Abstract

This work focuses on non-destructive examinations using array probe ultrasonic waves on complex materials generating a high structural noise on the examined area. During an ultrasonic examination, multiple scattering of the ultrasonic waves at the grain boundaries makes the distinction between this structurally induced noise and a potential defect challenging. The difficulty of the interpretation can moreover be increased in the near surface area because of the subsurface wave. In order to ease the analysis of these acquisitions, some numerical processing methods are proposed. Statistical properties of the imaging results (for instance, total focusing method or plane wave imaging) are first calculated on several sensor positions. These statistical properties are then used to post-process the imaging results and enhance any signal values that do not belong to the structural noise expected statistics. The method, called “CORUS,” has been successfully tested on cast austenoferritic stainless steel coarse-grained mock-ups, with several dB gain compared to the classical total focusing method. It is now integrated in a civa software plugin and in a prototype version of the real-time PANTHER-phased-array acquisition system from Eddyfi Technologies.

1 Introduction

Ultrasonic examinations can be strongly affected by the inspected material's microstructure [13]. A macrography of a coarse grain component is shown in Fig. 1. Depending on the used frequency and the grain size, the grain boundaries can generate an important multiple scattering signal, also called “structural noise” (Fig. 2). It is difficult to distinguish the numerous echoes originating from the microstructure from an echo that really comes from a defect. Moreover, the potential flaw signal can be strongly attenuated due to the ultrasonic energy spreading in the microstructure.

This structural noise has the same temporal (and spectral) signature as the signal generated by the searched defects; typical signal processing methods (filtering, wavelets, etc.) are therefore not applicable to suppress the structural noise.

To ease the defect detectability, a significant improvement can be obtained by using array probes. A review of array probes signal processing techniques in the presence of structural noise can be found in Ref. [4]. Figure 3 shows an example of array signal acquisition—one line of a full matrix capture (FMC)—in case of strong structural noise.

A signal processing method, called “CORUS,” is proposed here to increase the defect detectability on phased-array acquisitions.

The experimental evaluation process (mockup and sensors) is detailed in Sec. 2, and then, the method is fully described in Sec. 3, and the results are presented in Sec. 4. Section 5 concludes the paper.

2 Experimental Procedure

The experimental procedure is described in Fig. 4. FMC acquisitions have been performed on a coarse grain cast austenoferritic stainless steel mockup. The mockup is a rectangular parallel-piped 100-mm-thick block with side drilled holes (SDHs) of diameter 2, 1, and 0.5 mm at a depth varying from 2.5 mm to 30 mm.

The block has been inspected with a 128-element phased-array probe in contact mode with a central frequency of 2 or 5 MHz. The probe pitch is 0.5 mm.

3 Signal Processing Method

A two-step method is described here for processing FMC acquisitions from array probe. This method extends the algorithm proposed in Ref. [5] for time of flight diffraction (TOFD) acquisitions. A first step, described in Sec. 3.1, enables to homogenize the noise influence in the entire imaging area. The second step, described in Sec. 3.2, is a spatial filtering approach which enhances the flow detectability.

3.1 Noise Homogenization.

Let Ik(u, v) be the imaging result at coordinate (u,v) for sensor positions k (Fig. 5). Ik(u, v) can for instance be a “total focusing method” (TFM) result given by 
Ik(u,v)=|i=1Mj=1Mxk(n(u,v,i,j),i,j)|
(1)
where xk(n, i, j) is the full matrix capture acquisition (sensor position k, emitter element i, and receiver element j) and n(u, v, i, j) is the traveling time (expressed as a number of sampling times) of a signal:
  • sent by element i

  • reflected at position (u, v)

  • received by element j

Different TFM implementations exist. The raw signal (xk) in the summation can be replaced by its absolute value, its analytic transform, or its envelope.

A preliminary phase consists in learning the statistical structural noise influence on Ik(u, v); for a given imaging position (u, v), the structural noise mean A(u, v) and standard-deviation B(u, v) can be calculated using the imaging results at Ns sensor positions 
A(u,v)=1Nsk=1NsIk(u,v)
(2)
 
B(u,v)=1Ns1k=1Ns(Ik(u,v)A(u,v))2
(3)

Figure 6 shows examples of the obtained mean and standard deviation matrix obtained on the block mockup presented Fig. 1. Note that, due to the subsurface wave, the structural noise is much more important in the first millimeters.

These statistical properties are then used to post-process the imaging results and enhance any signal values that do not belong to the structural noise expected statistics. For any imaging results I(u, v), we build a spatially homogenized image Ihomog(u, v) given by: 
Ihomog(u,v)=|I(u,v)A(u,v)B(u,v)|
(4)

Ihomog(u, v) is statistically homogeneous and can be easily used for defect detection: in undisturbed areas, it has a similar mean (0) and standard-deviation (1) at any pixel location (u, v). An example of the noise homogenization result is given (Fig. 7).

3.2 Image Filtering.

After the noise homogenization, several image filtering methods can be used to enhance the flaw detectability. We propose to use a bilateral filter [6], a classical image processing tool, which allows to reduce the noise level while maintaining the image edges.

In our defect detection case, the bilateral filter enables to decrease the (homogenized) noise variability while preserving the potential defect signal value, thus improving the defect detectability.

To lighten the equations we call w=(u,v), the pixels coordinate in the noise-homogenized image 
Ihomog(u,v)=Ihomog(w)
(5)
The bilateral filtering equation is given by 
Ihomogfiltered(w)=neigbouringpixelsa(w,s)Ihomog(s)
(6)
where the filtering coefficients a(w,s) are given by 
a(w,s)=exp{sw22σspatial2}exp{|Ihomog(s)Ihomog(w)|22σamplitude2}
(7)

The first coefficient exp{sw2/2σspatial2} is a classical linear filtering coefficient, a Gaussian decay of the neighboring pixel distance. When the noise has been properly homogenized, an increase of σspatial increases the filtering, but also the computation time. σspatial can be set to the minimal search defect length.

The second coefficient exp{|Ihomog(s)Ihomog(w)|2/2σamplitude2} enables to average only the neighboring pixels with “similar” values, that is pixel s such as Ihomog(s)Ihomog(w) belongs to the (homogenized) noise expected values. The value of σamplitude is chosen to be three times the standard deviation of Ihomog.

An example of the bilateral filtering of a noise-homogenized TFM is provided in Fig. 8, where a 10-dB gain is observed.

4 Results

At intermediate depth, from 10 mm to 20 mm, the SDH is easily detectable with the classical TFM imaging method. We therefore focus on the near surface SDH (below 10 mm, Figs. 911), where the subsurface wave strongly degrades the imaging quality, and the deep SDH at 30 mm (Fig. 12), where the attenuation due to the coarse gain structure strongly decreases the defect signal.

In Fig. 11 some artefact are visible around the hole. They are not created by our processing; one can see them, with lower contrast, in the raw TFM image. These spurious echo lobes might be attributed to creeping waves surrounding the side drilled hole, which reemit head waves in the bulk material. The re-emitted head waves are shear polarized waves and then submitted to higher attenuation.

The performances are summarized in Table 1. The CORUS method enables a strong increase of the signal-to-noise ratio. More than 10-dB gain is obtained in the different configurations (large or small depth, high or low frequency).

5 Conclusion

A signal processing method is proposed to ease the diagnostic of component with heterogeneous microstructure generating high structural noise. A preliminary step consists in estimating the noise influence on the imaging results statistics. These statistics are then used to homogenize the noise in the imaging results. An image filtering step is then applied to reduce the remaining noise variability while preserving the indication contrast. This method has been successfully tested on a coarse grain cast austenoferritic stainless steel mockup where several dB have been gained compared to the classical TFM.

Our main perspective is the integration of CORUS on industrial products. CORUS has been integrated in the civa software packages thanks to a python plugin. An example of a civa result is given in Fig. 13 on the SDH at a depth of 5 mm and diameter of 0.5 mm. CORUS is also integrated in a prototype version of the real-time phased-array acquisition system PANTHER from Eddyfi (Fig. 14).

Acknowledgment

The author would like to thank Vincent Bergeaud and Stephane Leberre, from the CEA, for the integration a CIVA plugin, and Kombosse Sy, from Eddyfi Technologies, for the integration on a PANTHER prototype.

Funding Data

  • The ongoing integration of CORUS in civa software and PANTHER acquisition system has received funding from the Euratom research and training programme 2014–2018 under grant agreement No. 755500.

References

References
1.
Lhuillier
,
P. E.
,
Chassignole
,
B.
,
Oudaa
,
M.
, and
Kerhervé
,
S.-O.
,
2017
, “
Investigation of the Ultrasonic Attenuation in Anisotropic Weld Materials with Finite Element Modeling and Grain-Scale Material Description
,”
Ultrasonic
,
78
, pp.
40
50
. 10.1016/j.ultras.2017.03.004
2.
Van Pamel
,
A.
,
Brett
,
C. R.
,
Huthwaite
,
P.
, and
Lowe
,
M. J. S.
,
2015
, “
Finite Element Modelling of Elastic Wave Scattering Within a Polycrystalline Material in Two and Three Dimensions
,”
J. Acoust. Soc. Am.
,
138
(
4
), pp.
2326
2336
. 10.1121/1.4931445
3.
Van Pamel
Anton
,
Nagy
Peter B.
,
Lowe
Michael J. S.
,
2016
, “
On the Dimensionality of Elastic Wave Scattering Within Heterogeneous Media
,”
J. Acoust. Soc. Am.
,
140
(
6
), pp.
4360
4366
. 10.1121/1.4971383
4.
Lopez Villaverde
,
E.
,
Robert
,
S.
, and
Prada
,
C.
,
2016
, “
Ultrasonic Imaging of Defects in Coarse-Grained Steels With the Decomposition of the Time Reversal Operator
,”
J. Acoust. Soc. Am.
,
140
(
1
), pp.
541
550.
10.1121/1.4958683
5.
Paul
,
N.
,
Foata
,
M.
, and
Kassis
,
P.
,
2018
, “
Digital Processing for Ultrasonic Testing of Components Generating a High and Heterogeneous Structural Noise
,”
Proceedings of the 12th European Conference on Non Destructive Testing
,
Gothenburg, Sweden
,
June 11–15
.
6.
Tomasi
,
C.
, and
Manduchi
,
R.
,
1998
, “
Bilateral Filtering for Gray and Color Images
,”
Proceedings of the 1998 IEEE International Conference on Computer Vision
,
Bombay, India
,
Jan. 4–7
, pp.
839
846
.