Delphine S Caravaca, Colin R Bird, Kathryn Beamish and Stephen Maddox
Paper presented at 26th International Conference on Offshore Mechanics and Arctic Engineering, OMAE 2007, San Diego, California, 10-15 June 2007. Paper no. 29614.
Friction Stir Welding (FSW) of aluminium alloys is now an establish joining technique and there is increasing application of FSW to joining of critical components and structures. Much emphasis has been placed on optimising tool design and process parameters to ensure joint quality but flaws may still be created in the production environment if the limits of the 'process window' are exceeded. There is a requirement to understand the type of flaws that may be generated, and their causes, when welding conditions deviate from the optimum. This paper has taken a step forward in NDT knowledge. This paper presents the result of an integrated project correlating NDT performance against fatigue performance for AL 2024 T3 butt welds.
Friction Stir Welding (FSW) of aluminum alloys is now an established joining technique and there is increasing application of FSW to joining of critical components and structures. Much emphasis has been placed on optimising tool design and process parameters to ensure joint quality. However flaws may still be created in the production environment if the limits of the 'process window' are exceeded. As FSW is a machine/automated process, once the process has been designed and tested, defect production should not occur but, 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 is required for the detection and characterisation of production flaws. This paper is based upon a project correlating NDT Data against fatigue performance of FSW joints. A method for determining whether the weld has been correctly forged has been established and the correlation between the defect criteria level and the fatigue life has been demonstrated.
Friction stir welding
Friction Stir Welding is a solid state joining process, which produces a fully consolidated weld (Thomas et al 1991). Welds are made by plunging a rotating cylindrical tool into the joint to be welded. The heat generated by the friction is sufficient to locally soften the workpieces and the rotating motion of the tool stirs the workpieces together as the tool is advanced along the joint line.
The technique has many advantages. The process can be fully mechanised, allowing the production of welds of a consistently high quality. There are also practical advantages such as low shrinkage and distortion, and no porosity.(Kallee 2000) 
All welds were made using TWI's ESAB SuperStir friction stir welding machine. The tool probe was machined to be 0.15mm shorter than the total material thickness and the machine spindle was tilted 1.5° away from the direction of weld traverse. The weld was made at 350rpm, 210mm/min.
Weld flaws definitions
There is often confusion between terminology such as flaws or defects. To avoid misunderstandings this paper defines these differences.
A defect is defined as an imperfection that has been shown to compromise the integrity of the structure and its presence is therefore intolerable. A flaw is defined as an imperfection whose significance has not been established, and which could be possibly tolerated in the structure. Thus all the imperfections generated for the present study will be termed flaws. There are two categories of flaws in friction stir welds; volumetric flaws and joint line flaws(Leonard 2001)  .
Volumetric flaws (Voids)
Imperfections in a friction stir weld due to a lack of material are termed volumetric flaws or voids. Voids may be caused by inadequate material flow either due to tool features or selection of an excessive welding speed. Voids may also be caused by inadequate consolidation of the softened material due to a reduced forging pressure, inadequate material clamping (plates separate during the welding) or the presence of a significant gap along the joint line because of poor fit up.
Joint Line Flaws (JLF)
In friction stir welding, the original joint line can still be traced and it is referred to as a joint line remnant. Detailed examination shows that it consists of oxide particles delineating the original joint line. Thus, the presence of a joint line remnant should not be necessarily considered as flaw.
The most serious defect associated with the joint line remnant is those located at the weld root. The extreme condition is a lack of bond caused by a lack of penetration, which could provide by a shortened pin or by a poor control of tool position or force.
The most difficult flaw to quantify is a region of weakly bonded material in the root of the weld. Such regions can exist at the end of root flaws, but will always follow the path of the joint line remnant. They are very difficul tto detect non destructively since a bond exists.
From previous published work (Bird C)  and discussions with the aircraft industry it was clear that conventional flaws, for example voids and lack of bond, could be detected by current ultrasonic method. It is known that the vast majority of FSW joints are expected to be free defects but it is not always possible to assume that they are completely flaw-free. To improve the confidence in the design, manufacture and application of FSW joints, manufacturers are seeking data on the properties of welds containing flaws and require validated inspection techniques to detect those flaws on-line, after manufacture and in-service.
To establish a correlation between mechanical properties and NDE performance, a number of welded plates were produced. The project used 4mm thick aluminium 2024-T3. This project generated a number of defect types but this paper concentrates on the correlation of NDE results and fatigue life, and presents the results for the Joint Line Flaws (JLF) welds:
Plate 1: good control weld as shown in Fig.1, where the weld root is fully penetrated.
Plate 2: 0.3mm shortened pin.
Plate 3: 0.5mm shortened pin as shown in Fig.2. It can be seen that on the picture that the weld root has a similar undisturbed structure to that of the parent plate.
Fig.1. Picture of the weld nugget structure from a good weld
Fig.2. Picture of a joint line flaws showing an open crack generated by a 0.5mm shortened pin
It should be noted that the dimensions of the flaw produced by this approach will not necessarily correlate exactly with the length of the pin reduction, and may vary along the length of the weld within machine positional accuracy.
As reported previously (Bird) [4,5] it was clear that ultrasonic phased array inspection is a relevant method to inspect FSW.
Metallurgical properties affect ultrasonic transmission. The forging of the metal by the FSW tool refines and re-orientates the grain structure, resulting in reduced back-scattered amplitude from the grain structure of the weld nugget as compared to that of the parent metal.
The inspection technique is designed to detect both conventional flaws (volumetric and joint line) and the presence of the specific FSW flaw: joint line remnant. Although joint line remnant flaws are commonly so tight that they cannot be detected directly (unless they are so severe that they could be classified as lack of bond), weld quality can be assessed on the grounds of ultrasonic grain noise. The average value of the grain noise is calculated in both weld nugget and parent plate and compared. This comparison is used to measure the quality of the weld root region.
The phased array inspection was performed from the weld cap side of the plate. The characteristic grain structure difference between the parent material and the weld nugget was best provided with a 15MHz linear array probe using abeam angle of 70?, with 32 active elements having 0.2mm pitch. To achieve adequate ultrasonic coupling of the probe, the surface finish of the component was required to be good and water was required as an ultrasonic couplant. A schematic illustration of the ultrasonic phased array probe is shown in Fig.3.
Fig.3. Schematic illustration of ultrasonic phased array probe arrangement
The scans are displayed as a combination of images in the TomoView software. The forging of the metal by the FSW tool refines and re-orientates the grain structure, resulting in reduced back-scattered amplitude (or 'noise') from the grain structure of the weld nugget as compared to that of the parent material.
To provide a stable assessment of the root area of the weld, noise amplitude measurements was produced. By normalising the noise measurement against the noise measure in the parent material. An illustration of the data sample positions is shown in Fig.4.
Fig.4. Material noise level sampling areas
The through wall placement of the sample box is a compromise between detection of unwanted sample surface noise and detection of small flaws. As illustrated in Fig.4, the indication from a small surface scratch provides a signal to a depth of about 0.5mm from the bottom surface of the plate. This signal does not indicate that the scratch was 0.5mm deep, in fact it was <0.1mm. The arc signal is a measure of the UT beam width. For this reason, a compromise has to be reached between detection of welding flaws and detection of natural marks on the surface of the component. In this case the sample box was placed ata distance of 0.5mm from the bottom of the plate to minimise the signals from the plate surface but capture the signal from a flaw.
When the weld has been properly manufactured, the ultrasonic level of the weld root should be lower than that of the parent material. If the weld root has not been fully forged, then the noise level will be equal to that of the parent plate and if there is an open conventional defect such as lack of bond, the noise level will be higher than that of the parent material. By comparing the level inside the root to that of the parent material, the operator has a powerful tool for estimating the pin penetration and the likelihood of a flawed weld.
Ultrasonic inspection results
A sectional view and a plan view of the ultrasonic data are given in Figs 5a to 5c, where the black dashed line represents the position of the weld nugget.
Fig.5. Phased array sectional and plan view
: a) Good weld
; b) 0.3mm shortened pin; c) 0.5mm shortened pin
Table 1 gives the detailed noise analysis results for each plate at three positions along the weld (50, 100 and 150mm from scan start). In addition to the noise ratio measurements, the average and range is given for the weldroot of each weld.
Table 1. Ultrasonic data analysis, signal amplitude measurements
| ||Plate 1||Plate 2||Plate 3|
The ultrasonic data presented in Fig.5a shows no flaw indications. The weld root is fully penetrated on the full length of the plate. The average noise ratio for this plate is 0.28 with a range of 0.02. This analysis shows that the recorded grain noise in the weld root was constant and approximately one quarter of that of the parent material which indicates a highly grain-refined weld root. Furthermore, the range indicates that the welding quality was very consistent.
Figure 5b shows a sectional view where the weld appears to be fully forged and a further section where there is a flaw signal in the middle of the weld root. The plan view shows intermittent indications along the center line of the weld.
The average noise ratio for this weld is 0.63 with a range of 0.42. The noise analysis shows that the noise is approximately twice that of plate 1 with a large variation (0.42) of noise ratio with respect to axial position. Additionally where there is a point indication in the plan view, the sectional view shows a not fully forged weld with a darker zone in the root than for a good weld. Hence in addition to the positive indication there is evidence thatthe weld is not fully forged.
Figure 5c shows the ultrasonic image of the weld where a flaw appears in the weld root for the majority of the inspected FSW length. The average noise ratio for this weld is 4.0 with a range of 1.73. This defect is analysed by conventional data analysis as lack of bond and is highlighted on the plan view. The noise ratio measurement indicates that the noise in the weld is higher than that of the parent plate and for this reason there must be an open ultrasonic reflector. This result is consistent with the macro section shown in Figure 1 where a lack of bond flaw was visible in the root. The large range of noise ratio readings indicates that the flaw size varies considerably with axial position, this is also clear from a comparison of the two sectional views in Figure 5c.
Endurance fatigue testing
Fatigue test specimens were designed with the weld at the centre and oriented in the transverse direction with respect to the direction of loading and a minimum of 5 tensile tests were performed per weld. As noted previously, the weld cap were machined flush before extracting the specimens to avoid a life reduction due to the surface roughness.
Endurance fatigue test were conducted in a 100kN capacity Amsler Vibraphone testing machine. The tests were carried out under constant amplitude axial loading, in air and at room temperature. More details about the tests including the applied stress ranges, the cycling frequency and the applied stress ratio are given in Table 2 R= σmin / σmax , where σ min is the minimum and σ max is the maximum stress in each cycle). Stress ranges were chosen to give fatigue lives in the range 104 to 107 cycles, although in the event none of the specimens that failed from flaws gave lives above 106 cycles. The specimens were tested to complete failure or the occurrence of a through-thickness crack. A run-out endurance of around 107 cycles was adopted, at which point testing was stopped if the specimen showed no signs of fatigue cracking. Some such specimens were subsequently re-tested at higher stress ranges to increase the available database.
Table 2. Fatigue tests parameters implemented in fatigue testing of plate 1 to 3
stress range, Mpa
σmin / σmax
Fatigue test results
The fatigue tests results are presented in the Fig. 6 below. The majority of specimens containing flaws failed from a flaw on the weld center line. However, specimens from the flaw-free plate and some specimens from Plate 2 failed in the parent plate, away from the weld.
Nominally Flaw-Free weld, Plate 1
These specimens were tested at stress ranges between 100 and 200Mpa. Only three of the six specimens failed in the weld. Two of the other failed from the edge of the specimen in the parent plate away from the weld, while the third failed where the specimen was gripped in the wedge jaws of the testing machine. There are too few results to establish statistical limits and therefore a scatter band bounding the results has been estimated by eye.
Fig.6. Fatigue results from Plate 1, 2 and 3 specimens
0.3 Shortened Pin, Plate 2
Apart from one case of parent metal failure, all the specimens from this plate failed in the weld. The welding method produced definite evidence of Joint Line Flaws specially lack of bond therefore it is not surprising to find that fatigue cracking always initiated at such flaws on the weld root side. However the fatigue test results are compared with the reference scatterband for flaw free and a large variation in fatigue life can be observed. This large variation is thought to be a reflection of the intermittent nature of the flaws. The presence of the flaws has reduced the fatigue performance of the welded joints considerably, all the failures being obtained at stress ranges below the apparent fatigue limit of the flaw free weld. Compared with the reference scatterband, the reduction in fatigue strength ranges from about 30 to 40%. It is noticed that for the same stress range 75Mpa, the endurance varies greatly showing a highly variable flaw depths. This result correlates well with the NDE results as shown in the plan view Fig.5b, and with the weld ratio analysis which provides a large range.
0.5 Shortened Pin, Plate 3
The stress ranges for test on this welded joint were chosen on the basis of both the intended flaw size and the NDT results, which reported near uniform lack of bond. Thus, four specimens were tested at 50Mpa and one at 100Mpa for direct comparison with the Plate 2.
The fatigue test results are compared with the reference scatterband for flaw-free weld in Fig.6. Due to the near uniform Lack of bond of 0.5mm, all the welds failed from the flaw at the weld centerline. As it can be seen in Fig.6, there was very little scatter in the fatigue lives obtained from the four specimens tested at 50 MPa. It shows that the defect was constant all along the weld and confirms the consistently high noise ratio results and the large visible flaw in the plan view in Fig.5c.
To assist the discussion, a very simple summary of the results is presented in the Table 3 below. This table shows that the lowest average noise ratio provides the highest weld quality whereas higher average noise ratio provides lower fatigue strength.
Table 3. Summary of the NDT and fatigue test results
of flaw-free weld
||Equivalent to parent material
||Up to 0.3mm lack of penetration
||0.3 - 0.5mm lack of penetration
Plate 2 It can be stated that the flaws was produced using a FSW pin shortened by 0.3mm. It is fairly clear that the ultrasonic inspection method has the ability to detect the variation of a flaw size in FSW joint. Presence of a flaw is shown by an increase in the weld root noise, which ranges from 1. 5 to almost 3 times that of the good weld. But it is a closed flaw as the noise ratio is less than 1.0. The ultrasonic results agree with the fatigue results, both show a large scatter in the result obtained from the specimens tested.
The flaw size variation has been further confirmed by the analysis of the fracture faces from two specimens shown in Fig.7a, which show that the original flaw varies from 0.198 to 0.390mm.
Fig.7. Picture of the macro fracture face of the defected weld a) 0.3mm shortened pin (plate 2) b) 0.5mm shortened pin (plate 3)
Figure 7b shows a constant lack of bond about 0.5mm at the root of the weld, which was consistently detected with a high noise ratio by the ultrasonic phased array method. The result correlate with the fatigue test where a very little scatter in the fatigue lives was obtained from the four specimens tested at 50 Mpa, showing that the defect was constant all along the weld.
The results demonstrated a strong relation ship between the NDT results and the fatigue performance for the AL 2024 T3 FSW joints. The NDT results allow a quantitative assessment to be made about the stirring quality of the weld, inturn, the possible presence of joint line remnant.
The NDT method is able to discriminate between a correctly forged weld nugget and the parent material.
The NDT method is fully able to detect joint line remnants down to 0.2mm in height, flaws that cause a reduction in fatigue performance.
NDT method can demonstrate the variance in the weld quality.
A method has been established to measure the quality of the weld even if there is no direct signal from a flaw.
- Nicholas E D and Kalee S W: 'Friction Strir welding- A decade on' IIW Asian Pacific International Congress, Sydney, Australia, October 2000.
- Leonard A J: 'Flaws in aluminium alloy friction stir welds', TWI Members Report 726/2001, July 2001.
- Bird C R, 'Ultrasonic Phased Array Inspection Technology for the Evaluation of Friction Stir Welds', Insight, 2003.
- Bird C R, 'Quality control of Friction stir welds by the application of non destructive testing', 4th International symposium of friction stir welding, Utah, USA, May 2003.
- Thomas W M, Nicholas E D, Needham J C, Murch M G, Temple-Smith P and Dawes C J:' Improvements relating to friction stir welding', European Patent Specification 0615480B1, December 1991.