Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com

ScienceDirect
ScienceDirect
Procedia
Available Manufacturing
online 00 (2018) 000–000
atatwww.sciencedirect.com
Available online www.sciencedirect.com
Procedia Manufacturing 00 (2018) 000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
ScienceDirect
ScienceDirect
Procedia Manufacturing 26 (2018) 1287–1295
Procedia Manufacturing 00 (2017) 000–000
www.elsevier.com/locate/procedia
46th SME North American Manufacturing Research Conference, NAMRC 46, Texas, USA
46th SME North American Manufacturing Research Conference, NAMRC 46, Texas, USA
Background noise mitigation of dual microphone system for defect
Background noise mitigation of dual microphone system for defect
detection
Manufacturing Engineering in International
Society electrical Conference
cable connection
2017, MESIC 2017, 28-30 June
detection in electrical cable connection
2017, Vigo (Pontevedra), Spain
Neha S. Joshi, Sonali Singh, Matthew Krugh*, Laine Mears
Neha S. Joshi, Sonali Singh, Matthew Krugh*, Laine Mears
Costing models for
Clemson capacity
University Internationaloptimization in Industry
Center for Automotive Research, 4.0: Trade-off
Greenville, SC, USA
Clemson University International Center for Automotive Research, Greenville, SC, USA
between used capacity and operational efficiency
* Corresponding author. Tel.: +1-864-283-7229; fax:
a +0-000-000-0000a,*.
* Corresponding
E-mail address:author. A. Santana , P. Afonso , A. Zanin , R. Wernkeb
Tel.: +1-864-283-7229; fax: +0-000-000-0000 .
[email protected]
b
E-mail address: [email protected]
a
University of Minho, 4800-058 Guimarães, Portugal
b
Unochapecó, 89809-000 Chapecó, SC, Brazil
Abstract
Abstract
The automotive wiring harness forms the nervous system of the modern automobile and is dependent on electrical circuits joined
The automotive
together
Abstract by a myriadwiringof harness
electricalforms the nervous
connector system
designs. of theinmodern
If a defect a singleautomobile
connectionand is dependent
occurs on electrical
during vehicle assembly,circuits joined
the vehicle
together by a myriad of electrical connector designs. If a defect in a single connection occurs during vehicle
systems may not communicate correctly, and the entire vehicle can be sent for rework incurring a potentially significant added assembly, the vehicle
systems may notofcommunicate correctly,
cost. Detection
Under the concept these
of defects
"Industry 4.0",and
currently the on
relies entire
production vehicle
human can be sent
willandbefor
validation,
processes can rework incurring
be difficult
pushed a potentially
to betoincreasingly
maintain significant
consistent added
inspection
interconnected,
cost. Detection
reliability. In thisofwork,
thesethedefects currently
“clicking” sound relies on human
corresponding to validation,
the successfulandconnection
can be difficult
of to electrical
mating maintainconnectors
consistent isinspection
analyzed
information
reliability.
based
In this
on the
a real time basis
“clicking”
and, necessarily, much more connection
efficient. In this context, capacity optimization
in the presence of work,
noise to quantify sound corresponding
the ability to the successful
to detect and classify of mating
these types of connections withinelectrical connectors
a simulated is analyzed
work environment.
goes
in thebeyond
presence the
of traditional
noise to aim
quantify of
the capacity
ability to maximization,
detect and classify contributing
these types ofalso for organization’s
connections within a profitability
simulated work and value.
environment.
The work herein investigates a methodology to separate this sound signal from background noise using spectral subtraction, and
Indeed,
The worklean
introduces herein management
a classification a and
investigatesmethod, continuous
methodology improvement
to separate
demonstrated in a this soundapproaches
controlled signal suggest
from background
application.
capacity
Experiments noise optimization
using
using labspectral
instead and
sensingsubtraction,
systems were
of
maximization.
introduces to
performed The study
a classification of capacity
method,
process and analyze optimization
demonstrated
signals and costing
in a controlled
and the results models is an
application.reduction
showed a significant important
Experiments research
using
in noise, topic
lab sensing
effectively that deserves
systems were
distinguishing the
contributions
performed
signal from both
to process
of interest. the practical
and analyze signalsand
andtheoretical
the results perspectives. This paper
showed a significant presents
reduction andeffectively
in noise, discusses distinguishing
a mathematical the
model forinterest.
signal of capacity management based on different costing models (ABC and TDABC). A generic model has been
developed
© 2018 Theand it wasPublished
Authors. used to analyze idleB.V.
by Elsevier capacity and to design strategies towards the maximization of organization’s
©© 2018 The
2018TheThe Authors.
Authors. Published
Published by
by Elsevier
Elsevier B.V.
B.V. committee
value.
Peer-review trade-off
under capacity
responsibility maximization
of the scientific
Peer-review under responsibility of the scientific committee vs operational
ofofthe efficiency
NAMRI/SME.
46th SME North is American
highlighted and it is shown
Manufacturing Research that capacity
Conference.
Peer-review under
optimization might responsibility
hide operationalof the scientific committee of NAMRI/SME.
inefficiency.
Keywords:
© 2017 TheSpectral Subtraction;
Authors. Published Dual
bymicrophone;
Elsevier B.V.Automotive; Assembly; Connection.
Keywords: Spectral
Peer-review under Subtraction; Dualofmicrophone;
responsibility Automotive;
the scientific committeeAssembly; Connection. Engineering Society International Conference
of the Manufacturing
2017.

Keywords: Cost Models; ABC; TDABC; Capacity Management; Idle Capacity; Operational Efficiency

1. Introduction

The cost of idle capacity is a fundamental information for companies and their management of extreme importance
in
2351-9789 ©production
modern systems.
2018 The Authors. In general,
Published it isB.V.
by Elsevier defined as unused capacity or production potential and can be measured
in several©under
2351-9789
Peer-review ways: tons of production,
2018responsibility
The Authors. Published
of available
by Elsevier
the scientific B.V.hours
committee of manufacturing, etc. The management of the idle capacity
of NAMRI/SME.
Peer-review underTel.:
* Paulo Afonso. responsibility
+351 253 of the761;
510 scientific committee
fax: +351 253 604of741
NAMRI/SME.
E-mail address: [email protected]

2351-9789 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the Manufacturing Engineering Society International Conference 2017.
2351-9789 © 2018 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the scientific committee of the 46th SME North American Manufacturing Research Conference.
10.1016/j.promfg.2018.07.139
1288
2 NehaSonali
Neha S. Joshi, S. Joshi et al.Procedia
Singh/ / Procedia Manufacturing0026(2018)
Manufacturing (2018) 1287–1295
000–000

Nomenclature connection between the plug and the receptacle defeats

the function of the electronic equipment as electronic
b(t) impulse response signals do not have a communication pathway at the
d(t) additive noise connector. Incomplete connection may occur because
D(m,n) STFT of additive noise of misconnection or incomplete connection during
f(t) output of FIR filter assembly, and is one of the major causes of wiring
g(t) square of output of FIR filter harness assembly defect costs. It is therefore
h(t) moving window standard deviation imperative that the completed harness is verified for
G(m,n) STFT of gain proper connection before being handed over to the
P(t) final binary output signal customer [2]. If wiring errors are not discovered until
Q(m,n) STFT of relative signal level the completed harness is tested, it can either be
s(t) clean signal reworked while still in the assembly area or sent to an
S(m,n) STFT of clean signal offline rework for repair and correction whose
x1(t) mic 1 (wrist mic) data consequences can be felt through line efficiency drop
x2(t) mic 2 (arm mic) data ultimately leading to loss of time and money [2-3].
X1(m,n) STFT of wrist mic signal This is compounded as vehicle assembly continues, as
X2(m,n) STFT of arm mic signal parts can be placed in the way of access to already-
y(t) processed signal installed connectors necessitating disassembly of
Y(m,n) STFT of processed signal sections of the vehicle to access a missed connection.

2. Background and approach

1. Introduction
Automotive manufacturing facilities have
Modern day vehicles employ varied electrical
employed numerous methods in assembly lines to
actuators such as motorized mirrors, windows, seat
address electrical connection defects. These methods
adjusters, trunk hatches, navigational CRT displays,
range from simple visual checks to a method as
and more. The demand for new features is changing at
complex as Augmented Reality (AR) [4]. To move one
an ever-increasing rate. This surge in electrical devices
step towards Industry 4.0, we investigate a sound
imposes the use of more wiring harness branches and
detection system for electrical connectors and analyze
associated electrical connectors at which each are a
signal through one background noise mitigation
point of potential manufacturing-related failure in the
methodology. The purpose of this system is to detect
vehicle.
electrical connection activity during assembly
The upsurge of desire for electronic inclusion in
operations in the presence of background noise.
automobiles has necessitated the development of
The proposed system uses two microphones to
smart devices to interconnect the entire production
measure the sound level of the ‘click’ sound generated
system wirelessly. These smart devices are slowly and
from the male-female terminals mating at complete
steadily bringing about a new wave of awareness and
locking to define the successfulness of the connection
digitalization in the manufacturing environment by
task. The proposed system has an advantage over
emphasizing increased human and machine
vehicle diagnostics checks performed at selected key
interaction, and thereby paving a way for the fourth
points along the assembly line in that it aims to detect
industrial revolution (i.e., Industry 4.0). The fourth
the connection error immediately and indicate to the
revolution will shift the use of disparate manufacturing
associate the connection status thereby allowing the
entities to a connected ecosystem in the coming years
associate to correct the defect before it leaves the takt.
[1].
Adoption of new human sensing modalities will
Electrical connectors at the basic level consist of a
enable previously unseen data streams from and to the
plug (male connector) and a receptacle (female
associate and aid in incorporating the future assembly
connector). Complete connection of the automotive
associate into emerging industry 4.0 manufacturing
connector is typically determined based on a “click”
systems.
sound (the target signal of interest) generated upon the
One of the challenges associated with
mating of the two connectors, and may sometimes
implementation of a noise detection system within a
include an additional locking feature. Incomplete
vehicle manufacturing assembly line is the mitigation
 Neha S. Joshi et al. / Procedia Manufacturing 26 (2018) 1287–1295 1289
Neha S. Joshi, Sonali Singh/ Procedia Manufacturing 00 (2018) 000–000 3

of background noise. Noise is pervasive in the the source will pick up both desired signal and noise.
manufacturing plant acoustic environment and can be The secondary microphone, located away from the
as high as 125 decibel (dB) for short periods of time source will pick up primarily the background noise and
[5]. The sound recorded by shop floor level will be used for noise spectrum estimation [15].
microphone is generally contaminated with Resorting to two microphones can greatly help in noise
background noise emanating from sources not reduction while enhancing the sound quality. The dual
pertinent to the process of interest. The background microphone based system used for background noise
noise interferes with and degrades the quality of mitigation is as depicted in Fig. 1.
recorded sound, reducing the chances of detection [6].
In such noisy environments, signal de-noising is Wrist
crucial to improving the signal-to-noise ratio (SNR) Microphone
and preserving the shape and characteristics of the
original signal [7].
Click Sound
Over the past several years, significant efforts have +
been taken in the field of mobile phones and hearing Background
aid machines resulting into a plethora of techniques Noise
that have been proposed to enhance the quality and
Short-Time
intelligibility of speech degraded by additive
Spectral
background noise. Low-pass filtering, Wiener
Subtraction
filtering, spectral subtraction and comb filtering are
some of the techniques used for speech enhancement
[8]. Beamforming is another potential solution. Background
Noise
However, in the presence of significant reverberation,
it fails to improve the SNR [9]. The drawback of
traditional wiener filtering is that it requires a prior
Arm
estimate of clean signal [10]. Short-Time Fourier
Microphone
Transform (STFT) based spectral subtraction is one of
the most widely accepted approaches over the variety
of algorithms that have been proposed [8, 11-12]. Due Fig. 1. Dual microphone sensor system.
to its simplicity, versatility and computational
efficiency, spectral subtraction algorithm is selected to Short-time spectral subtraction approach in [15] is
address noise mitigation problem considered in this implemented in this paper using MATLAB. The
paper. assumptions considered in this approach are that the
background noise is additive and uncorrelated with
3. Methodology clean signal.
The noisy ‘click’ signal, 𝑥𝑥1 (𝑡𝑡) , measured using
Spectral subtraction involves use of two magnitude primary microphone can be represented as the sum of
clean signal s(t) and additive noise d(t).
frequency spectrum estimates: noisy desired signal
spectrum and noise spectrum [13]. Clean signal
𝑥𝑥1 (𝑡𝑡) = 𝑠𝑠(𝑡𝑡) + 𝑑𝑑(𝑡𝑡) (1)
spectrum is obtained by subtracting an estimate of
noise spectrum from noisy desired signal spectrum
Let, X1(m, n) be the STFT representation of
based on the gain function. The background noise
primary signal 𝑥𝑥1 (𝑡𝑡). Here, m and n are time index and
spectrum can be estimated during the pauses between frequency index respectively used for STFT
‘click’ signal in case of single microphone if spectral calculation. Let, X2(m, n) be the STFT representation
magnitude of the noise prior to the signal and during of background noise measured using secondary
the signal are assumed to be the same. However, this microphone. The STFT of noisy ‘click’ signal based
assumption is not valid for the non-stationary on Eq. (1) can be expressed as:
background noise at the manufacturing plant. To
account for this problem, a dual-microphone system is
adopted [14]. The primary microphone, placed close to
1290
4 NehaSonali
Neha S. Joshi, S. Joshi et al.Procedia
Singh/ / Procedia Manufacturing0026(2018)
Manufacturing (2018) 1287–1295
000–000

𝑋𝑋1 (𝑚𝑚, 𝑛𝑛) = 𝑆𝑆(𝑚𝑚, 𝑛𝑛) + 𝐷𝐷(𝑚𝑚, 𝑛𝑛) (2)

The STFT magnitude square for Eq. (2) would be,

|𝑋𝑋1 (𝑚𝑚, 𝑛𝑛)|2 = |𝑆𝑆(𝑚𝑚, 𝑛𝑛)|2 + |𝐷𝐷(𝑚𝑚, 𝑛𝑛)|2 + 𝑆𝑆(𝑚𝑚, 𝑛𝑛) ∗

𝐷𝐷(𝑚𝑚, 𝑛𝑛)∗ + 𝑆𝑆(𝑚𝑚, 𝑛𝑛)∗ ∗ 𝐷𝐷(𝑚𝑚, 𝑛𝑛) (3)

The cross term between clean signal and noise signal

can be statistically set to zero since both are Fig. 2. Schematic representation of STFT based spectral
uncorrelated. Estimated clean ‘click’ signal power subtraction [15].
spectrum |Ŝ (m, n) |2 can be expressed as:
Here, f(t), b(t) and y(t) are the output of FIR filter,
2 impulse response and input to the filter respectively.
|𝑆𝑆̂(𝑚𝑚, 𝑛𝑛)| = |𝑋𝑋1 (𝑚𝑚, 𝑛𝑛)|2 − |𝐷𝐷(𝑚𝑚, 𝑛𝑛)|2 (4) The impulse response b(t) is taken as the time-reversed
signal template corresponding to the average clicking
The real positive gain G (m, n) as explained in [15] can sound pressure. Implementation of FIR filter was
be represented as: followed by squaring the output of FIR filter to further
enhance the SNR.
|𝑆𝑆̂ (𝑚𝑚,𝑛𝑛)|
𝐺𝐺(𝑚𝑚, 𝑛𝑛) = (5)
𝑋𝑋1 (𝑚𝑚,𝑛𝑛)

From Eq. (4) and (5), gain G (m, n) can be defined as:

1
𝐺𝐺(𝑚𝑚, 𝑛𝑛) = √1 − (6)
𝑄𝑄(𝑚𝑚,𝑛𝑛)

Where, Q (m, n) is the relative signal level.

|𝑋𝑋1 (𝑚𝑚,𝑛𝑛)|2
𝑄𝑄(𝑚𝑚, 𝑛𝑛) = |𝐷𝐷(𝑚𝑚,𝑛𝑛)|2
(7)

Signal from the secondary microphone will be used for

noise power spectrum estimation [14].

|𝐷𝐷(𝑚𝑚, 𝑛𝑛)|2 = 𝛽𝛽|𝑋𝑋2 (𝑚𝑚, 𝑛𝑛)|2 (8)

In this paper, the controlling parameter of spectral

subtraction, β is determined based on the heuristics.
The STFT of the processed signal, Y (m, n) can be
Fig. 3. Schematic representation of background noise
obtained as: mitigation and signal detection algorithm.

𝑌𝑌(𝑚𝑚, 𝑛𝑛) = 𝐺𝐺(𝑚𝑚, 𝑛𝑛) ∗ 𝑋𝑋1 (𝑚𝑚, 𝑛𝑛) (9) The duration of click signal was approximately 4
milliseconds that corresponds to a window of 40 data
Finally, the filtered signal in time domain y(t) can be points. The standard deviation in this window around
obtained by performing inverse STFT of Y (m, n). Fig. the click signal is very high as compared to that around
2 shows schematic representation of STFT based only noise. Therefore, standard deviation over a
spectral subtraction. window of 40 data points was selected as a
distinguishing criterion for click signal detection. A
To further improve the SNR of y(t), MATLAB’s moving window standard deviation was calculated
built-in 1-D digital filter function was implemented as over the entire length of squared output signal to
a matched filter. The filter design can be depicted as: identify the time at which the click occurs. A threshold
value was considered for the moving window standard
f(t) = filter [b(t), 1, y(t)]
 Neha S. Joshi et al. / Procedia Manufacturing 26 (2018) 1287–1295 1291
Neha S. Joshi, Sonali Singh/ Procedia Manufacturing 00 (2018) 000–000 5

deviation of the entire signal. The threshold was pin connector, one 8-pin connector, and two 2-pin
determined based on the type of the connector and the connectors.
corresponding signal level. When the standard Recordings were made for two conditions of
deviation exceeded threshold value, an output of unity noise: no noise and with white noise. Two sets of test
was generated. This identified the presence of click recordings were made for each of the two
signal. When the standard deviation was below the experimental conditions with five connections in each
threshold value an output of zero was generated, run. The test was performed in the following
indicating the absence of click signal. This final output chronology:
signal is binary and it could be used as an input to an
actuator such as light or an alarm to indicate that the Table 1. Experimental conditions for recordings
electrical connection has been established correctly. # Noise Condition Runs Total
1 No noise 2 12
4. Experimental set-up 2 Noise 2 12

To assess the effectiveness of background noise White noise was introduced into the room through
mitigation and signal detection algorithm, the tests a loudspeaker which was kept 1.5 meters away from
were conducted first in a semi-anechoic chamber the sound signal. The location of noise was chosen to
(sound pressure level = 30 dB) and then in a noisy maximize spatial separation of the speech and noise.
environment (sound pressure level = 87 dB). The
sound pressure level in noisy environment represented 5. Results and discussion
the upper level of expected short term environmental
noise in a factory setting [16]. The duration of the click sound was found to last
The detection system consisted of two pre- on average 4 milliseconds. The eDAQ with sampling
polarized condenser microphones connected to an frequency 10,000 Hz registers 40 data points in 4
eDAQ data collection system. The primary milliseconds, which were sufficient to capture the
microphone (wrist microphone) was placed on a glove click signal characteristics. The parameters of STFT
to keep it close to the signal of interest while the and inverse STFT were experimentally determined to
secondary microphone (arm microphone) was placed achieve a balance between time resolution and
50 cm from the first one in order to pick up the frequency resolution. Maximum noise reduction
background noise. The 50-cm distance was chosen to without distortion of signal was achieved using control
replicate a factory setting wherein an associate wears parameter value of 5. This value was generalized for
one microphone on the wrist and the other on the upper all the connectors under consideration in order to
arm. achieve maximum possible noise reduction. These
The equipment used in the experiments are as parameters are summarized in Table 4.
follows: Condenser microphones (PCB 377B02),
Signal Amplifier (PCB 426E01 ICP), Signal Table 2. Parameters of signal processing algorithm
Gain Estimator
Conditioner (SoMat EHLS-ICP-B), DAQ (SoMat Control parameter (β) 5
eDAQ). STFT and Inverse STFT parameters (Hanning Window)
Window
1024 Overlap 50%
Length
Sampling Samples Per
10000 1024
Frequency Frame

The sound pressure data recorded by both wrist

and arm microphone for all four connectors in white
noise environment (87 dB) is depicted in Fig. 5 and
Fig. 6, respectively. The plots are centered on a
Fig. 4. Microphones placed 50 cm apart. respective click event.
For testing, six different types of connectors were
used: one 3-pin connector, one 4-pin connector, one 6-
61292 Neha S. Joshi,
NehaSonali Singh/
S. Joshi et al.Procedia Manufacturing
/ Procedia 0026
Manufacturing (2018) 000–000
(2018) 1287–1295

Fig. 5. White noise environment: unprocessed sound pressure data Fig. 7. White noise environment: processed (output of Spectral
x1(t) recorded by wrist microphone Subtraction) sound pressure data y(t) with reduced background
noise.

In figure 8, focusing further in on the processed vs

unprocessed click it can be noted that the
characteristics of the click signal has not been
significantly distorted during spectral subtraction
processing while reducing the noise content.

Fig. 6. White noise environment: unprocessed sound pressure data

x2(t) recorded by arm microphone

After employing the spectral subtraction

algorithm outlined above, significant reduction in the
background noise can be observed for all the
connectors in Fig. 7. The click event centered in each Fig. 8. White noise environment: processed (y(t) - solid line) and
plot is readily apparent. unprocessed (x1(t) - dashed line) sound pressure signals
 Neha S. Joshi et al. / Procedia Manufacturing 26 (2018) 1287–1295 1293
Neha S. Joshi, Sonali Singh/ Procedia Manufacturing 00 (2018) 000–000 7

The noise reduction methodology was also tested

The difference between spectrogram of wrist for a quiet environment (30 dB background noise)
microphone before and after processing, suggesting a recordings to further validate the noise reduction
significant reduction in noise, is evident in the Fig. 9 without significant distortion or removal of the signal
and Fig. 10. of interest. As depicted in Fig. 11, even for the quiet
environment, the signal of interest has been
maintained after the spectral subtraction process with
the note that the background noise content was already
low.

Fig. 9. White noise environment: spectrogram of unprocessed

sound pressure data recorded by wrist microphone.

Fig. 11. Quiet environment: processed or y(t) (solid line) and

unprocessed or x1(t) (dashed line) sound pressure signals

After validating the spectral subtraction method

did not remove the click signal content, a rudimentary
signal classification step was included to demonstrate
a contrived lab application which determines the
presence of a click event.
The processed signal obtained in the previous
section was used as an input to a matched filter for
classification. From Fig. 11, the output of the matched
filter depicts an increased signal amplitude. However,
the amplitude of the noise has also increased. The
squaring process reduces this noise. The moving
window standard deviation plot for squared output of
matched filter is shown in Fig. 12.

Fig. 10. White noise environment: spectrogram of processed sound

pressure data recorded (output of Spectral Subtraction).
 8 Neha S. Joshi, Sonali Singh/ Procedia Manufacturing 00 (2018) 000–000
1294 Neha S. Joshi et al. / Procedia Manufacturing 26 (2018) 1287–1295

Fig. 11. Output of FIR filter f(t). Fig. 13. Final binary output signal P(t) of the proposed algorithm.

6. Conclusion

The performance of the proposed signal

processing strategy for detection of successful
electrical connections was assessed for a limited set of
connectors within a controlled noise environment.
A dual microphone spectral subtraction noise
mitigation and subsequent signal classification
through average template matching was shown to
significantly reduce the noise content and allow
effective process classification within a lab
environment under white noise. The target sound
signal was shown to not be significantly compromised
in either quiet or noisy environments. Noise mitigation
and signal detection was reliably achieved for four of
the five electrical connectors tested, with each signal
processed by signal matching filter. The sound level of
the 2-pin connector was outside of the detectable range
of the microphone; therefore, classification of such a
signal may need to rely on other detection means such
as inertial measurement, ultrasonic measurement, or
Fig. 12. Moving window standard deviation g(t) plot. The dashed
signal fusion processing methods in order to extract
line indicates the threshold standard deviation value
key features. This example highlights the need for
further investigation into alternate sensing and signal
In order to classify the signal, any data point
fusion strategies for the many varied types of electrical
above the threshold was designated with level 1,
connectors used in modern automotive manufacturing.
indicating that it was part of the target event, and any
This is one of the first papers to assess spectral
data point below the threshold was designated with
subtraction as a sound denoising method for human-
level 0, indicating that it did not belong to the event. A
process assembly interaction data. Human-process
sample output plot from this classification method is
interaction data will increasingly become more
shown in Fig 13.
 Neha S. Joshi et al. / Procedia Manufacturing 26 (2018) 1287–1295 1295
Neha S. Joshi, Sonali Singh/ Procedia Manufacturing 00 (2018) 000–000 9

prevalent as ubiquitous sensing, resulting from real- Implants,” Trends Amplif., vol. 16, no. 2, pp. 102–
world Industry 4.0 implementations, increases the 116, 2012.
number and variety of data streams coming from and [7] U. Shrawankar and V. Thakare, “Noise estimation
going to humans. and noise removal techniques for speech
Future work includes comprehensive recognition in adverse environment,” in
experimentation with variable background noise. The International Conference on Intelligent
experiments in this work were conducted under a Information Processing, 2010, pp. 336–342.
constant noise level environment but the soundscape [8] J. S. Lim and A. V. Oppenheim, “Enhancement
of an assembly line is highly variable in nature with and Bandwidth Compression of Noisy Speech,”
the background noise level changing between low to Proc. IEEE, vol. 67, no. 12, pp. 1586–1604, 1979.
short periods of high background noise, and departure
[9] S. Karimian-Azari and T. H. Falk, “Modulation
from Gaussian behavior. A wider range of connector
spectrum based beamforming for speech
styles and locking methods, as well as characteristic
enhancement,” in 2017 IEEE Workshop on
signatures of “good” and “bad” connections should be
investigated in order to gain a better understanding on Applications of Signal Processing to Audio and
how to advance the existing strategy and quantify its Acoustics (WASPAA), 2017, pp. 91–95.
limits for eventual industrial deployment. [10] M. A. Abd El-Fattah, M. I. Dessouky, S. M. Diab,
and F. E.-S. Abd El-Samie, “Speech
Enhancement Using an Adaptive Wiener Filtering
References Approach,” Prog. Electromagn. Res. M, vol. 4,
pp. 167–184, 2008.
[1] BMW Group, “New perspectives in production: [11] R. W. Morris, J. A. Arrowood, and M. A.
BMW Group draws on digitalization to advance Clements, “Comparison of Autoregressive
production network,” Press Release, 2015. Parameter Estimation Algorithms for Speech
[Online]. Available: Processing and Recognition,” in Proceedings.
https://www.press.bmwgroup.com/global/article/ (ICASSP ’05). IEEE International Conference on
detail/T0229627EN/new-perspectives-in- Acoustics, Speech, and Signal Processing, 2005.,
production:-bmw-group-draws-on-digitalization- 2005, vol. 1, no. 5, pp. 229–232.
to-advance-production-network?language=en. [12] N. Evans, J. Mason, W. Liu, and B. Fauve, “On
[Accessed: 11-Nov-2016]. the fundamental limitations of spectral
[2] B. Stringham, “Wire Processing: Guided Harness subtraction: An assessment by automatic speech
Assembly and Testing,” Assembly Magazine, recognition,” Proc. EUSIPCO, pp. 2–5, 2005.
2010. [Online]. Available: [13] S. Boll, “Suppression of acoustic noise in speech
https://www.assemblymag.com/articles/88714- using spectral subtraction,” IEEE Trans. Acoust.
wire-processing-guided-harness-assembly-and- Speech Signal Process., vol. 27, no. 2, pp. 113–
testing. [Accessed: 23-Jan-2016]. 120, 1979.
[3] B. C. F. de Oliveira, R. C. C. Flesch, A. L. S. [14] H. Gustafsson, I. Claesson, S. Nordholm, and U.
Pacheco, and M. B. Demay, “Vision system for Lindgren, “Dual-Microphone Spectral
detection of defects in the electrical connector of Subtraction,” 2000.
electric motors: test rig and algorithms,” J. Appl. [15] S. Amornwongpeeti, N. Ono, and M.
Instrum. Control, vol. 5, no. 1, pp. 34–42, 2017. Ekpanyapong, “Design of FPGA-based rapid
[4] M. D. Mura, G. Dini, and F. Failli, “An Integrated prototype spectral subtraction for hands-free
Environment Based on Augmented Reality and speech applications,” 2014 Asia-Pacific Signal
Sensing Device for Manual Assembly Inf. Process. Assoc. Annu. Summit Conf. APSIPA
Workstations,” Procedia CIRP, vol. 41, pp. 340– 2014, 2014.
345, 2016. [16] OSHA, “Noise,” Occupational Safety and Health
[5] D. H. F. Liu and B. G. Liptak, Environmental Administration Technical Manual, 2013.
Engineers’ Handbook, Second Edition, Second. [Online]. Available:
CRC Press, 1997. https://www.osha.gov/dts/osta/otm/new_noise/in
[6] K. Kokkinakis, B. Azimi, Y. Hu, and D. R. dex.html. [Accessed: 04-Jun-2017].
Friedland, “Single and Multiple Microphone
Noise Reduction Strategies in Cochlear