Xiaoliang Zhao1

Intelligent Automation, Inc., 15400 Calhoun

Drive, Rockville, MD 20855
e-mail: xzhao@i-a-i.com

Venugopal K. Varma
Oak Ridge National Laboratory, Nuclear Science
& Technology Division, Oak Ridge, TN 37831

Gang Mei
Bulent Ayhan
Chiman Kwan
Intelligent Automation, Inc., 15400 Calhoun
Drive, Rockville, MD 20855

In-Line Nondestructive Inspection

of Mechanical Dents on Pipelines
With Guided Shear Horizontal
Wave Electromagnetic
Acoustic Transducers
Circumferential guided ultrasonic Shear Horizontal (SH) wave Electromagnetic Acoustic
Transducer (EMAT) pairs mounted on a mobile fixture in a through-transmission mode
were used for detection and characterization of mechanical dents on the outer surface of
a pipe wall from inside the pipe. The dents were created on a 12 in. diameter standard
seamless steel pipe by hydraulically pressing steel balls of various sizes into the pipe
wall. n1 mode SH wave was directed through and along the wall of the pipe. Multiple
measurements were obtained both from the dents and from the no-flaw region of the pipe
using the EMAT pair. Dent features were extracted with a Principal Component Analysis
(PCA) technique and classified into cup and saucer types using Discriminant Analysis (DA). The overall approach is able to detect and classify dents of depth 25% through
wall or deeper, which should meet the needs of the pipeline safety inspection community
(U.S. Department of Transportation, Research and Special Program Administration).
Preliminary dent depth estimation potential is also shown via an amplitude correlation
approach.
DOI: 10.1115/1.1991879

Introduction

Underground pipelines used for transporting natural gas and

petroleum fluid are often subject to outside force damages such as
third party mechanical damages or earthquakes/floods. The supply
of energy has too often been disrupted by local pipeline leaks due
to the damages. Historically, mechanical damage is the single
largest cause of failures on pipelines 1. It deforms the shape of
the pipe, scrapes away metal and coating, and changes the mechanical properties of the pipe near the damage. The mechanical
dents formed as a result of the damage can usually be divided into
two basic groups, namely, cup and saucer dents. The saucer
dents are smooth, typically noninjurious; however, for cup dents
that are abrupt, certain range of field conditions including soil
type, stress, cathode potential, coating conditions, and temperature, etc. may lead to a catastrophic failure via the coincidental
metal loss. Axial and circumferential field magnetic flux leakage
MFL in-line inspection ILI smart Pipe-Inspection-Gears PIG
and compression wave ultrasonic transmission devices, currently
used for detecting metal loss and other defects like Stress Corrosion Cracks SCC, are limited by a localized point-by-point inspection approach and have the difficulty of reliably detecting the
very injurious cup dents and associated coincidental metal loss.
There is a great need for an in-line inspection technology that can
detect, classify, and characterize the mechanical dents on the outer
surface of the pipelines.
Ultrasonic guided waves are elastic waves propagating along a
thin walled structure or structural boundaries 24. Shear Horizontal SH wave, a special type of guided wave whose particle
displacement is parallel to the structure surface, has the advantage
of simple wave structure, less mode conversion, and less attenu1
Corresponding author.
Contributed by the Pressure Vessels and Piping Division for publication in the
JOURNAL OF PRESSURE VESSEL TECHNOLOGY. Manuscript received: February 28, 2005;
final manuscript received: April 6, 2005. Review conducted by: Sam Zamrik.

304 / Vol. 127, AUGUST 2005

ation to fluid load and coating, etc. This makes it an ideal choice
for defect detection in pipes. For example, Hirao and Ogi 5
proposed a circumferential SH-wave EMAT technique for detecting corrosion defects on the outer surface of steel pipelines with
and without protective resin coating, the amplitude change of the
SH wave signal inferred the presence of a corrosion; Gauthier 6
used multi-mode SH waves generated by EMATs to form B-scan
images to detect notches on a pipe; Zhao and Rose 7 calculated
the reflection and transmission coefficients of SH waves passing
through a two-dimensional surface-breaking groove or a stringer
like internal inclusion in a pipe using Boundary Element method;
and Luo, Rose, and Kwun 8 investigated the SH magnetostrictive transducers mounted on the outer wall of the pipe for axial
crack detection and sizing.
In this paper, circumferential guided ultrasonic SH wave EMAT
sensor pairs in a through-transmission mode are proposed for the
mechanical dents detection and characterization. Mechanical
dents were created on a 12 in. diameter standard seamless pipe by
hydraulically pressing various size steel balls into the pipe wall.
n1 mode SH wave was directed through and along the wall of the
pipe. The mechanical dents within the wave propagation path introduced their unique signatures. By thoroughly analyzing the
multimeasurement wave forms collected from the EMAT pair at
multilocations, defect features were extracted with a Principal
Component Analysis PCA technique and classified into cup
and saucer types with the Discriminant Analysis DA. It is seen
that this overall approach can detect and classify the dents of
depth 25% through wall or deeper. Preliminary dent depth estimation potential is also shown via an amplitude correlation approach.

Circumferential SH Wave EMAT Technology

Since the envisioned dent inspection system is mounted on a

PIG that travels along the pipe, a circumferential direction guided
wave in a pipe automatically realizes the two-dimensional scanning of the pipe wall. The theoretical development of the circum-

Copyright 2005 by ASME

Transactions of the ASME

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Fig. 1 Circumferential SH wave dispersion curves in a 12 in.

standard pipe

ferential SH waves in a pipe can be found in 9, where the dispersion relation under a traction free boundary condition was
written as
Jk1kTa Jk+1kTaY k1kTb Y k+1kTb Jk1kTb
Jk+1kTbY k1kTa Y k+1kTa = 0

In which a and b are the inner and outer radius of the pipe,
respectively. kT = / cT is the wave number of the shear wave.
Jkx and Y kx are, respectively, the first and second kind Bessel
functions of order k, and k is the normalized nondimensional wave
number. For a 12 in. diameter standard pipe, the dispersion curves
of the circumferential SH wave is shown in Fig. 1. In our study, n1
mode SH wave was used due to its high excitation and reception
efficiency compared with the n0 mode 5.
The basic components of EMAT consist of a face coil and magnets. It works under the Lorentz force principle in a nonferromagnetic metal 10. The oscillating current in the face coil induces
oscillatory eddy current at the surface of a metal close to the face
coil. Under the magnetic field of the permanent magnets or electromagnets, the induced eddy current will exert vibration force to
the lattice of metal microstructure, which in turn induces mechanical vibration in the material. The EMAT receiver is just the reverse of that process. It is of a great advantage that EMAT does
not require couplant for transmitting energy into the material like
conventional piezoelectric transducers. And it can be easily put
onto and taken off from the structure and has very good measurement repeatability. Figure 2 shows the diagram of the SH EMAT
principle and a sample SH EMAT probe used for this study. The
length of the alternating magnets in the wave propagation direction determines the slope of the excitation line that runs from the
origin across the dispersion curve see Fig. 1 dashed line, i.e.,
= 2d =

Cp
f

Fig. 2 a Principle of the Lorentz force SH wave EMAT and b

SH EMAT probe designed for this study

hydraulic press with support braces under it. A fixture attached to

the ram on the hydraulic press secured the steel balls. By controlling the amount of displacement of the ram, the dents of different
depth were created. The larger diameter balls were used to create
saucer dents while the smaller ones produced the cups. The
differentiation of a cup and a saucer is based on the severity of the
slope of the defect. Figure 3 shows the diagram of the pipe specimen with the mechanical dents created. Figure 4 gives an example
of the cup and saucer created on the 12 in. diameter pipe. It can be
seen that the cup dent is relatively smaller in size but much
sharper, while the saucer dent is flat and large in area. There are
also some minor deformations around the dents seen from the
photo.
To obtain a representative sample of the EMAT response to
varying dent sizes, multiple dents were created. When the steel
ball penetrates the 12 in. pipe and creates a dent, the material is
displaced, but due to the elasticity of the pipe, the material springs
back when the ram is released. The amount of spring back is

where is the wavelength of the SH wave to be generated, d is

the distance between the center of the two adjacent alternating
magnets, C p is the phase velocity, and f is the excitation frequency.

Experimental Setup and Data Collection

3.1 Mechanical Dents Creation. A 12 in. diameter standard

wall thickness 9.525 mm seamless steel pipe sample was selected to study the feasibility of detecting mechanical dents using
EMATs. The dents on the pipe were created using a hydraulic
press. Chrome finish steel balls of 5, 3.75, 1, and 0.75 in. diameter
were used for the creation of dents. The pipe was placed on the
Journal of Pressure Vessel Technology

Fig. 3 Three rows of cup and saucer dents created on a 7 feet

long, 12 in diameter schedule 40 seamless pipe. They are 1 foot
apart from each other or the pipe end for the same row.

AUGUST 2005, Vol. 127 / 305

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Fig. 5 Diagram of the dent and the induced deformation

Fig. 4 Sample cup and saucer dents created by a 0.75
and 5 diameter steel ball

difficult to predict and control. The depth of the dent created is

dependent on how the pipe is secured and where the supports are
relative to the ram. In some cases while the ram was pushing the
ball into the pipe, the opposite end of the pipe lifted upward. To
eliminate all these variables, measurements are based on the dent
depth after they have been created and not during the creation of
dent. The depths were referenced with the original no-dent condition, i.e., the deepest point of the dent to the original pipe surface.
Although the ball was spherical, due to the curvature of the pipe
circumference, the dents created were ellipsoidal. Table 1 gives
values for the depth and area of the dents, as well as their major
and minor diameters. Note that around the area of the dents, there
was a region that experienced a flattening or gradual slope to the
actual depression of the cup and saucer. This area is referred
to as the area of deformation. At the area of deformation, the
curvature of the pipe is reduced a bit. Figure 5 explains the shape
around the dents.
3.2 EMAT Data Collection. After the dents were created on
the pipe, two of the EMAT probes shown in Fig. 2 were used to
collect the circumferential SH wave through-transmission data.
Figure 6 shows the receiver and transmitter EMAT inside the 12
in. pipe. The flaws were positioned between the two EMATs. The
spacing between the transmitter and receiver could be adjusted in
the fixture for the particular measurement of interest. Note that for
field use, even though the flaw position is not known a priori,
since the EMAT probes are bi-directional, they can still inspect the
whole circumference of the pipe with a one-dimensional scan. For
all the tests here, the center-to-center transmitter-receiver spacing
was set at 165 mm. The frequency of the pulser was set at 263.4
kHz with a pulse width of 22 microseconds and a repetition rate of
12 milliseconds. The signals are amplified by 93 dB before being
collected by the data acquisition unit.
Since EMAT measurement in this configuration is not an abso-

lute measurement of the dents, baseline readings were also obtained to reference the signal deviation from the norm. To obtain
the baseline readings, the EMATs were positioned on the pipe
with no visible flaws. Multiple measurements were taken to obtain
a good sample of the no-flaw signal. Figure 7 shows the sample
wave forms for the normal condition, a cup dent and a saucer dent
collected by the EMAT sensors. It is seen that the amplitude of the
direct through signal for the cup and saucer dents was reduced.
Moreover, the signal from the saucer dent becomes dispersed due
to a multipath effect around the dent, which is less so for the cup
dents because of the smaller size.

Mechanical Dents Detection and Classification

Once the EMAT data were collected for all the 16 dents and
some normal pipe conditions, a time domain correlation analysis
method was used for dents detection. The correlation value is
calculated by the following procedures:
1 Use the first normal scan as the reference sample.
2 For any new sample, shift it in the time domain so that the
correlation value iMainlobeRe f iNew Samplei is maximized.
3 The outcome correlation value is recorded as this new
sample to the reference sample.
During this calculation, we noticed that the wave forms for shallow dents i.e., depth less than 2.5 mm did not show much visual
difference from the normal cases, and the correlation method also
did not give correct detections of those dents. This could be potentially viewed as a detection limitation of the EMAT sensors
since the dents are too small to bear any changes to the through
transmission wave form. For dents that are deeper than 2.5 mm,
we collected repeatedly 150 data on three cup dents Nos. 2, 3,
and 5 in Table 1 and 150 data on the saucers Nos. 8, 10, and 11
in Table 1, 50 for each dent. The correlation value of their wave
forms to the reference signal is shown in Fig. 8.
By choosing an appropriate threshold as shown in a dash-dot
line in Fig. 8, we can determine the presence of dents whose

Table 1 Dent number and size information all dimensions are

in inches

Fig. 6 EMAT configured inside the test pipe

306 / Vol. 127, AUGUST 2005

Transactions of the ASME

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Fig. 7 Sample waveforms of EMAT through-transmission signal from a normal pipe section, a cup dent, and a saucer dent

depths are larger than 2.5 mm. This depth is much less than 12.7
mm of which the OPS pipeline safety inspection regulation requires 1. Table 2 shows the detection results of the collected
data.
In order to classify the different types of dents as cup and
saucer, we used a PCA-DA based algorithm which is composed
of three parts: Feature extraction by Power Spectral Density
PSD, feature dimension reduction by Principal Component
Analysis PCA, and data classification using Discriminant Analysis DA. The block diagram of the applied classification method
is depicted in Fig. 9.
The features used in the classification are extracted by one of
the power spectral density methods: Periodogram 11. Since the
PSD consists of large dimensions of data, only its principal components which contain more than 95% of the original information

Fig. 8 Time domain correlation value for the normal condition,

cup dents, and saucer dents. The first 90 samples are for normal conditions.

Journal of Pressure Vessel Technology

were extracted through a minimum-loss transformation algorithm

called PCA. Details of the PCA algorithm can be found in the
book by S Haykin 12. Finally, the dimension-reduced feature
vectors were trained and used to classify different dent types via
Discriminant Analysis see Appendix for the algorithm details
13.
The 300 dent-data sets were divided into two sets: Training data
set 100 Cup, 100 Saucer and testing data set 50 Cup, 50 Saucer. The testing data were from a dent that was not used for

Fig. 9 Block diagram of the PCA-DA based classification

method

AUGUST 2005, Vol. 127 / 307

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Table 2 Results for the testing data sets regarding the dent
detection

Table 3 Results for the testing data sets regarding the cup/
saucer classification

training. A total of 14 principal component features which shows

good computational accuracy were used, and the test results are
shown in Table 3.
We can see that with the PCA-DA algorithm, we can successfully detect cup dents deeper than 2.5 mm with no error. This is
critical since cup dents are considered more harmful to the safety
of the transmission lines. The classification rate for saucer dents is
not as good as cup dents, however, with more EMAT data and
more dents specimen, we could produce a table or a probability of
detection versus false alarm rate curve for each type of dents so
that the inspection personnel have a better picture of how effective
the EMAT sensor for dent detection and what the inspection limitation is.

different sizes of dents, and the EMAT data collection conditions.

This will be studied more in the future. Relatively, the saucer
dents amplitude decreases more for the same depth compared to
the cup dents, which complies with our intuition since saucers
have larger dented area and block more wave energy that is coming through.
With the trend of the dent signal amplitude calibrated for a pipe,
the signal amplitude of a newly collected signal could potentially
be used for the dent depth estimation, albeit not exactly, upon
successful separation of cup or saucer types. This will certainly provide useful information for the pipeline inspectors.

Dent Depth Estimation Potential

Once the dents were classified into two groups, we would like
to estimate their depths which could potentially help inspection
engineers to schedule repair or replacements. To examine how the
dent depth increase correlates to the EMAT through-transmission
signal, all the 9 groups of normal data, 6 groups of cup dents
data and 10 groups of the saucer data were calculated for their
signal envelop. The peak value of the direct through signal envelop were then plotted versus the dent depth. The result is shown
in Fig. 10. Note that for each dent, 10 frames of data were used.
Due to the EMAT signal strength variations and background
noises, the peak amplitudes of the signals have some variances,
thus each forming a cluster in the figure. Two dashed lines were
drawn in the figure to roughly fit the trend of the clusters, showing
that the overall trends of the amplitudes of the cup and saucer
dents decrease with the increase of dent depths. However, there
are some local amplitude fluctuations that may be due to the constructive or destructive interferences of SH wave passing through

Conclusions

Circumferential guided ultrasonic Shear Horizontal SH wave

has been utilized for dent inspection in a pipe. n1 mode SH wave
generated by EMATs mounted on a mobile fixture inside a pipe
can successfully detect and classify mechanical dents of depth
25% through wall or deeper on the outer surface of the pipe wall.
Preliminary study shows that the overall through-transmission
wave form amplitude decreases with the increase of dent depth;
saucer dents have relatively larger variance of signal amplitude
compared with cup dents and decrease more with depth due to
the larger dent area.

Acknowledgments
This work is supported by U. S. Department of Transportation,
RSPA/OPS under Contract No. DTRS57-04-C-10053. The authors
would also like to thank Mr. James Merritt of RSPA for his technical suggestions.

Fig. 10 EMAT through-transmission signal amplitude versus the dent depth showing the overall decreasing signal strength with
deeper dents. Saucer dents experienced a relatively sharper decrease compared to the cups. This information can be used
inversely to estimate the dent depth after cup and saucer separation.

308 / Vol. 127, AUGUST 2005

Transactions of the ASME

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use

gix = xtWix + witx + i0 ,

Appendix: Discriminant Analysis

Suppose there is a set of n dimensional samples, D
= x1 , , xN, where each data sample is represented by a data
vector, x j. The samples in D belong to a total of c classes and Di
will be used as the notation to represent the data samples in class
i, where i = 1 , , c. Let mi represent the class mean, and ni represent the number of samples in class i. The relation between
within-class scatter matrix and scatter matrices can then be depicted by A1.
c

SW =

S,

A1

i=1

where the scatter matrices can be expressed by A2.

Si =

x m x m .
i

A2

xDi

1
m=
n

1
x,
ni xD

A3

1
x=
n im i .
n i=1

A4

Using these definitions, the total and between class scatter matrices are introduced in A5A7.
ST =

x mx m ,

A5

SB =

n m mm m ,
i

A6

i=1

ST = SW + SB .

A7

The DA based classifier is based on computing the c discriminants

for the given c classes in the data set. In the computation of the
discriminant functions within this classifer, it will be assumed that
the covariance matrices of each class, i, are different, and i is
the vector corresponding to the mean of each class. The resulting
discriminant functions are quadratic, and mathematically described in A8:

Journal of Pressure Vessel Technology

Wi = 2 i1 ,

A9

i1i ,

A10

wi =

i0 = 21 iti1i 21 lni + ln Pi.

A11

Note that in A11, Pi represents the prior probability of class

i. In this study, it is assumed that each class has the same prior
probability. There are c discriminant functions, where gix denotes the discriminant function of class i. The class label of
sample x is determined by comparing the values of the c discriminant functions with respect to x. The maximum value of the c
discriminant functions determines the decided class label, *, for
x, as depicted in A12.

The class mean and total mean vectors can be computed as A3

and A4, respectively.
mi =

A8

where Wi is mathematically introduced in A9, wi in A10, and

i0 in A11:

* = arg maxgix.
i

A12

References
1 U. S. Department of Transportation, Research and Special Program Administration, 2003, Pipeline Safety: Pipeline Integrity Management in High Consequence Areas Gas Transmission Pipelines, Federal Register 49 CFR Part
192 Docket No. RSPA007666; Notice 4 RIN 2137AD54.
2 Viktorov, I. A., 1970, Rayleigh and Lamb Waves, Plenum Press, New York.
3 Auld, B., 1990, Acoustic Fields and Waves in Solids, Krieger Press, Malabar,
Florida.
4 Rose, J. L., 1999, Ultrasonic Waves in Solid Media, Cambridge University
Press, Cambridge, UK.
5 Hirao, M., and Ogi, H., 1999, An SH-Wave EMAT Technique for Gas Pipeline Inspection, NDT & E Int. 32, pp.127-132.
6 Gauthier, J., Mustafa, V., Chabbaz, A. and Hay, D. R., 1998, EMAT Generation of Horizontally Polarized Guided Shear Waves for Ultrasonic Pipe Inspection, ASME International Pipeline Conference 1, 327-334.
7 Zhao, X., and Rose, J. L., 2003, Boundary Element Modeling for Defect
Characterization Potential in a Waveguide, Int. J. Solids Struct. 40, pp. 2645
2658.
8 Luo, W., Rose, J. L. and Kwun, H., 2004, Circumferential Shear Horizontal
Wave Axial-Crack Sizing in Pipes, Res. Nondestruct. Eval., 154, pp. 123.
9 Zhao, X., and Rose, J. L., 2004, Guided Circumferential Shear Horizontal
Waves in an Isotropic Hollow Cylinder, J. Acoust. Soc. Am., 115, pp. 1912
1916.
10 Thompson, R. B., and Alers, G. A., 1972, Application of Direct Electromagnetic Lamb Wave Generation to Gas Pipeline Inspection, Proceedings of the
IEEE Ultrasonic Symposium, pp. 9194.
11 Hayes, M. H., 1996, Statistical Digital Signal Processing and Modeling, John
Wiley & Sons, New York.
12 Haykin, S., 1990, Neural Networks, Prentice-Hall Press.
13 Duda, R. O., Hart, P. E., and Stork, D. G., 2001, Pattern Classification, John
Wiley & Sons, New York.

AUGUST 2005, Vol. 127 / 309

Downloaded From: http://pressurevesseltech.asmedigitalcollection.asme.org/ on 08/30/2015 Terms of Use: http://www.asme.org/about-asme/terms-of-use