In-line Nondestructive Inspection of Mechanical Dents on Pipelines ...

Zhao et al. ASME J. Pres. Ves. Tech.

In-line Nondestructive Inspection of Mechanical Dents on Pipelines with Guided Shear Horizontal Wave Electromagnetic Acoustic Transducers

Xiaoliang Zhao1, Venugopal K. Varma2, Gang Mei1, Bulent Ayhan1 and Chiman Kwan1 1

2

Intelligent Automation, Inc., 15400 Calhoun Drive, Rockville, MD 20855

Oak Ridge National Laboratory, Nuclear Science & Technology Division, Oak Ridge, TN 37831

Corresponding author: Xiaoliang Zhao Email: [email protected]

Abstract Circumferential guided ultrasonic Shear Horizontal (SH) wave Electromagnetic Acoustic Transducer (EMAT) pairs mounted on a mobile fixture in a throughtransmission 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-inch diameter standard seamless 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 from the dents and 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 2.5mm (0.1 inch) or larger, which should meet the needs of the pipeline safety inspection community [1]. Preliminary dent depth estimation potential is also shown via an amplitude correlation approach.

1


1. 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 non-injurious; 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.

2

Zhao et al. ASME J. Pres. Ves. Tech. Ultrasonic guided waves are elastic waves propagating along a thin walled structure or structural boundaries [2-4]. 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 attenuation 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-inch 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 multi-measurement waveforms collected from the EMAT pair at multi-locations, defect features were extracted with a Principal Component Analysis (PCA) technique and

3

Zhao et al. ASME J. Pres. Ves. Tech. 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 2.5mm (0.1inch) or larger. Preliminary dent depth estimation potential is also shown via an amplitude correlation approach.

2. 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 circumferential SH waves in a pipe can be found in [9], where the dispersion relation under a traction free boundary condition was written as [ J k) −1 ( kT a ) − J k) +1 ( k T a )][Yk) −1 ( kT b) − Yk) +1 ( kT b)] − [ J k) −1 ( kT b) − J k) +1 ( kT b)][Yk) −1 ( kT a ) − Yk) +1 ( kT a )] = 0

(1)

In which a and b are the inner and outer radius of the pipe, respectively. kT = ω / cT is the wave number of the shear wave. J k) (x ) and Yk) (x ) are, respectively, the first and

) ) second kind Bessel functions of order k , and k is the normalized non-dimensional wave number. For a 12-inch 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 non-ferromagnetic metal [10]. The oscillating current in the face coil induces oscillatory eddy current at the surface of a metal close to

4

Zhao et al. ASME J. Pres. Ves. Tech. 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.

λ = 2d =

Cp f

(2)

where λ 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.

3. Experimental Setup and Data Collection 3.1 Mechanical dents creation

A 12-inch diameter standard (wall thickness 9.525mm) 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-inch, 3.75-inch, 1-inch and 0.75-inch diameter were used for the creation of dents. The pipe was placed on the 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

5

Zhao et al. ASME J. Pres. Ves. Tech. 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 “cup”s. 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-inch 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–inch 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 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

6

Zhao et al. ASME J. Pres. Ves. Tech. 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. Table 1 Dent number and size information (all dimensions are in inches)

Diameter

Dent Number

Axial

Circumferential

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.395 0.5 0.444 0.415 0.545 0.386 0.72 1.35 0.595 1.46 1.46 0.73 0.615 0.875 0.886 0.745

0.378 0.488 0.435 0.415 0.534 0.375 0.65 0.995 0.526 0.855 0.894 0.715 0.547 0.775 0.75 0.62

Depth

Area of deformation

0.085 0.218 0.124 0.065 0.20 0.062 0.033 0.118 0.04 0.134 0.121 0.094 0.070 0.179 0.123 0.076

0.63 x 0.55 1.4 x 0.75 0.95 x0.75 0.625 x 0.625 1.35 x1.2 0.55 x 0.51 1.25 x 0.875 3x3 1.25 x 0.95 2.4 x 1.56 1.6 x 1.3 1.1 x 1.1 0.95 x 0.65 1.75 x 1.3 1.5 x 1 1.1 x 0.85

Ball Diameter

0.75

1

3.75

5 3.75 3.75 5 5

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-inch 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 prior, since the EMAT probes are bi-directional, they can still inspect the whole circumference of the pipe with a 1dimensional scan. For all the tests here, the center-to-center transmitter-receiver spacing

7

Zhao et al. ASME J. Pres. Ves. Tech. 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 absolute 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 noflaw signal. Figure 7 shows the sample waveforms 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 were reduced. Moreover, the signal from the saucer dent becomes dispersed due to a multi-path effect around the dent, which is less so for the cup dents because of the smaller size.

4 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

∑ Re f

i

• NewSamplei is maximized.

i∈Mainlobe

(3) The outcome correlation value is recorded as this new sample to the reference sample.

8

Zhao et al. ASME J. Pres. Ves. Tech. During this calculation, we noticed that the waveforms for shallow dents (i.e., depth less than 2.5mm) 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 waveform. For dents that are deeper than 2.5mm, we collected repeatedly 150 data on three cup dents (No. 2, 3 and 5 in Table 1) and 150 data on the saucers (No. 8, 10 and 11 in Table 1), 50 for each dent. The correlation value of their waveforms 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 depths are larger than 2.5mm. This depth is much less than 12.7mm of which the OPS pipeline safety inspection regulation requires [1]. Table 2 shows the detection results of the collected data.

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

Normal Dents

Number of samples 80 300

Labeled as Normal 80 0

Labeled as Defect 0 300

Accuracy 100% 100%

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.

9

Zhao et al. ASME J. Pres. Ves. Tech. 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 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 dimensionreduced 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 training. A total of 14 principal component features which shows good computational accuracy were used, and the test results are shown in Table 3. Table 3 Results for the testing data sets regarding the Cup/Saucer classification.

Cup Saucer

Number of samples 50 50

Labeled as Cup 50 17

Labeled as Saucer 0 33

Accuracy 100% 66%

We can see that with the PCA-DA algorithm, we can successfully detect cup dents deeper than 2.5mm 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.

10


5. 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 throughtransmission 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 Figure 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 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.

11


6. 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 2.5mm or larger on the outer surface of the pipe wall. Preliminary study shows that the overall through-transmission waveform 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 number DTRS57-04-C-10053. The authors would also like to thank Mr. James Merritt of RSPA for his technical suggestions.

Appendix 1. Discriminant Analysis

Suppose there is a set of n dimensional samples, D = { x1 ,..., x N } , where each data sample is represented by a data vector, x j . The samples in D belong to a total of c classes and D i will be used as the notation to represent the data samples in class i , where

i = 1,..., c . Let m i 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). 12

Zhao et al. ASME J. Pres. Ves. Tech. c

SW = ∑ Si ,

(A1)

i=1

where, the scatter matrices can be expressed by (A2). Si =

∑ (x − m )(x − m ) i

x∈D i

t

i

.

(A2)

The class mean and total mean vectors can be computed as (A3) and (A4), respectively.

m=

1 ni

∑x,

(A3)

1 1 c x = ni m i . ∑ n∑ n x i =1

(A4)

mi =

x∈D i

Using these definitions, the total and between class scatter matrices are introduced in (A5-A7). ST = ∑ (x − m)(x − m)t ,

(A5)

x

c

S B = ∑ ni (m i − m)(m i − m)t ,

(A6)

i =1

ST = SW + S B .

(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, Σi , are different, and µ i is the vector corresponding to the mean of each class. The resulting discriminant functions are quadratic, and mathematically described in (A8): gi (x) = xt Wi x + w ti x + ωi 0 ,

(A8)

where Wi is mathematically introduced in (A9), w i in (A10), and ωi 0 in (A11):

13


1 Wi = − Σi−1 , 2

1 2

(A9)

w i = Σi−1µ i ,

(A10)

1 2

(A11)

ωi 0 = − µti Σi−1µi − ln Σi + ln P (ωi ) .

Note that in (A11), P (ω i ) represents the prior probability of class ωi . (In this study, it is assumed that each class has the same prior probability.) There are c discriminant functions, where gi (x) denotes the discriminant function of class ωi . 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, ω ∗ , for x , as depicted in (A12).

ω ∗ = arg max ( gi (x)) . 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. RSPA–00–7666; Notice 4] RIN 2137–AD54. [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.

14

Zhao et al. ASME J. Pres. Ves. Tech. [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. Sol. Struct. 40, pp. 2645-2658. [8] Luo, W., Rose, J. L. and Kwun, H., 2004, “Circumferential Shear Horizontal Wave AxialCrack Sizing in Pipes,” Research in Nondestr. Eval., 15(4), pp. 1-23.

[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., Alers, G. A., 1972, “Application of Direct Electromagnetic Lamb Wave Generation to Gas Pipeline Inspection,” Proceedings of the IEEE Ultrasonic Symposium, pp. 91-94. [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.

15

Zhao et al. ASME J. Pres. Ves. Tech. Figure Captions Figure 1: Circumferential SH wave dispersion curves in a 12-inch standard pipe

Figure 2: (a) Principle of the Lorentz force SH wave EMAT and (b) SH EMAT probe designed for this study

Figure 3: Three rows of cup and saucer dents created on a 7 feet long, 12 inch diameter schedule 40 seamless pipe. They are 1 foot apart from each other or the pipe end for the same row.

Figure 4: Sample “cup” and “saucer” dents created by a 0.75” and 5” diameter steel ball

Figure 5: Diagram of the dent and the induced deformation

Figure 6: EMAT configured inside the test pipe

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

Figure 8: Time domain correlation value for the normal condition, cup dents and saucer dents. The first 90 samples are for normal conditions.

Figure 9: Block diagram of the PCA-DA based classification method.

16


Figure 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

17


10.0 9.0

n2

Phase velocity (km/s)

8.0

n1

7.0 6.0 5.0 4.0

Operation point

n0

3.0 2.0

EMAT excitation line

1.0 0.0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frequency (MHz)

Figure 1: Circumferential SH wave dispersion curves in a 12-inch standard pipe

18


S

N

S

N

S

N

N

S

N

S

N

S

Magnet

Current Coils

Shear

f=JxB

Metal plate

(a)

Frame

Magnets EMAT Coil

(b)

Figure 2: (a) Principle of the Lorentz force SH wave EMAT and (b) SH EMAT probe designed for this study

19


1 foot

7 feet

cup dents

saucer dents

Figure 3: Three rows of cup and saucer dents created on a 7 feet long, 12 inch diameter schedule 40 seamless pipe. They are 1 foot apart from each other or the pipe end for the same row.

20


(a) “cup” dent

(b) “saucer” dent

Figure 4: Sample “cup” and “saucer” dents created by a 0.75” and 5” diameter steel ball

21


No-Deformation

Dent

Deformed region

Figure 5: Diagram of the dent and the induced deformation

22


EMATs

Inside pipe wall

Rollers EMAT positioning fixture

Figure 6: EMAT configured inside the test pipe

23


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

24


1.4

Normalized Correlation Value

1.2 1.0 0.8 0.6 0.4 0.2 0.0 0

50

100

150

200

250

300

350

400

Sample Number

Figure 8: Time domain correlation value for the normal condition, cup dents and saucer dents. The first 90 samples are for normal conditions.

25


Figure 9: Block diagram of the PCA-DA based classification method.

26


cup

Saucer

Figure 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

27