Developments in Ultrasonic Inspection II

Accurate Defect Sizing using Phased Array and Signal Processing

M.Moles, Olympus NDT, Canada; L. Wesley, T. Sinclair, University of Toronto

ABSTRACT
Accurately sizing cracks has been a main objective of ultrasonic nondestructive evaluation for
decades. Early sizing techniques based on echo amplitude have repeatedly been shown to be deficient.
Diffraction techniques offer a significant improvement, but are also limited by pulse length and
reliance on operator interpretation. In particular, Time-Of-Flight Diffraction is limited by pulse length
in terms of the smallest midwall defect it can size. Tip back diffraction works well with phased array
S-scans, but is also limited by similar factors to those limiting TOFD. Overall, diffraction techniques
(TOFD and back diffraction) currently have a typical accuracy of + 1 mm, with best practice pushing
the accuracy down to ~0.5 mm.
The University of Toronto and Olympus NDT Canada have been collaborating on the
development of digital signal processing techniques to improve the time resolution of ultrasonic
signals, based on Weiner filtering combined with autoregressive spectral extrapolation. Early results
on synthetic signals and simple laboratory test specimens showed promise, and the technique has now
been incorporated into S-scans on single fatigue cracks. The results showed a significant improvement
over current sizing accuracies using traditional S-scans.

INTRODUCTION
Accurate defect sizing is essential for Fitness-For-Purpose (FFP) applications, where defect height is a
key parameter. Since the arrival of ultrasonic nondestructive evaluation (NDE), about fifty years ago,
this technology has offered the potential for very accurate defect sizing compared to radiography.
With this potential, plus the ability to tailor inspections to the geometry and type of defect, ultrasonic
testing has been steadily encroaching on radiographys hold on weld inspection. This has accelerated
with the arrival of Automated Ultrasonic Testing (AUT), which permits the operator to save all data
for post-processing, audit the results and adjust some signal processing parameters after acquiring the
data. Despite these inherent advantages, there have been significant stumbling blocks to accurate
defect sizing using ultrasonics.

Simple Amplitude Sizing

The first ultrasonic sizing techniques used direct amplitude measurements as an estimate of size.
These approaches became formalized into two well-known codes the DGS system in Germany [1],
and the American Welding Society D1.1 code in the USA [2]. Both approaches were designed to be
very simple to use, and do not require any significant amounts of subjective judgment or interpretation
on the part of the operator.
These simple approaches have many technical limitations. For example, the ultrasound beam
must be centered on the defect, the material should conform to the American Welding Society
(AWS) attenuation ratings, the defect must be smaller than the beam, and must be correctly oriented.
However, natural defects tend to be rough, curved, twisted, tight, branched and characterized by many
other complicating features - all of which tend to produce variations in amplitude. These amplitude
approaches work quite well for simple workmanship applications where ductile steels and design
redundancy are used. However, they are not really applicable to advanced designs, high strength
steels, and other demanding materials and applications.

Advanced Amplitude Sizing Techniques

Several years later, more sophisticated and flexible amplitude sizing techniques were developed. For
these amplitude-drop methods, the beam is scanned over the defect, which is then sized by
measuring the relative decrease in signal magnitude compared to a reference value. A variety of
amplitude-drop criteria can be used, typically -20 dB, -6 dB and -3 dB. While these techniques do not
require specific transducers (as with AWS for example), they are still limited by other factors such as
defect morphology, orientation and operator ability. Figure 1 (top) shows defect through-wall sizing
results from the Defect Detection Trials in the UK, and compares ultrasonically measured depths from
all techniques used with true values obtained from metallographic measurements. Figure 1 (bottom)
indicates that only Time-of-Flight-Diffraction techniques showed consistently good correlation
between ultrasonic estimates and true defect depth. By contrast, the amplitude-drop sizing accuracy is
generally inadequate for Fitness-For-Service calculations.

Figure 1 - Measured defect depths vs. destructive depths for DDT trials [3]. Top: defect sizing in DDT
Trial plate 1 using all UT techniques; bottom: defect sizing using TOFD only.

TOFD (Time-Of-Flight Diffraction)

TOFD was developed by the UKAEA Harwell (now AEA Technologies) in the late 1970s by
Maurice Silk and associates [4]. Harwell developed an early automated ultrasonic system called
Zipscan, which used signal averaging to improve the signal-to-noise ratio. TOFD came to prominence
during the nuclear PISC II and DDT trials [3, 4], where it showed not only good detection capability
(especially in the midwall regions), but also significantly better sizing than conventional ultrasonic
testing (see Figure 1 bottom), where the TOFD data has been extracted from all the DDT Plate 1
sizing results.
Since the early 1980s, TOFD has become standard practice in the nuclear industry. In the last
few years, TOFD has also been accepted by the petrochemical and other industries.

Back Diffraction
Crack tip diffraction techniques can be applied using a variety of set-up geometries and wave modes in
addition to the longitudinal-longitudinal pattern in a pitch-catch configuration as used in standard
TOFD. Other possibilities include:
Shear-shear diffraction
Longitudinal-shear diffraction
Single transducer diffraction, called back diffraction or the tip echo method in Japan [5]
Twin transducer TOFD with both transducers on the same side of the defect/weld.
Of these many alternative configurations to standard TOFD, the most popular for sizing applications is
back diffraction, which has recently become more prevalent with the arrival of phased arrays [6].
The concept is shown in Figure 2 for surface-breaking flaws.

With monocrystal transducers, the biggest problem usually is identifying the crack tip. With phased
array True Depth S-scans, the tip is usually easy to identify as it is located on top of the corner signal
(for a vertical surface-breaking defect). In addition, the arrival of piezo-composite arrays has
significantly improved the tip resolution over the older piezo-ceramic arrays. The S-scan display
allows the operator to optimize the sizing as well [7]. Figure 3 shows an S-scan with tip backdiffracted signal arrowed.

Figure 2 - Back diffraction concept and waveforms for surface-breaking defects [6].

Figure 3 - Phased array S-scan with crack tip arrowed [12].

Defect Sizing Accuracy using Diffraction
TOFD Sizing Accuracy
Typical estimates of TOFD sizing are + 1 mm, though quotes of accuracies to + 0.2 mm have been
made [8]. The reality is that TOFD is significantly more accurate and more reliable if performed
correctly than amplitude sizing techniques. However, values of + 0.2 mm are unlikely to be achieved
on everyday inspections.
Some special applications have produced impressive results. For example, using 20 MHz
shear wave TOFD in thin-walled zirconium pressure tubes hydride blister cracks has yielded defect
size estimates to + 0.1 mm [9]. Again, these are not typical conditions.
Tip Back Diffraction Sizing Accuracy
Not surprisingly, back diffraction can size to similar accuracies to TOFD, but can also size smaller
defects. In general, these errors are ~1 mm, though careful practices have reduced the error to ~0.5
mm, as shown in Figure 4 [10].

Figure 4 - Sample accuracies from on ferritic steel headers. (Courtesy of OPG)

Digital Signal Processing Techniques for Improved Accuracy

The limiting accuracy seems to be data analysis by the ultrasonic testing operator, since ultrasonic
pulses feature a wide beam with a spectrum of frequencies. A quick look at the S-scans in this paper
shows the corner and tips signals to be of finite size, not Dirac delta functions. To that end, the
Olympus NDT and the University of Toronto are working on digital signal processing techniques for
sharpening the time resolution of the ultrasonic echoes in S-scans [11]. When proven and
commercialized, this should give operators with standard training the ability to accurately and reliably
size defects to fractions of a millimetre.
The approach selected is a semi-empirical technique using Weiner filtering combined with
autoregressive spectral extrapolation, derived from experience in the geophysical field. The technique
is described in more detail elsewhere [12], and only typical results will be shown here. A significant
amount of work was initially performed on artificial reflectors such as notches and drilled holes, but
the data presented here will also include the first results from actual cracks. The signal processing
algorithm functions independently, without user input of parameters or operator intervention, needing
only the raw data and a reference echo from the probe to complete an enhancement of an ultrasonic
signal. Figure 5 shows a sample data progress set from a side drilled hole. There is a significant
improvement in readability as the signal processing takes place.

Figure 5 - A specimen with a 0.88mm diameter hole drilled to 9.67mm of the opposite surface. The
original signal (a), the Wiener-filtered signal (b), AR order extrapolation result (c), and the analytic signal
magnitude of the ranged AR extrapolation (d).
When applied to actual S-scan back diffraction signals on actual samples with cracks, preliminary
signal processing results are shown in Figures 6-9. Figure 6 shows an original OmniScan S-scan with
both crack tip and corner echoes, reconstructed in MatLab where all calculations were performed;
Figure 7 shows the processed S-scan. Figures 8 and 9 show zoomed views of Figures 6 and 6,

respectively. Figures 7 and 9 show a significant tightening of the image, which is much better for
sizing estimates. This data set comes from a 10% deep crack (2.5 mm) using a 10L64 phased array.

Figure 6 - Original S-scan with a tip and corner, reconstructed in MatLab

Figure 7 - Processed data from MatLab. The corner and tip are visible.

Figure 8 - Zoomed, unprocessed data from MatLab S-scan.

Figure 9 - Zoomed S-scan after processing, with some artefacts visible

Actual crack measurements were performed on six artificially-grown fatigue cracks manufactured by
Sonaspection, with nominal depths of 5, 10, 15, 20, 25 and 30% on a 25.4 mm bar. The cracks were
later oxidized and broken open for accurate measurements. The ultrasonic estimates of crack size
using the new digital signal processing (DSP) technique were compared with standard back diffraction
measurements (see Tables 1 and 2 for 10 MHz and 5 MHz results). In general, the DSP measurements
had noticeably smaller errors than the corresponding back diffraction measurements, by almost a
factor of two on average. A couple of the DSP measurements with the 5 MHz transducer displayed
unexpectedly large errors work is underway to refine the signal processing technique to eliminate
this problem. In general, however, this type of crack sizing operation is usually carried out with
transducers with a central frequency close to 10 MHz, corresponding to the data of Table 1.The
uncertainty of the six (laboratory) back diffraction measurements in Table 1 was ~+ 0.4 mm, while
that from the DSP measurements was ~+ 0.16 mm.
Specimen ID: Nominal
Crack Depth
30%
25%
20%
15%
10%
5%

Crack Depth Error

Estimated from Raw
S-Scan
-0.7033 mm
+0.4267
+0.3499
-0.4101
+0.3656
-0.0275

Crack Depth Error

from Processed
Signal
-0.1054 mm
-0.2321
-0.2422
+0.2602
-0.0750
-0.0205

Table 1 - Differences between distances measured with ultrasonic techniques (10 MHz transducer) and
actual distance measured after specimen destruction. (All distances are in mm; a negative distance
implies the ultrasonic signal undersized the actual distance.)
Specimen ID: Approximate
fractional Crack Depth
30%
25%
20%
15%
10%
5%

Crack Depth Error

Estimated from Raw
S-Scan
-0.3234 mm
+0.5885
-0.3161
+0.1986
+0.2842
+0.6261

Crack Depth Error

from Processed
Signal
-0.6686 mm
+0.2109
-0.2324
-0.2054
-0.0032
+0.7103

Table 2 - Differences between distances measured with ultrasonic techniques (5 MHz transducer) and
actual distance measured after specimen destruction. (All distances are in mm; a negative distance
implies the ultrasonic signal undersized the actual distance.)

DISCUSSION
Even though these are preliminary results from a limited data set, they indicate significant potential
improvement in sizing accuracy. However, there is a considerable amount of work to do before
commercialization:
1. Determine the limits of the application of the technique, related to set-up parameters, crack
depths, proximity of signals, etc.
2. Improve the DSP algorithms, including the displays and analysis.

3. Optimize the selection of the autoregressive order number a parameter internal to the signal
processing scheme.
4. Commercialize the process, first by porting it from Matlab, and second by speeding it up.
5. Last, once commercial, the DSP routine will need to be vigorously tested by external sources
to develop credibility.

CONCLUSIONS
1. A novel signal processing technique based on a semi-empirical approach of Weiner filtering
with autoregressive spectral extrapolation has been developed for defect sizing using
ultrasonic B- and S-scans.
2. Early results with this technique showed an improvement of a factor of about two in sizing
errors over standard phased array back diffraction techniques.
3. A significant amount of development will be required before this technique is commercially
viable.

REFERENCES
1)
2)
3)
4)
5)
6)
7)
8)
9)

10)

11)

12)

DIN EN 583-2, April 2001. Non Destructive Testing Ultrasonic Examination. Part 2:
Sensitivity and Range Setting, English Version of DIN 583-2.
AWS, American Welding Society D1.1, Structural Welding Code, 2006.
Charlesworth J.P. and J.A.G. Temple, 1989, Ultrasonic Time of Flight Diffraction, Research
Studies Press.
Silk M.G., 1979, Defect Sizing using Ultrasonic Diffraction, British Journal of NDT, p. 12.
K. Harumi, Japanese Tip Echo Handbook, First Edition, 1989.
F. Jacques, F. Moreau and E. Ginzel, Ultrasonic back diffraction sizing using phased array
developments in tip diffraction flaw sizing, Insight, Vol. 45, No. 11, November 2003.
J. Mark Davis and Michael Moles, Resolving Capabilities of Phased Array Sectorial Scans
(S-Scans) on Diffracted Tip Signals, Insight, Vol. 48, no. 4 April 2006, p. 1.
D. Dechene, Practical Applications of the Time-Of-Flight Diffraction (TOFD) Ultrasonic
Techniques, FabTech, Las Vegas, Oct 6-8, 2008.
K. Lindenschmidt and M. Moles, Crack Depth Measurements in Thin-Walled Tubing by
Time-Of-Flight, Review of Progress in Quantitative NDE, Vol. 11, 1991, P. 2093, Ed. D.O.
Thompson and D.E. Chimenti, Plenum Press, New York, 1992.
P. Ciorau, D. Gray and W. Daks, Phased Array Ultrasonic Technology Contribution to
Engineering
Critical
Assessment
(ECA)
of
Reheater
Piping
Welds,
http://www.ndt.net/article/v11n05/ciorau2/ciorau2.htm
L.J.G. Wesley, A.N. Sinclair, M. Moles, Enhanced Ultrasonic Defect Sizing Techniques
using Weiner Filtering and Autoregressive Spectral Extrapolation, International Conference
on Fracture, ICF12, July 12-17, 2009, Ottawa, ON, Canada.
L.J.G. Wesley, M.A.Sc. Thesis, Image Enhancement of Ultrasonic Sectorial Scans Using
Autoregressive Spectral Extrapolation, University of Toronto, 2008.