as follows [,3]: t b W ( a, b) = f ( t) ψ ( ) dt a a where ψ(t) is the wavelet kernel function along with the continuous scaling arameter a and the time shifting arameter b. W(a, b) ers to the CWT coefficient. The signal f(t) can be obtained from the wavelet coefficients by the inverse WT, only when it satisfies the admissibility condition ( C ψ < ): dadb f ( t) = C W ( a, b) ψ ( t) 0 a () ψ ()

4 FIG. (A) the time history and (B) the amlitude sectrum of the Morlet wavelet. where constant value C ψ is defined by: C = π dωψˆ ( ω) ω (3) ψ where ψ ˆ ( ω) is the Fourier transform of the wavelet kernel function ψ (t). Equation 3 requires that ψ ˆ (0) =0, which equals to ψ ( t) dt = 0, and wavelet kernel functions have a zero average in the time domain [3]. In addition, after normalization, ψ ( t ) dt = (A) (B) is required as well. To normalize wavelet kernel function ψ (t), a wavelet atom φ γ (t) is roosed: t b t) = ψ a b ( t) = ψ ( ) a a φ γ (4) (, where γ is a multi-index arameter erring to a and b. Morlet wavelet was used in this study for noise analysis. Morlet wavelet was derived from the Gaussian function, and it can be defined as: t ψ mor ( t) = ex( ) ex( jηt) (5) / 4 ( τ π ) τ where τ is the arameter to control the frequency bandwidth, and η reresents the central frequency of this kernel function. Figure A shows the time history of the real art of the kernel functions of Morlet wavelet. Morlet wavelet is symmetric about y-axis in the time domain. The frequency sectrum of the Morlet wavelet function is shown in Figure B, in which the f c is central frequency, and the f l and f h ers to the low and high frequency boundaries at half amlitude, resectively. As described in Equation 5, the f l and f h can be defined and adjusted by the arameter τ, and the f c is controlled by the arameter η. In the frequency domain, it behaves like band-ass filters, and can extract the localized frequency details of transient signals. Furthermore, the WT yields out the decomosition of the acoustic signal, and its coefficients alternatively can be reresented in the sound ressure level (SPL) as: where 6 P = 0 0 Pa. D. Characterization of comlex noise ws ( b) ws ( b) SPL ( b) a = 0 log0( ) (6) P Filters with frequency-deendent gain have been introduced to mimic the frequency resonse of the human auditory organ. The A-weighting and C-weighting filters are most commonly used for noise evaluation. Figure 3 shows the frequency resonses of the A-weighting and C-weighting filters. The A- weighting filter has narrower bandwidth than the C-weighting filter. Secifically, at frequencies less than 000 Hz, the A- weighting filter quickly decays, while the C-weighting filter doesn t decay. Consequently, an A-weighted noise metric such as L Aeq does not count low frequency comonents in a noise signal. Such a metric is not suitable for accurate evaluation of the comlex noise, which contents strong low frequency comonents. In contrast, the C-weighting filter is relatively flat within the normal audible frequency range, and it should be considered as a more suitable metric for the evaluation of the comlex noise in worklaces. * TABLE I. SUMMARY OF DIFFERENT NOISE LEVELS MEASURED AT THREE LOCATION IN THE FIELD. FIG. 3 illustration diagram of the frequency resonses of A-weighting and C- weighting filters. Locations L Aeq, db L Ceq, db L max, db Level Level 3½ Level

5 (A) (B) FIG.4 (A) waveform with five second duration and (B) amlitude sectrum of a reresentative comlex noise signal measured in the field. In this study, new noise metrics have been roosed to evaluate the comlex noise measured in mining fields. These metrics include C-weighted equivalent SPL L Ceq, WT based equivalent L WTeq, eak SPL L max, and Kurtosis kurt(ω). Generally, the equivalent SPL L eq can be defined as [4]: L eq t A = 0log0( dt) t t (7) t where 0 is the erence ressure level 0uPa. A is the acquired sound ressure in Pa. The time eriod for data samles ranges from t to t. When alying the A-weighting and C-weighting filters to the acoustic signals, equivalent SPL are described as L Aeq and L Ceq, resectively. In addition, the WT based equivalent SPL is roosed to reveal the time history of the different frequency comonents. The WT based L WTeq is calculated as a function of frequency as [9]: T eq ( ω ) = ( ( t)) (8) T 0 where ( t) is the time histories of sound ressure obtained at central frequencies for each /3 octave band. L WTeq (ω) can be calculated as [9]: eq ( ω) LWTeq ( ω ) = 0 log0( ) (9) The eak SPL L max is defined as [4]: eak L max = 0 log0( ) (0) 0 FIG.5 T-F reresentation of the SPL of a reresentative noise signal measured in the field. The statistical metric Kurtosis is a arameter to describe the imulsiveness and the sectral comosition of a comlex noise. The Kurtosis directly lects the strength and skewness of the eak ressure, which are usually taken as the metric to assess the risk of hearing loss. Generally, it can be defined as the ratio of the fourth moment to the squared second moment. In this study, the Kurtosis as a function of frequency is calculated from the SPL time history of each wavelet scale [9]. Kurt( ) = N ( N N N ( ( ( t) ( t))4 ω () ( t) ( t)) ) III. RESULTS AND DISCUSSION A. Noise waveform and sectrum using regular FFT. The L Aeq, L Ceq, and L max of the comlex noise measured by the develoed system at three locations in the field are summarized in Table. At all three locations, the C-weighted SPL L Ceq are about 0 db higher than the A-weighted SPLs L Aeq. Moreover, Figure 4 shows a reresentative waveform (5 second time duration) and amlitude sectrum of a noise signal measured in the field. It shows the noise signal measured in the field is a comlex noise (Non-Gaussian noise). In the amlitude sectrum, higher amlitudes can be found at low frequency range (< 00Hz). It indicates that the comlex noise measured in the field contains very strong low frequency comonents. Such low frequency comonents are not counted into the L Aeq due to the frequency resonse of an A-weighting filter. While these low frequency comonents are included in C-weighted SPL calculation. This is the main reason that causes 0 db difference between L Ceq and L Aeq summarized in Table. The L Aeq is the metric currently used in government regulation for assessment of the comlex noise in mining fields. It is generally considered aroriate for continuous Gaussian noise but not for comlex noise. Such criteria may significantly underestimate the risk of high level comlex noise in mining fields.

6 FIG.6 Time history of the /3 octave SPL comonents of a field noise signal. Each time history was obtained by alying the AWT to the noise with the center frequency at the frequency shown in the figure. For examle, 0.5 khz time history shown in the figure aroximates the field noise that assed through a /3 octave filter of 0.5 khz center frequency. B. AWT based noise characterizaiton The Morlet WT was alied to decomose the measured noise signals into the T-F domain. In order to decomose the noise signal following the characteristic of human hearing system, the central frequencies and bandwidths on the frequency axis were adjusted according to /3 octave bands in the WT. At each scale, the SPL is integrated along the time axis based on the wavelet coefficients. Figure 5 shows the T-F reresentation of SPL of a noise signal measured by the develoed system in the field. The 3D structure in the T-F domain lects the distribution of SPL of the noise signal Hz) are illustrated in Figure 6. The lowest frequency band centered at 00 Hz contains higher energy around 0dB level. Corresondingly, the highest frequency sectrum centered at 0 khz shows lower energy around 90dB level. However, the SPL distribution for 0 khz demonstrates several high eaks more than 0dB. Those eaks cannot be found in the ower sectrum obtained by regular FFT, but they may cause hearing loss. FIG.8 the frequency of the WT based equivalent SPL L WTeq(ω) of a noise signal measured in the field. FIG.7 the frequency distribution of Kurtosis of a noise signal measured in the field. The SPL time histories of a field noise signal at selected /3 octave bands (central frequencies at 00, 500, 500, and C. WT based new nosie metrics: L WTeq (ω) and Kurtosis

7 Kurtosis was roosed as a new noise metric to assess the comlex noise. According to Equation, the Kurtosis kurt(ω) as a function of frequency can be calculated from ressure wavelet coefficients ( t) in the T-F domain. Figure 7 shows the frequency distribution of Kurtosis of a noise signal measured in the field. Overall, the Kurtosis at entire frequency range is greater than 3, which is the Kurtosis value of the Gaussian noise. It indicates the noise signal measured in the field is a non-gaussian comlex noise. In addition, the Kurtosis is increasing from 0 to 60 with the frequency increasing. It means that the noise signal contents imulsive comonents, secifically, at high frequency. Another new metric roosed in this study is the WT based equivalent SPL L WTeq (ω), which can be calculated from ressure wavelet coefficients according to Equations 8 and 9. Figure 8 illustrates the frequency distribution of L WTeq (ω) calculated from a noise signal measured in the field. The results indicate that the L WTeq (ω) is a function of frequency, and it decreases with the frequency increasing. L WTeq (ω) lects the averaged energy at different /3 octave bands. As shown in Figure 8, the energy is concentrated at the lower frequency bands (less than 500 Hz). It is also can be found that the L WTeq (ω) of high frequency bands ranging from 5 khz to khz are above 85dB level. In addition, the L WTeq (ω) could be associated with hearing loss at different frequency bands in the cochlear, and it will be a meaningful metric for the evaluation of comlex induced noise loss. IV. CONCLUSION In this study, we develoed new noise measurement system and roosed new noise metrics for the investigation of the comlex noise in high noisy industry fields. The develoed system not only can measure the noise levels, but also can record detailed waveform of noise signals in the field. The system has been successfully validated both in our lab and in a coal mining field. Moreover, the Morlet WT algorithm has been alied to characterize the comlex noise measured in the field. Such technology can rovide detailed information of a comlex noise in the T-F domain, and lead to the develoment of new noise metrics. In addition, new noise metrics for the evaluation of the comlex noise have also been roosed and discussed in this aer. The reliminary results demonstrated the roosed new metrics can more accurately evaluate the risks of high level comlex noise. In our future work, a new metric evaluating system will be established, and more field measurements will conducted in varied industrial fields (e.g., coal rearation and handing lants, surface mines, and other industrial fields). REFERENCES [] X.D. Zhu, and J. Kim, Alication of analytic wavelet transform to analysis of highly imulsive noises, Journal of Sound and Vibration, vol.94, , July 006. [] W. A. Ahroon, R. P. Hamernik, et al., COMPLEX NOISE EXPOSURES - AN ENERGY ANALYSIS, Journal of the Acoustical Society of America, vol.93, ,993. [3] R. P. Hamernik, W. Qiu, et al., Hearing loss from interruted, intermittent, and time varying non-gaussian noise exosure: The alicability of the equal energy hyothesis, Journal of the Acoustical Society of America vol., , 007. [4] D. Havelock, S. Kuwano, and M. Vorländer, Handbook of signal rocessing in acoustics (Vol. ). NY: Sringer, 008. [5] R. P. Hamernik, and W. Qiu, Energy-indeendent factors influencing noise-induced hearing loss in the chinchilla model, Journal of the Acoustical Society of America, vol.0, , 00. [6] R. P. Hamernik, W. Qiu, et al., Cochlear toughening, rotection, and otentiation of noise-induced trauma by non-gaussian noise, Journal of the Acoustical Society of America vol.3, , 003. [7] J. Starck, E. Toila, et al., Imulse noise and risk criteria, Noise & Health, vol.5, , 003 [8] J. Kim, et al., Time-frequency characterization of hand-transmitted, imulsive vibrations using analytic wavelet transform, Journal of Sound and Vibration, vol.308,. 98-, Nov 007. [9] X. Zhu, et al., Develoment of a noise metric for assessment of exosure risk to comlex noises, The Journal of the Acoustical Society of America, vol.6,.703-7, 009. [0] D. Henderson, and R.P. Hamernik, Imulse noise critical review, Journal of the Acoustical Society of America, vol.80, , 986. [] R.P. Hamernik, et al., Audiometric and histological differences between the effects of continuous and imulsive noise exosures, Journal of the Acoustical Society of America, vol.93, [] I. Daubechies, Ten Lectures on Wavelets. Philadelhia: SIAM, 99. [3] S. Mallat, A wavelet Tour of Singal Processing. Academic Press, 997. ACKNOWLEDGMENT This roject was funded by the Illinois Clean Coal Institute with funds made available by the State of Illinois.