The phased array inspection of friction stir welded aluminium plant
C R Bird and D Kleiner
Proceedings of OMAE 2004: 23rd International Conference on Offshore Mechanics and Arctic Engineering, Vancouver, Canada 20-25 June 2004.
Friction stir welding (FSW) is a relatively new welding process compared to electron beam or arc welding. Unlike most other welding processes there is no liquid state for the weld pool. For this reason the potential defect types present within the weld are quite different from conventional welding defects such as voids or lack of fusion. These can still be present, but defects such as slag or hot tearing due weld pool shrinkage cannot. Other defects more akinto those associated with resistance welding (joint line remnants) are more likely and can be more serious. TWI has run and taken part in a number of projects including the European 'Qualistir' project lead by R/D-tech. The object of these projects was to develop a reliable inspection method for determining the quality of FSW in a butt and lap welded configurations. This paper describes the novel method developed by TWI for the detection of the joint line remnant defects. The paper describes the use of back scattered noise analysis for determining whether the weld has been correctly forged and whether the metallurgical structure indicates a good weld. Further to the difficult FSW specific flaws the paper also describes how the inspection procedure detects the more conventional defects.
Friction stir welding (FSW) is a relatively new welding process compared to electron beam or arc welding. Unlike most other welding processes there is no liquid state for the weld pool. For this reason the potential defect types present within the weld are quite different. Conventional welding defects such as voids or lack of fusion can still be present, but defects such as slag or hot tearing due weld pool shrinkage cannot. But other defects more akin to those associated with resistance welding can be present.
As for all welding processes, if incorrectly designed or controlled, welding defects can be generated. As FSW is a machine/automated process, once the process has been designed and tested defect production should not occur. However, like all processes, process control can become out of limits or unforeseen circumstances can affect the quality of the weld. For this reason, good NDT to aid the quality control process is required.
This paper concentrates on the NDT aspects of the 'Qualistir' TM project and TWI exploratory work for the on-line quality control of FSW in Aluminium [1,2] . Qualistir was a collaborative project led by R/D-tech with co-industrial partners Vermont, Isotest, Neos Robotics and Gatwick Fusion, and research partners TWI, GKSS and TUS. The object of the project was to develop an on-linemethod for determining the quality of FSW in a butt-welded configuration. The industrial interest in Qualistir guided the project towards materials and geometries used in the aircraft industry and subsequent work has broadened the TWI knowledge into other areas of industry including shipping and rail. From this knowledge base it is believed that the basic inspection technology could be adapted for aluminium pipelines constructed by FSW. The techniques described in this paper detect conventional defects and provides a method for determining whether the weld has been correctly forged.
Friction stir welding
Friction stir welding (FSW) is a solid-state bonding technique, which uses the heat generated from a milling-type tool to plasticise and bond two sheets of metal ( Fig.1). This welding technique has many advantages. In particular, no melting means less contamination - this particularly applies to aluminium, where oxides are readily generated.
Fig.1. Illustration of friction steer weld process
Like many welding processes, FSW can be incorrectly performed resulting in defects. The defect that is particular to FSW is a Joint Line Remnant (JLR), which is defined below.
Some flaws may be acceptable to design codes in which case they could be termed an acceptable flaw. Others, depending upon size and severity, may be unacceptable, in which case the flaw could be termed a defect. Figure 2 below identifies four conventional flaws. Figure 3 below identifies a JLR. The following flaw definitions are used in this paper.
Fig.2. Macro-section through FSW containing voids, faying defects and lack of penetration
- Void. This is a flaw that is totally subsurface, volumetric (not planar) and contains no material. These flaws are usually aligned with the welding direction.
- Worm hole. This is a void which is aligned in the through wall direction.
- Lack of penetration (LOP). This flaw is present where the full thickness of the weld has not been forged leaving the original parent plate butting surfaces unbroken or undisturbed. This type of flaw usually has a small but finite gap between the adjacent parent plate surfaces and is primarily caused by the FSW pin not penetrating deeply enough into the weld.
- Faying surface flaw. This flaw is surface breaking. The flaw can contain oxide and is metallurgically similar to that of a rolling lap.
- Joint Line Remnant (JLR). This flaw is the most difficult to detect and has been called a 'kissing bond or entrapped oxide'. This flaw emanates from an incorrectly broken and stirred fusion face, leaving a semi-continuous layer of oxide in a plane parallel to the weld. This flaw is fully bonded in that there is no air between adjacent surfaces and it provides some mechanical strength. The severity of this flaw depends upon its planar extent and proximity of the adjacent oxide particles. These flaws have also been termed lazy S flaws. By studying the microstructure of the sample shown in Fig.3 it can be seen that the oxide layer slopes, hence some forging has taken place. This compares with the weld in Fig.2 where the defect is vertical. Furthermore, by studying the sample at high magnification, it can be seen that the oxide is not continuous and that some of the grains cross the apparent line of the flaw. There is little disruption to direct passage of sound or electricity through the flaw and there is negligible foreign matter of a lower density that would enable X-ray detection. For this reason, direct detection with any NDT method is extremely difficult.
Fig.3. Weld T7 macro showing weld nugget structure (top) and coarse grain structure of the weld root (bottom). It can be observed that the pin penetration was ~0.8mm from the bottom surface during the process (leading to the creation an entrapped oxide defect)
From previous published work and discussions with the aircraft industry it was clear that conventional flaws, e.g. voids and lack of penetration, could be detected by current ultrasonic and radiographic methods. The flaws that were evading reliable direct detection were joint line remnants (JLR). For this reason the NDT development concentrated on the detection of JLR and a method for in-line quality control. Other work in particular work performed by DLR  , shows that, using high frequency (20MHz) focused probes, it is possible to detect very small (<0.5mm) LOP/JLR defects in the weld roots.
Because the work of TWI concentrates on industrial application of NDT methods the emphasis of the development was on the rapid inspection of FSW. For this reason phased array ultrasonic inspection was believed to be a suitable method of applying focused ultrasonic beams to the components. Before phased array technology was applied, conventional focused probes were used for a number of technical and economic reasons. These conventional probe trials provided the scanning patterns and probe frequencies required for the application which was then converted to the phased array system.
To develop inspection methods, defects of a known size are required. Development of the process to generate controlled JLR was difficult, requiring a large number of destructive measurements following the controlled use of incorrect welding parameters. After many trials, a process was developed, using a specifically designed welding tool, which generated perfect joints when operated within the correct welding parameters and controlled JLR with small parameter changes.
Early development work and choice of inspection system
It was desirable to use phased array technology because of its inspection speed but, at the time of the Qualistir project phased array probes and instruments were limited to a maximum inspection frequency of about 15MHz. This has now been extended to instruments that are effective up to about 25MHz. In addition, phased array probes were relatively expensive and the availability of high frequency phased array probes was limited.
From early experiments it was clear that direct ultrasonic detection of these defects by back-reflected energy could not be achieved reliably. On some occasions small signals were detected in the weld roots but it was not clear whether these were as a result of the LOP, JLR or general material noise. These early trials included inspection with focused 10MHZ to 30MHz immersion probes working at very high inspection sensitivities.
At a fundamental level, metallurgical properties affect ultrasonic transmission. Because the material bonding is weaker in the region of a flaw and contains foreign matter, it was decided to investigate other ultrasonic properties, including frequency filtering and velocity changes, between the parent plate, weld nugget and weld root region.
This project's early development work used conventional focused immersion probes. Once the basic inspection concept had been developed, the technique was transferred to that of phased array probes, which benefit from increased in-line inspection speed. The phased array probe arrangement and scanning pattern is illustrated in Fig.4.
Fig.4. Illustration of phased array scanning pattern
The development results are divided into four categories:
- Conventional defect detection
- Ultrasonic velocity measurements
- Ultrasonic frequency measurements
- Ultrasonic noise distribution measurements
Conventional flaw detection
Although this paper concentrates on JLR, the development programme also collected data on the more conventional but equally important flaws, as detailed earlier. Further TWI understands that some amount of JLR may be acceptable where a component is under static (no fatigue) conditions or where the JLR does not lie in a crack propagation direction.
This project used 6.25mm thick aluminium 7075. GKSS deliberately generated a number of welds containing conventional defects including voids, lack of penetration and faying defects. Figure 5 shows the micro-section of a weld containing voids and tight LOP. It further shows the clear difference between the weld nugget and the parent plate. Further it can be seen that the weld root has a similar undisturbed structure to that of the parent plate.
Fig.5. Micro-section through conventional weld defects
Figure 6 shows an ultrasonic sectional image generated from the inspection of the GKSS weld shown in Fig.5. This image, generated by Tomoview software, shows the two voids and the weak signal from the LOP. High signal amplitudes are represented by red and low by pale blue.
Fig.6. Ultrasonic data from sectional view of GKSS weld
In this weld the LOP was under compressive stress, allowing a high percentage of the energy to transmit through the flaw. The horizontal dark blue line represents the top of the plate and the horizontal red line the bottom of the plate. The dashed line surrounds the weld nugget region. This region is lighter blue (lower signal amplitude) than the parent plate. The super-imposed red box is discussed later.
Velocity measurements were made, both shear and longitudinal (S & L), through both the parent plate and the weld region. The direction and mode of these measurements are shown in Fig.7. Neither TUS's nor TWI's measurements revealed any measurable velocity variations due to the presence of entrapped oxide, or a significant change in velocity due to the weld. Furthermore, the measurements of the horizontal longitudinal (L2) velocity at several weld depth positions revealed no significant velocity variations.
Fig.7. Velocity and FFT scanning directions
Ultrasonic frequency measurements
Data for all the welds was collected in both rectified and un-rectified forms. The un-rectified data was digitised at 100MHz and used for Fast Fourier Transform (FFT) analysis on the back-scattered energy. Furthermore, TUS performed through-transmission scans in both the L2 and S directions. Figure 7 provides an example of the RF data collected on a good weld. From both the TUS and TWI results it was clear that the forging of the weld nugget had refined the grain structure to such an extent that it had become highly transparent to ultrasonic frequencies up to 20MHz. This, in turn, provided little back-scattered energy or filtering of the energy by the grain structure. In contrast, the parent plate provided interference with the transmission of the ultrasound. For the two alloys examined, the mean back-scattered frequencies from the parent plate was approximately 8MHz. The 8MHz energy was also present in the root of incorrectly forged welds. Frequency analysis was investigated as a means of determining whether defectiveness could be determined in this way. To provide a frequency spectrum of a digitised waveform a sample equal to the full pulse length is required. The pulse length of the probes used was approximately two wave lengths requiring a digitised length of 0.6mm. For this reason, the FFT approach was not taken forward but again the data showed this very strong contrast between correctly forged and incorrectly forged weld roots, emphasising that UT could, in principle, be used as a measure of grain refinement. An example of the RF frequency spectrum is shown in Fig.8. Although the relationship between back scattered energy and grain size is well known it is rarely used as a quality control method.
Fig.8. Screen dump of RF data collection showing noise distribution and FFT analysis
Noise distribution results and analysis
During the data collection, it became clear that there was a clear noise pattern associated with the FSW nugget, as shown in Fig.9.
Fig.9. Sectional image through Weld T4 with a joint line remnant
The pale blue zone shows the grain scatter noise from the weld nugget, whereas the darker blue zones correspond to the parent plate. In this sample the low noise zone does not extend to the full depth of the parent plate. Additionally a high noise zone can be seen corresponding to the Thermo-Mechanically Affected Zone (TMAZ). The TMAZ generates ultrasonic back-scatter noise from the relatively large, vertically orientated grains in this zone. These relatively high amplitude signals can be above the normal aircraft industry reporting threshold, causing false defect calls.
From consultation within TWI and with other welding engineers it was determined that, where the weld nugget is correctly forged through to the weld root, JLR defects should not be present but the oxides will be present within the volume of the nugget. Hence, if a method could be developed for determining the depth of the correctly stirred zone, a quality control method could be provided to ensure that the welding process was in control ensuring a low probability of JLR.
Following this early idea, a programme of inspecting many welds containing JLR, plus control welds containing no defects, was undertaken to determine whether noise could reliably determine the extent of correct root forging. To provide reliable evidence, multiple weld manufacturers and materials were required to determine whether the ultrasonic properties were stable with respect to the welding process. TWI, GKSS and Alenia all manufactured welds for this project. GKSS and TWI used aluminium alloy 7075 and Alenia used 2219. Further to this, TWI has performed a self-funded project on Lap welds to see whether the concept can be applied to other alloys and geometries.
The design of the weld tool is critical to the weld quality and the grain structure within both the weld nugget and the TMAZ. All three manufacturers used different tool designs providing the required process stability test.
Figure 10 shows a macro-section through a weld containing a 0.4mm deep JLR. Figure 9 shows the clear contrast in back-scattered noise between the weld nugget and the parent plate. Furthermore, a high noise level can be seen in the weld root for a distance of approximately 0.4mm from the bottom of the plate. Figure 11 provides a C-scan (plan view) of a weld in aluminium 2219 with JLR. This linear scan was performed with a 10MHz phased array probe.
Fig.10. Macrograph of weld t4 containing entrapped oxide
Fig.11. C-Scan view of weld and parent plate at a depth of 2mm to 3mm from the weld root
To provide a quantitative inspection method, a stable measure of the mean noise level within the weld root was required. Initially the ratio of weld root noise to weld nugget noise was studied and reported  . From further study, the ratio of the parent plate noise to the weld root noise was chosen as the most stable measure, this ratio normalising all the inspection parameters to that of the parent plate material noise. The statistical results from the Qualistir trials are discussed in Ref.  showing a stable method for both 7075 and 2219 butt welds. To apply this basic method to a production situation a semi automated quantitative method was required for the sentencing of the data.
Figure 9 provides a sectional view through the ultrasonic data showing two boxes, M1 and M2, these illustrate where the ultrasonic data was extracted for analysis. Figure 11 illustrates the plan view of the data extraction boxes in yellow. The data within these boxes is analysed for back-scatter noise amplitude. FSW produces relatively large changes in microstructure with respect to axial position along the weld further the parent plate showed ultrasonic back scatter noise variations with respect to plate axial position. The plate grain structure variations are caused by the change in either roller, die or material temperature when initially rolled or extruded. The signal amplitude changes due to the JLR are small hence the variation in material properties with respect to axial position prevented a simple threshold being placed on the weld root noise or even a single sample point for the parent plate noise. Because FSW is a machine process if an out of control welding parameter develops, it is likely to be relatively long (>5mm). This feature provided the opportunity to perform statistical processing to stabilise the small variations in noise ratio due to the local grain structure changes providing a stable measurement of quality. An algorithm was developed which imports 3D volumes of data out of the raw inspection data from the specified areas within the weld and parent plate. It then performs a volumetric moving average type function to smooth out the individual noise amplitude values. This method also minimises the effect of spurious indications in the scan data, which generally appear singularly and not in clusters. The algorithm then compares the manipulated noise data of the weld root with that of the parent plate. The ratio of the two isa quantitative indicator of the weld quality.
Figure 12 shows a graph of the back-scattered noise data collected from a weld with no JLR or natural defects. The yellow line provides the moving average ultrasonic back scattered amplitude of the weld root, the red line the parent plate and the black line the noise ratio/ weld quality number. It can be seen that the noise ratio achieved in the weld root, was about 0.5. A noise ratio of 0.5 means that the noise of the weld root is half that of the parent plate, whereas at 230mm along the weld where a 0.2mm spark eroded notch had been machined the noise ratio is 1.2. A noise ration of 1 indicates no root forging i.e. the weld root has the same grain structure as the parent plate, and the likely presence of JLR or LOP. Figure 13 shows a back scattered noise graph from a weld with 1.0mm JLR but with no LOP. Here the noise ratio lies almost entirely over the value of 1, detecting a high noise level at the weld root and indicating an incorrectly forged nugget.
Fig.12. Noise ratio graph of FSW containing a 0.2mm spark eroded notch
The development work has concentrated on an in-line method for providing data as part of the overall quality control process for FSW The developed inspection and data analysis method can clearly and precisely detect conventional defects, e.g. voids and LOP, with a through wall height of 0.1mm with a good signal-to-noise ratio. Conventional data analysis uses signal amplitude, as a defect discriminator. FSW can, if not correctly controlled, have serious defects with signal amplitudes smaller than that of the TMAZ.
Fig.13. Noise ratio graph of FSW containing a 1.0mm joint line remnant
Application of developed method
The developed inspection method requires inspection frequencies in the region of 10MHz to 15MHz. At this frequency, water is the only practical medium for providing coupling between the ultrasonic probe and the component. To provide water coupling, a scanning head Fig.14, was developed which was applied to both a flat bed FSW machine at TWI and a robotic arm at GKSS. This head provides a mobile water bath for the phased array probes without spilling the water onto the FSW machine. Figure 15 shows the scanning head about to scan a curved weld at GKSS. The water is supplied to an inner chamber and removed from an outer chamber by a vacuum system. The water is then continuously recycled through the scanning head, pumps and vacuum chamber.
Fig.14. Photograph of developed probe pan
Fig.15. Phased array inspection head on curved weld at GKSS
Demonstration of inspection system
The system was demonstrated both at TWI and at GKSS in both cases the welds contained a mixture of good regions and deliberate defects. Figure 16 shows the noise ratio graph of the flat bed weld generated for this demonstration. This weld had four zones; the first zone from 0 to 200mm was a fully penetrated region with no JLR. This zone has a noise ratio of0.4 to 0.5; the second zone from 200mm to 450mm has a JLR 0.7mm deep with a noise ratio of 0.6 to 0.8; the third zone again was a good region; the fourth zone again had a JLR/LOP to a depth of 1mm with the noise ratio of 0.7 to 1.0.Themacrographs of regions one and two are shown in Fig.17 and 18 and plan view of the ultrasonic data is shown in Fig.19. The ultrasonic image is from the root of the weld and clearly shows the four weld zones.
Fig.16. Noise ratio of flat bed demonstration weld
Fig.17. Macrograph of zone 1 of flat bed weld
Fig.18. Macrograph of partially penetrated weld containing a joint line remnant
Fig.19. Pan view of ultrasonic data showing the four weld zones and the differing noise levels
The second demonstration was at GKSS. Again the weld was a butt weld in Al 7075 but this time the weld was curved and the NDT head was scanned over the weld by a robot arm. The noise ratio graph is given in Fig.20 and a plan view of the ultrasonic data is given in Fig.21. From the graph it can be seen that the beginning and end of the weld has noise ratio exceeding 0.7 whilst the central portion has a noise ratio of about 0.5. The central portion of the weld was defect free and the ends had a mixture of defects. Where the noise ratio exceeded 1.2 the welds contained LOP and lack of fusion defects due to tool miss-alignment.
Fig.20. Graph of noise ratio with respect to axial position for GKSS weld showing defective areas at the beginning and end of weld
Fig.21. Plan view of GKSS demonstration weld containing lack of fusion and lack of penetration defects
The inspection technique is flexible and can be modified and optimised for different series of aluminium and different weld geometries. TWI is currently investigating 2000, 5000, 6000 and 7000 Al alloy series FSW butt, lap, T and box weld configurations. The technique is applicable to other aluminium components such as oil pipe lines and drilling risers. Different alloys and configurations will require a considerable amount of technique development and validation.
On thin awall components, the welds are subject to small amounts of distortion after welding which makes ultrasonic tracking of the weld nugget difficult and could lead to false calls. TWI is developing software for FSW lap welds to track the data analysis box with respect to the true position of the weld this will enable an accurate analysis of data from small volumes.
- An ultrasonic method has been developed for the reliable determination of forging depth within a friction stir weld.
- The project has developed a phased array method for on-line quality control of butt welded friction stir welds.
- It is believed that the concept of determining degree of forging in addition to direct defect detection provides a unique quality control method.
- This basic method is now being developed for other weld alloys and weld configurations
- The EC for providing 50% of the Qualistir development funds.
- Business partners in the Qualistir project R/D-tech, Vermont, GKSS, Gatwick Fusion, Neos Robotics and TUS.
- AleniaSpacio for providing valuable weld samples.
- EC project, 'Qualistir' TM , Project No: CRAF-1999-70641.
- TMS conference proceedings 2003, New Developments of the Ultrasound Phased Array for the Evaluation of Friction Stir Welds, Colin R Bird TWI, Olivier Dupuis R/D-tech, Andre Lamarre R/D-tech.
- Insight Journal, Vol. 46 No1, January 2004 Ultrasonic Phased Array Inspection Technology for the Evaluation of Friction Stir Welds, C R Bird, TWI.
- 'Non destructive inspection of aluminium alloy friction stir welds', Workshop on NDI of Friction Stir Weld, DLR, Cologne, December 2003.