Fully non-contact hybrid NDT inspection for 3D 3 defect reconstruction using an improved SAFT 4 algorithm 5

: Non-destructive testing of metallic objects that may contain embedded defects of different sizes is an important application in many industrial branches for quality control. Most of these techniques allow defect detection and its approximate localization, but very few give enough information for its 3D reconstruction. Here we present a hybrid laser – transducer system that 21 combines remote laser-generated ultrasound excitation and non-contact ultrasonic transducer 22 detection. This fully non-contact method gives access to separating scan areas on different object’s 23 faces and defect details from different angles/perspectives can be analysed. This hybrid system can analyse the whole object’s volume data and allow a 3D reconstruction image of the embedded defects. As a novelty for the signal processing improvement, we use a 2D apodization window filtering technique, applied along with the synthetic aperture focusing algorithm in order to remove the undesired effects of side lobes and wide-angle reflections of propagating ultrasound waves, 28 thus, enhancing the resulting 3D image of the defect. We provide both qualitative and quantitative 29 volumetric results with high accuracy and resolution compared with conventional techniques.


34
Non-Destructive Testing (NDT) techniques are widely used for embedded crack detection inside 35 solid materials, with effective results in the quality control and material inspection strategies in 36 modern industry [1]. Ultrasound transducers are considered the most common devices for NDT 37 inspection thanks to their wide frequency band selection, small sizes and ability to work in harsh 38 environments [2,3]. The ultrasound transducers can be used in two major modes of exciter/ receiver 39 operation, namely pulse-echo or pitch-catch modes [4,5]. Depending on the typical application and 40 availability of transducers; either of these two modes is favoured. However, the use of ultrasonic 41 transducers both as exciters as well as receivers of ultrasound has always faced some limitations, contact transducers fixed in a single position on the object's surface. Although in contact mode, the 48 received signal has a higher signal to noise ratio (SNR) with higher gain and less attenuation, the 49 contact mode has an important drawback when the same detector has to be used to scan a certain 50 area. When the detector is moved from one point to another, the coupling factor changes, making 51 impossible a quantitative comparison of the two measurements. Moreover, an automatic scanning is 52 not possible to implement with contact sensors. On the contrary, a non-contact transducer can be 53 programmed to automatically scan a larger area, with the drawback of signal attenuation due to the 54 airgap and the weaker sensitivity to detect small variations in the ultrasound signal.

55
The signal processing usually implemented to contact or non-contact ultrasonic techniques is

62
The Fourier transform has the ability to detect the frequency response of the detected signal without 63 giving detailed information about the corresponding TOF of a certain frequency component. STFT

64
and WT can detect information about both frequency and TOF with a certain level of uncertainty.

65
However, WT is more flexible and accurate than STFT due to the fact that the window size in the WT 66 changes with frequency and time, whereas the window size, in the case of STFT, is fixed, which limits 67 the resolution of the algorithm [16]. Moreover, time reversal techniques rely on the principle of 68 detecting the ultrasound wave field using a receiver and resending the same wavefield to the source, 69 but reversed in time, producing a convergence of the signal towards the initial source position. If the 70 source of this signal is a defect, then the convergence of the reversed signal will occur at the defect 71 position allowing its visualization. [11]. Finally, the SAFT technique relies on the principle of delay 72 and sum (DAS) that generates a focused image of the defect out of multiple unfocused images. This 73 focused image has a much higher amplitude at the defect position compared with healthy positions 74 inside the object of interest. The SAFT has the advantage of being able to visualize the whole volume 75 providing 2D or 3D information about the object depending on the resolution and the number of 76 scans performed on the object [17]. In the recent literature, the SAFT is used with ultrasonic 77 transducers for generating an image of the embedded defects in the sample under test representing 78 a viable solution for the signal processing [12,14,18].

79
As an alternative to ultrasonic methods, all-optical systems for ultrasound generation and 80 detection have emerged. Laser generated ultrasound (LGU) technique is used for excitation and 81 optical interferometry setups are used for signal detection. These all-optical methods have a very 82 high resolution and the ability to send and receive signals remotely at a high range of power intensity 83 that is not achievable in the case of conventional transducers [19,20]. This fact implies a higher 84 penetration of the ultrasound to deeper areas inside the object and ability to detect deeper embedded 85 defects. However, the main drawback of the interferometric optical detection, apart from the fact that 86 it is very expensive, is the stability criterion where the system needs to be isolated against 87 surrounding vibrations, making it difficult to apply in practical industrial applications [19]. Due to 88 these drawbacks, all-optical NDT methods are only justified when the defect size is at a micrometre 89 scale.

90
In this work, we propose a hybrid system which combines laser-induced ultrasound as an exciter 91 and conventional non-contact transducers as a receiver. This method combines the advantages of 92 both technologies to obtain a full contactless 3D reconstruction of a defect with a very good resolution 93 and to improve the feasibility of the system in industrial applications [21]. In the proposed method, 94 both the exciter (the laser spot) and the receiver (only one non-contact transducer) are programmed the 3D reconstruction of the defect at different perspectives/angles. The recorded ultrasound signals 97 are processed using an improved SAFT algorithm for 3D defect reconstruction. In addition to the 98 SAFT method, an enhanced apodization function is implemented to eliminate the effect of the 99 ultrasound side lobes generated by the exciter as an artifact in the input signal [22]. This technique 100 results in an enhanced quality of the reconstructed image with less redundant or unnecessary 101 shadows. To our knowledge, the apodization function for signal processing using the SAFT 102 algorithm was previously applied to 1D geometries for 2D defect reconstruction. Here, we developed

126
For an arbitrary excitation point, T, that generates an ultrasound wave propagating in the 127 volume and an arbitrary receiver, R, which detects the reflected signal by the volume points, one of 128 which is the point of interest, P, we can define the TOF of the signal as [12,14,18]:

145
In order to reach more accurate results avoiding uncertainties, we propose the use of an 146 apodization function to enhance the resulting image by weighting down the amplitudes that would 147 result from the propagation of side lobes that can produce secondary echoes that influence and distort 148 the main echoes generated by scatterers (Eq.3).

154
In the SAFT analysis, if the scan is performed in a line of scan to generate a reconstruction image 155 in a 2D plane, i.e. scan in one line and inspection of the depth of the object in the plane containing 156 that line, we use a 1D apodization window which represents the aperture line at each point of the 157 scatters in the cross-section depth of the object. Extending this to the 3D case, when the SAFT scan is 158 performed in a 2D area to perform a 3D reconstruction image, means that the aperture of the 159 apodization window is also becoming a 2D area. To make it clearer, the dimension of the apodization 160 window is the same as the dimension of the scan line/area as it depends on the position of the scan 161 sensors. A drawback of apodization is the possible change of the main lobe width, which affects the 162 lateral resolution. Hence a good selection of the apodization function and its width helps get the 163 optimum results with a minimum lateral resolution deficiency.

164
To give a clearer understanding of the apodization function, we will explain this concept shortly 165 in the case of a 1D scan line at X-axis and scatter plane XZ, and later, we will expand the explanation 166 to the general case of 2D scan area and scatterer volume.  X , and neglecting all other measurements outside this threshold (Eq.6).
where X-X' is the horizontal shift between the position of the scan point and scatterer.

179
The most commonly used types of the apodization function are the rectangular or Hanning function Fine tuning to the threshold criterion value, α, should be applied to obtain the optimum results. In 189 order to apply the same apodization function to the whole volume with a 2D scan area, it is required 190 to put into consideration that the beamwidth angle becomes a solid angle instead of a planar angle.

191
To make it simpler, we will have two angles

266
The results of our full non-contact experiment, for each scan area, are shown in Figure 4. The defect

296
Combining all the three faces views and representing them at the scale of the cube, we may 297 generate the 3D reconstruction of the defect as shown in Figure 5. Here, a unique universal Cartesian 298 axis reference XYZ is used for all the 3 faces with the origin in the corner of the cube object itself. The

299
relative displacement between the scan faces is considered when superimposing the scan faces all 300 together on the 3D reconstruction. We applied a thresholding filter to reject all data below a certain 301 amplitude threshold to keep only the high intensity data to represent the defect position and shape.

302
Applying these reject thresholds with proper fine tuning of the threshold limit also helps reduce the 303 size error since some data with lower intensity is rejected, improving the accuracy of the algorithm. applying a reject threshold for data with an intensity below 65%, while Figure 5b shows the front 306 view resulting from face 1 scan area inspection by applying a reject threshold for data with an 307 intensity below 65%. Figure 5c,d shows the isometric view and side view representations 308 respectively for face 2 with the same threshold condition. Figure 5e,f shows the isometric view and 309 top view representations respectively for face3. Figure 5g shows the isometric view including the 310 intersection between reconstruction results from the three faces with a filtering threshold of intensity 311 reject for values less than 65%. Figure 5h shows the isometric view including the intersection between 312 reconstruction results from the three faces with a more restrictive threshold of intensity to reject data 313 with an intensity below 85%. We superimpose the true cylindrical defect shape on the reconstructed 314 defect images to see how close the detected information is, with respect to true information. It is clear  view/X1Y1) scan area inspection by applying a reject threshold for data with an intensity below 65% b) Front 329 view resulting from face 1 scan area inspection by applying a reject threshold for data with intensity below 330 65% c) Isometric view resulting from face2 (Side view/X2Y2) scan area inspection by applying a reject threshold 331 for data with an intensity below 65% d) Side view resulting from face 2 scan area inspection by applying a 332 reject threshold for data with an intensity below 65% e) Isometric view resulting from face3 (Top view/X3Y3) 333 scan area inspection by applying a reject threshold for data with an intensity below 65% f) Top view resulting 334 from face 3 scan area inspection by applying a reject threshold for data with an intensity below 65% g)

335
Isometric view resulting from the 3d reconstruction by applying a reject threshold for data with an intensity 336 below 65% h) isometric view by applying a stricter reject threshold for data with an intensity below 85%.

337
The 3D reconstruction of the defect in Figure 5 clearly shows that the improved SAFT method used 338 in this work has powerful advantages of visualizing the defect in the 3D isometric view. The

339
projections of the SAFT planes make it easier to distinguish the location of the defect in a 3D manner 340 instead of just getting two dimensional results. The SAFT algorithms using a hybrid technique allows 341 scanning over large objects without losing much information. In fact, we get a much better localized 342 quality image of the defect with larger objects compared with smaller ones due to the aforementioned 343 boundary conditions.

344
The precision of the results obtained here can be further improved using a non-contact transducer 345 with a smaller size, adapted to the horizontal size of the defect. The resolution would be enhanced 346 and calculated accurately with less error. Another significant qualitative enhancement for using the 347 non-contact transducer is the cleaner signal with a higher intensity contrast due to having the 348 receiving transducer closer to the exciter. However; there are also some drawbacks for using the non-349 contact transducer. For example, positioning the non-contact transducer close to the exciter is 350 obligatory, otherwise, there would be much attenuation in the received signal and it will be very 351 difficult to detect the presence of embedded defects. Also, the airgap affects the quality of the signal, 352 but this is rectified in our experiment by using a high-power pulsed laser exciter and high-gain 353 preamplifier.

354
An important advantage of the proposed method is that there is no need to use a healthy reference 355 sample for comparison with the resulting reconstruction image of the unhealthy sample to detect the 356 presence of the defects. However, in the case of an object that has no symmetrical dimensions or has 357 a composite structure of different materials, it would be better to use the healthy reference sample.

358
This is because the internal reflections of the internal boundaries inside the object can be 359 misinterpreted as a defect while they, in fact, represent the object structure.

361
We have proved that using a hybrid method composed by LGU, as an exciter, and non-contact