Subscribe to our newsletter to receive the latest news and events from TWI:

Subscribe >
Skip to content

Girth Weld Fatigue in Full-Scale Pipes and Strip Specimens

   

Comparison of Fatigue of Girth-Welds in Full-Scale Pipes and Small-Scale Strip Specimens

Stephen J Maddox and Yan-Hui Zhang
TWI Ltd, Granta Park Great Abington, Cambridge, CB21 6AL, UK

Paper presented at Proceedings of the 27th International Conference on Offshore Mechanics and Arctic Engineering AE2008 June 15-20, 2008, Estoril, Portugal

Abstract

As part of a study of fatigue in girth-welded steel pipes, tests  were performed under constant amplitude loading on both full-scale pipes and strip specimens cut from such pipes. Significant differences were found in their high-cycle fatigue lives, which extended to around 108 cycles, and apparent fatigue endurance limits, the small-scale strips displaying superior fatigue properties. The reasons for this were investigated considering the fatigue crack initiation site, weld geometry, type of pipe, loading conditions, residual stresses, the re-testing of unfailed specimens and size effects. Fracture mechanics fatigue crack growth calculations were also performed using a K solution specially calculated by FEA for the girth weld. Conclusions are drawn about the suitability of strip fatigue test specimens for representing the fatigue behaviour of full-scale girth welded pipes and the scope for re-testing unfailed full-scale pipes.

1. Introduction

In the context of risers for deep-water oil and gas recovery, it is common practice to require fatigue tests to be performed on the girth welds to validate the fatigue strength assumed in design. Ideally, such tests should be carried out on full-scale girth welded pipes, the usual method involving the application of a rotating bending moment by resonance (1,2). However, in view of the expense and limited availability of suitable test facilities, an attractive alternative is the axial testing in standard fatigue testing machines of strips cut from girth welded pipes. Indeed, tests on such strip specimens may be the only practical option if, for example, it is required to carry out corrosion fatigue tests or tests under variable amplitude loading. Unfortunately, a potential problem is that strip specimens can give optimistic fatigue lives, especially in the high-cycle regime approaching the fatigue limit (3,4). Clearly, one explanation for this is that a strip specimen may not include the poorest, from the fatigue resistance viewpoint, part of the girth weld. However, another is that a difference in the level of tensile residual stress from welding is a significant factor. On this basis, it has been concluded that strips should be representative of full-scale pipes as long as they are fatigue tested under high tensile mean stress (3,5), to simulate the presence of the high tensile residual stress existing in the full-scale pipe.

A recent project dealing with the fatigue behaviour of girth welds included constant amplitude tests on full-scale pipes and strips cut from them. The emphasis was on their high-cycle fatigue performance and definition of their fatigue limits. Consequently, the results offer an ideal opportunity to re-examine the comparative fatigue performance of girth welds in large- and small-scale specimens and to check the validity of conclusions drawn to-date.

2. Experimental Details

2.1. Test Specimens

The fatigue tests were conducted on full-scale girth welded pipes (Fig.1) and strip specimens extracted from them (Fig.2). These were fabricated from 508mm outside diameter by 22mm wall thickness steel UOE pipe to API 5L-X65 specification. The welds were made from one side by GMAW, in the 5G position, with temporary copper backing. Each specimen was instrumented with strain gauges, as indicated in Fig.1 and 2, to determine the nominal stress close to the region of fatigue failure in the case of the full-scale specimens and also to enable any misalignment-induced secondary bending stress to be established in both specimen types.

2.2. Fatigue Testing

The 7.5 metre long full-scale specimens were fatigue tested in resonance at loading frequencies between 25 and 30Hz. The rigs subject the pipes to rotating bending moments so that each girth weld experiences alternating (tension-compression) loading. However, an axial tensile mean stress of 125N/mm2, achieved by pressurizing the pipes internally with water, was also applied. Thus, the stress ratio R, the minimum/maximum applied stress, increased with decrease in applied stress range.

The strip specimens were fatigue tested under axial loading in conventional fatigue testing machines at the same tensile mean stress of 125 N/mm2 and frequencies between 3 and 8Hz. Apart from establishing the basic S-N curve, there was particular interest in the very high-cycle fatigue of the girth welds, including definition of the fatigue limit. Thus, applied loading conditions were selected accordingly. In the event, some tests extended beyond 108 cycles.

2.3. Analysis of Fatigue Test Results

The fatigue test results were considered in terms of the local stress range close to the region of fatigue crack initiation that included any secondary bending stress due to joint misalignment. This was characterized in terms of the stress magnification factor km, defined at the ratio of the local to nominal stress. The strain measured on the outside of the full-scale pipe was used to calculate the nominal stress on the inside surface by assuming a linear stress distribution across the pipe wall thickness. In the strips, the nominal stress was the applied force divided by the cross-sectional area of the specimen.

In the case of the full-scale specimens, axial misalignment measurements made adjacent to the failure position after testing were used to calculate km . The joints were reasonably well-aligned and therefore the following equation was used (2,6):

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 1
[1]

where tmin ,max = pipe wall thicknesses at location analysed and e = axial misalignment (or centre-line mismatch).

In the strip specimens, km was calculated directly from the measured strains assuming:

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 2
[2]

where εr is the strain measured on the pipe surface of interest and εt is the strain on the opposite surface.               

2.4. Fatigue Test Results

The test results obtained from the full-scale pipes and the strip specimens are presented in Tables 1(a) and 2(a) respectively. Fatigue cracks always initiated at the toe of the weld root bead and propagated through the pipe wall thickness. The fatigue life was defined as the number of cycles required to produce a through-wall fatigue crack. The results are plotted together in Figure 3(a) in terms of the local stress range. Both sets of data agree reasonably well with the mean S‑N curve for BS7608 (7) fatigue design Class E, the classification widely adopted for riser girth welds, until they start to deviate towards different fatigue limits (the ‘high-cycle regime’). Indeed, a distinct difference between the full-scale pipes and strips is that the apparent fatigue limit for the strips is higher by a factor of two or more than that for the full-scale pipes. It will also be evident that the fatigue limit for the full-scale pipes corresponds to endurance greater than 107 cycles, the value widely assumed in fatigue design rules. In contrast, that for the strips appears to be reached close to 2x106 cycles. Further investigation was performed in an attempt to explain this large difference in the high-cycle fatigue performance of the full-scale and strip specimens containing nominally identical girth welds.

2.5. Factors Affecting Relative Fatigue Performance of Full-Scale Pipes and Strip Specimens

Differences in fatigue crack initiation site Referring to Fig.3(a), it will be noted that a particularly high fatigue life was obtained from the most highly loaded full-scale pipe (specimen1). A difference between this and most of the other full-scale specimens concerned the presence of longitudinal seam welds in the UOE pipe used to make them. In general, rather poor weld root bead profiles were produced at the junction of a seam weld and the test girth weld, with a relatively high weld bead, up to 1.3mm, a sharp corner at the weld toe and cold lap-type flaws due to local incomplete fusion. An example is shown in Fig.4. These features were thought to arise because the slight peaking of the pipe circumference at a longitudinal weld seam inhibits fit-up between the copper backing and the inside surface of the pipe during girth welding. The same problem has been observed if there is a significant mismatch between the inside surfaces of the abutting pipes (a ‘hi-lo’) (2). Clearly, a combination of this with peaking at the longitudinal seam weld would intensify the problem of achieving a favourable weld root bead profile and freedom from cold lap flaws. In the present context, the fatigue cracks initiated at the girth/seam weld junction in all but two of the full-scale specimens, 1 mentioned earlier and 9, which also exhibited better fatigue performance than other specimens tested at similar stress levels. In both these specimens, fatigue cracks initiated remote from the girth/seam weld junction at locations where the weld bead profiles were judged to be favourable with bead heights of around only 0.5mm and no evidence of cold laps.

The significance of the girth/seam weld junction was not appreciated at the start of the project and consequently no particular attention was paid to ensuring its presence in the strip specimens, only one of which could be extracted from a single girth weld in any case. In the event, the junction was present in only one strip specimen (no. 2-1). Noting that it failed at a stress close to the apparent fatigue limit for the strips, it is clearly possible that strip specimens containing the girth/seam weld junction could fail at lower stresses, closer to the fatigue limit obtained from the full-scale pipes. Some additional fatigue tests were performed on strips to investigate this possibility.

Differences in residual stress state Apart from potential differences between the geometries of the weld beads in the full-scale and strip specimens, there is also likely to have been a difference in the level of residual stress due to welding. Residual stresses were not measured in the present specimens but past experience from measurements and calculations suggests that there could have been relatively low compressive residual stresses near the girth weld on the inside of the full-scale pipe. However, residual stress measurements in girth welds have shown that they can be widely scattered (2) and vary in sign, even for a single weld. Thus, they may have been tensile or compressive in the present fullscale girth welded pipes. Cutting out the strip specimens would be expected to relax them and thus leave lower residual stresses in the strips. This would be detrimental to fatigue if compressive residual stresses in the full-scale pipe were relaxed in the strips, which is at odds with the actual high-cycle fatigue performance of the strips. However, the opposite would be the case if tensile residual stresses in the full-scale pipe were relaxed in the strips. Having said this, the fact that both types of specimen were tested under a tensile mean stress of 125 N/mm2 would be expected to diminish the significance of any difference in residual stress level. Nevertheless, some additional fatigue tests were performed on strips at higher tensile mean stresses to see if they gave fatigue lives closer to those obtained from full-scale pipes at stress levels below the apparent fatigue limit for the strips.

Further fatigue tests In order to explore the significance of the girth/seam weld junction in strip specimens, two additional specimens that included the junction were produced and fatigue tested at stress ranges below the apparent fatigue limit for the strips. One of them was also used to investigate the significance of residual stress differences between the strips and full-scale pipes by testing it under the more severe tensile mean stress of 365 N/mm2. In addition, a further strip specimen that did not include a girth/seam weld junction was also tested under a high tensile mean stress, 295 N/mm2 in this case. Finally, in order to expand the database, particularly near the transition between the high-cycle results for full-scale pipes and the apparent fatigue limit for strips, further full-scale and strip specimens that had survived previous fatigue testing at lower stresses without cracking were re-tested.

All the additional fatigue test results are presented in Tables 1(b) and 2(b), as appropriate. In order to clarify their comparison with the previous results, scatter-bands enclosing the data in Fig.3(a) were produced, as shown in Fig.3(b). Initially, regression analysis was performed on the results obtained from full-scale and strip specimens that appeared to lie on a single line, shown enclosed in boxes in Fig.3(b), to fit an S-N curve. From observation of the remaining results, it was then assumed that the mean fatigue limit for the strip specimens corresponded to N=2x106 cycles, giving the value 126N/mm2, while that for the full-scale pipes was 50N/mm2, corresponding to N=3.24x107 cycles. Scatter limits parallel to the S-N curve were then drawn to enclose all the data used to calculate the mean S-N curve. These proved to lie 2.4 standard deviations of log N either side of the mean curve which, for the number of test results involved, were slightly wider than the 95% confidence intervals corresponding to 97.7% probability of survival (lower limit) or failure (upper limit). Finally, scatter-bands were also drawn for the fatigue limits on the basis that they should be the same width in the stress axis direction as that enclosing the results for failed specimens. In other words, the scatter in fatigue strength was assumed to be the same for endurances above and below the fatigue limits.

The new results are shown with the scatter-bands from Fig.3(b) in Fig.5. Starting with the additional results obtained by re-testing previously tested specimens it will be seen that their fatigue performance appears to have been enhanced rather than degraded as a result of the previous fatigue testing at lower stresses. This is especially significant in the case of full-scale specimen 12b which gave a fatigue life over eight times higher than expected from the mean S-N curve fitted to the results from new specimens. None of the re-tested specimens failed from a girth/seam weld junction and clearly this might explain their superior fatigue performance. However, the weld bead profiles at the fatigue crack initiation sites were all judged to be comparable with those that did fail from that location. The more likely explanation is that their fatigue performance has been enhanced as a result of either 'under-stressing' or 'coaxing' (8). The former normally refers to cyclic stressing for an appreciable period at a stress below the fatigue limit, while the latter refers to situations in which gradually increasing stresses less than the final one are applied in small increments. Gurney (8) notes that both techniques, but especially the latter, can produce increases in fatigue strength of the order of 30%,   which would correspond to a two-fold increase in life for the present specimens. The present results suggest that the effect is even greater in the high-cycle regime. The exact mechanism for the coaxing effect is not very clear, but work hardening or  strain ageing have been suggested. However, in view of the fact that the fatigue behaviour of welded joints is remarkably insensitive to the strength of the material, it is difficult to see why mechanisms like these that increase the tensile strength of steel would be influential. It is more likely that they referred to fatigue crack initiation at mild stress concentration features, when material strength would be relevant.

Turning to the significance of having the girth/seam weld junction in the strip specimens, both of the additional  specimens failed where they were gripped in the testing machine rather than from the girth weld. Detailed examination of sections at the girth/seam weld junctions showed that the weld bead profiles were relatively poor with evidence of local cold lap-type flaws, as shown in Fig.4. There was no evidence of fatigue crack growth in specimen 3-1 but there did appear to be what could have been a fatigue crack, propagating from the cold lap flaw, in specimen 5-1.5. Thus, specimen 5-1.5 may have failed eventually if it had not failed prematurely where it was gripped. However, although the result would lie below the apparent fatigue limit for the strips it would still be above the results obtained from the full-scale pipes. Thus, although the girth/seam weld junction is clearly significant, it does not seem to be the only factor to explain the difference between the fatigue performance of the full-scale and strip specimens.

The final variable considered was the significance of differing levels of residual stress in the two types of specimen. In an attempt to eliminate any influence of residual stress, two strip specimens were fatigue tested under very high tensile mean stress conditions. However, neither specimen failed from the weld. One of them was specimen 3-1, containing a girth/seam weld junction. As noted above, this showed no evidence of fatigue cracking after an endurance close to the mean S-N curve for full-scale specimens, indicating either that its fatigue life would have been considerably longer or that the applied stress was below the fatigue limit. The other specimen, 3-7, did not include a girth /seam weld junction and indeed the weld profile was relatively good. This might account for its  very high endurance, some 10 times greater than expected from the full-scale specimen results. However, full-scale specimens with comparable weld geometries failed in shorter lives. Thus, again possible differences in residual stress between the full-scale and strip specimens have some effect but they do not explain fully the large difference in high-cycle fatigue performance.

Statistical effect of variations in initial flaw size In analyzing the difference in fatigue limits between small and large components, Haagensen (9) cited the greater likelihood of finding a large defect in a large volume of material than in a small one. When flaws in a material are assumed to be randomly distributed and of random severity, an equation with the following form was proposed to describe the effect of the volume of highly stressed material on the fatigue limit:

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 3
[3]

where Sof is the known fatigue limit for a full-scale specimen of volume Vf, Sos is the unknown fatigue limit for a small-scale specimen of volume Vs. From SAE (10), a suitable value for b  is 0.034. Haagensen (9) took the highly stressed volume to be proportional to the cube of a characteristic length such as the diameter or thickness, changing the exponent to 3b, and demonstrated a good prediction.

The same approach was applied in the present study in an attempt to quantify the likely difference in fatigue limit  between full-scale and strip specimens. The assumed characteristic dimension was specimen width, so that Eq.3 becomes:

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 4
[4]

where Wf and Ws are the full-scale and strip specimen widths. Substituting Ws =80mm and Wf =1457mm (inside circumference of the pipe), Eq. 4 indicates that the fatigue limit for the full-scale specimens should be 0.74 times that for the strips, that is 0.74 x 126 N/mm2 = 93N/mm2. In fact, it was around 50 N/mm2, a much greater difference than predicted. This adds support to the previous indications that variations in weld geometry are not sufficient to explain fully the large difference in fatigue limits between the full-scale and strip specimens.

2.6. Fracture mechanics analysis

The effect of possible differences between full-scale and strip specimens on their relative fatigue performance was also investigated theoretically using fracture mechanics. In particular, the approachwas used to calculate the S-N curve for the girth weld for different assumptions about the stress concentration effect of the weld geometry, initial flaw size and residual stress level. As detailed in BS 7910 (6), use was made of the Paris law relationship between the stress intensity factor range, ΔK, and the rate of fatigue crack growth, da/dN:

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 5
[5]

where, for a crack of depth a at the toe of a weld subjected to applied stress range Δσ,

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 6
[6]

In this equation, Y is a function of crack size and shape while Mk is the magnification factor that allows for the influence of the stress concentration effect of the weld. Integration of Eq.5 gives the number of cycles N for a specified flaw (ai ) to propagate to failure (a=af , commonly assumed to correspond to through-thickness cracking, as in the present fatigue tests) under applied stress range Δσ:

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 7
[7]

Knowing the relevant input parameters, Eq.7 can be used  to calculate the S-N curve for a weld detail on the basis that the life corresponds to the number of cycles required to propagate a pre-existing flaw to failure. There is a threshold value of ΔK, ΔKth , below which da/dN is negligible. Thus, if the  combination of flaw size, detail geometry and applied loading results in ΔK< ΔKth, fatigue failure will not occur. Clearly, the lowest applied loading to produce this condition corresponds to the fatigue limit for the detail concerned.

In order to apply Eq.7, it was necessary to define the relationship between ΔK and the girth weld geometry, flaw size and shape and applied loading. Although most of the relevant information is provided in BS 7910 (6), the weld toe magnification factor Mk for the very narrow weld beads produced in girth welds is not. Therefore, a special solution  was calculated for the present girth weld geometry (Fig. 6) using finite element analysis (FEA). The pipe was modelled as axi-symmetric, and an elastic analysis was carried out for axial tensile loading. Second order displacement field elements were used, with fine meshing around the crack. Based on sections of the actual girth welds, the assumed weld root bead geometry was as shown in Fig.6, with α = 70o and h = 0.3 or 1.3mm representing the range of weld root bead heights.  Mk was  found to decrease with decreasing weld root bead height but an increase in the angle beyond 70o had little effect. As anticipated, the Mk for the narrow girth weld root bead did turn out to be less than that of wider welds, as shown in Fig. 7.

Fracture mechanics models were developed on the basis of different assumptions about the initial flaw size and shape, the severity of the stress concentration effect of the weld root bead and the residual stress state. The aim was to produce models that reflected possible differences between the full-scale and strip specimens that might explain differences in fatigue limit while still producing the same S-N curve. Input parameters related to weld and flaw geometry were based on examinations of the actual welds tested and limited observations of fatigue crack growth in strip specimen tests. In the case of the full-scale specimens it was assumed that flaws were larger than in the strips and that the stress concentration effect (SCF) of the weld geometry was at its highest. To obtain the same S-N curve for the strips with a smaller assumed flaw and a less severe stress concentration effect, the flaw was assumed to adopt a more severe aspect ratio (lower depth/surface length). This is not unreasonable in that a deep flaw at a seam/girth weld junction in the full-scale case would be more localized than a shallower one elsewhere along the girth weld.

Residual stress considerations were reflected in the assumed values of ΔKth , a parameter that is highly dependent on the stress ratio, which itself depends on a combination of the applied minimum and maximum stresses and the residual stress. In the case of the full-scale pipes, the lower bound value of ΔKth = 63N/mm3/2 from BS7910, corresponding to very high stress ratios, was assumed. However, for the strips the threshold value that resulted in a fatigue limit close to that obtained was estimated and the likely residual stress state deduced.

Following trial-and-error calculations, the actual S-N curve for the girth weld was produced assuming the following:

a) Full-scale specimens - semi-elliptical flaw of depth ai =0.25mm and 2.5mm surface length; Mk for h=1.3mm; average km =1.04; ΔKth =63N/mm3/2.

b) Strip specimens - semi-elliptical flaw of depth ai =0.15mm and 12mm surface length; Mk for h=0.3mm; average km =1.095; ΔKth =value deduced from actual fatigue limit.

A suitable crack growth law was (N and mm units):

Comparison of fatigue of girth-welds in full-scale pipes and small-scale strip specimens - equation 8
[8]

This is close to the mean ‘simplified’ law recommended in BS 7910 for steels for R ≥ 0.5.

In support of these assumptions, Fig.8 shows a comparison of the calculated progress of a fatigue crack in a strip specimen with that measured by alternating current potential drop. Similarly, Fig.9 shows the good agreement between the calculated S-N curves and the fatigue endurance data obtained from the full-scale and strip specimens and, in the case of the full-scale specimens, the fatigue limit. The calculations produce the average fatigue limit of 126N/mm2 for the strips if it is assumed that ΔKth =120N/mm3/2. Based on the database used to establish the recommendations in BS 7910 (11), this clearly corresponds to a much lower stress ratio than the value chosen for the full-scale specimens, but the scatter in threshold data is such that its precise value cannot be defined.

Nevertheless, this analysis has confirmed that the difference between the fatigue limits for the full-scale and strip specimens is consistent with reasonable assumptions about possible variations in initial flaw size, weld geometry and effective stress ratio, the latter reflecting differences in residual stress, between the two types of specimen.

3. Discussion

The fatigue performance of the two types of specimen was found to agree reasonably well at high stress ranges and could be represented by the Class E mean curve. However, there was a significant difference in apparent fatigue limit between the full-scale and strip specimens, the former being only around one half of that for the latter. Various factors were investigated in an attempt to explain this difference, specifically the presence or not of a girth/seam weld junction, the general variation in weld toe flaw size and weld profile, residual stress differences and a statistical size effect, but no single explanation was found.

Examination of the general appearance and sections of the girth welds tested revealed that the girth/seam weld junction inevitably contained weld defects at the root bead toe. Their sizes varied but appeared to be larger at locations where the weld root bead was relatively high (>0.7mm) with a poor peaky profile. Furthermore, the weld root bead height also varied around the weld circumference. It was found that almost all those specimens with better weld root bead profiles (height < 0.7mm) gave fatigue lives significantly higher than specimens with poorer quality roots. Attempts to obtain fatigue failures at stresses below the apparently high fatigue limit for the strip specimens from additional strip specimens containing the girth/seam weld junction were not successful, even under very high applied tensile mean stresses. Indeed, the fact that one of the specimens survived without any evidence of fatigue damage for over 108 cycles at a stress range some 30% higher than those that had caused failure of full-scale specimens suggests that difference between the fatigue limits for the full-scale and strip specimens cannot be attributed solely to the presence or not of a seam/girth weld junction. Nevertheless, it is an undesirable feature if it results in weld root geometries of the kind observed in the present specimens. One factor that is likely to encourage a poor weld is alignment of the seams of the two pipes at the joint, a practice that was not applied to the present specimens and is generally avoided by pipeline fabricators. However, even when the seams are not in line, definite peaking at a seam may require some remedial grinding to ensure as close a fit as possible between the inside of the pipe and the copper backing. The issue is unlikely to be so significant in the case of girth welds made without copper backing but the authors are not aware of any experimental confirmation of this.

It was speculated that the welding-induced residual stress could be partly or even fully released during extraction of a strip specimen from the original welded pipe. A consequence of having lower residual stresses in the strip specimens is that the effective mean stress, resulting from the superposition of the applied and residual stress, would have been lower than in the full-scale specimens. As is well known, mean stress can have a significant effect on fatigue, especially in the high-cycle near‑fatigue limit regime, with fatigue damage increasing with increase in tensile mean stress. However, the residual stresses in the full-scale girth welds would need to have been tensile for this to improve the fatigue performance of the strips, whereas available evidence suggests that they are more likely to have been compressive. In any case, testing both types of specimen under high tensile mean stress conditions should have eliminated a significant effect of any differences in residual stress unless they were extremely high in one type of specimen. Nevertheless, the fracture mechanics analysis indicated that although the large difference in fatigue limits for the full-scale and strip specimens could be attributed partly to rational differences in assumed flaw size and weld geometry, it was also necessary to assume that the effective stress ratio, from the combination of applied and residual stresses, was positive and considerably higher in the full-scale specimens, suggesting that the residual stresses in the strip specimens were either very low in tension, or compressive. However, in view of the many assumptions made in a fracture mechanics analysis, a different conclusion might be drawn on the basis of other assumptions. Residual stress measurements on each specimen would help to resolve this.

Consideration of the statistical size effect, based on the assumption that a severe flaw was more likely to be present in a large-scale than a small-scale specimen, provided a positive indication, at least qualitatively, that the fatigue limit should be higher in the strips than the full-scale specimens. However, using a relationship based on SAE recommendations the effect was underestimated considerably.

Thus, although a number of possible differences between the full-scale and strip specimens could have contributed to an increase in the fatigue limit for girth welds when estimated from tests on strip specimens, no single factor could explain the present difference. This issue deserves further attention. Meanwhile, the present results throw doubt on the validity of assuming that strip specimens provide a realistic representation of full-scale girth welded pipes as long as they are fatigue tested under conditions of high tensile mean stress. Although the agreement may be good at relatively low endurances, the high-cycle fatigue performance, particularly the fatigue limit, can be very seriously over-estimated.

One final finding which also has practical implications with regard to validation fatigue testing of girth welds concerns the practice of re-testing unfailed specimens. The present results indicated that the previous testing at lower stress could improve fatigue performance, possibly as a result of ‘coaxing’. In an extreme case, the life of one re-tested full-scale girth weld was increased by around eight times. The mechanism responsible is not fully understood but, whatever the explanation, it is clear that fatigue results from re-tested girth welds should not be relied upon as a basis for design.

4. Conclusions

Comparative fatigue tests were performed on nominally identical girth welds in 508mm OD x 22mm wall thickness API 5L-X65 UOE steel pipe, either as full-scale pipe specimens or small-scale strips cut from such pipes. Differences in fatigue performance were investigated by considering possible differences in weld geometry, flaw size, residual stress and applied mean stress on the basis of physical examination, additional fatigue tests, fracture mechanics calculations and statistical evaluation. The following conclusions were drawn:

  • The fatigue performance of both the full-scale and strip specimens above their fatigue limits agreed well and was consistent with BS 7608 Class E mean.
  • However, the high-cycle fatigue strength and the fatigue limit of the full-scale specimens were significantly lower than those for the strip specimens
  • The junction of the girth and longitudinal seam weld in the UOE pipe proved to be the main site for fatigue crack initiation in the full-scale pipe specimens. Inadequate fit-up between the copper backing and the pipe in this region during welding led to higher stress concentrations and larger welding flaws at the weld root bead toe. Care is needed to avoid this problem in practice.
  • Attempts to explain the difference in fatigue limit between the full-scale and strip specimens suggested that it was due to a combination of differences in flaw size, weld root bead quality and residual stress, but a definitive explanation could not be found from the information available.
  • The stress intensity factor (K) solution for a crack at the root of a girth weld, produced by FEA, confirmed the influence of weld root profile (notably weld root bead height) on K, and hence on fatigue performance.
  • Strip specimens cannot be relied upon to establish the S-N curve in the high-cycle regime or the fatigue limit of a girth welded pipe, even if they are tested under conditions of high applied tensile mean stress.
  • Re-testing of specimens previously tested at lower stresses produced unrepresentatively high results, indicating that the pre-fatigue testing was beneficial. Consequently, data from re-tests of unbroken fatigue test specimens must be treated with caution and should not be used as a basis for design.

5. Acknowledgements

Most of the work described was supported by BP Exploration and Production Ltd, Chevron Energy Technology Company, ExxonMobil Upstream Research Company, Heerema marine Contractors, Petrobras, Total SA and the UK Health and Safety Executive. The authors are grateful for their support and permission to publish the findings.

6. References

  1. Buitrago J, Weir M S and Kan W C: 'Fatigue design and performance verification of deepwater risers', Paper OMAE2003-37492, Proc. Offshore Mechanics and Arctic Engineering Conference, ASME,2003.
  2. Maddox S J, Razmjoo G R and Speck J B: 'An investigation of the fatigue performance of riser girth welds', Paper OMAE2006-92315, Proc. Offshore Mechanics and Arctic Engineering Conference, ASME, 2006.
  3. Maddox S J and Razmjoo G R: 'Fatigue performance of large girth welded steel tubes', Paper OMAE98-2355, Proc. Offshore Mechanics and Arctic Engineering Conference,  ASME, 1998.
  4. Buitrago J and Zettlemoyer N: 'Fatigue design of critical girth welds for deepwater application', Paper OMAE98-2004, Proc. Offshore Mechanics and Arctic Engineering Conference, ASME, 1998.
  5. Salama M M: 'Fatigue design of girth welded pipes and the validity of using strips', paper OMAE99-2003, Proc. Offshore Mechanics and Arctic Engineering Conference, ASME, 1999.
  6. BS 7910:2005: 'Guide to methods for assessing the acceptability of flaws in metallic structures', British Standards Institution, London, 2007.
  7. BS 7608:1993: 'Fatigue design and assessment of steel structures', British Standards Institution, London, 1995.
  8. Gurney T R: 'Fatigue of welded structures', Cambridge University Press, UK, 1979.
  9. Haagensen P J, Slind T and Orjasaeter: 'Size effects in machine components and welded joints', Proc. Offshore Mechanics and Arctic Engineering Conference, ASME, 1988.
  10. SAE ‘Fatigue design handbook', Society of Automotive Engineers, Warrendale, PA., USA, 1969.
  11. King R N: 'A review of fatigue crack growth rates for offshore steels in air and seawater environments', HSE Report OTH 511, London, Health & Safety Executive Books, 1998.

Table 1 Fatigue test results obtained from full-scale pipe specimens

a) Original tests

Specimen number. km Applied stress range, N/mm2 Stress ratio, R Endurance, cycles Failure location and position relative to the seam weld
Nominal Local
1 1.05 181 190 0.16 1.24x106 At weld root, remote from seam weld
2 1.03 49 50 0.67 4.03x107 At weld root, at seam weld centre
3 1.06 68 72 0.57 8.63x106 At weld root, at seam weld centre
4 1.02 52 53 0.66 2.12x107 At weld root, at seam weld centre
5 1.00 92 92 0.46 4.17x106 At weld root, at edge of a seam weld
6 1.09 49 53 0.67 8.04x107 At weld root, at seam weld centre
7 1.07 56 60 0.63 2.34x107 At weld root, at seam weld centre
8a 1.03 51 53 0.66 1.49x108 Unbroken
9a 1.02 54 56 0.65 1.29x108 At weld root, remote from seam weld
10 1.07 56 60 0.63 2.17x107 At weld root, 10mm from seam weld edge
11a-weld A 1.03 36 37 0.75 1.57x108 Unbroken
11a-weld B 1.03 36 37 0.75
12a 1.05 42 44 0.71 1.28x108 Unbroken

b) Additional tests on specimens previously tested at lower stress

Specimen number. km Applied stress range, N/mm2 Stress ratio, R Endurance, cycles Failure location and position relative to the seam weld
Nominal Local
8b 1.03 158 163 0.23 1.65x106 At weld root, remote from seam weld
11b-weld A 1.03 105 108 0.41 6.55x106 Unbroken
11b-weld B 1.03 107 110 0.40 At weld root, 15mm from seam weld edge
12b 1.00 94 94 0.45 3.38x107 At weld root, remote from seam weld

Table 2 Fatigue test results obtained from strip specimens

a) Original tests

Specimen number km Applied stress range, N/mm2 Stress ratio, R Endurance, cycles Comments1
Nominal Local

6-3.5

1.01

102 104 0.42 3.01x107 Unbroken

6-9

1.11

92 102 0.46 3.31x107 Unbroken

10-1

1.01

255 257 -0.01 3.05x105  

10-3

1.18

130 153 0.32 7.97x105  

2-4.5

1.13

101 115 0.43 6.09x106  

1-11

1.15

102 118 0.42 1.63x107  

2-1

1.17

107 125 0.40 2.11x106 Failed at girth/seam weld junction

2-9

1.16

96 111 0.46 1.86x106  

4-2

1.09

112 123 0.38 2.32x106  

1-4

1.16

104 120 0.41 1.02x108  

3-2.5

1.05

196 207 0.12 4.61x105  

1-9

1.13

137 154 0.29 9.85x105

b) Additional tests

Specimen number km Applied stress range, N/mm2 Stress ratio, R Endurance, cycles Comments1
Nominal Local

3-7

1.07

70 75 0.79 1.13x108 295 N/mm2 mean stress; unbroken

5-1.5

1.04

78 81 0.52 1.79x107 Failed from jaw; specimen included girth/seam weld junction.

3-1

1.01

69 69 0.83 1.12x107 365 N/mm2 mean stress; failed from jaw; specimen included girth/seam weld junction.

5-5

1.09

130 142 0.32 4.80x106 Re-tested after previous testing at lower stresses

3-11

1.06

150 158 0.25 2.17x106 Re-tested after previous testing at lower stresses

Note: 1 Failure from toe of weld root bead unless stated otherwise.

Figure 1 Full-scale girth welded specimen, including locations of strain gauges (gauges 6, 7 and 8 on other side of pipe, opposite to gauges 4, 3 and 2 respectively).
Figure 1 Full-scale girth welded specimen, including locations of strain gauges (gauges 6, 7 and 8 on other side of pipe, opposite to gauges 4, 3 and 2 respectively).
Figure 2 Small-scale strip specimen
Figure 2 Small-scale strip specimen
Figure 3 Fatigue test results obtained from full-scale and strip specimens a) Original results
Figure 3 Fatigue test results obtained from full-scale and strip specimens a) Original results
Figure 3 Fatigue test results obtained from full-scale and strip specimens b) Scatterband produced on the basis of the mean S-N curve fitted to the enclosed results
Figure 3 Fatigue test results obtained from full-scale and strip specimens b) Scatterband produced on the basis of the mean S-N curve fitted to the enclosed results
Figure 4 Examples of girth weld profiles and the local geometry at the weld root bead toe a) Example of girth weld profile at a seam/girth weld junction
Figure 4 Examples of girth weld profiles and the local geometry at the weld root bead toe a) Example of girth weld profile at a seam/girth weld junction
Figure 4 Examples of girth weld profiles and the local geometry at the weld root bead toe b) Example of girth weld profile remote from a seam weld
Figure 4 Examples of girth weld profiles and the local geometry at the weld root bead toe b) Example of girth weld profile remote from a seam weld
Figure 5 Fatigue test results obtained from extra tests performed to investigate reasons for difference between high-cycle fatigue performance of full-scale and strip specimens
Figure 5 Fatigue test results obtained from extra tests performed to investigate reasons for difference between high-cycle fatigue performance of full-scale and strip specimens
Figure 6 Geometry modelled for calculation of Mk for a girth weld root bead toe crack by finite element analysis
Figure 6 Geometry modelled for calculation of Mk for a girth weld root bead toe crack by finite element analysis
Figure 7 Mk solutions produced by FEA for crack at toe of 5mm wide girth weld root bead
Figure 7 Mk solutions produced by FEA for crack at toe of 5mm wide girth weld root bead
Figure 8 Comparison of actual progress of weld root bead toe fatigue cracks, at two locations A and B along weld root, and fracture mechanics calculation
Figure 8 Comparison of actual progress of weld root bead toe fatigue cracks, at two locations A and B along weld root, and fracture mechanics calculation
Figure 9 Comparison of S-N curve for girth welds tested calculated using fracture mechanics with actual data
Figure 9 Comparison of S-N curve for girth welds tested calculated using fracture mechanics with actual data

For more information please email:


contactus@twi.co.uk