Comparison of X-ray computed tomography and immersion ultrasonic non-destructive testing techniques in the case of qualitative and quantitative assessment of brazing quality level

The main aim of this work is to juxtapose directly two non-destructive testing techniques used to evaluate quality of brazed joints between metallic parts, namely X-ray Computed Microtomography (μCT) and immersion ultrasonic inspection (UT). From a general point of view, the application of ultrasound is a more efficient way to assess information about brazing filling imperfections (shorter time needed to obtain results, lower cost of measurements). However, in cases when connected parts have specific shape or geometry that is unfavourable for the application of ultrasound, an alternative approach has to be used such as porosity analysis of 3D data obtained from μCT technique. Hence, both techniques were employed in order to analyse brazed interfaces between copper and stainless steel pipe fitting in eleven cylindrical samples. The proposed methodology as well as developed data processing algorithm and results obtained are presented in this study.


Introduction
In the field of non-destructive testing of materials, quality control of joints connecting metallic parts plays an important role, especially at the European Organization for Nuclear Research (CERN) where inside high-energy particle colliders, such parts are subjected to extreme conditions and their failure may lead to costly consequences.Therefore, a conformity procedure was developed based on suggested quality levels given in ISO 18279:2003 European standard [1] in order to check brazed metallic components before their application.Among all possible defects that could occur in a brazed joint, cracks, voids and other filling imperfections are of particular interest, as they may strongly decrease heat transfer, electrical conductivity and worsen mechanical properties resulting in crack initiation and propagation.It is also of key importance to avoid filling imperfections in order to prevent leaks in hermetic seal applications and ultra-high vacuum environments.
Abovementioned brazing quality levels (Bstringent, Cintermediate, Dmoderate), used to define acceptance requirements, set limits for filling percentage, derived as the percentage of the total projected area covered with filler metal, to: >80%, >70%, >60% for quality levels B, C and D, respectively.
In most cases, these acceptance criteria concern brazes with internal imperfections, thus they have to be validated using well-known volumetric NDT techniques [2]: X-ray radiography, immersion ultrasonic inspection (UT) [3] and X-ray Computed Microtomography (µCT) [4], which allow to investigate internal structure of materials.
The first one is the fastest, but it is mostly limited to flat brazed joints.The latter two, in turn, are much more versatile techniques offering high capabilities at an expense, however, of more demanding measurement procedure.Both of them provide similar data, namely: position (x, y, depth) and ultrasonic wave reflection amplitude in the case of UT, and position (x, y, z) and X-ray linear absorption coefficient in the case of µCT.Hence, both have been recently compared in the case of extraction of delamination damage [5].Both methods have their own limitations though.
As application of µCT technique strictly depends on absorption properties of an inspected material, it is favourable and convenient that most of brazing filler metals (silver, gold, cobalt, copper) are highly absorbent (dense and with high atomic number Z), implying good contrast amongst filling layer, base metals and potential defects, even for such joined pieces like stainless steel or copper.On the other hand, penetration of X-ray beams generated by laboratory X-ray tubes is limited, reducing feasibility of µCT measurements of dense materials (copper, stainless steel) to few centimetres of thickness.This restriction goes hand in hand with a second size constrain coming from the fact that smaller data voxel size requires higher magnification and, thus, reduced maximum of object under inspection.
Another point to be noted is post processing and analysis procedure of tomographic 3D data obtained.It is complex in the case of estimation of filling percentage, as it involves the following steps: correction of measurement artifacts, determination and plane projection of brazing surface (unrollment, "thick slab" processing), application of an appropriate algorithm for segmentation of defects (thresholding) and porosity calculations using binarized image.
Immersion ultrasonic testing, combined with modern equipment allowing C-scan measurements, provides, in turn, direct estimation of filling percentage.It is also less restricted by total object thickness than µCT.However, it is limited in terms of spatial resolution: small, isolated pores may not be detected, while individual bigger pores contour may not be resolved correctly because of smoothed, overlapping boundaries and, thus, be considered as one large void.Moreover, appropriate conditions have to be met in order to conduct reliable measurements, as unfavourable geometry of volume of interest can reflect transmitted ultrasound in a wrong direction and thus prevent reception by the transducer.Consequently, the brazed interface has to be parallel to the object's external entry surface.Furthermore, the ultrasound beam must enter the object in a propagation direction perpendicular to the entry surface in order to transmit only a compression wave with no angular deviation (refraction) or conversion into different wave modes.
Taking into account the reasons abovementioned, brazed samples analysed in this work consisted of two thin metallic tubes exhibiting cylindrical symmetry.These brazed pieces are used in input-output module plate of re-designed version of radiation tolerant high power converters (Figure 1), which are developed at CERN in the frame of the Radiation To Electronics (R2E) project.These R2E converters provide [+4kA +6kA +8kA] +08V power output to power superconductive magnets.They will be installed in the Large Hadron Collider (LHC) at CERN during its next long shutdown (LS2).
Having UT and µCT measurements conducted on these samples, the aim of this study is to compare both techniques in terms of qualitative interpretation of images representing the brazed joint (UT C-scan amplitude image vs unrolled surface extracted from µCT volume) as well as estimated quantitative information about filling percentage given from these images.The filling percentage was provided directly by UT from a single scan, whereas several image analysis and processing steps were required to provide the filling percentage based on µCT scan.These steps are described in more details in Chapter 3. Chapter 4 presents comparison of the results obtained, which leads to the conclusion that both techniques give consistent and matching filling percentages.Moreover, µCT technique is considered as an appropriate technique to be used for non-destructive evaluation of brazing quality when high resolution image of filling defects is demanded.

Experimental procedure
Eleven samples developed for the LHC were inspected by UT and µCT.Each of them consists of a copper tube (5 mm in diameter) joined to a stainless steel fitting by brazing, which was done with an automatic induction system.The thickness were 1 mm for the copper and 2 mm for the stainless steel parts and the length of the brazed overlap was 8 mm.The brazed area is marked by green rectangle in Figure 1.Ultrasonic measurements were performed using a Krautkramer USPC 2100 immersion system and Harisonic I3 1006T (Ø15 mm, 10 MHz, spherical focus) probe, generating a longitudinal 10 MHz wave.An automated polar scan was programmed in order to cover the entire length of the brazed joint around 360° (length step 0.2 mm, circumferential resolution 1.5°), with ultrasound entering perpendicular to the outer copper surface.Gain was set such that the reflection from the inside surface of the copper tube (at a position where the braze was filled and therefore allowed ultrasound to pass through) reached an amplitude of 100% screen height.This provides a reference reflection amplitude to which any reflections from the brazed interface are comparedany material attenuation in the copper is negligible due to the small thickness.The limit of the area of imperfections was estimated using the -6dB method, i.e.where the reflection signal amplitude drops to 50% screen height.
µCT measurements were performed with a Zeiss Metrotom 1500 CT scanner (Figure 2) equipped with a 225 kV X-ray tube, using: accelerating voltage of 215 kV, accelerating current of 72 µA and 1.5 mm of copper filter.Over 1850 X-ray projections were collected with 40 cm x 40 cm detector (2048 x 2048 pixels, gain: 16, no image averaging, projection integration time: 2 s) and reconstructed using Feldkamp algorithm with Shepp Logan filter applied.Resulting 3D volume has voxel size of 15 µm.

Image processing and analysis of µCT data
Analysis of obtained 3D volumes was conducted using Volume Graphics StudioMax 3.2 software [7] and ImageJ freeware, open source distribution [8].The former was applied in order to align sample orientation vertically, prepare 2D and 3D views and determine a cylinder fitted to the brazing surface.For proper visualisation reasons, Figure 3 presents an example of cylinder (pink lines) located next to the brazed interface.The final cylinder was, however, positioned exactly inside the braze.Each surface was then unfolded into a flat plane using unroll algorithm (Figure 3d).In most cases the thickness of the unrolled cylinder was increased to 0.1 -0.2 mm using thick slab mode processing in order to include all defects and to reduce the influence of slight deviations of brazed interface from an ideal cylindrical shape.In these cases the minimum grey values from the slab thickness were selected.
The 2D grey level image obtained by unrolling the brazed surface contains information about filling imperfections.Unfortunately, in most cases the filling percentage cannot be directly quantified with a simple thresholding because of the influence of background grey level gradients.Therefore, a segmentation algorithm was developed in order to reduce grey level artifacts and present a binarized black and white picture (example shown in Figure 4d and Figure 5c).
Two image processing methods have been successfully tested and applied.First method consists of: histogram equalization, median filtering for noise reduction, Phansalkar's local thresholding, whereas the second uses Fourier Transform (FT) bandpass filtering combined with global thresholding.Both methods are then followed by binary morphological corrections.Finally, the data obtained are additionally transformed (adjustment of pixel size ratio, rescaling, cropping) to match the region of interest determined from ultrasonic C-scans (1074x 195 px)see examples in Figure 4 and Figure 5.The described processing was conducted using ImageJ software.Based on prepared binary images, filling percentage can be calculated as a ratio of area covered by white pixels to total area of the image.

Comparison between µCT and UT
Among all the inspected samples, five were selected to present direct comparison of brazed joints resulting from UT and µCT measurements.The comparison takes into account two cases when filling imperfections are almost negligible and when the brazed joint contains large defects.
Examples of UT C-scans obtained for the amplitude of echoes at the brazed interface as well as an unrolled brazing surface obtained from µCT are shown in Figures 6 -8, the colour scale of Figure 6 applies to all C-scans.The colours of the C-scan map correspond to ultrasound reflection amplitude (as percentage of screen height at the chosen gain level) and percentage of total area with amplitude over 50%, respectively.All amplitudes above 50% (red and yellow colours) are counted as filling imperfection.Presented images show good agreement between brazed joints derived from µCT and UT scans.Large defects are represented by both techniques with comparable accuracy; only in the case of sample 2 there is a more noticeable mismatch because of the large defect located in the middle of the image.At the same time, it has to be admitted that smaller imperfections are much better represented, thanks to high resolution measurements, by µCT technique.Defects smaller than 50 mm in diameter are missed in UT scans due to limited frequency of ultrasound.
This difference, nevertheless, does not affect in a great manner estimation of braze filling percentage, as revealed by the following quantitative analysis performed based on the presented images.
Table 1 presents filling percentage obtained from both techniques as well as associated highest quality level within EN ISO 18279.The differences between obtained results are usually less than few percent.This is the reason why µCT and UT give the same quality level -B to all the samples inspected.
In this context, it should be noted that area estimation using -6dB drop method has inherent error due to variables which can affect the amplitude of any received ultrasound reflection, including the reference reflection, such as quality of the entry and inside reflecting surface.In practice, manual variation of the reference amplitude level and reflections received from the brazed interface can alter the percentage filling estimation by around +/-5%.At the same time, the highest difference in filling percentage, reported for sample 11, is equal to 6%.

Conclusion
Several brazed joints between copper tubes and stainless steel pipe fittings were analysed using immersion ultrasonic and X-ray Computed Tomography measurements with an aim to compare both techniques in terms of estimation of filling percentage of braze metal.For this reason, special image processing methods have been developed and applied to µCT 3D data in order to unfold brazed interfaces, extract filling imperfections and calculate filling percentage.As it can be seen based on figures presented in this work, the proposed algorithms offer high accuracy of segmentation process which, in turn, allows detailed qualitative and quantitative characterization of the braze quality.
Comparison of calculated filling percentages (see Table 1) reveals that both techniques provide very similar results, especially one takes into account allowable tolerances.It is, therefore, concluded that they can be used interchangeably for nondestructive evaluation of brazing quality, depending on particular size and geometry of a brazed joint.
In addition, it has been shown how high resolution data obtained from µCT can shed more light on smaller filling imperfections that may be missed by ultrasonic inspection.

Figure 1 :
Figure 1: Example pictures of an inspected sample (on the left) and R2E power converter [6] (on the right)

Figure 2 :
Figure 2: One of the inspected samples placed inside tomograph chamber in front of X-ray tube

Figure 3 :
Figure 3: Visualisation of example µCT volume obtained using 2D cross-sections: top (a) and right (b), 3D model (c) and 2D unrolled surface representing the brazed interface (d)

Figure 4 :Figure 5 :
Figure 4: Example of an application of the first developed image processing procedure which is based on local thresholding

Figure 6 :Figure 7 :
Figure 6: Two representations of a same brazed joint containing one large defect.C-scan from UT inspection (above) with colour scale (right).Unrolled surface by µCT (below)

Figure 8 :
Figure 8: Brazing surface area containing small defects obtained from UT (first row) and µCT (next two rows)

Table 1 :
Filling percentage obtained from µCT and UT measurements