![SAFT-DGS assessed SAFT result [21].](https://www.researchgate.net/profile/Johannes-Vrana/publication/319701866/figure/fig1/AS:616359226511375@1523962899527/SAFT-DGS-assessed-SAFT-result-21_Q320.jpg)
Content uploaded by Johannes Vrana
Author content
All content in this area was uploaded by Johannes Vrana on Sep 14, 2017
Content may be subject to copyright.
JOHANNES VRANA1, KARSTEN SCHÖRNER2, HUBERT MOOSHOFER2,
KARSTEN KOLK2, ALEXANDER ZIMMER3, KARL FENDT4
ULTRASONIC COMPUTED TOMOGRAPHY – PUSHING THE
BOUNDARIES OF THE ULTRASONIC INSPECTION OF FORGINGS
Abstract
Forgings, being usually one of the most critical components especially in power
generation machinery, require intensive volumetric inspection to guarantee a sufficient
lifetime. This is usually accomplished by manual or automated ultrasonic testing.
We are reporting about a game changer in ultrasonic testing: Ultrasonic Computed
Tomography uses analytics (i.e. a mathematical algorithm) to reconstruct the volume (In fact
it uses a linearized diffraction tomographic approach for the solution of the inverse problem).
This does not only allow to display indications spatially and visually correct in the 3D
volume, but also improves the signal to noise ratio significantly, allowing an increase of
sensitivity by up to an order of magnitude. The method is based on the Synthetic Aperture
Focusing Technique (SAFT). The applied software is a brand-new implementation of SAFT
with a strong focus for a large scale industrial application: the complete 2D as well as 3D
reconstruction of ultrasonic inspections of heavy rotor forgings.
This paper shows the working principle of the method along with the first results and
computation times. Ultrasonic Computed Tomography was also awarded by the Werner von
Siemens Award as one of the Top 15 ingenuity programs.
Keywords
NDE, NDT, Ultrasonic Testing, UT, Synthetic Aperture Focusing Technique, SAFT,
Reconstruction, CT, Computed Tomography
1. Introduction
Ultrasonic testing of forgings started right after the commercial availability of the first
ultrasonic systems around 1950 [1, 2, 3] and became one of the most important tools to
guarantee a sufficient lifetime (see Fig. 1). Material and NDE Engineers got a view into the
components and Engineers learned how to interpret signals and how to classify indications.
This information feedback from the final parts also helped improving the melting, heat
treatment and forging processes.
In the 1960s fracture mechanics was introduced and starting around 1970 sizing of
indications became a requirement for most large forgings. For indications larger than the
beam profile, sizing by probe travel (echo dynamic sizing, e.g. -6 dB drop method) is used
and for indications smaller than the beam profile area amplitude based sizing methods are
used (like DAC or DGS). Those methods compare the reflectivity of indications with the
reflectivity of artificial reflectors like Flat Bottom Holes, Disc Shaped Reflectors, or simply
1 VRANA GMBH
2 SIEMENS AG
3 SAARSCHMIEDE GMBH
4 FRIEDRICH-ALEXANDER-UNIVERSITÄT ERLANGEN-NÜRNBERG
back wall reflections. The size of an indication is finally quantified by reporting the diameter
or area of an equivalent artificial reflector. By comparing the noise level with the reflectivity
of equivalent defect sizes, this also allows to measure the achieved sensitivity during a
particular inspection, reported as a minimum detectable defect size. Nowadays sensitivities
are typically equivalent to ~0.7 – 2 mm diameter disc shaped reflectors, depending on the
size of the inspected part and the component material.
Fig. 1: Ultrasonic examination of a heavy rotor forging in 1953 [4] (left)
and automated inspection system from 1995 at Saarschmiede [2] (right).
2. Automated Data Recording
Around 1995 automated inspection systems (see Fig. 1) became available allowing a
more precise control of key factors of the ultrasonic inspection. Moreover data recording,
enhanced data processing and off-site data-review became possible.
Typically, when using automated ultrasonic inspection systems, the component is
rotated continuously (scan direction) and the probe holder is indexed along the radial or axial
axis (index direction). Due to a certain pulse repetition rate the data is recorded in a certain
grid – also called the scanning grid [5, 6]. At each point of the grid the time signal (reflected
amplitude over time) is recorded. The inspection data can then be displayed as A-Scans (x-
axis: sound propagation direction, y-axis: Amplitude), B-Scans (one axis: scan or index
direction; other axis: sound propagation direction; color coding: amplitude) and C-Scans (one
axis: scan direction; other axis: index direction; color coding: amplitude).
However, the time signal does not only contain the information from the center beam
(region “below” the probe), but is an integrated value across the beam profile of the probe
(see Fig. 2). This means that at each probe location we have accurate information about the
distance between indication and probe, but not about the direction.
Fig. 2: Beam profile of a normal probe (left)
and of a narrow beam probe (right).
On the other hand, indications are typically not only showing in the signal from one
probe location, but in the signals from multiple adjacent probe locations. As in B-Scans the
indications are plotted along the main beam axis this results in a typical hyperbola shape of
indications (see Fig 3).
The achieved sensitivity is limited both by the sound attenuation and by noise. Noise
consists of stochastic noise (e.g. due to the electrical noise of the UT flaw detector or
electromagnetically induced spikes from machines nearby) as well as echo signals caused by
the grain structure of the component material. Those grain noise signals however are
integrated over the beam profile which means that a wider beam creates a higher grain noise
level (see Fig. 2 – dark gray area).
Both facts (noise level and hyperbola shape of indications) are good arguments for
focusing the sound beam either by using larger probe sizes [7] or by using higher probe
frequencies. Higher probe frequencies will, however, lead to higher sound attenuation and
increased noise and hence reduce the inspection sensitivity. Larger probes require a more
sophisticated probe design and are more difficult to couple properly. Furthermore, they
increase the required inspection time by one order of magnitude if performed according to
standards [8] (instead of 3 angles to cover a 21° zone (7°, 14°, 21°) a much higher number of
angles and thus scans are necessary).
A more favorable option is synthetic aperture focusing realized by computer based
postprocessing of the recorded UT data. This will be shown in the following.
Fig. 3: One indication (big black dot) is found by 7 probe locations, for each probe location the distance is
known but not the angle (left); in typical B-Scans indications are plotted at the location of the probe using the
known distance; this leads to crescent shaped indications (middle); B-Scan of a real indication (right).
3. Synthetic Aperture Focusing
The knowledge how the hyperbola shaped indications are created can be used to derive
a simple heuristic method to reconstruct the real indication location. A straight forward way
would be to draw half circles at the beam exit points of the probes using the distance to the
indication as radius. These circles will intersect at the location of the indication as shown by
Fig. 3.
Mathematically a different approach is superior. The volume of the component is
divided into voxels (a 3-dimensional grid with a grid size one order of magnitude smaller
than the wavelength) and for each voxel with the coordinates 𝑟=𝑥,𝑦,𝑧 all signals are
considered which could have its origin in that voxel (𝑟
!∈𝑆!"). All those signals 𝑎𝑟
!,𝑡,
with probe location 𝑟
!, time 𝑡=!
!
𝑟
!−𝑟, and sound velocity 𝑐, are added for each voxel
(best is to use the non-rectified high-frequency signals). This will lead to constructive
interference (synthetic aperture focusing) at the location of indications and to destructive
interference in the surrounding area (see Fig. 4) [9]:
𝐼𝑟=
𝑥
𝑦
𝑧
=𝑎𝑟
!,𝑡=
2
𝑐𝑟
!−𝑟
!!∈!!"
.
(1)
Fig. 4: Constructive (left) and destructive (right) interference.
This means for every voxel within a 3D volume a two-dimensional summation of all
signals from the scan surface has to be conducted. This approach is called 3D-SAFT [10]. To
reduce calculation time a different approach can be taken by separating the complete 3D
volume into individual 2D slices. This reduces the computational effort to a situation that for
every “voxel” within the 2D slice a 1-Dimensional summation of the signals from one single
scan line has to be conducted [10], but with a reduced signal to noise ratio and no resolution
improvement in the direction of the “omitted” axis. The result of this 2D approach can still be
displayed in a 3D volume and by other authors is often referred to as a 2.5D imaging.
The result of both reconstruction methods is one amplitude value for each voxel, just
like in X-Ray Computed Tomography results. This is why SAFT is also called Ultrasonic
Computed Tomography.
The Synthetic Aperture Focusing Technique (SAFT) has its origin in the Synthetic
Aperture Radar (SAR) from 1967 [11]. The first implementation by D.W. Prine in 1972 [12]
was using optical components. The first digital system for NDT was realized by J.R.
Frederick in 1976 [13]. Despite the SAFT method being known for long time it required the
progress of computer technology and the advent graphics board usable for general-purpose
calculations until it could be applied for large component inspection. In 2010 Siemens and
BAM implemented SAFT for the inspection of large rotor forgings [14, 15].
Fig. 5 demonstrates the improvement of both the imaging of indications and the
sensitivity (shown by the blue noise in the classical result).
Fig. 5: Classical result of the ultrasonic inspection (left)
of a rotor forging with an outside diameter of 1460 mm
in comparison to the SAFT reconstruction (right).
4. Benefits of SAFT
The most important benefit of the presented method is the significant improvement of
the signal to noise ratio (SNR) (see Fig. 5). Both mentioned sources of noise, stochastic noise
and noise caused by grain structure can be effectively reduced. Stochastic noise is reduced by
a factor of 𝑛 , with 𝑛 different probe locations building the reconstruction result, as SAFT
can be considered an intelligent averaging method with a phase-corrected superposition of
the signals. Grain noise is reduced due to the focusing effect of SAFT (as discussed in
paragraph 2).
A second, extremely important benefit is the improvement of the lateral sizing and
resolution (see Fig. 5). Due to the focusing effect of SAFT a lateral resolution 𝐹𝑊𝐻𝑀!"# of
half the probe diameter 𝐷! can be reached [16, 17]:
𝐹𝑊𝐻𝑀!"# =
𝐷!
2.
(2)
In classical ultrasonic inspections, this resolution is only possible at the focal point of the
beam and it will even worsen with increasing metal travel path due to the increase in beam
diameter. Typical metal travel paths in large forgings (1000mm) result in very large beam
diameters (130mm) for classical standard probes.
Due to the cylinder geometry, the lateral resolution is improved further in radial scans for
locations closer to the center. For a disk without inner bore the best resolution (a quarter of
the wavelength 𝜆) is reached in the center:
𝐹𝑊𝐻𝑀!"# =
𝜆
4.
(3)
As shown in Fig. 6 the improvement of the lateral resolution also helps resolving group
indications.
The improvements in SNR in combination with the improved lateral resolution will
significantly reduce the false call rate in case of grouped indications detected on long metal
travel paths. At the same time, the determination of the indication location is simplified.
Conventionally, the determination of the indication location is challenging (in particular for
angle beam inspections) as only the distance to the indication is known and not the angle.
SAFT automatically displays the indications at their correct location.
Fig. 6: Conventional result of the ultrasonic inspection of two reflectors in close proximity (left)
in comparison to the SAFT reconstruction (right).
SAFT is easy to perform: existing automated inspection systems can be used and the
imaging of the results is similar to the imaging of the classical ultrasonic data. A-Scan, B-
Scan and C-Scan like views of the results are possible. Moreover, SAFT can be used both for
longitudinal waves and shear waves.
Finally, during classical inspections, the classification, evaluation, and localization of
indications might require additional (manual) scans to establish a reliable result. Due to the
improved determination of indication location and lateral sizing by SAFT those additional
scans can be reduced.
5. Benefits of the Presented SAFT Implementation
The presented SAFT implementation is developed for production use in the forging
industry [18, 19]. This brings certain boundary conditions (which are also applicable for any
implementation for any other industrial use). The software must be easy to use. This enables
inspectors in the forging shops to perform the reconstruction reliably and to review the data
similarly to classical UT data. The reconstruction process has to be validated and must be
faster than the actual inspection. This allows to trust the inspection results and to finish one
reconstruction while collecting the data. The aim is to keep the time delay between the last
finished scan and the review as short as possible, to finish the review of one part, before the
next one is loaded on the machine. Finally, the file sizes must be manageable.
All those goals are achieved: the results are within ±2 𝑑𝐵, the reconstruction time is,
even for 3D SAFT, below half the data acquisition time (by switching from CPU to GPU
reconstruction a speed-up by a factor of ~2000 was accomplished), and, due to combined
data compression algorithms, all necessary data is maintained.
Usually indications in the SAFT results show oscillations (see Fig. 7). Instead of using
the measured echo signals directly for SAFT reconstruction, it is possible to convert them
into complex numbered echo signals with same amplitude and phase (analytical signal) by
using the Hilbert Transformation:
𝑎𝑟
!,𝑡=𝑎𝑟
!,𝑡+𝑖∙ℋ𝑎𝑟
!,𝑡.
(4)
By performing the SAFT reconstruction on the complex numbered echo signals and
stripping the signal phase afterwards the oscillations are eliminated and the signal envelope is
displayed. This leads to a reconstruction result which is easier to interpret in particular as the
maximum amplitude is easier to determine. Moreover, the scanning grid can be coarser which
reduces the inspection time.
Fig. 7: SAFT result after reconstruction using real numbers (left)
and SAFT result after reconstruction using complex numbers (right).
6. SAFT: A Quantitative Technique
SAFT indications larger than the lateral resolution 𝐹𝑊𝐻𝑀!"# (which is still smaller
than the beam profile) can be sized by imaging tools. This leaves the challenge of measuring
the sensitivity of SAFT and to size small indications. For years, SAFT was considered just as
an imaging tool and up to the moment the improvement of the signal-to-noise ratio discussed
in chapter 4 could not be used to improve the inspection.
However, just like it is done in classical ultrasonics the amplitude of the SAFT signal
can also be used to gain size information [20, 21]. This is possible by comparing the
amplitude sum of the SAFT reconstruction with the amplitude sum of a reconstructed
simulation of artificial reflectors. Therefore, this method is like the DGS method – just that
the SAFT-DGS table needs to be calculated for a certain component geometry.
As shown in Fig. 8 the SAFT implementation presented in this paper includes a SAFT-
DGS assessment meaning the amplitude of each voxel represents the equivalent reflectivity
of a disc shaped reflector (DSR). This makes the determination of the achieved sensitivity
and of the indication size straight forward and due to the its capability of small defect sizing
SAFT grows from an imaging tool to a full-fledged quantitative measurement technique. This
also allows quantifying the improvement of the signal-to-noise ratio.
Fig. 8: SAFT-DGS assessed SAFT result [21].
8. Summary
The first production inspections using SAFT show a sensitivity improvement of up to
one order of magnitude compared to classical ultrasonic inspections. In the centre of the
forging, which is usually one of the most critical areas from a fracture mechanics viewpoint,
the largest sensitivity improvement can be achieved.
Combined with the better resolution / spatial separation of defects (which simplifies
indication classification) and fast reconstruction times SAFT has proven to be a very reliable
and versatile addition to automated inspection systems and a game changer in ultrasonic
testing.
References
[1] J. Vrana, A. Zimmer, et al., “Evolution of the Ultrasonic Inspection Requirements of
Heavy Rotor Forgings over the Past Decades”, Review of Progress in QNDE (2010, 29),
edited by D.O. Thompson, D.E. Chimenti, pp. 1623-1630
[2] A. Zimmer, J. Vrana, J. Meiser, and N. Blaes, “Evolution of the Ultrasonic Inspection of
Heavy Rotor Forgings Over the Last Decades”, Review of Progress in QNDE (2010, 29),
edited by D.O. Thompson, D.E. Chimenti, pp 1631-1638
[3] J. Vrana, A. Zimmer, et al., “Evolution of the Ultrasonic Inspection over the Past Decades
on the Example of Heavy Rotor Forgings”, World Conference on Non-Destructive
Testing, Munich (2016, 19), Th-2-A-5
DSR$0$…!1!mm
DSR$0.98!mm
[4] R. Schinn, “Erfahrungen mit der Ultraschallprüfung bei der Abnahme von Rotoren für
Turbosätze“, Metall (1953, 7), pp. 502-506
[5] J. Vrana, et al., “Determination of an Optimal Examination Grid for the Automated
Ultrasonic Inspection of Heavy Rotor Forgings”, European Conference on Non-
Destructive Testing, Prague (2014, 11)
[6] DGZfP, “Guideline for the determination of the scanning grid for the automated
ultrasonic testing of large forgings”, DGZfP, Berlin (2017)
[7] G. Maes, P. Tremblay, D. Devos, N. Hoshi, H. Nimura, H. Narigasawa, “Improved
Inspection Technique for Large Rotor Shafts, Using a Semi-Flexible Phased Array Probe”,
Int. Conf. on NDE in Relation to Structural Integrity for Nuclear and Pressurized
Components, Jeju (2015, 11)
[8] “Inspection and Testing of Large Forgings and Castings for Steam and Gas Turbine
Generator Sets”, VGB, Essen (2015)
[9] K. Fendt, “Ultraschall-Charakterisierung von Defekten”, Doctoral Thesis (2017, to be
published)
[10] H. Mooshofer, U. Büchner, W. Heinrich, K. Kolk, H.P. Lohmann, J. Vrana, “3D SAFT-
Analyse von großen zylindrischen Schmiedeteilen”, DACH-Jahrestagung, Graz (2012)
[11] W.M. Brown, “Synthetic aperture radar”, IEEE Trans. Aerosp. Electron. Syst. (1967,3),
pp. 217–229
[12] D.W. Prine, “Synthetic Aperture Ultrasonic Imaging”, Proceedings of the Engineering
Applications of Holography (1972) pp. 287-294
[13] J.R. Frederick, J.A. Seydel, R.C. Fairchild, “Improved Ultrasonic Non-Destructive
Testing of Pressure Vessels”, Nuclear Regulatory Comission Report NUREG-0007-1
(1967)
[14] R. Böhm, W. Heinrich, “Störsignalminderung bei der Ultraschallprüfung durch ein
Auswerteverfahren in Anlehnung an die Synthetic Aperture Focussing Technique
(SAFT)”, DACH Jahrestagung, St. Gallen (2008), P40
[15] H. Mooshofer, M. Goldammer, W. Heinrich, K. Kolk: “Untersuchung von Robustheit
und SNR der SAFT-Analyse”, DGZfP Jahrestagung, Bremen (2011), Mi.1.A.
[16] K.T. Fendt, H. Mooshofer, S.J. Rupitsch, R. Lerch, H. Ermert: “Investigation of the
Synthetic Aperture Focusing Technique resolution for heavy rotor forging ultrasonic
inspection”, IEEE International Ultrasonics Symposium, Prague (2013), pp. 1869 - 1872
[17] H. Mooshofer, K. Fendt, U. Büchner, W. Heinrich, K. Kolk, H.P. Lohmann, J. Vrana:
“Anwendung von SAFT im Energiemaschinenbau”, Seminar des DGZfP Fachausschuss
Ultraschallprüfung, Berlin (2013)
[18] B. Müller, “Transparent Turbines”, Pictures of the Future (2016,3)
[19] V. Steger, “The Transparent Shaft”, Siemens YouTube Channel (2016,5)
[20] R. Boehm, K. Fendt, W. Heinrich, H. Mooshofer, “Verfahren und Vorrichtung zur
Defektgrößenbewertung”, Patent DE102013211616 (2014)
[21] H. Mooshofer, R. Boehm, W. Heinrich, K. Fendt, M. Goldammer, K. Kolk, J. Vrana,
“Amplitudenbasierte Fehlergrößenbewertung mit SAFT: Auf dem Weg von der bildlichen
Darstellung zum Messverfahren”, DGZfP Jahrestagung, Koblenz (2017), Di.1.C.3
