You are on page 1of 7

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/250323551

Grinding Acoustic Emission Classification in


Terms of Mechanical Behaviours
Article in Key Engineering Materials January 2007
DOI: 10.4028/www.scientific.net/KEM.329.15

CITATIONS

READS

36

2 authors:
Xun Chen

James Marcus Griffin

Liverpool John Moores University

Coventry University

87 PUBLICATIONS 810 CITATIONS

19 PUBLICATIONS 64 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,


letting you access and read them immediately.

SEE PROFILE

Available from: James Marcus Griffin


Retrieved on: 08 September 2016

Key Engineering Materials Vol. 329 (2007) pp. 15-20


online at http://www.scientific.net
(2007) Trans Tech Publications, Switzerland

Grinding Acoustic Emission Classification


in Terms of Mechanical Behaviours
Xun Chen a and James Griffin b
School of Mechanical, Materials and Manufacturing Engineering
University of Nottingham, UK
a

xun.chen@nottingham.ac.uk, epxjmg@nottingham.ac.uk

Keywords: Grinding; Acoustic emission; Short time Fourier Transfer; Wavelet transfer; Neural
network.

Abstract. The material removal in grinding involves rubbing, ploughing and cutting. For grinding
process monitoring, it is important to identify the effects of these different phenomena experienced
during grinding. A fundamental investigation has been made with single grit cutting tests. Acoustic
Emission (AE) signals would give the information relating to the groove profile in terms of material
removal and deformation. A combination of filters, Short-Time Fourier Transform (STFT), Wavelets
Transform (WT), statistical windowing of the WT with the kurtosis, variance, skew, mean and time
constant measurements provided the principle components for classifying the different grinding
phenomena. Identification of different grinding phenomena was achieved from the principle
components being trained and tested against a Neural Network (NN) representation.
Introduction
When an object is subjected to an external force, it will generate elastic waves due to material
particle displacements and these waves can propagate in material media [1]. The elastic wave, also
called acoustic emission (AE), is mechanical vibration. When different external forces act on the
same material or the same external force acts on different materials, the elastic waves will have
different characteristics. Hence elastic waves can be used for material non-destruction tests or
monitoring many machining processes [2-4].
AE monitoring for grinding is more challengeable than other machining due to irregular
distribution of grits. Effective grinding monitoring is necessary for ensuring grinding quality and
improving grinding productivity. The thermal expansion accompanied by grinding may trigger
acoustic waves. Using AE for detecting grinding burn on line is a promising method [4-6]. A
comparison of the different behaviours of mechanical and thermal AE signals has been reported
recently [7]. However the details of grinding AE propagation in relation to grit mechanical
interference, such as rubbing, ploughing and cutting, are far from fully being understood. This paper
presents an investigation of grinding AE in relation to grit mechanical interference.
The AE wave is a type of non-stationary stochastic signal. In material tests, AE signals are
traditionally presented as root-mean-square (RMS), event count, count rate, energy factor, amplitude
of dominant-frequencies and power of dominant-frequency bands. This manner of the parameter
record can save a lot of storage space so that it can be broadly used for monitoring events in a long
period such as days. For grinding monitoring, a data stream is necessary, because different events may
happen in a very short time. The Fast Fourier Transform (FFT) is often used for extracting some
features of AE signals in the frequency domain. However, grinding acoustic emission is the transient
elastic energy spontaneously released when materials undergoing deformation or fracture or both. For
these non-stationary signals, FFT will make the frequency composition average over the duration of
the signals. As a result, FFT cannot adequately describe the characteristics of the transient signals in
the frequency domain [8].

All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the
written permission of the publisher: Trans Tech Publications Ltd, Switzerland, www.ttp.net. (ID: 128.243.220.21-27/05/07,21:02:22)

16

Advances in Abrasive Technology IX

Short time Fourier transform (STFT) gets round the problem by introducing time duration
information. However, it is difficult to balance both time and frequency resolution requirements when
using the STFT. Wavelet Transforms (WT) uses a family of orthonormal bases to overcome the
limitations of FFT in representing non-stationary signals through time-frequency analysis. It may
extract AE features of grinding events with limit time duration more effectively. The WT need not
have infinite assumption support as the FFT does and can be nonzero for only a short duration of time.
The WT can translate a time-domain signal into a representation that is localized both in the time and
frequency domains. The multi-resolution analysis of WT offers a solution of representing AE signals
which contain high frequency bursts of short duration and low frequency components of longer
duration [9-10]. This is extremely useful for grinding process monitoring.
The present investigation is motivated by the expectation that AE features of grinding defects can
be extracted clearly. The mechanical features of grinding AE are particularly interested. The
following is the investigation objectives of this paper:
To extract mechanical AE features through single grit scratching tests.
To identify the difference of mechanically induced AE in relation to grinding phenomena of
rubbing, ploughing and cutting.
Acoustic Emission in Grinding
As discussed previous, acoustic emission is a stress release process. The acoustic emission in grinding
may come from elastic and plastic shear stress due to material removal or thermal deformation caused
by grinding heat. In order to examine the acoustic emission in grinding, grinding tests were carried out
on a Makino A55 machine centre. An Inconel 718 sample was ground by using an alumina wheel
under the condition: grinding speed vs = 35 m/s, workspeed vw = 1000 mm/min, depth of cut ap = 1.0
mm. A PAC WD AE sensor was used to detect the AE signals. The sampling rate was 2 MHz. By
using a bandpass filter with cut-off frequency of 50 kHz to 1200 kHz most of the noise generated by
machine vibration and wheel rotation were easily eliminated. Fig. 1 illustrates the time-frequency
features of the AE signals of a grinding process. It can be seen in this case that the main features of the
AE signal are concentrated at the ranges around 100 kHz, 240 kHz and 550 kHz. In figure 1, it is
difficult to distinguish the individual chip formation in grinding. It is necessary to scrutinise the detail
phenomena of acoustic emission during grinding.
Grinding process is a material removal process utilizing a large number of grits as cutting media.
Removal of materials during grinding depends on the cutting action of each grit involved in the
grinding operation. A ground surface is created as a result of a series of cutting actions from the grits.
During grinding, randomly spaced grits on the wheel surface engage with the workpiece sequentially.
Depending on the cutting depth, the grit may experience rubbing, ploughing and cutting three stages
engaging with workpiece materials. The proportion of rubbing, ploughing or cutting in a grit pass
depends on the amount of grit engagement that is determined by the grit position on the wheel surface
and grinding kinematics. The engagement of a grit in grinding can be illustrated by undeformed
grinding chip shape, which is commonly described by chip thickness and chip length. It was noted that
the stress under the grit depends on the undeformed chip thickness. The larger the undeformed chip
thickness, the higher the force is needed to remove the chip. Therefore higher stresses would be
initiated.
In order to understand each individual grit performance in grinding single grit scratch tests were
carried out. The scratch tests were carried out by feeding an Al2O3 grit towards the rotational flat
sample. With a grit in-and-out stroke, a scratch groove will be formed on the flat sample surface. The
maximum scratch depth is about 1.5 m, which is a typical value of grinding chip in high efficiency
grinding. The scratching speed is 35 m/s. By using short time Fourier transform, the features of

Key Engineering Materials Vol. 329

17

acoustic emission at both time-frequency domains can be illustrated as shown in Fig. 2. It can be seen
that the AE feature frequencies of the scratches are in the range 100 ~ 550 kHz, which are similar to
the AE feature frequencies in grinding tests as shown in Fig. 1. This confirms the grinding acoustic
emission is aggregation of acoustic emission from individual grit grinding actions. Higher frequency
components are relevant to grinding thermal behaviour [7]. The low amplitude of high frequency
elements indicates that the thermal effect in a single grit scratch is very small. It should be noted that
the AE frequency features are changing with the time during a scratch action. This may be the
important information that presents the differences between rubbing, ploughing and cutting.
Therefore, the AE propagation should be considered in both time and frequency features.
5

10

x 10

10

x 10

20
9

-10

-20

-30

-20

6
5

-40

4
-60

Frequency (Hz)

Frequency (Hz)

-40

-50

-60
4
-70
3
-80
2

-90

-80
1

0.02

0.04

0.06
0.08
Time (S)

0.1

0.12

-100
0

4
Time (S)

8
-4

x 10

Figure 1. Short time Fourier transfer spectrum of Figure 2. Short time Fourier transfer spectrum of
grinding acoustic emission signals
single grit scratch acoustic emission signals

Acoustic Emission Features of Rubbing, Ploughing and Cutting


Occurrence of rubbing, ploughing and cutting in a grinding process depends on the undeformed chip
thickness in front of each grit pathway. In the rubbing case, there is surface friction between grit and
test piece. Only elastic deformation involves in the rubbing action. However, for ploughing and
cutting, the energy is consumed from material plastic deformation, which would provide different AE
energy signatures. On the other hand the cutting is different from rubbing and ploughing in terms of
material removal. This suggests therefore that different AE signatures should be apparent between
these different phenomena. Ploughing and cutting are somewhat similar in that the materials are either
pushed/slided to sideway or removed respectively. As these predominately cause material
deformation they both have plastic material deformation effects. Whereas rubbing does not remove or
slide any material away instead it touches the surface with no visible markings which has elastic
material characteristics in that the material deforms and returns back to its original state after single
grit pass has occurred.
In short the boundaries are much closer in terms of ploughing and cutting AE distinguishing
features. Ploughing and cutting are perhaps the most difficult phenomena to separate based on this
assumption. The technique of both STFT and WT needs to be as accurate as possible to ensure the
time constants as well as the sharpness of the waveforms are accurately represented when trying to
distinguish the frictional energy and material deformation energy in the form of plastic or elastic
energy. The difference between STFT and WT is that the short-time window for STFT represents a
specific frequency band with respect to time and the user can not get an exact time reading at a
particular point of phenomenon. The WT on the other hand breaks the signal up into smaller

18

Advances in Abrasive Technology IX

frequency components albeit high-short and low-long frequency components and the user is able to
read the exact point of phenomenon start or finish. Therefore the WT resolution is much better than
that of STFT. However if an optimised window (not too small for reduction of frequency resolution
and not too big for reduction of time resolution) is applied, good STFT results can still be obtained.
In comparison to STFT, Fig. 3 and 4 display the WT energy spectrums at 3 levels which signify
cutting and rubbing respectively. These results were taken from the same scratch (single grit cut) and
the same piece of material (CMSX4). In looking at the energy spectrums in terms of features and
amplitude levels it is possible to segregate the phenomena however there are a lot of overlaps in the
signals that have to be ignored from presenting similar characteristics across one phenomenon border
to another. All AE extracted signals were tested against an AE pencil lead break test [11] where the
AE signals could be normalised on a day to day basis. The rubbing case in Figure 4 and others not
displayed here were found to be not pure rubbing due to slight indentation marks which does not
occur during rubbing. No scratch or mark is based on the rubbing phenomenon's elastic material
properties. This phenomenon is more of rubbing with little plastic deformation.
4

x 10

-6

Detail L1
1.5

x 10

-4

Detail L2

x 10

-5

Detail L3

3
1
2

4
0.5

7
8
Frequency (Hz)

0
2.5

10
x 10

3.5
4
Frequency (Hz)

4.5

5
x 10

1.5

2
Frequency (Hz)

2.5
x 10

Figure 3. WT energy spectrums of cutting AE signal at the levels 1~3

2.5

x 10

-6

Detail L1
7

x 10

-5

Detail L2
1

x 10

-4

Detail L3

0.8
5

1.5

0.6

4
3

0.4

2
0.5

0.2

1
0

7
8
Frequency (Hz)

10
x 10

0
2.5

3.5
4
Frequency (Hz)

4.5

5
x 10

1.5

2
Frequency (Hz)

2.5
x 10

Figure 4. WT energy spectrums of rubbing AE signal at the levels 1~3


A set of back-propagation neural networks (NN) were developed to identify different phenomena
involved in single grit scratch tests. The inputs of the NN are 256 neurons for the STFT data. There
were a total of 6 scratches made 2 mm apart for each material therefore a total of 24 scratches existed
for all the materials. The materials used in the following tests consisted of EN8 steel, Inconel 718,

Key Engineering Materials Vol. 329

19

CMSX4 and MAR-M002. Looking at Figure 5(a) the first initial results for STFT is very encouraging
albeit the data set is limited in size and the ploughing and cutting phenomenon has been combined.
This experiment gives a confidence milestone and paves the way for classification between all single
grit scratch phenomena, cutting, ploughing and rubbing. Figure 5(b) displays results that are again
very encouraging with only 1 misclassification. The NN structure was based on two hidden layers
with the input layer having 256 inputs (length of the STFT vector inputs) and the two hidden layers
being set to one and half times that amount. The output layer was set to one to ensure a crisp output
answer was obtained. Again the results were very encouraging with only one misclassification.

(a)
(b)
Figure 5. Classification results of the NN with STFT AE signal data
Considering the WT may provide better interpretation of the AE signals, a modified NN was
developed for the tests. The inputs of the NN are the combination of statistical windows of the WT
with the kurtosis, variance, skew, mean and time constant measurements. Using a combination of WT
and statistical windowing as the NN inputs it was possible to get the salient principle components and
obtain 100% classification for identifying both the two phenomena (Fig. 6(a)) and three phenomena
(Fig. 6(b)).

(a)

(b)

Figure 6. Classification results of the NN with WT and statistics AE signal data

20

Advances in Abrasive Technology IX

Conclusion
Monitoring of grinding using AE signals requires good understanding of the AE features in relation to
different grinding behaviour. This paper presents an investigation of using Acoustic Emission signals
to identify the grinding phenomenon of cutting, ploughing and rubbing. The paper has demonstrated
that both the STFT and WT and Statistics methods can extract useful features to distinguish the
different AE signals between the phenomena of cutting, ploughing and rubbing. With less resolution
the STFT displays similar information albeit less accurate than WT. A hybrid technique of both
techniques could prove more accurate for grinding process monitoring system. Looking at the NN
results both STFT and WT have a high confidence in segregating the different phenomena.
References
[1]

D. Royer, E. Dieulesaint: Elastic Waves in Solids I, II. (Springer-Verlag Berlin Heidelberg,


New York, 2000).

[2]

J.Webster, I. Marinescu, R. Bennett: Acoustic emission for process control and monitoring of
surface integrity during grinding, Annals of the CIRP, Vol. 43/1 (1994), pp. 299-304.

[3]

M. Chen, B.Y. Xue: Study on acoustic emission in the grinding process automation, The ASME
International Mechanical Engineering Congress and Exhibition, MED, (Nashville, TN, USA,
ASME, Fairfield, NJ, USA. 14-19 Nov. 1999).

[4]

Q. Liu, X. Chen, N. Gindy: Fuzzy pattern recognition of AE signals for grinding burn,
International Journal of Machine Tools and Manufacture, Vol. 45/7-8 (2005), pp. 811-818.

[5]

P. R. De Aguiar, J. F. G. De Oliveira: Production grinding burn detection using acoustic


emission and electric power signals, Abrasives (Dec 1998/Jan 1999), pp. 8-11.

[6]

Z. Wang, P. Willett, et al.: Neural network detection of grinding burn from acoustic emission,
International Journal of Machine Tools and Manufacture, Vol. 41/2 (2001), pp. 283-309.

[7]

Q. Liu, X. Chen, N. Gindy: Investigation of acoustic emission signals under a simulative


environment of grinding burn, International Journal of Machine Tools & Manufacture, Vol. 46
(2006), pp. 284292.

[8]

X. Li: A brief review: Acoustic emission method for tool wear monitoring during turning,
International Journal of Machine Tools and Manufacture, Vol. 42/2 (2002), pp. 157-165.

[9]

X. Li, J. Wu: Wavelet analysis of acoustic emission signals in boring, Proceedings of the
Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, Vol. 214/5
(2000), pp. 421-424.

[10] W.J. Staszewski, K.M. Holford: Signal processing of acoustic emission data, Key Engineering
Materials, Vol. 204-205 (2001), pp. 351-358.
[11] M. Barbezat, A.J. Brunner, P. Flueler, C. Huber and X. Kornmann: Acoustic emission sensor
properties of active fibre composite elements compared with commercial acoustic emission
sensors, Sensors and Actuators Vol. 114 (2004), pp. 13-20.