TOFD Inspection with Phased Arrays

Charles Brillon,1 , ∗ Tim Armitt,2 and Olivier Dupuis1

1 Olympus NDT Inc.
505,2 boul. du Parc-Technologique, Québec (Québec), G1P 4S9, Canada
Lavender International NDT Consultancy Services Ltd.
Unit 7, Penistone Station, Sheeld, S36 6HP, UK
(Dated: April 2008)
In the last decades two major inventions changed the face of ultrasonic NDT viz., time of ight
diraction (TOFD) and phased array ultrasonic testing (PAUT). When used in conjunction, these
two techniques allow for easier and faster inspections, normally requiring a great amount of time
and/or experienced UT inspectors. To do TOFD however (even TOFD by itself), it is necessary to
have dierent sets of probes with dierent probe separations when one wishes to inspect very thick
materials.
With the advent of more sensitive transducer arrays and faster electronics, we can now envisage,
using PA probes, to perform multiple TOFD inspections (the dierent apertures, osets and re-
fracted angles all being changed electronically) in one run and with a minimum number of props.
These dierent TOFD scans may also be merged a posteriori so that a simple global picture nally
emerges from the acquired data.
In this paper we will present some results achieved with this method as well as both theoretical
and practical aspects of the problem.
Keywords: Phased array, TOFD, sizing

I. INTRODUCTION

The TOFD technique was developed in the 70's. It was devised as a tool for accurately sizing and monitoring
defects. Indeed, since Pulse-Echo (PE) techniques allow for sizing based on amplitude of the reected signal, it is
very much dependent on the orientation of the defects. A slight change in angle from normal reection can induce a
large error in the size or can result in the defect being missed altogether.
TOFD on the other hand can spot defects no matter what their orientation, and is pretty robust as far as the skew
of the probes is concerned.
In this paper we show that PAUT traditionally used with PE techniques can be used to perform more complete
TOFD scans by allowing dynamical steering of the UT beams.

II. TOFD BASICS

FIG. 1: This gure illustrates a basic TOFD setup. A pulse leaves the probe on the left. The beam spreads
(illustrated in grayish blue) and diracts o the defect with two dierent times. Assuming this defect is positioned
in the center of the weld, we can deduce its height and depth from the resulting A-Scan.

TOFD relies on the sole diracted signals inside a material to resolve the position and size of the indications.
Consider a point indication located at a depth d from the top of the surface and located in the center of the weld1
 2

(see gure 1). The two probes straddling the weld are separated by a distance 2s, also known as the Probe Separation
Center (PCS). One of the probes (for instance the one on the left) emits a sound burst. The pulse will spread inside
the material just like a wave at the surface of a pond. The pulse will take a time t to reach the defect and the same
time after diracting o that defect to reach the receiving probe. From an A-Scan signal it is then possible to work
out d. Indeed, using Pythagoras' theorem with the knowledge of s and c, the velocity of sound in the inspected
material, we get2

q
2
d= (ct) − s2 . (1)

Applying this equation (1) for both top and bottom of the indication in gure 1, we get

h = dbt − d
q
2
q
2
, (2)
= (ctbt ) − s2 − (ct) − s2

where tbt and dbt stand for the time of ight (TOF) to the bottom tip and the depth of that same tip respectively.
Collecting A-Scans at constant intervals along an axis allows one to pick up patterns that would have been very
dicult to interpret otherwise. Indeed, gure 2 shows a D-Scan or non-parallel scan and corresponds to the scanning
motion illustrated in gure 6a3 .

FIG. 2: Typical non parallel scan (D-Scan) image. The A-Scan gives a slice of the gray scale image on the left. One
can identify the lateral wave and back wall reection on both images. On the left-hand side of the image an
indication is identied near the bottom of the piece. The red cursor represents the position of the A-Scan on the
right. The scan is that of a steel plate 20 mm deep. A 60 degree wedge was used to focus at two third of the
thickness of the material. The probe frequency was 5 MHz.

III. OBSTACLES AND SOURCES OF ERROR.

Once again TOFD is a very powerful tool but is not perfect. It might be very dicult to make sense of an indication
depending on where the defect is located in the weld volume. If accurate measurements are required, it is very possible
that this defect will require another scan with dierent parameters. In this section we list some weak points of TOFD
and what is usually done to improve on those.
 3

A. Timing errors

FIG. 3: This gure illustrates timing error with respect to the angle at which the tip is hit. The higher the
attacking angle, the larger the error. We have used a 5:1 ratio digitizing rate (the ideal minimum) to nd ∆t, the
timing error for each frequency.

Error calculations allow us to determine the behavior of the uncertainty on the depth as a function of the angle
at which the tip of the defect is being hit. Figure 3 gives us the errors with respect to the incident beam angle for
dierent frequencies in steel (c ' 5.9mm/µs). The lower the angle, the smaller the error. We also see that higher
frequency probes help reduce the error. In other words, to minimize the uncertainty on the depth measurement, one
needs to reduce s and increase frequencies and digitization rate. However both operations will reduce the coverage
since the beam spread for higher frequencies are smaller. The lower edge angles in the beam spread will also be hitting
the bevel higher, so that there will be less coverage of the back wall.
In order to have a good coverage of the whole area of interest, it is sometimes necessary to manually change the
PCS and perform another scan.

B. Lateral wave dead zone

Since the pulse has a certain width, it is sometimes impossible to resolve indications that are very close to one
another. Most generally, since the lateral wave (LW) has an extension in time and is present on every A-Scan (if
there is no surface breaking defect), it will potentially hide any signal popping up in that time window. This is the
so-called lateral wave dead zone. This is a very serious limitation of TOFD. Typically around one third of the rst
part of the weld volume can be buried in there. Algorithms have been, and are continually being further developed
to remove the LW and nd indications "underneath". However these algorithms are never perfect and it is easy to
misinterpret the data. Spurious structures sometimes show up where nothing is to be found in the actual sample.
If care can be taken before the acquisition takes place to minimize the range of the lateral wave, a lot of time could
be saved in processing and analyzing.
The expression for the lateral wave extent involves the pulse width and the PCS. Figure 4 illustrates that relation-
ship. Of course, we notice that higher frequencies give a better resolution (that is, if the number of cycles during
which the pulse rings is kept constant when we compare frequencies).
Then again, decreasing s and increasing the frequency f is the solution to reduce the impact of the lateral wave
ringing "inside" the material, and again we must take the coverage area into account.
The pulse width can be improved upon by using suciently damped transducers. Here also there is a compromise
to be made : Damping makes the probes less sensitive. In order to pick up diracted signals, one needs to crank up
the gain, which will also unfortunately bring up the noise level.
 4

FIG. 4: Lateral wave extent as a function of PCS. We see that a smaller pulse width (1/frequency × number of
cycles) helps shorten the lateral wave's "penetration".

C. O axis errors

FIG. 5: The indications located along an equal time elliptical curve will be displayed at the same position inside the
resulting A-Scan. Orientation, position and height of a defect, cannot therefore be accurately measured with a single
pass along the scan axis (D-Scan).

In section II we assumed that an indication was positioned in the center of the weld. In fact, in most cases this is
not true (as can be seen from gure 5) : defects like lack of sidewall fusion (LSF) will be located on the side of a weld
bevel, and cracks have dierent orientations. Since only TOF is used to measure positions, every sound path with
the same length is equivalent. This is the denition of an ellipse, whose foci are the exit point of the beam below the
wedge. Every point thus lying on this ellipse will be displayed at the same position in an A-Scan. Indeed, the defect
which is assumed to be located in the center of the weld in gure 5, could very well be located at either positions 1
or 2, for instance.
When defects are found in the volume of the weld, typical depth errors are around 3%. Generally, when a beam is
aimed at a certain depth, at the edge of the ultrasonic beam for this depth, the maximum error is around 8%.
At the back wall then, defects must be around 8% of the thickness of the specimen above the back wall (and inside
the beam spread) position for them to be detected. Otherwise, they will be hidden below the back wall signal. This
region is the so-called back wall blind zone. In that particular case, increasing the PCS will allow for defects buried
near the back wall to stand out.
All in all, if the exact position of the indication needs to be found in the specimen, the use of a B-Scan (parallel
scan) (gure 6b) will be required at a position along the scan axis. Again, this is rather time consuming.
 5

(a) D-Scan (perpendicular scan) (b) B-Scan (parallel scan)

FIG. 6: These two types of scan allow for a better resolution of defects and positioning of defects.

D. Resolution errors

To resolve the height of defects, one must be able to see both top and bottom tips. As in the case of the lateral
wave, it is important to minimize the eective extent of pulse width. This can be achieved by shortening the PCS,
increasing the probe frequency and digitization rate, and using more highly damped probes. Focusing the center of
the beam at a given depth d will also improve the resolution at that particular depth, since the beam spread for the
highest frequencies is quite narrow and as soon as we stray from the center, we start losing those frequencies and
thus resolution. Bottom line : if we want to better discriminate top from bottom tip, focusing directly on the area of
interest is a good solution. The coverage is very small however, so generally this is another form of scanning that will
usually be performed once a rst "detection" scan was done. Another pass again means more time spent on a sample.

E. Other error sources

Some other error sources can come in the way.

1. Human error
When sizing or calibrating, the cursors must be set by hand. This operation can induce an error in sizing, especially
in the upper part of the scan.

2. Index point migration

When linearizing the scale with the aid of the lateral wave and back wall TOF, we have assumed the beams exit at
the same point under the wedge. Doing so, wedge delays are equal. This is in general not the case. Although usually
very small, the beam spread inside the probe shoe will result in more than one exit point, and the delays associated
with each beam will be dierent. One must ensure that the probes are small enough so that the value of the PCS
doesn't vary too much, or else the assumption will not hold. For most TOFD applications this is generally not of
major concern.

IV. SOLUTIONS WITH PHASED ARRAYS

All of the problems encountered above have a common solution : vary the parameters in order to get the best possible
coverage. This is done by performing multiple scans. This is where phased arrays come in handy. A transducer array
is comprised of many single element transducers.
 6

FIG. 7: The individual elements are pulsed at dierent times. This allows to focus and/or steer the ultrasonic beam.

Figure 7 shows the principle behind phased array ultrasonics. The individual crystals are excited at dierent times.
This creates a constructive interference that allows to steer and/or focus the beam in many directions. Delays are
usually also applied when receiving the signal to make sure that the maximum energy possible is displayed in the
A-Scan.
Phased arrays are mostly used in PE mode. They can also be used in through transmission mode (one probe on
the OD and one probe on the ID) or the so-called tandem mode (to detect vertical aws via skips o the bottom of a
piece). However, they are seldom used to detect diracted signals, which are rather faint.

A. Sectorial TOFD

In a PE sectorial scan, dierent angles are used to make sure that the defect is hit as perpendicularly as possible.
In TOFD, a sectorial scan can be used to focus at dierent depths in the weld. Figure 8 is an example of a sectorial
scan, performed on a 20 mm thick steel plate. Two arrays of piezoelectric crystals (see gure 9) are held together just
like in a regular TOFD setup. One of the probe emits beams at dierent angles (four with a 10 degrees separation in
the gure above.). The four quadrants on the right-hand side of the image represent one of the four beams illustrated
on the End View on the left-hand side. This view is just a guide to the eye and there is no point in trying to make
sense of the real measurements (index or ultrasonic) on that part. The Side Views on the right-hand side, on the
other hand, have already been calibrated on the lateral wave and the back wall. All 16 elements are being red
here. The pitch distance between elements being 0.6 mm an aperture of 16 elements, gives us a rectangular area of
9.6mm × 10mm (where 10mm refers to the so-called probe elevation). This will not give huge beam spread, however
the advantage is concentrated energy resulting in stronger diracted signals. Use of multiple angles alleviate the need
for a single wide beam.
If we follow the defects at the top (number 1) and the bottom (number 2) of the plate, we notice that the top
one becomes more and more visible as we increase the angle. This is because there is more energy to hit it at those
higher angles. You can see that the software gain had to be cranked up a lot in order to pick up the lateral wave.
Speaking of the lateral wave, even though it is not very well dened, we can see it "penetrates" the material a lot,
whereas for other angles it is shorter. If we go back to III B and have a look at the lateral wave range as a function of
PCS and frequency (gure 4 on page 4), we see that the lateral wave extent depends on the pulse width and on the
PCS. Assuming that our PCS is constant (no index point migration), we conclude that the pulse width is changing.
Actually, the dominant frequency is lower for the lateral wave in that area. In fact, a fast Fourier transform (FFT)
of the lateral wave for the 40 degrees beam gives a dominant frequency of around 1.91 MHz (we get around 3.9 MHz
for the 70 degrees beam) because of the beam spread which is wider for low frequencies.
Defect number 1, which was hidden below the lateral wave then becomes apparent when the angle increases. Defect
number 2, on the other hand, disappears when we increase the angle : not enough energy is reaching the bottom of
the plate.
For the purpose of demonstration, we used only four beams, but we could have put many more. For instance, one
beam every degree. Doing so, we are sure to hit the whole weld bevel with optimum frequency and intensity. In fact,
nothing prevents the use of the power of phased array to focus the beam (using Huygens' principle) along a vertical
line in the middle of the weld.
 7

FIG. 8: Sectorial beams. 10 degrees apart. PCS constant Beam focused at dierent depths. There are four angles:
40-50-60-70 degrees, with their B-Scans on the right-hand side, respectively in the top left, top right, bottom left
and bottom right quadrants. The green arrow points to an indication, labeled "1" (probably a lack of fusion), while
the red arrow points to another one, labeled "2". This plate is the same one we inspected with conventional TOFD
(gure 2). The scans were performed and viewed with the TomoView— software by Olympus NDT® .

FIG. 9: PA TOFD setup. 5 MHz probes. Maximum aperture of 16 elements.

 8

FIG. 10: Principle behind the linear scan. Some elements are open to create a virtual aperture (VA). Once these
elements have red, we turn o some of them at the beginning of the VA (typically 1 or 2) and turn on the same
number of elements on the other end of the VA. They re again. Then, we proceed again with the permutation of
the elements. This will create an eective displacement of a focal law along a certain distance.

FIG. 11: Illustration of varying PCS to better cover the weld center. 5L64-A2— probes by OlympusNDT® have
been used. As the focus is set deeper the VA of the probe is increased for more intensity

B. Varying the PCS

Instead of using a sectorial scan, the same angle can be used but we can vary the PCS electronically. This is done
by using a linear scan (explained in caption of gure 10 with an angled wedge, the exit point of the beam changes
with the red beam position. The receiving probe will have symmetric electronic movement so that s will be the same
on both sides of the weld center. Figure 11 illustrates the scanning. In this conguration we cannot use the whole
array as a VA. The electronic scanning requires closed elements on both sides of the current VA.
Here, we have a combination of linear and sectorial scanning. Dierent angles are being used but this is to better
cover the whole piece (keeping only one angle and varying the PCS will give us only limited focused coverage of the
center of the weld. One can notice that the PCS is dynamically changing for each beam (just below the probe and
before entering the shoe).

C. Raster TOFD

So far, we have concentrated our eorts on varying the PCS and beam angles dynamically to have better response
and resolve defects more eciently. One last aspect we have not yet covered in this section is the depth error due
to o-axis positioning of defects (see section III C on 4). The conventional method to do this is to perform a B-Scan
(gure 6b). With long enough arrays, this can be performed dynamically as can be seen in gure 12.
 9

FIG. 12: Illustration of varying PCS to better cover the weld center. 5L64-A2— probes by OlympusNDT® are
displayed. As the focus is set deeper, the VA of the probe is increased for more intensity.

The setup from last section (gure 11) is used again. This time the PCS is kept constant. Basically the foci from
the ellipse locus in gure 5 will be displaced from left to right. The true depth of the indication will then show up as
the minimum position. The oset of the defect will be given by the index oset of the focal law which picked up the
minimum position.

D. Merging the data

Performing all these scans in one pass will generate a huge amount of data. In PA softwares like TomoView— it
is possible to merge the data that is redundant (like when sectorial scans are overlapping for instance). The image
below shows a mere view of the data acquired in gure 8. The image is not scaled but the idea is there. Now this
would represent a side view (i.e. for a single index value). With the rastering it would be possible to obtain several
values of index for the entire volume of the weld. Basically, we would have a volumetric TOFD view.
We can also note that at a xed scan axis position value for the probes, the merging of the data is not really
inuenced by the variation of the couplant thickness. Therefore, there is no need to synchronize the lateral wave like
we sometimes need to do with D-Scans.

E. Comment on probes

Probes used for TOFD in PA should be made of traditional piezo-ceramic elements. Their main advantage is that
they can be dampened more easily than other types of probes. It is then easier to select the right amount of damping
material in order to nd the appopriate pulse width for top and bottom tip resolutions and minimum extent of the
lateral wave.

V. CONCLUSION

It has been found that PAUT techniques can be used with TOFD with little modications. A PA-TOFD system has
all the forces of regular TOFD but none of its weaknesses. It allows for better localization of defects and everything is
done in one pass and kept in one data le. Sectorial scans and linear scans can be combined and merged a posteriori
to better analyze the data at hand and in a more user-friendly fashion. No call to rescan a certain area needs to be
madethe system does it all in one pass. The problem of the lateral wave extent can be optimized by tweaking the
pulse width with single crystal probes that can be more easily dampened.
 10

FIG. 13: This is a merged view of the data displayed in gure 8. Courtesy of NSG® Japan.

VI. ACKNOWLEDGMENTS

Charles Brillon would like to thank the people at Lavender's International for their hospitality while part of this
work was performed. Special thanks go to Tim Armitt, Steve Hewling and Nicola Dodsley.

∗
Electronic address: charles.brillon@olympusndt.com
1
This is seldom the case as we will see in section III C.
2
In fact, these times are the ones when the wedge delay 2t0 (t0 in only one probe) have been removed, using the time taken
by the lateral wave (LW) and the back wall (BW) in the calibration process.
3
This nomenclature can be somewhat confusing. Indeed, most of the time, a D-Scan, also called an End View, represents a
view of a section looking in the direction of the scan axis. In this paper, we will use the TOFD nomenclature unless otherwise
specied. A B-Scan is also referred to as a Side View. In TOFD however, a B-Scan is a parallel view. Instead of moving the
probes along the scan axis direction, the encoders are placed so that the probes are moving along the index axis direction.