Upload
vivek-anandan
View
13
Download
2
Embed Size (px)
Citation preview
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6, pp. 709-715 JUNE 2016 / 709
© KSPE and Springer 2016
Prediction of Surface Roughness using Spectral Analysisand Image Comparison of Audio Signals
R. Panneer1,#, S. P. Harisubramanyabalaji1, C. A. Sribalaji1, A. Vivek1, and G. Vigneshwaran1
1 School of Mechanical Engineering, SASTRA University, Thirumalasamudram, Thanjavur, Tamilandu, 613401, India# Corresponding Author / E-mail: [email protected], TEL: +91-9566730200
KEYWORDS: Audio signals, Histogram, MATLAB, Microphone, Spectrogram, Surface roughness
The aim of this work is to design an off-line system, method and experimental set-up for predicting surface roughness (Ra) of metal
surfaces with the help of audio signals. The frictional contact between a metal surface and sharp pencil like scratching tool will
produce audio signals which vary based on the roughness of the surface. The samples considered to design and validate the concept
are work pieces machined with metal cutting processes such as Turning and Grinding. Several audio signals are generated from
various types of metal surfaces produced by these processes after the completion of the machining process away from the machining
area in an enclosed chamber. The audio waves are captured with the help of a microphone fixed inside the chamber. These audio
signals are processed to generate the surface pattern of the relevant surface. The audio signals are then converted to spectrogram
and normalized histogram plots with the help of MATLAB, based on which the roughness of the surfaces is predicted. An experimental
set-up is designed which provides a sound-proof environment to capture and record the audio signals. The proposed system, method
and set-up are validated with the actual surface roughness of the chosen surfaces measured with the help of a surface roughness
measurement instrument.
Manuscript received: December 10, 2015 / Revised: February 1, 2016 / Accepted: February 11, 2016
1. Introduction
Quality of any surface is specified with the help of three major
parameters such as surface roughness, waviness, and lay. Surface
Roughness is the very small deviation of the surface from the intended
flat ideal plane. The most important aim of manufacturing industry is
to produce surfaces with relevant surface finish suitable to the function
of the surfaces because apart from dimensions and geometry, surface
roughness plays a major role in realizing the function of the surface.
Therefore both online and offline measurement of surface finish is
required in the manufacturing industries. There are many parameters
available for the measurement of surface finish among which the Average
Roughness (Ra) is broadly used by the Industry.1 For measurement of
Ra both contact and non-contact methods are used. Nano indenters,
Rutherford Backscattering Spectroscopy, Dektak Talysurf, are various
methods/devices which employ the contact method.2-4 Generally the
contact type methods use probes which glide over the concerned metal
surface to measure the Ra value.5,6 The limitation of the contact method
is that the probe used for measurement is very petite, fragile and of
high cost. It wears out quickly if it is continuously used on a metallic
surface of high surface roughness. Optical Microscopy, Interferometry,
Confocal and Atomic Force Microscopy, Surface Topography, Image
Processing, Holographic Contouring, and Ultrasound Reflectometry are
various methods/devices which employ the non-contact method.7-16
Non-Contact methods are considered to be the nondestructive type of
measurements in which, mostly a light source like laser or sound is
made to fall on the surface and the backscattered waves and light are
absorbed and analyzed to get the Ra value of the surface. It is observed
that the cheapest available contact type surface roughness tester in the
market is Mitutoyo 178-561-02A Surftest SJ-210 costs around INR 0.14
million. The cheapest available non-contact type Microscope that can
measure surface roughness costs a minimum of INR 0.20 million which
NOMENCLATURE
Ra = Arithmetic Mean Deviation of the Profile
µm = micrometer
DOI: 10.1007/s12541-016-0088-7 ISSN 2234-7593 (Print) / ISSN 2005-4602 (Online)
710 / JUNE 2016 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6
may not be economical to small scale industries for offline measurement
of non-critical surfaces. The proposed device will cost only around INR
0.02 million for a single device. In case of mass production, the cost of
the device will further get reduced. In the proposed method, sound is
generated by the dry friction contact between the finished work piece
and a sharp pencil like tool in an experimental set-up outside the
machine that encloses the surface to be measured and the microphone
in a sound-proof chamber. The sound can be captured with the help of
microphone without the noise interruption and analyzed using
MATLAB.17 Though the proposed method is a contact type
measurement, it can be deployed in the areas of surfaces which are not
functionally critical. Hence this method will reduce the cost of
measurement without compromising quality.
2. Methodology
The method used is based on the theory that when two surfaces are
rubbed against each other, audio signals are generated whose
characteristics depend on the roughness of the two rubbing surfaces.18
Since this method uses audio signals, a microphone is used to acquire
the sound waves generated from the dry friction contact between a tool
and the sample surface considered. The microphone is placed at a
distance of 3 cm from the contact point of the tool and the sample. The
dry friction contact is made for a sampling length of 10 cm. The
microphone is selected with less attenuation value so that the sound
that is to be captured is not attenuated. The noise from the surrounding
is removed with the help of microphone parameter adjuster so that
there will not be any mixing of external noise. The force applied on the
tool and speed of movement of the tool should be uniform to get a good
quality audio signal. Moving the tool over the surface manually may
lead to the non-uniform speed and force which may lead to deep cuts
(cuts > 1 mm) and damage to the surface. The intensity of pressure of
the tool on the surface at various points from beginning to the end of
the tool-metal contact should not vary. Hence the metallic surface is
moved under the tool with constant velocity and applied force for a pre-
determined distance without any deep cuts using a motor. Once the audio
signals are generated because of the frictional contact between the
sample surface and the tool, they are transferred to MATLAB software.
To know the intensity range of the audio signals, spectrogram plot is
generated using MATLAB. Since different surface roughness give rise to
sounds of different intensities, spectrogram plot is made to show that
difference. To get the surface roughness values of the surfaces, a histogram
plot is generated. Histogram plot gives a plot in the form of number of
pixels vs. intensity of the color values, from the spectrogram plot. This
method uses image comparison technique to give the error value between
the known roughness value of a sample and unknown roughness value
of a sample. The roughness value of the surface for which the error is
least is predicted as the roughness of the unknown sample.
3. Experimental Set-Up
Fig. 1 shows a hollow rectangular base which is raised to a height
to avoid external vibrations and to provide rigid support. This base
contains two parallel groves which holds set of steel balls. These steel
balls provide smooth movement of the movable base plate which is
placed over them. As shown in Fig. 2, a lead screw is fixed in between
the parallel groves and bottom of the base plate. The lead screw is
operated by a motor fixed at one end of the lead screw. When the lead
screw is rotated, the movable base plate translates along the length of
the lead screw. The sample is fixed using a clamping arrangement fixed
to the movable base plate. The clamping arrangement is provided with
fixed and adjustable jaws that enables the mounting of samples of
various lengths. Fig. 3 shows a semi sound-proof chamber which
remains stationary around the testing area. Fig. 4 shows the upper
sound-proof chamber which provides insulation from external sound.
This upper sound proof chamber is attached to the raised supports on
the base plate. It is capable of moving in the vertical direction. The
upper soundproof chamber when brought down encloses the lower
stationary semi sound-proof chamber and provides double insulation
from external sound at the base level and enhances the quality of the
sound captured by the microphone. A microphone and a tool attached
to the upper sound-proof chamber as shown in Fig. 5 reaches the base
level when the chamber is lowered. The tool and the microphone
remain stationary while the movable base plate along with the metallic
sample moves and creates the sliding friction contact between the tool
and sample.
Fig. 1 Sliding mechanism
Fig. 2 Movable base plate attached to the sliding mechanism
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6 JUNE 2016 / 711
4. Experimentation
For this work, sample Mild Steel Cylindrical Pieces of Ø 25 mm ×
150 mm are machined using cylindrical grinding and turning operation.
Totally 10 samples are made, 5 pieces ground in Cylindrical Grinding
Machine and 5 pieces Turned in Conventional Lathe. Each sample is
kept in the prototype of the device and sound is generated by making
dry friction contact with the pencil like sharp tool as explained below.
Each sample is placed in the clamping mechanism inside the semi-sound
proof chamber and clamped. After the upper sound proof chamber is
brought down to cover the sample and semi-sound proof chamber, the
motor is operated which rotates the lead screw and thereby moves the
sample against the bottom of the tool. The tool is placed perpendicular
to the sample’s axis to generate a sound of superior quality. The sound
produced by the movement of the sample surface under the tool due to
friction is recorded by the microphone. This procedure is repeated for
all the 10 samples and the audio signals are recorded, which is then
processed in MATLAB. The major parameters of measurement are the
tool sharpness and speed of movement. As the measurement is made
only for 10 samples, the sharpness is assumed to be constant and linear
speed is 0.01 m/sec. The sound is generated three times for each sample
for a sampling length of 10 cm on the same spot and audio signals are
generated. It is observed that the signals are almost same.
The surface roughness values of all the ten samples are determined
using a portable “Touch” type surface finish measuring instrument
manufactured by “Mitutoyo, Japan”. The roughness values are determined
on the same spots where audio signals are generated and the average
value of three measurements is considered. One sample each from turning
and grinding is assumed as unknown sample but with knowledge that
the value of its roughness lies between the ranges of roughness values
of other four known samples. A spectrogram is plotted for each of the
predetermined eight known samples (four each from each operation) and
also for the unknown samples. The spectrogram of the unknown sample’s
signal is compared with the spectrogram of the known sample’s signal
by comparing their normalized histogram values. The Euclidian distance
is calculated for each case and the known sample with the least error
is predicted to have the roughness of the unknown sample.
5. Results and Discussions
5.1 Spectrogram plots
The Spectrogram plot19 is made taking frequency along Y-axis and
Time along X-axis. The portion which is Red in colour indicates the
area where the intensity of sound is higher and the portion which is
Yellow in colour indicates the area where the intensity of sound is
comparatively less. The Spectrogram plots of audio signals obtained
during dry friction of Tool on ground surfaces are presented in Figs. 6
to 10. (10 represents the plot for which Roughness value is assumed
unknown). Similarly the Spectrogram plots of the audio signals
obtained during dry friction of tool on turned surfaces are presented in
Figs. 11 to 15. (15 represents the plot for which Roughness value is
assumed unknown).
5.2 Histogram plots
Depending on the density of the Red color in the Spectrogram plot,
the final normalized histogram plot is generated. The comparison of the
unknown sample’s histogram plot with the known sample’s histogram
plot gives the difference in intensity of the plot as the error value. The
comparison of the pair in which a least error value obtained is taken as
the nearest value of the Surface Roughness of the unknown sample. The
Fig. 3 Semi Sound-proof arrangement remains stationary around the
testing area
Fig. 4 Upper sound-proof chamber and microphone attached to it
Fig. 5 Emery tool attached to the upper sound-proof chamber beside
the microphone
712 / JUNE 2016 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6
Fig. 6 Spectrogram plot for ground surface - sample 1
Fig. 7 Spectrogram plot for ground surface - sample 2
Fig. 8 Spectrogram plot for ground surface - sample 3
Fig. 9 Spectrogram plot for ground surface - sample 4
Fig. 10 Spectrogram plot for ground surface - unknown sample
Fig. 11 Spectrogram plot for turned surface - sample 1
Fig 12 Spectrogram plot for turned surface - sample 2
Fig. 13 Spectrogram plot for turned surface - sample 3
Fig. 14 Spectrogram plot for turned surface - sample 4
Fig. 15 Spectrogram plot for turned surface - unknown sample
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6 JUNE 2016 / 713
Figs. 16 to 19 represents the Histogram plot obtained by comparison of
the Unknown ground sample with the known ground Samples.
5.3 Prediction of surface finish
Table 1 represents the values of Surface Roughness of the ground
samples obtained from the Roughness Measurement Instrument and the
error values as compared to the unknown sample obtained from the
Normalized Histogram Plot. From Table 1, it is observed that the least
error value (1.2446×10-4) is for Sample 1 and hence the Surface
Roughness Value of the unknown ground sample is predicted as 0.23
µm.
Table 2 represents the values of Surface Roughness of the turned
samples, obtained from the Roughness Measurement Instrument and
the error values as compared to the unknown sample obtained from the
Normalized Histogram Plot. From Table 2, it is observed that the least
error value (0.49776×10-4) is for Sample 2 and hence the Surface
Roughness Value of the unknown turned sample is predicted as, 2.98
µm.
Fig. 16 Comparison of histogram plots of the unknown sample with
sample 1 (ground surface)
Fig. 17 Comparison of histogram plots of the unknown sample with
sample 2 (ground surface)
Fig.18 Comparison of histogram plots of the unknown sample with
sample 3 (ground surface)
Fig. 19 Comparison of histogram plots of the unknown sample with
sample 4 (ground surface)
Fig. 20 Comparison of histogram plots of the unknown sample with
sample 1 (turned surface)
Fig 21 Comparison of histogram plots of the unknown sample with
sample 2 (turned surface)
Fig. 22 Comparison of histogram plots of the unknown sample with
sample 3 (turned surface)
Fig. 23 Comparison of histogram plots of the unknown sample with
sample 4 (turned surface)
714 / JUNE 2016 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6
6. Validation
The unknown ground sample’s surface roughness is measured using
a surface roughness measurement instrument and observed that the
roughness value is 0.25 µm. From Table 1, it is predicted that the
Roughness Value of the unknown ground sample is 0.23 µm which is
nearly equal to the actual Roughness of sample 1. The unknown turned
sample’s surface roughness is measured using a surface roughness
measurement instrument and observed that the roughness value is 3.14
µm. From Table 2, it is predicted that the Roughness Value of the
unknown turned sample is 2.98 µm which is nearly equal to the actual
Roughness of sample 2. Table 3 presents the actual and predicted
roughness values of the unknown samples.
There is a good correlation between the measured surface roughness
and signals frequency. This correlation scientifically proves that the
undulations in the surface caused by the type of manufacturing process
will vary the frequency of the audio signals generated. Based on this
scientific principle it can be established that using the audio signals for
predicting surface roughness is a reliable method within an accuracy
level of ±8%, which can be further improved by increasing the number
of samples.
7. Conclusions
The audio signals produced by a pair of tool and metallic surface
vary in frequency and intensity based on the surface roughness which
can be used to generate spectrograms and histograms. The histograms
of surfaces with known roughness values can be compared with
Histograms of unknown roughness values. Based on the errors observed,
the surface roughness value of the unknown surface can be predicted.
Even though this method does not measure the surface roughness like
a measuring instrument, it is very useful to predict/compare the surface
roughness during the manufacturing process before the final machining
is done, or as an in process measurement. It can be also used to predict
roughness of surfaces which are not functionally critical where
inaccuracy up to ±10% can be tolerated. This method is a simple and a
economical method to predict surface roughness. The major limitation
is that it cannot be used inside the machine shop as it needs sound proof
environment. Further, if we make the system perfectly sound proof,
accuracy of the output signal can be increased and thereby the accuracy
of roughness measurement can be increased. The accuracy of this
method also increases with more number of known samples.
REFERENCES
1. Benardos, P. G. and Vosniakos, G.-C., “Predicting Surface
Roughness in Machining: A Review,” International Journal of
Machine Tools and Manufacture, Vol. 43, No. 8, pp. 833-844, 2003.
2. Elmas, S., Islam, N., Jackson, M., and Parkin, R., “Analysis of
Profile Measurement Techniques Employed to Surfaces Planed by
an Active Machining System,” Measurement, Vol. 44, No. 2, pp.
365-377, 2011.
3. Mignot, J. and Gorecki, C., “Measurement of Surface Roughness:
Comparison between a Defect-of-Focus Optical Technique and the
Classical Stylus Technique,” Wear, Vol. 87, No. 1, pp. 39-49, 1983.
4. Davinci, M. A., Parthasarathi, N., Borah, U., and Albert, S. K.,
“Effect of the Tracing Speed and Span on Roughness Parameters
Determined by Stylus Type Equipment,” Measurement, Vol. 48, pp.
368-377, 2014.
5. Bjuggren, M., Krummenacher, L., and Mattsson, L., “Noncontact
Surface Roughness Measurement of Engineering Surfaces by Total
Integrated Infrared Scattering,” Precision Engineering, Vol. 20, No.
1, pp. 33-45, 1997.
6. Devillez, A., Lesko, S., and Mozer, W., “Cutting Tool Crater Wear
Measurement with Infrared Scattering,” Wear, Vol. 256, No. 1-2, pp.
56-65, 2004.
7. Bhuiyan, M. S. H., Choudhury, I. A., and Dahari, M., “Monitoring
the Tool Wear, Surface Roughness and Chip Formation Occurrences
using Multiple Sensors in Turning,” Journal of Manufacturing
Systems, Vol. 33, No. 4, pp. 476-487, 2014.
8. Lee, C. S., Kim, S. W., Yim, D. Y., and Tönshoff, H., “An in-
Process Measurement Technique using Laser for Non-Contact
Monitoring of Surface Roughness and Form Accuracy of Ground
Surfaces,” CIRP Annals-Manufacturing Technology, Vol. 36, No. 1,
pp. 425-428, 1987.
9. Tanner, L. H., “A Comparison between Talysurf 10 and Optical
Measurements of Roughness and Surface Slope,” Wear, Vol. 57, No.
1, pp. 81-91, 1979.
10. Figgis, D. L. and Sarkar, A. D., “Wear Results from Talysurf
Traces,” Wear, Vol. 51, No. 2, pp. 317-326, 1978.
11. Persson, U., “Measurement of Surface Roughness on Rough
Table 1 Error values from histogram pair- ground samples
Sample
No.
Machining
process
Measured surface
finish (Ra)
Obtained error values from
histogram pair ×10-4
1 Grinding 0.23 µm 1.2446
2 Grinding 0.19 µm 2.5931
3 Grinding 0.30 µm 1.4136
4 Grinding 0.32 µm 7.8885
Table 2 Error values from histogram pair-turned samples
Sample
No.
Machining
process
Measured surface
finish (Ra)
Obtained error values from
histogram pair ×10-4
1 Turning 5.1 µm 7.3936
2 Turning 2.98 µm 0.49776
3 Turning 2.60 µm 2.44310
4 Turning 2.85 µm 0.72949
Table 3 Comparison of measured and predicted Ra
SampleMeasured/actual
surface finish (Ra)
Predicted surface
roughness (Ra)
Ground sample 0.25 µm 0.23 µm
Turned sample 3.14 µm 2.98 µm
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 17, No. 6 JUNE 2016 / 715
Machined Surfaces using Spectral Speckle Correlation and Image
Analysis,” Wear, Vol. 160, No. 2, pp. 221-225, 1993.
12. Jeyapoovan, T. and Murugan, M., “Surface Roughness
Classification using Image Processing,” Measurement, Vol. 46, No.
7, pp. 2065-2072, 2013.
13. Xiang, H. Z., Lei, Z., Jiaxu, T., Xuehong, M., and Xiaojun, S.,
“Evaluation of Three-Dimensional Surface Roughness Parameters
based on Digital Image Processing,” The International Journal of
Advanced Manufacturing Technology, Vol. 40, No. 3-4, pp. 342-
348, 2009.
14. Pancewicz, T. and Mruk, I., “Holographic Contouring for
Determination of Three-Dimensional Description of Surface
Roughness,” Wear, Vol. 199, No. 1, pp. 127-131, 1996.
15. Mitri, F. G., Kinnick, R. R., Greenleaf, J. F., and Fatemi, M.,
“Continuous-Wave Ultrasound Reflectometry for Surface
Roughness Imaging Applications,” Ultrasonics, Vol. 49, No. 1, pp.
10-14, 2009.
16. Gao, Z. and Zhao, X., “Roughness Measurement of Moving Weak-
Scattering Surface by Dynamic Speckle Image,” Optics and Lasers
in Engineering, Vol. 50, No. 5, pp. 668-677, 2012.
17. Harisubramanyabalaji, S. P., Sribalaji, C. A., Vivek, A.,
Vigneshwaran, G., Abhishek, S., et al., “Development of a
Theoretical Model for Prediction of Surface Roughness of Metallic
Surfaces using Acoustic Signals,” Indian Journal of Science and
Technology, Vol. 8, No. 22, pp. 1-7, 2015.
18. Singh, S. K., Srinivasan, K., and Chakraborty, D., “Acoustic
Characterization and Prediction of Surface Roughness,” Journal of
Materials Processing Technology, Vol. 152, No. 2, pp. 127-130,
2004.
19. Jiaa, C. L., “Spectrogram Analysis of Random Laser Texture Pattern
Media,” Surface and Coatings Technology, Vol. 123, No. 2-3, pp.
140-146, 2000.