NDT&E International 38 (2005) 605–614

www.elsevier.com/locate/ndteint

Fault diagnosis of internal combustion engines using visual dot patterns

of acoustic and vibration signals
Jian-Da Wua,*, Chao-Qin Chuangb
a
Institute of Vehicle and Rail Engineering, National Changhua University of Education, 1 Jin-De Rd, Changhua City, Changhua 500, Taiwan, ROC
b
Department of Mechanical and Automation Engineering, Da-Yeh University, 112 Shan-Jiau Rd, Da-Tsuen, Changhua, 515 Taiwan, ROC
Received 3 January 2005; revised 14 February 2005; accepted 14 February 2005
Available online 19 March 2005

Abstract
An investigation of the fault diagnosis technique in internal combustion engines based on the visual dot pattern of acoustic and vibration
signals is presented in this paper. Acoustic emissions and vibration signals are well known as being able to be used for monitoring the
conditions of rotating machineries. Most of the conventional methods for fault diagnosis using acoustic and vibration signals are primarily
based on observing the amplitude differences in the time or frequency domain. Unfortunately, the signals caused by damaged elements, such
as those buried in broadband background noise or from smearing problems arising in practical applications, particularly at low revolution, are
not always available. In the present study, a visual dot pattern technique is proposed to identify the acoustic emission and vibration signals for
fault diagnosis in an internal combustion engine and drive axle shaft. Experiments are carried out to evaluate the proposed system for fault
diagnosis under various fault conditions. The experimental results indicate that the proposed technique is effective in the fault diagnosis of an
internal combustion engine and drive axle shaft.
q 2005 Elsevier Ltd. All rights reserved.

Keywords: Fault diagnosis; Visual dot pattern; Internal combustion engine; Drive axle shaft

1. Introduction spectrum level measured from the fan blades of an internal

Interest in the application of acoustic emissions and
vibration signals for fault diagnosis in rotating machinery has
grown significantly, having advanced with the progress in
digital signal-processing technology and algorithms in the
last two decades [1–5]. The acoustic and vibration signals of
rotating machinery, such as the cooling fan and internal
combustion engine normally consists of a combination of
basic frequencies with discrete or narrowband frequency
components and various harmonics, most being related to the
revolution of the machinery [6]. Diagnostic technology using
acoustic and vibration signals already exists in the form of
techniques for applying the fast Fourier transform (FFT) to
time-axis signals and analyzing changes in the spectrum of
energy. An example result of a sound emission power
* Corresponding author. Tel.: C886 47115443.
E-mail address: jdwu@cc.ncue.edu.tw (J.-D. Wu).
0963-8695/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ndteint.2005.02.007
combustion engine cooling system is shown in Fig. 1. The
energy is increased when the fan blade is damaged.
Therefore, most of the conventional fault diagnosis tech-
niques are used to observe the time-domain amplitude or
frequency-domain spectrum difference for damage diagno-
sis. However, in these conventional methods, a smearing
problem generally arises in practical applications, particu-
larly at low and other levels of machine revolutions.
Meanwhile, the signals caused by the damaged element are
not always available, such as those buried in broadband
background noise or the signal-to-noise (S/N) ratio of the
sound components used to detect the fault is too low. In these
conditions, an adaptive line enhancement system can be used
for improving the response of the fault signal [7]. Some other
fault diagnosis techniques such as order tracking diagnosis
have also been developed and implemented for fault
diagnosis in rotating machinery based on different faults
that can be observed on different order of signals [8–10].
In the present study, a fault diagnosis technique based on
visual dot pattern of acoustic and vibration signals is
proposed. According to the previous work by Pickover [11],
606 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614
100 Shaft rotation frequency
80
SPL (dB) 60
40
20
0
0 100 200
Fig. 1. Power spectrums of acoustic emission from engine cooling system. A solid line depicts cooling fan blades without damage; broken lines depict blade of
which one is damaged.
there exists a symmetrized dot technique for speech-
waveform analysis. The signal-processing algorithm has
also been demonstrated in DeRosier’s research for crunchy
cereal food [12] and Shibata’s research for fault diagnosis of
rotating machinery [3]. In the present study, a dot pattern
image automatic matching and fault identification system is
developed. The signals of acoustic emission and vibration
signals are measured from a four-stroke, four-cylinder,
internal combustion engine. Obviously, internal combustion
is one of the types of classical rotating-machineries that can
be measured using the acoustic and vibration signals for
fault diagnosis. When an engine component is damaged in
the combustion process, this malfunction may be obvious
through listening to the acoustic emission and vibration
signals due to the unstable cylinder pressure. However, the
acoustic emission and vibration signals that are caused by
faults are not always available due to high level of
background noise signals during practical engine operation.
In order to evaluate the proposed fault diagnosis technique,
four faults in the engine, five faults in the cooling fan
and three faults in the drive axle shaft under various
operations are designed on the test platform for evaluating
the proposed technique. The principle of the visual dot
pattern and the technique of digital image processing are
described in the following sections.
2. Visual dot pattern of acoustic and vibration signals
The acoustic emissions or vibration signals generated by
rotating machinery can be represented as the superposition
of sinusoidal signal. The proposed visual dot pattern is a
method for visually expressing the problem in an easy-to-
understand figure. The display characterizes signal wave-
forms using patterns of dots and requires very limited
computational times. Changes in the amplitude and
frequency of sound and vibration signals can be easily
identified. It involves the transformation of the sound
pattern into a set of dots having mirror symmetry, in which
Blade passing frequency
Twice blade passing frequency
300 400 500 600 700 800
Frequency (Hz)
the time waveform of the sound signal is visualized as a
snowflake-shaped pattern of six-fold symmetry. This pattern
provides a stimulus by which local visual correlations are
integrated to form a global percept. It can potentially be
applied to the detection and characterization of significant
features of any sound signal.
Fig. 2 shows the principle for plotting visual dot patterns
of acoustic and vibration signals. The visual dot pattern is
obtained by transforming the correlation of value F(t) of the
time-axis signal at time t and its value F(tCn) at time tCn
into polar coordinates P(r(n), f(n), Q(n)). One set of data
(F(n), F(tCn)) is plotted so as to be linearly symmetrical
with the initial line of the polar coordinate graph. Then, a
figure symmetrical with a point is created by repeatedly
rotating the axis of linear symmetry and plotting. The
transformation from the discrete signal F(n) at a fixed
sampling time n to polar coordinates is given by the
equation


FðnÞ K L
rðnÞ Z g; (1)
H KL
where r(n) is the radius of the visual dot pattern, F(n) is a
sound or vibration signal, H is the maximum value of F(n),
L is the minimum value of F(n), g is a gain value of signal, n
is the number of dots. All of the variable parameters of the
Fig. 2. Principle of visual dot pattern.
 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614 607

Table 1 fixed sine wave of 100, 300 and 500 Hz and random noise
Notation list of equations signals. The input signal F(n) is first normalized to the range
Variable Definition through finding the high (H) and low (L) values for the n
r(n) Radius of symmetrized dot pattern points of data in the window, so overall amplitude is not a
f(n) Angle of dot in symmetrized dot pattern factor in the characterization. The pattern is similar to the
Q(n) Negative angle of dot in symmetrized dot pattern result of Pickover’s paper [11].
F(n) Acoustic or vibration signal
i Time lag coefficient for plotting
g Gain value of input signals
H Maximum of signal waveform 3. Experimental verification of fault diagnosis systems
L Minimum of signal waveform
q Rotation angle of initial line
n The number of dots in signal waveform In this section, a fault diagnosis system that used the
f(x, y) Size of image within M!N proposed visual dot pattern algorithm is described. The
w(x, y) Size of sub-image within J!K experimental work is focused on the application of fault
s A variable on x axle of image size diagnosis in the internal combustion engine, cooling fan and
t A variable on y axle of image size
drive axle shaft of vehicles. As a preliminary test of the fault
w The average value of the pixels in w(x, y)
fðx; yÞ The average value of f(x, y) in the region coincident diagnosis system, an engine with no faults is operating and
with the current location of w the acoustic and vibration signals are measured and
recorded. Furthermore, the acoustic and vibration signals
are also measured with four different faults in the engine,
equations are summarized in Table 1. The angle of dots in five faults in the cooling fan and three faults in the drive axle
pattern are defined as shaft. The patterns from the above conditions are obtained

 using the proposed visual dot pattern algorithm and saved as
Fðn C iÞ K L a databank of automatic image processing for the purpose of
fðnÞ Z q C g; (2)
H KL fault diagnosis.


Fðn C iÞ K L 3.1. Experimental set-up and applications
QðnÞ Z q K g; (3)
H KL
In the experimental investigation, the acoustic and
where Q(n) is negative angle of the dot in pattern, q is vibration signals of automotive engine are measured to
rotation angle of the initial line and i is the time lag verify the proposed fault diagnosis system. The complete
coefficient for the purpose plotting. The algorithm maps the experimental arrangement for measurement, including the
normalized time waveform into visual dot space creating a accelerometer (PCB 353B15), condenser microphone
scatter plot of neighbor amplitudes on a polar graph. A point (ACO P4012), four-cylinder internal combustion engine
in the time waveform is mapped onto a radial component, and data recording system are shown in Fig. 4. For the
and the adjacent point is mapped to an angular component.
Fig. 3 shows the visual dot patterns of synthesized signals in

Fig. 3. Visual dot patterns of synthesized signals. (a) 100 Hz sine wave,
(b) 300 Hz sine wave, (c) 500 Hz sine wave, (d) random noise. Fig. 4. Measurement of acoustic and vibration signals in an automobile.
608 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614

Fig. 5. Acoustic signal patterns of IC engine in different operating conditions.

engine fault diagnosis, the signals are recorded from
the engine operation conditions of a normal operating
state and four different faults at engine speeds from 750
to 3000 rpm. The faults include one cylinder miss; two
cylinders miss firing, lag of spark ignition, and a leaking
of the intake manifold. The acoustic emission and
vibration signals will obviously increase in an engine
with faults because of change in the balance of the
engine and fluctuations in pressure [13]. The acoustic
signal dot patterns of an engine without any faults at
different speeds are displayed in Fig. 5. The acoustic and
vibration dot patterns of engines with different faults are
summarized in Figs. 6 and 7 as a databank for the
template matching image of the fault diagnosis system.
In the proposed fault diagnosis system, the faults of the
engine cooling fan are also considered. Typically, the

Fig. 6. Patterns of engine acoustic signals in different fault conditions.

 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614 609

Fig. 7. Patterns of engine vibration signals in different fault conditions.

Fig. 8. Experimental set-up of fault diagnosis in engine cooling fan.

610 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614

Fig. 9. Patterns of cooling fan acoustic signals in different operation conditions.

engine cooling fan noise has both random and repetitive
component in its frequency domain. For the proposed test
system, the centrifugal fan has five backward curved
blades. The acoustic signals are recorded for the fan’s
normal operating state and also when it has five different
faults. Fig. 8 illustrates the experimental set-up for the
fault diagnosis of the engine cooling fan. In this
experiment, the acoustic signal of the fan is measured
with a condenser microphone (ACO P4012). Fig. 9
shows the acoustic signal dot patterns and databank of
the cooling fan under different operating conditions.
The other experimental application of the proposed
diagnosis system is for a Rzeppa-type CV joint in a
vehicle drive axle shaft. The Rzeppa-type CV joint is the
most commonly used outer joint on most front-wheel-
drive vehicles. The outer joint must allow a turns of up
to 408 while the front wheels continue to move up and
down and transmit engine power to drive the front
wheels [14]. However, the outer joints often have to
endure variable and difficult road conditions. When a CV
joint wears or fails such as when the boot is torn, the
grease leaks or the assembly wears, the most common
symptom is noise while driving. In order to evaluate the
proposed fault diagnosis technique, the acoustic signals
are recorded from the axle shaft under normal operating
conditions and three different fault conditions. The shaft
revolution is varied from 200 to 1200 rpm. The faults
include the boot being torn, grease leaking, and the balls
and cage being worn or loose. Fig. 10 shows the
experimental apparatus, including an accelerometer
(PCB353B15), a condenser microphone (AOC P4012),
an optical fiber sensor (LM339), and a DC motor.

Fig. 10. Experimental set-up of measurement in drive axle shaft.

 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614 611

Fig. 11. Patterns of drive axle shaft acoustic signal in different fault conditions.

The acoustic and vibration dot patterns of different faults
are summarized in Figs. 11 and 12 and are used as a
database for the template matching system.
3.2. Principle of the image template matching system
Interest in digital image processing methods stems
from two areas of application: improvement in pictorial
information for human interpretation, and the processing
of scene data for autonomous machine perception. In this
study, we conclude our coverage of digital image
processing by developing a technique for image template
matching for fault diagnosis. The template matching
system for the pattern is shown as Fig. 13. Template
matching is one digital image processing method that can
be used to compare the pattern of images. In Fig. 14, we
consider an image correlation as the basis for finding
matches of a sub-image w(x, y) of size J!K within
an image f(x, y) of size M!N, where we assume that
J%M and K%N. Although the correlation approach can
be formulated in vector form, working directly with an
image or sub-image format is more intuitive. In
its simple form, the correlation between f(x, y) and
w(x, y) is
XX
cðs; tÞ Z f ðx; yÞwðx K s; y K tÞ (4)
x y
where sZ0, 1, 2,.,MK1, tZ0, 1, 2,.,NK1, and the
summation is taken over the image region where w and
612 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614

Fig. 12. Patterns of drive axle shaft vibration signals in different fault conditions.

P P
f overlap. In this procedure, we ensure that the origin x

y ½f ðx;yÞK f ðx;yÞ½wðxKs;yKtÞK w

of f(x, y) is at its top left and the origin of w(x, y) is at gðs;tÞZ P P  2
P P 
x y ½f ðx;yÞK f ðx;yÞ x y ½wðxKs;yKtÞK w
its center. Any value of (s, t) inside f(x, y) that applies
(5)
in Eq. (4) yields a value of c. As s and t can vary w(x,
y) moves around the image area yielding the function
c(s, t). The maximum value of c(s, t) indicates the where sZ0, 1, 2,.,MK1, tZ0, 1, 2,.,NK1, w is the
position where w(x, y) is the best match f(x, y). Note average value of the pixels in w(x, y) (computed only
that accuracy is lost for the values of s and t near the once), f(x, y) is the average value of f(x, y) in the
edges of f(x, y), with the amount of error being region coinciding with the current location of w, and
proportional to the size of w(x, y) [15]. The correlation the summations are taken over the coordinates common
function given in Eq. (4) has the disadvantage of being to both f and w, The correlation coefficient g(x,t) is
sensitive to change when it is within the amplitude of scaled in the range K1 to 1, independent of scale
f(x, y) and w(x, y). For example, doubling all values of changes in the amplitude of f(x, y) and w(x, y).
f(x, y) doubles the value of c(s, t). An approach For the template matching system, all of the signal
frequently used to overcome this difficulty is to patterns must be changed to an unusual pictures file
perform matching via the correlation coefficient, and each of the patterns must be the same in the picture
which is defined as center prior to the match. For the fault diagnosis, the
 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614 613

Fig. 13. Template matching of visual dot patterns in fault diagnosis.

unknown signal pattern was input into the template
matching system from the data record system. According
to the unknown pattern property, the matching system will
find the best matching pattern from the databank. The best
match pattern is the fault condition pattern for the
experiment.
was developed for the purpose of fault diagnosis. The
experimental results indicated that the proposed algorithm
was effective in fault diagnosis for experimental cases.
y
N

4. Conclusions t

The fault diagnosis technique is used to prevent early Origin

faults in a mechanical system. Most of the conventional K
methods for fault diagnosis using acoustic and vibration
signals are primarily based on observing the amplitude
difference of the time or frequency domain. Unfortunately, x M s J
the signals caused by damaged elements in practical
application, such as those buried in broadband background (s, t)
noise or those caused by smearing problems, are not always
w(x-s, y-t)
available. In this paper, a visual dot pattern technique using
f(x, y)
acoustic and vibration signals is used to identify faults in an
internal combustion engine, cooling fan and drive axle shaft
in a vehicle. An automatic image template matching system Fig. 14. Arrangement for obtaining the correlation of f(x, y) at point (s, t).
614 J.-D. Wu, C.-Q. Chuang / NDT&E International 38 (2005) 605–614
Acknowledgement
This study was supported by the National Science
Council of Taiwan, the Republic of China, under project
number NSC-93-2218-E-018-002.
References
[1] Lee SK, White PR. Higher-order time–frequency analysis and its
application to fault detection in rotating machinery. Mech Syst Signal
Proc 1997;11:637–50.
[2] Lin J, Qu L. Feature extraction based on Morlet wavelet and its appli-
cation for mechanical fault diagnosis. J Sound Vib 2000;234:135–48.
[3] Shibata K, Takahashi A, Shirai T. Fault diagnosis of rotating
machinery through visualization of sound signals. Mech Syst Signal
Proc 2000;14:229–41.
[4] Wang WJ, Lin RM. The application of pseudo-phase portrait in
machine condition monitoring. J Sound Vib 2003;259:1–16.
[5] Frank PM. Fault diagnosis in dynamic system using analytical and
knowledge-based redundancy: a survey and some new results. IEEE
Trans Automatic Control 1990;26:459–74.
[6] Wu JD, Bai MR. Application of feedforward adaptive active noise
control for reducing blade passing noise in centrifugal fans. J Sound
Vib 2001;239:1051–62.
[7] Wu JD, Hung RJ. Design of an adaptive line enhancement system
using a variable step-size affine-projection algorithm. NDT&E Int,
2005;38:387–93.
[8] Vold H, Leuridan J. High resolution order tracking at extreme slew
rates, using Kalman filters. SAE Paper 1993;931288:219–26.
[9] Vold H, Mains M, Blough J. Theoretical foundations for high
performance order tracking with the Vold–Kalman tracking filter.
SAE Paper 1997;972007:1083–8.
[10] Bai MR, Jeng J, Chen C. Adaptive order tracking technique using recur-
sive least-square algorithm. Trans ASME J Vib Acoust 2002;124:
502–11.
[11] Pickover CA. On the use of symmetrized dot patterns for the visual
characterization of speech waveforms and other sampled data.
J Acoust Soc Am 1986;80:955–60.
[12] DeRosier BL, Normand MD, Peleg M. Effect of lag on the
symmetrised dot pattern (SDP) displays of the mechanical
signatures of crunchy cereal foods. J Sci Food Agric 1997;75:
173–8.
[13] Halderman JD. Diagnosis and troubleshooting of automotive
electrical, electronic, and computer systems. New Jersey: Prentice-
Hall; 1997.
[14] Halderman JD. Automotive chassis systems: brakes, steering,
suspension, and alignment. New Jersey: Prentice-Hall; 1996.
[15] Gonzalez RC, Woods RE. Digital image processing. New York:
Addison-Wesley; 1992.