Professional Documents
Culture Documents
Abstract
1 Introduction
All measurement results have an associated element of uncertainty about their true value. In
general terms, this is known as measurement uncertainty, and is attributed to factors that
influence the final results of the measurements. Measurement uncertainties can come from the
measuring instrument, from the item being measured, from the environment, from the operator,
from subjects used and from other sources. The errors of measurement can be expressed by
the measurements uncertainty value. This value can be used to quantify the confidence limits of
the measured results and allows comparison of measurements carried out by same or different
laboratories and for same or different products. The uncertainty calculated here is based on
GUM1 and other publications [2], [3], [7] and [8]. Uncertainty calculation for Hearing Protector
Device (HPD) noise attenuation measurement can be carried out in different situations such as:
a) One specified brand of HPD measured in one laboratory: which is the most important case
that should be reported by that laboratory at the final results report. Usually a hearing
protector’s manufacturer or user asks a laboratory to make measurements for noise
attenuation of a certain model of HPD. Also this can be extended to measurements of
several times the same HPD, in the same laboratory, using different subjects, which is very
common case of periodic measurements of the same HPD in the same laboratory;
b) Specified type of HPD: in this case the laboratory measured the noise attenuation of
different model of HPD (for example a number of different model of Earmuff HPD) and
should combine the results to give the uncertainty of all these models measured. Figure 1
show different types and models of HPD;
c) Inter-laboratories measurements: for example, inter-laboratories round measurements of
noise attenuation of a specific model of HPD. In this case a specific model of HPD goes
around different laboratories, where the noise attenuation is carried out using the same
standard.
Most of publications and also at the standards like ISO 4869-1 [4] or ISO 4869-5 [5] and ANSI
S12.6 [6] do not consider these different cases separately. The first case is the most important
one since we are usually interested to know for a specific model measured in a specific
laboratory, what is the uncertainty of the noise attenuation measurement of this HPD.
This paper developed a detailed calculation of uncertainty of measurements results of noise
attenuation of one specific model of HPD measured in one laboratory and extend that to
repeated periodic measurement of the same HPD in the same laboratory several time. This
paper is refinement of our previous paper published [7], where calculation is presented in more
details with justifications.
2
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
Ai is the attenuation of a subject considering the two trials of a frequency band; CT and OT is
the closed and open threshold of hearing of the first and second trial of a subject for a frequency
band. Considering “N” test subject, the overall average attenuation (Af) will be the arithmetic
mean for a given frequency band.
𝑁
1 2
𝑆𝑓 (𝑓) = √∑(𝐴𝑖 (𝑓) − 𝐴𝑓 (𝑓)) (2)
√𝑁 − 1 𝑖=1
3
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
2
𝑢𝐴𝑡𝑡𝑒𝑛𝑢𝑎𝑡𝑖𝑜𝑛 = √𝑢𝑆𝑢𝑏𝑗𝑒𝑐𝑡 2 2 2 (3)
𝑅𝑒𝑠𝑝𝑜𝑛𝑠𝑎 + 𝑢𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 + 𝑢𝐴𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑆𝑡𝑒𝑝 + 𝑢𝑆𝑜𝑢𝑛𝑑 𝐹𝑖𝑒𝑙𝑑
4
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
𝑐𝑜𝑣(𝑐𝑙𝑜𝑠𝑒𝑑, 𝑜𝑝𝑒𝑛, 𝑓)
𝜌(𝑐𝑙𝑜𝑠𝑒, 𝑜𝑝𝑒𝑛, 𝑓) = (4)
√𝜎(𝑐𝑙𝑜𝑠𝑒𝑑, 𝑓). 𝜎(𝑜𝑝𝑒𝑛, 𝑓)
Where ρ(Closed,Open,f) is the correlation coefficient between the closed and open threshold of
hearing for a frequency band; cov(Closed,Open,f) is the covariance between the closed and
open threshold of hearing for a frequency band; σ(Closed,f) and σ(Open,f) is the standard
deviation of the closed and open threshold of hearing for each frequency. Data gathering from
LAEPI since 2008 was used to calculate the correlation coefficient between the closed and
open threshold of hearing. It was used 15360 closed and open threshold of hearing to calculate
the correlation coefficient for each frequency band. The correlation coefficient found for each
frequency band is show in Table 1.
Table 1: Correlation coefficient between open and closed threshold of hearing
The correlation coefficient between the closed and open threshold of hearing affects the way
that the combined standard uncertainty of one trial for one subject attenuation is estimated. If
the correlation coefficient is 0 it shows that they are statistically independent. Since there is a
correlation between the closed and open threshold of hearing the following equation must be
used to estimate the combined standard uncertainty of one trial for one subject:
𝑢 𝑇𝑟𝑖𝑎𝑙 (𝑓) = √𝑢2 (𝑐𝑙𝑜𝑠𝑒𝑑, 𝑓) + 𝑢2 (𝑜𝑝𝑒𝑛, 𝑓) − 2. 𝑢(𝑐𝑙𝑜𝑠𝑒𝑑, 𝑓). 𝑢(𝑜𝑝𝑒𝑛, 𝑓). 𝜌(𝑐𝑙𝑜𝑠𝑒𝑑, 𝑜𝑝𝑒𝑛, 𝑓)
(5)
Equation (5) will estimate the standard uncertainty for one trial for one subject for one frequency
band. The next step is to estimate the standard uncertainty of one subject for one frequency
band, this will be determined by combine the standard uncertainty of the two trials for each
subject for one frequency band. It is necessary to evaluate the correlation between the two trials
of each subject for each frequency band. Data gathering from LAEPI since 2008 was used to
calculate the correlation coefficient between the two trials of each subject for each frequency
band. 7680 subject were used, each subject with two trials for each frequency band. The
correlation coefficient found for each frequency band is show in Table 2.
Table 2: Correlation coefficient between two trails for each frequency band
Table 2 shows that the two trials have a strong correlation between then. Therefore, the
following equation is appropriate:
5
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
For the subject response variation for all the subjects used in the test can be estimate. The
following equation can be used to estimate the standard uncertainty of the subject response
variation for each frequency band:
𝑁
1 2
𝑢𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑅𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝑉𝑎𝑟𝑖𝑎𝑡𝑖𝑜𝑛 (𝑓) = . √∑ 𝑢𝑆𝑢𝑏𝑗𝑒𝑐𝑡 (𝑖, 𝑓) (7)
𝑁
𝑖=1
The experimental standard deviation of the mean is sometimes called a Type A standard
uncertainty, N is the number of subjects of the attenuation test, and the experimental standard
deviation of the mean needs to be calculate for each frequency test band according to equation
(2) and equation (8).
6
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
The coverage factor is obtained by calculating the effective degrees of freedom obtained from
the Welch-Satterthwaite formula, for the attenuation measurements the effective degrees of
freedom can be obtained as follows:
4
𝑢𝐶𝑜𝑚𝑏𝑖𝑛𝑒𝑑 (𝑓)
𝜐𝑒𝑓𝑓 (𝑓) = 4 4 4 4 (12)
𝑢𝑆𝑢𝑏𝑗𝑒𝑐𝑡 (𝑓) 𝑢𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 (𝑓)
𝑢𝐴𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 (𝑓) 𝑢𝑆𝑜𝑢𝑛𝑑 𝐹𝑖𝑒𝑙𝑑 (𝑓)
+ + +
𝑁−1 𝑁−1 𝑁−1 𝑁−1
N is the number of subjects in a given test. Since the standard uncertainty of the amplitude step
and sound field have infinite degrees of freedom, these terms in equation (12) can be
neglected. The value of the Student coefficient (k) corresponding to the number of effective
degrees of freedom with confidence level of 95% and can be obtained from references [1] and
[9].
3 Case study
In order to check the methodology described above, it was applied to HPD measurements on
an earmuff using ANSI S12.6, method B [6]. This methodology can be applied using the ISO
4869-1 [4] or ISO 4869-5 [5] standard too. It is important to mention that these uncertainty
values and coverage factor can change for each HPD test, each measurement system used
and each standard used. In order to estimate the uncertainty for the measurement, it is
necessary to calculate the uncertainties of each trial of each subject for each frequency band.
Since this procedure would be exhaustive if it were presented for each trial. The uncertainty
calculation of the subject response variation and the attenuation calculation are demonstrated
for just one subject in one frequency band selected randomly. The 8th subject at 2000 Hz test
7
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
frequency will be present. The first step is to calculate the uncertainty of the subject response
variation. Table 3 below shows the uncertainty of each threshold of the 8th subject at 2000 Hz
test frequency, this data will be used to calculate the uncertainty of the subject response
variation.
Table 3: Threshold uncertainty of the 8th subject at 2000 Hz frequency band
Applying equation (5) with the correlation coefficient of Table 1 for 2000 Hz test, the uncertainty
of the first and second trial is 4.00 dB and 3.14 dB. The standard uncertainty of subject
response variation is calculated using equation (6) and the result is 3.57 dB. The same
procedure described above is used to calculate the uncertainty of each subject at each
frequency band, Table 4 below shows these values. To obtain the overall uncertainty of the
variation in subject response for each frequency band it is necessary to apply equation (7).
Table 4: Subject response variation uncertainty of each frequency band, all values in dB
The next step is to calculate the standard deviation between subjects. Table 5 below presents
the threshold measurements. It is necessary to use equation (1) to calculate the attenuation of
each trial for each frequency band and the mean attenuation of the two trials.
Table 5: Threshold measurements and mean attenuation of the 8th subject at 2000 Hz frequency
band
8
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
Table 6 shows the mean attenuation for all subjects in all frequency band and the average
attenuation of each frequency band by using equation (1), the standard deviation of each
frequency band by using equation (2) and the uncertainty of the standard deviation between the
subjects of each frequency band by using equation (8).
Table 6: Attenuation of each subject, mean attenuation, standard deviation and uncertainty of the
standard deviation between subjects for each frequency band, all values in dB
Table 7 present all uncertainty for each frequency band. Equation (3) is used to calculate the
combined standard uncertainty for each frequency band. The next step is to calculate the
Student coefficient by calculating the effective degrees of freedom using equation (12) and the
expanded uncertainty using equation (11) for each frequency band and express the result in
terms of the mean attenuation value and the expanded uncertainty with 95% of confidence
level.
Table 7: Standard uncertainty, combined standard uncertainty, Student coefficient and expanded
uncertainty for each frequency band, all values in dB except the Student coefficients
9
22nd International Congress on Acoustics, ICA 2016
Buenos Aires – 5 to 9 September, 2016
4 Conclusions
The largest contribution of uncertainty in hearing protector noise attenuation measurements is
the subject response variation and standard deviation between subjects. Therefore, for each
measurement of a HPD in one laboratory, the uncertainty calculation should be carried out. If
the whole measurements are repeated for the same HPD, even for the same subjects,
calculation should be carry out again because of the different in the subject responses used.
This calculation should be presented in the final report of measurements.
The subject response variation distribution need to be study more since it is not clear if it has a
rectangular distribution. The subject response variation distribution may have a bimodal
distribution since all the peak are near each other and all the valleys are near each other in the
trace of the subject. Therefore, is important to determine what kind of distribution this type of
uncertainty has.
Acknowledgments
This research was carried out with the support of the Brazilian funding bodies (CNPq, CAPES
and FINEP) and Laboratory of Personal Hearing Protector Equipment (LAEPI) of NR
Consultancy Ltda. The guidance and orientation of Prof. Armando Albertazzi of POSMEC for
the uncertainty calculation used her is very much appreciated.
References
[1] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML, Guide to the Expression of Uncertainty in
Measurements, Geneva, 1995.
[2] Bell, S. A beginner’s Guide to Uncertainty of Measurements, Measurement Good Pratice Guide No.
11, NPL, 1999.
[3] Craven, N. J.; Kerry G. A Good Practice Guide on the Source and Magnitude of Uncertainty Arising in
the Practical Measurement of Environmental Noise. The University of Salford, 1997.
[4] International Organization for Standardization, ISO 4861-1: Acoustics – Hearing protectors – Part 1:
subjective method for measurement of sound attenuation, Geneva, 1990.
[5] International Organization for Standardization, ISO 4861-5: Acoustics – Hearing protectors – Part 5:
Method for estimation of noise reduction using fitting by inexperienced test subject, Geneva, 2006.
[6] Acoustical Society of America, ANSI 12.6: Methods for Measuring Real-Ear Attenuation of Hearing
Protectors, Melville, 2008.
[7] Lima, F.; Gerges, S.; Zmijevski, T.; Bender D.; Gerges, R. Uncertainty Calculation for Hearing
Protectors Noise Attenuation Measurements by REAT Method, Journal of the Brazilian Society of
Mechanical Sciences and Engineering, Vol 32 (1), 2010, pp 28-36.
[8] Lima, F. Development and Metrologic Validation of a Hearing Protector Device Attenuation
Measurement System by the Subjective Response Method. Doctor Thesis, Federal University of
Santa Catarina, Brazil, 2003.
[9] Albertazzi, A.; Sousa, R. Fundamentals of scientific and industrial metrology. Manole, Barueri (Brazil),
1st edition, 2008.
10