EFFECT OF SURFACE CURVATURE ON BACKSCATTERED
ULTRASONIC GRAIN NOISE IN TITANIUM FORCINGS
F. J. Margetan and R. B. Thompson
Center for Nondestructive Evaluation, Iowa State University, Ames, IA 50011, USA
Andrei Degtyar and Harpreet Wasan
United Technologies Pratt&Whitney, East Hartford, CT 06108,USA
ABSTRACT. Six specimens, having roughly similar microstructures, were machined from different
angular positions within a forged disk of Ti 6-4 alloy. The "upper" surfaces of the six specimens each
had a different cylindrical curvature (3 concave, 2 convex, 1 flat). The opposite "lower" surfaces were
flat. Using a focused ultrasonic transducer, backscattered grain noise measurements were first made
through the (flat) "lower" surfaces to quantify the effects of the minor microstructural differences
between specimens. Grain noise was then measured through the "upper" surfaces to assess the effect of
surface curvature. Measured root-mean-squared (rms) and gated-peak noise levels, corrected for
microstructural differences, were compared to the predictions of a single-scatterer noise model. Model
predictions of grain noise levels were generally found to be in good agreement with experiment.
INTRODUCTION
The Engine Titanium Consortium is studying the ultrasonic properties of titaniumalloy forgings used in the fabrication of rotating jet-engine components. The goal is to
better understand the influence of the titanium microstructure on inspectability, and to use
that understanding to improve ultrasonic inspection practices. One limiting factor in the
detection of small or subtle internal defects is backscattered "grain noise" which arises from
the scattering of sound by microstructural boundaries within the forging. In this paper we
investigate the interplay between the observed grain noise level and the curvature of the
forging surface. In pulse/echo immersion inspections, any surface curvature will modify
the incident sound profile within the forging, thus causing each metal grain to be insonified
in an altered manner. This, in turn, changes the statistical properties of the backscattered
grain noise. Backscattered grain noise measurements were carried out on six titanium
forging specimens having roughly similar microstructures but different surface curvatures.
The results of those measurements are presented and discussed, and compared with the
predictions of the "independent scatterer" grain noise model [1].
SPECIMENS AND NOISE MEASUREMENTS
In order to isolate the effect of surface curvature on backscattered noise, it is
necessary to have a suitable set of specimens. As depicted in Figure 1, our specimens were
machined from different angular positions within a forged disk of Ti-6Al-4V alloy. Such
jet-engine forgings typically have complex microstructures possessing several length scales
[2]. The overall manufacturing process (cast ingot -> reduced diameter billet -> forging)
CP657, Review of Quantitative Nondestructive Evaluation Vol. 22, ed. by D. O. Thompson and D. E. Chimenti
© 2003 American Institute of Physics 0-7354-0117-9/03/$20.00
1363
Surface/.
Fiat:
1
FIGURE 1. Fabrication of the six "noise curvature" specimens from a forged disk of Ti 6-4 alloy.
typically produces a microstructure that varies strongly in the radial direction, less strongly
in the axial direction, and only weakly in the circumferential direction. Thus specimen-tospecimen microstructural differences were minimized by our construction plan, although
the microstructure was not uniform within any given specimen. Five specimens were
initially cut from the disk, three having cylindrical concave upper surfaces and two having
convex upper surfaces, as enumerated in Figure 1. The radii of curvature were chosen based
on the sonic shapes of existing and potential forging designs. A sixth specimen with a flat
upper surface was machined at a later time from the remaining disk material. The lower
surfaces of each specimen were machined flat and contained several small flat-bottomed
holes (FBHs), These l/64"-diameter, 0.10" deep holes were not used in the present study.
In addition to the six so-called "noise curvature blocks", 8 smaller rectangular coupons
(1.25" x 1.25" x 2.00") were cut from the same disk for basic ultrasonic property
measurements. Those measurements of sound velocity, attenuation, and backscattered
noise capacity (FOM) were described in an earlier article [3].
The noise measurements performed using the curvature blocks are depicted in
Figure 2a. The flat lower surfaces were scanned to assess microstructural differences
between specimens. The curved upper surfaces were then scanned to assess the effect of
surface curvature on grain noise. The transducer and inspection water paths used in our
measurements were similar to those used in actual UT disk inspections. Transducer focal
characteristics were determined by on-axis profile measurements similar to those of Ref.
[4]. The 10-MHz transducer had a measured effective diameter of 0.344" and a geometric
focal length in water of 3.95". Two-dimensional, surface-following scans were performed at
normal incidence using two water paths : 2.4" (focused slightly below the entry surface)
and 1.0" (focused about 0.5" deep). In all cases, the surface area scanned was 0.5" by 3.0"
in steps of 0.015". The measurements were conducted in an industrial setting over a threemonth period as scheduling allowed. A reference echo from a flat-bottomed hole in a finegrained Ni-alloy block was acquired each day that data was taken. This was used to correct
for day-to-day differences in the operating efficiency of the measurement system.
1364
DO
IP
OO
ID
FIGURE 2. Noise measurement (immersion) setup (a) and examples of noise data, (b) Typical total noise
A-scan. (c) Typical rms-total-noise-versus depth curve. Measured (d) and model-simulated (e) gated-peak
noise C-scan images for inspection of the flat specimen at a 2.4" water path.
The grain noise measures considered in this paper are illustrated in Figure 2b-d. At
each scan position, RF (A-scan) data was acquired at a 100-MHz sampling rate and stored.
For a given flat or curved sound entry surface, root-mean-squared (rms) noise values were
obtained as described in Ref. [1]: the RF noise voltages observed at a given time relative to
the front-wall echo were squared, averaged over transducer position within the 0.5" x 3"
scan area, and the square root was taken. This spatial averaging process was repeated at
each observation time to obtain a curve displaying the dependence of noise on depth below
the entry surface. Examples of raw and smoothed rms-noise-versus-depth curves are shown
in Figure 2c. The smoothing operation replaces all values within a 0.5 usec window (or
0.061" depth zone) by their simple average. The observed total noise is a combination of
backscattered grain noise and electronic (equipment) noise, with the rms values related by
(RMS
l)2
tota
=
(RMS
n)2
grai
+ (RMS electronic )
(1)
The rms electronic noise was directly measured by acquiring noise A-scans when no
specimen was present in the water tank. All absolute noise values reported in this work
have been rescaled to a fixed, absolute equipment gain setting of 57 dB. At that setting, the
measured rms grain noise was 0.32 % of the oscilloscope's full screen height (FSH).
In addition to rms noise, the stored noise A-scans were used to construct C-scan
images of rectified gated-peak noise (GPN) amplitudes. For a given choice of the time
gate, the largest absolute noise voltage within the gate was determined for each transducer
position in the scan region. The peak-amplitude-versus-position data was then displayed as
a 2D image, as illustrated in Figure 2d. Four time gates with the same center point but
different durations were used, corresponding to depth zones of 0.8"-1.0" (Gatel), 0.7"l.l"(Gate 2); 0.6"-1.2" (Gate 3), and 0.5"-1.3" (Gate 4), respectively. For each C-scan
image, the average amplitude, the maximal amplitude, and the standard deviations of the
amplitudes about their mean were computed.
1365
MEASUREMENTS THROUGH THE LOWER (FLAT) SURFACES
For our forging specimens, available information indicated that the actual
microstructure varied only modestly in the axial and hoop directions, but varied greatly with
radial position. Measurements through the lower (flat) surfaces were used to quantify these
variations. The variation with radial position is obvious from C-scan images such as that of
Figure 2d. Horizontal profiles of the C-scan noise images made through the lower surfaces
are shown in Figure 3a for one choice of water path and time gate: the amplitudes in each
vertical row of pixels in the C-scan image were averaged to construct a profile. For each
specimen, the GPN profile amplitude increases markedly from the OD to the ID portion of
the scan region. The upper-right-hand panel in Figure 3 shows the average of the six
profiles in the upper-left-hand panel, renormalized to have an average value of unity. Also
shown is a 4-th order polynomial fit that summarizes the radial variation. This is used as an
empirical input to the noise model described later.
In an ideal study, the microstructures of the six specimens would be nearly identical,
and noise measurements through the lower surfaces would produce similar average results.
In our case, clear coupon-to-coupon differences in average noise levels can be seen in the
radial profiles of Figure 3a. Similar coupon-to-coupon differences are seen for various
noise quantities averaged over the entire scan through the lower surface. As one example,
Figure 3c displays the absolute rms grain noise curve for each specimen for the 1.0" water
path, after using Eq. (1) to remove electronic noise from the measured total noise. The
divergences at the beginning and end of the depth range are due to the influences of the
front-wall and back-wall echoes respectively. In the absence of other effects, the absolute
noise level tends to increase as the sonic beam is focused [1]: fewer grains are insonified,
®m €«n
v®m§«
f%f i&
Flit
Pstt; Gate 4
No!$e M88&UH&n$nts tmj Lower (Rag) 8mf
1,!f water &&rt 57 dl Sain
A*nj»8fc»i0 Polite
U" Water Fift;; G*& 4
$ t& Skte Inspection; 1 ,0"
GPN
10-
0.0
(C)
0.2
0.4
0.8
OJ
1>0
1,4
1.6
Depth limits)
FIGURE 3. Examples of noise variations due to microstructural differences between specimens. All results
shown are for measurements through the flat (bottom) sides of the specimens at 1.0" water path, (a): Radial
profiles of C-scans using time gate #4. (b): Polynomial fit to the average radial profile of panel (a), (c): RMSnoise-versus-depth curves, (d): Average gated-peak noise amplitudes seen in full-area C-scans.
1366
but each produces a larger echo resulting in a net noise increase. This effect is very clear for
the 1.0" water path which produces a beam focus about 0.5'" deep. Specimen-to-specimen
variations are also evident in Figure 3d which shows the average GPN amplitude for each
choice of the time gate. As the gate is enlarged there are more opportunities to observe
large GPN amplitudes, and the average consequently increases.
The coupon-to-coupon differences seen in Figure 3 were used to adjust like
quantities measured through the curved surfaces so as to approximately correct for the
microstructural differences between specimens. Due to space limitations, the adjustment
procedure will not be described in detail. In brief, if a measurement through the lower (flat)
surface of a given coupon found the noise amplitude at depth Z to be 10% lower than the
average for all coupons, then the noise amplitude measured at depth T-Z from the curved
side was adjusted upward by 10% (T = coupon thickness).
NOISE MODELING
Before showing experimental results for measurements through the upper (curved)
surfaces, we will briefly discuss the independent scatterer noise model whose predictions
will be compared with experiment. The model assumes that the observed noise is an
incoherent superposition of the echoes produced by the individual grains, and multiplescattering events are ignored. A flow chart for the model calculations is shown in Figure 4.
Ref. [1] and the references cited therein describe the use of the "basic" model to predict the
absolute rms noise level seen in a given inspection, and the subsequent prediction of gatedpeak noise amplitude distributions from rms noise-vs-time curves. The absolute noise level
will depend on properties of the metal specimen and details of the inspection system. Metal
UT properties that are input to the model include the density, sound speed, attenuation, and
noise-FOM. The latter is a frequency-dependent measure of the noise generating capacity
of the microstructure, equal to the square root of the backscatter power coefficient.
The average ultrasonic properties (i.e., averaged over the 0.5" x 3" x 1.6" scan
volume) assumed in our calculations were largely determined from prior measurements on
the smaller rectangular "property" coupons, and are listed in the upper portion of Figure 4.
Those prior measurements are described in Reference [3]. The model calculations assume
that the microstructure of each curvature block is the same, and does not vary in the axial or
hoop direction, but does vary in the radial direction. The rms-noise-versus-depth curve for
the "average" (homogeneous) microstructure was first predicted using the model. Then this
curve was scaled by a factor that depended on radial position. The assumed scaling factor
Electronic
Empirical date
on how
wtttt
focal
fdJameter, QFL)
UT
from
F8H
FIGURE 4. Flow chart for noise model calculations.
1367
was shown earlier in Figure 3b. At each scan position (model image pixel), the portion of
the predicted rms-noise-vs-depth curve within the gate of interest was used to predict the
gated-peak noise amplitude. A Monte-Carlo simulation was used, governed by Eq. (4) of
Ref. [1]. The distance the transducer must be scanned before a statistically independent
noise A-scan was obtained was assumed to be 0.045", and this served as the size of the
model pixels. This estimate of the so-called "noise spatial correlation length" was based on
earlier measurements on other Ti 6-4 forging specimens. The result of the basic MonteCarlo GPN calculation is a simulated C-scan image, like that shown earlier in Figure 2e,
from which amplitude statistics (mean, maximum, standard deviation.) can be calculated.
For each inspection scenario, 1000 simulated noise C-scans were calculated, and their
amplitude statistics were averaged. The results were then compared with experiment.
COMPARISONS OF MEASURED AND PREDICTED GRAIN NOISE
In this section, representative results from the many measurements and analyses will
be presented. Figure 5 compares measured and predicted rms-noise-versus-depth curves,
Noise Measurements thru Flat Surfaces
Noise Measurements thru Flat Surfaces
2.4" water path; 57 dB
0.0
0.2
0.4
(a)
0.6
0.8
1.0
1.2
1.4
1.6
Metal Depth (inches)
0.0
(c)
0.0
0.2
0.4
0.6
0.8
0.4
1.2
1.4
1.0
1.2
1.4
1.6
-*- 0.75" Concave -*-2" Concave
-*~ 8" Concave
-x- 4" Convex
1.6
0.4
0.6
0.8
-A-10" Convex
1.0
1.2
Depth in Metal (inches)
Depth in Metal (inches)
Effect of Curvature - 2.4" water path - Experiment
-e- 0.75" Concave
-»~ 8" Concave
-x- 4" Convex
0.8
Effect of Curvature -1.0" water path - Theory
-*-2" Concave
-A-10" Convex
1.0
0.6
Metal Depth (inches)
Effect of Curvature -1.0" water path - Experiment
- 0.75" Concave
-8" Concave
- 4" Convex
0.2
(b)
Effect of Curvature - 2.4" water path - Theory
~»~2" Concave
-*~ 10" Convex
0.0
Depth in Metal (inches)
0.2
0.4
0.6
0.8
1.0
1.2
Depth in Metal (inches)
FIGURE 5. Measured and predicted rms grain noise levels for inspections through the flat (a-b) and
curved (c-f) surfaces of the Ti 6-4 forging specimens.
1368
for averages over the full 0.5" x 3.0" scan region. For all cases shown, Eq. (1) was used to
subtract electronic noise from the measured total noise to obtain an experimental estimate
of backscattered grain noise alone. The upper two panels in Figure 5 compare rms noise
levels for inspections through the flat sides of the specimens at each water path. There, the
model prediction is compared to the average of the measured rms noise curves for the six
specimens; e.g. the experimental curve shown in Figure 5a is the simple average of the six
curves shown in Figure 3c. Model predictions are seen to be in good agreement with
experiment within the 0.3" - 1.3" depth range where the data values are most trustworthy.
Outside of this range, front-wall, back-wall, and flat-bottomed hole echoes interfere with
the grain noise data. Note that for 1.0" water path, the focal maximum in the noise level
curve occurs at sufficient depth to be well resolved from the front-wall ring down.
The lower two panels of Figure 5 show the measured rms noise levels for each
curved coupon, divided by the value seen at the same depth through the flat coupon. This
ratio is a direct measure of the influence of surface curvature on the observed noise. Prior
to division, the measured values have been corrected for specimen-to-specimen
microstructural differences in the manner described earlier. Again the model predictions
are in good overall agreement with experiment. The surface curvature primarily acts to
Curved Side Inspection; 1 .0" water path; Average GPN
Curved Side Inspection; 1 .0" water path; Average GPN
20
Experiment
Theory
^ESEE^^
/>i/^
^
1
S~~~~~~~~*^~~~~~^.
r 15
/ /
_____ _^___
^^-~~-~^*
1
E
10
S/&/
<£^>^S
^Gatel
^^^
-c^Gatel
X^>^
-*-Gate2
X^^
°^
-.-Gate 2
••••*••• Gate 3
~*-Gate3
-*-Gate4
-•-Gate 4
(Corrected tor microstructural differences)
0
/
\
0.75" 2.0" 8.0" flat
x"/
concave concave concave
10.0" 4.0"
convex
convex
0.75"
(b)
2.0" 8.0" flat
concave concave concave
10.0" 4.0"
convex
convex
Curved Side Insp; 1 .0" water path; Std.Dev./Ave. GPN
Curved Side Insp; 1.0" water path; Std.Dev./Ave. GPN
0.50
Theory
^$r
*
% 0.30
-*-Gate2
•••*••• Gate 3
-«- Gate 4
O
3 °-25
0.20
0.75"
(c)
2.0" 8.0" flat
concave concave concave
10.0" 4.0"
convex
convex
L_
o oo o oo o
to
ho
ww
5 0.35
»-.
/*\
^^
^'i^^^S^:^
^"^
!>°-40
'Q£ std- Dev- / Average GPN
0,5
* > • * > • bi
Experiment
'
Curved Side Insp.; 1 .0" water path; MaxVAve. GPN
Experiment
-o-Gate1
-*• Gate 2
-*- Gate 3 _
0.75"
2.0" 8.0" flat
concave concave concave
10.0" 4.0"
convex
convex
Curved Side Insp.; 1 .0" water path; MaxVAve. GPN
Theory
/\
i3-
^£^*~^£
f==^*^ ^* ——. 1
•£ 2 —
|2-
-o-Gate1
-»-Gate2
-*- Gate 3
-•-Gate 4
%
0.75"
fQ\
,
concave
2.0" 8.0" flat
concave
concave
10.0" 4.0"
convex
-o- Gate 1
-*- Gate 2
-*- Gate 3
-*- Gate 4
X
convex
0.75"
(i\
concave
2.0" 8.0" flat
concave
concave
10.0" 4.0"
convex
convex
FIGURE 6. Measured and predicted statistics of gated-peak noise C-scans for inspections through the
curved surfaces of the Ti 6-4 forging specimens at 1.0" water path.
1369
modify the focal pattern of the sonic beam in the axial-hoop plane of the specimens, with
the focal depth in the metal being decreased (increased) for concave (convex) curvature. For
concave curvature with a water path less than the focal length, for example, the net effect of
the curvature is usually to boost the grain noise level seen at shallow metal depths, and to
decrease it deeper in the metal.
Measured and predicted values of gated-peak noise (GPN) statistics are compared in
Figure 6 for selected cases involving measured or simulated C-scans over the full 0.5"x 3.0"
scan area. Experimental values were adjusted to correct for measurement system efficiency
differences and for specimen-to-specimen microstructural differences. In this case, the
measured electronic noise was added to the model predicted rms grain noise, before GPN
C-scans were simulated. For the cases shown, the time gate was centered 0.9" below the
sound entry surface; i.e., beyond the focal maximum for either water path. The average
GPN amplitude was found to depend significantly on surface curvature, tending to be
largest for the near-flat curvatures, with the curvature dependence more pronounced for the
1.0" water path than for the 2.4" water path. As can be seen in Figure 6a-b, the manner in
which the (absolute) average GPN amplitude varied with surface curvature and time-gate
duration was well predicted by the model.
One measure of the noise variability within a gated-peak noise C-scan image is the
dimensionless ratio of the standard deviation of the noise amplitudes over their mean. For
the cases considered in our study, the model calculations predict that this ratio is primarily
determined by the radial variation in the noise level (see Figure 3a), and hence is
approximately independent of surface curvature and the duration of the time gate. This
p'rediction is in reasonable agreement with experiment, as shown in Figure 6c-d for the 1.0"
water path. Another dimensionless ratio of interest to the UT-inspection community is the
maximum noise amplitude seen in a gated-peak C-scan image divided by the mean noise
amplitude. The model calculations also predict that this ratio is approximately independent
of surface curvature and time gate duration. As illustrated in Figure 6e-f, the model
predictions for this ratio tended to be slightly low, perhaps because microstructural
variability in the axial and hoop directions was ignored in the model calculations.
Taken as a whole, the model predictions were found to be in generally good
agreement with experiment. This indicates that the Independent Scatterer Noise Model is
sufficiently accurate to serve as a helpful simulation tool for investigating and optimizing
UT inspections of Ti 6-4 forgings.
ACKNOWLEDGEMENT
This material is based upon work supported by the Federal Aviation Administration
under Contract #DTFA03-98-D-00008, Delivery Order IA029 and performed at United
Technologies Pratt&Whitney and at the Iowa State University Center for NDE as part of the
Center for Aviation Systems Reliability program.
REFERENCES
1. F. J. Margetan, I. Yalda and R. B. Thompson, Review of Progress in QNDE, Vol. 15,
eds. D. O. Thompson and D. E. Chimenti (Plenum, New York, 1996), p. 1509.
2. P.D. Panetta, R. B. Thompson and F. J. Margetan, Review of Progress, in QNDE, Vol.
17, op. cit. (1998), p. 89.
3. L. Yu, F. J. Margetan, R. B. Thompson and Andrei Degtyar, Review of Progress in
QNDE, Vol. 21, op. cit. (2002), p. 1510.
4. F. J. Margetan, K. Y. Man, I. Yalda. S. Goettsch and R. B. Thompson, Rev. of Prog, in
QNDE, Vol. 14, op. cit. (1995), p. 2129.
1370