Hot-Spot Stress S-N Curves for FPSO Fatigue Assessment

TWI Limited, Granta Park, Great Abington, Cambridge CB1 6AL, UK

Published in the International Journal of Offshore and Polar Engineering (Paper also given at 10 ^th International Offshore and Polar Engineering Conference (ISOPE), Stavanger, Norway, 17-22 June 2001)

Abstract

FPSOs are critical structures in which there is a high risk of fatigue cracking and therefore comprehensive design guidance is required. An approach based on the hot-spot stress is expected to be the most suitable for treating welded joints in which weld toe cracking is the likely mode of failure. As part of a recent JIP addressing the fatigue design of FPSOs, available hot-spot S-N data obtained from weld details relevant to FPSOs, including results obtained as part of the JIP, were evaluated as the basis for hot-spot stress design S-N curves.

Introduction

Floating Production, Storage and Offloading (FPSO) units are critical structures in which there is a high risk of fatigue cracking and therefore comprehensive design guidance is required. In fact, there are several well-founded fatigue design specifications for welded connections in steel that should prove to be suitable to some extent (BSI 1993, DNV 1998, DNV 2000). However, a feature of such specifications which is increasingly regarded as a disadvantage is that they are based on the use of the nominal applied stress. This proves to be a problem in the design of some structural configurations because of the difficulty of defining nominal stresses. The same problem frequently arises when stress information is obtained by finite element analysis (FEA). In view of this, there is growing interest in the use of hot-spot stresses for fatigue design. Such an approach is expected to be more suitable for designing FPSO structures.

The hot-spot stress approach has been in routine use in the design of steel tubular structures for over 25 years (HSE, 1995). However, the approach is not yet fully developed for application to plate structures like ships. Two issues still need to be addressed:

• The definition of the hot-spot stress and how it is derived from stress analysis of the structure.
• The choice of hot-spot stress design S-N curves.

Both issues were addressed in a major Joint Industry Project (JIP) 'FPSO - Fatigue Capacity', which included extensive FEA and fatigue testing of typical FPSO weld details (Lotsberg 2001). The present paper addresses (b), partly on the basis of fatigue test results obtained in the JIP (Kim 1999) but mainly from relevant published fatigue test data for a wider range of structural connections and dimensions.

A basis for hot-spot stress design S-N curves

The hot-spot stress fatigue design approach is only applicable to situations where the potential mode of failure is by fatigue crack growth from the toe of a weld. In general, three types of weld toe failure can be identified (Fricke 2001), as illustrated in Fig.1. In two cases, (a) and (c), it is generally accepted that the stress distribution approaching the weld toe depends, amongst other things, on the plate thickness. Consequently, the hot-spot stress is either the stress at a point that is some proportion of plate thickness away from the weld toe, or it is obtained by extrapolation from stresses located at distances that are some proportions of plate thickness away from the toe. However, in the case of type (b), edge cracking from a weld toe or end, the stress distribution approaching the weld toe does not usually depend on the plate thickness. Consequently, the methods used to address hot-spot types (a) and (c) are not appropriate; alternatives are still under investigation. In view of this, the present review is confined to type (a) and (c) hot-spots.

Fig. 1. Types of hot-spot in welded structures

a) Weld at end of longitudinal attachment (weld toe or end on loaded plate surface)
b) Weld on or around a plate edge (weld toe on plate edge)
c) Weld transverse to loading (weld toe on loaded plate)

By definition, the hot-spot stress includes all the stress concentrating features of a welded joint except that due to the local weld toe geometry. Thus, the most logical basis for the hot-spot stress design S-N curve is that for transversely-loaded butt welds. In such details, the only source of geometric stress concentration is the weld itself.

In practice, another source of stress concentration arises in butt welded joints if they are not aligned. Under axial loading, misalignment can increase the stress experienced by the weld as a result of the introduction of secondary bending. However, there is now a substantial database obtained from butt welded specimens instrumented with strain gauges to measure the effect of any misalignment (Maddox 1997). Many of these were made deliberately misaligned, with plate-centre line offset up to 100% of the plate thickness. As a result, the database encompasses a wide range of weld profiles, from generally favourable in aligned joints to profiles resembling fillet rather than butt welds in the severely misaligned joints. Similarly, there would have been some variability in local weld toe geometry as a result of the choice of welding process. In this respect it is well known that submerged arc welding can produce very unfavourable profiles. In spite of this, as seen in Fig.2, there is good correlation between the data regardless of welding process when plotted on the basis of the measured nominal stress near the weld. Apart from weld profile, the stress concentration effect of the weld toe can also be expected to be influenced by plate thickness, being more severe in thick plates than thin ones. Therefore, the plate thicknesses are indicated in Fig.2. In the context of FPSOs, a reference plate thickness, up to which the design S-N curves would apply directly, of 25mm would encompass most applications. As will be seen, there is no consistent influence over the range of the data, from 13 to 23mm. Thus, the database provides a basis for a design S-N curve for perfectly-aligned butt welded joints in plates up to 23mm thick (rounding this to 25mm would clearly be acceptable), which could also be used as the hot-spot stress S-N curve. For thicker plates, a design penalty would be imposed. One option would be that specified in DNV (2000). In effect, this entails reducing allowable stress ranges for plates thicker than 25mm by the factor (25/ t) ^0.1, where t = plate thickness in mm. However, the whole issue of what allowance to make for a size effect when using the hot-spot stress approach to fatigue design has yet to be resolved.

Regression analysis of the data, 243 results in all, produced a mean S N curve with a slope of m = 3.24, where the equation of the S-N curve has the form:

S ^mN = A     Eq.1

and	S = stress range N = fatigue life A = constant

For consistency with the usual design S-N curves, for which m = 3, the results were re-analysed assuming m = 3. The resulting 95% confidence intervals, two standard deviations of log N either side of the mean, are included in Fig.2. Using the terminology adopted in the International Institute of Welding (IIW) fatigue design recommendations (Hobbacher 1996), FAT 'number' where the number is the fatigue strength in MPa or N/mm ² at 2 x 10 ⁶ cycles, the lower 95% confidence limit would be FAT 94. The closest current design curve (e.g. IIW, DNV) to this is FAT 90, which is very close to the UK Class D, upon which it is based, and the same as category D in DNV (2000). Assuming a standard deviation of log N of 0.2, as recommended in DNV (1998), Fig.2 includes the FAT 90 mean and design curves. Both are just below the mean and lower 95% confidence limit to the data and therefore FAT 90 would be a suitable choice as the hot-spot stress design S-N curve. Support for this conclusion was investigated by reference to available hot-spot stress S-N data.

Fig. 2. Fatigue test results obtained from transverse butt welds expressed in terms of the measured stress range

In reviewing the hot-spot stress S-N data attention was focused on just three of the various definitions of the hot-spot stress. These are discussed in more detail in Fricke (2001) and illustrated in Fig.3. The first is the linear extrapolation method recommended in IIW (1996), based on extrapolation to the weld toe from stresses 0.4t and 1.0t from the toe, where t = plate thickness. A second linear extrapolation method, favoured by some Classification Societies, involves extrapolation from 0.5t and 1.5t. Finally, in view of its greater simplicity, the use of the stress at a single point close to the weld toe, namely 0.5t, is of interest.

Fig. 3. Three definitions of hot-spot stress considered in present evaluation

Published hot-spot stress fatigue data

Literature review

A survey of the literature revealed only nine published papers containing hot-spot stress S-N data relevant to structural details in FPSO structures (Yagi 1992, Kawano 1993, Dexter 1994, Koskimaki 1995, Marquis 1995, Huther 1996, Andrews 1997, Dahle 1997 and Miki 1997). In addition, results from the present JIP (Kim,1999) and previously unpublished results obtained in a TWI/EWI joint industry project (Maddox, 1993) are considered. The weld details covered are shown schematically in Fig.4, while Table 1 provides a summary of other relevant information.

Fig. 4. Range of welded plate specimens and structural components for which hot-spot stress S-N data were available

(i) Transverse and (ii) longitudinal fillet welded attachments (or gusset)

(iii) Transverse load-carrying cruciform joint

(iv) Doubler plate(s)

(v) Bracket to beam connection tested by Dexter

(vi) Corner joint of the type tested by Yagi and Kawono

(vii) Tubular corner joint tested by Dahle

(viii) Hopper corner structure tested by Kim

(ix) Corner connection tested by Marquis

Table 1 Summary of sources of hot-spot stress S-N data

Reference	Weld detail	Mode of loading	Specimen or components	Thickness mm	Stress analysis		Hot-spot stress			Other
					FEA	Strain gauges	IIW	0.5/ 1.5T	0.5t
Yagi, 1992	a) Gusset at corner between I-sections b) Plate with longitudinal gussets c) Plate with doubler pad on one side	Bending Axial Axial	Component Specimen Specimen	16 15 and 30 15 and 30	- - -	✓ ✓ ✓	* * *	✓ ✓ ✓	* * *	✓ ✓ ✓
Kawono and Inoue, 1993	a) Gusset at corner between I-sections b) Plate with longitudinal gussets c) End bracket at corner between I-sections	Bending Axial Bending	Component Specimen Component	22 25 26 and 32	- - -	✓ ✓ ✓	* * -	* * -	* * *	- - -
Dexter et al, 1993	Angle sections fillet welded to top flange of box section beam (failure from toe of transverse fillet weld)	Bending and tension on angle	Component	10	-	✓	-	-	✓	-
Koskimaki, 1995	Plates with longitudinal gussets (ferritic, austenitic and duplex stainless steels)	Axial	Specimen	10	-	✓	✓	-	-	-
Marquis, 1995	a) Plate with longitudinal gussets b) T-joint between box beams, with corner gussets	Axial In-plane bending	Specimen Component	10	-	✓	✓	-	*	-
Huther, 1996	Plates with longitudinal gussets	Bending	Specimen	20	✓	✓	✓	*	*	-
Miki, 1996	Cope hole details in steel bridges	Bending	Component	16	✓	✓	-	-	-	✓
Andrews, 1997	Cruciform joint	Axial	Specimen	13	-	✓	*	-	✓	-
Dahle, 1997	Tube to rectangular box section (Failure from toe of transverse fillet weld). Constant and variable amplitude loading	Bending	Component	16	✓	-	✓	-	-	-
Kim, 1999	a) Plate with longitudinal gussets b) Plate with doubler plates on both sides c) Hopper corner structure	Axial Axial Bending	Specimen Specimen Component	10	✓	✓	✓	✓	✓	-
Maddox, 1993	a) Transverse butt weld b) Plate with transverse non-load carrying fillet weld c) Plate with longitudinal gussets d) Transverse load-carrying fillet weld e) Plate with doubler plates on both sides f) Plate with doubler plate on one side	Axial Axial Axial & bending	Specimen Specimen Specimen and component	13 13 13	✓ ✓ ✓	✓ ✓ ✓	✓ ✓ ✓	* * *	* * *	✓ ✓ ✓
* Can be deduced from stress distribution obtained

Another significant publication was Partenen (1995). This presents a very extensive database (180 results) of hot-spot stress S-N data obtained over many years in Finland from specimens of around 10mm thickness. The document provides a case for adopting FAT 100 as the hot-spot stress design S-N curve for plates up to 10mm thick. Assuming the plate thickness correction (t _ref/t) ^0.1, this would translate to FAT 90 for t = 25 mm. Unfortunately, the actual test data were not presented in the document and therefore they cannot be evaluated in the same way as the other published information.

Although hot-spot stresses were determined by FEA in some cases, this was not generally the case and greater reliance was placed on measured stresses. Therefore, in the present assessment the fatigue test results were evaluated in terms of hot-spot stresses estimated from measured strains, FEA results only being used in one case, (Dahle, 1997), when measured stresses were not quoted. An advantage of using measured stresses is that allowance is included for any secondary stresses in the test specimens, for example due to misalignment, which would not be included in FEA of an idealised geometry. Consequently, scatter in the fatigue data should be reduced. However, against this it needs to be borne in mind that the resulting hot-spot stress S-N curve will be less conservative as a design curve than one based on lower-bound FEA estimates of the hot-spot stresses, as used in the evaluation of fatigue data from the present JIP by Fricke (2001).

In most cases it proved possible to determine the IIW extrapolated hot-spot stress and the stress 0.5t from the weld toe. There was not usually sufficient information to determine the third hot-spot stress, by extrapolation from stresses 0.5t and 1.5t from the weld toe, but where possible this was also estimated.

Preliminary assessment of individual test series

The hot-spot stress S-N data detailed in Table 1 refer to a range of weld details in specimen or component form. Just over half refer to hot-spot type (c) with most cases of hot-spot type (a) being in the form of plates with longitudinal gussets.

In a preliminary assessment of the available data each set of results was analysed for comparison with the proposed FAT 90 design curve. S-N curves were fitted by regression analysis (excluding unbroken specimen results) assuming a slope m = 3 to facilitate comparison with the design curve. The lower 95% confidence intervals (approximately two standard deviations of log N below the mean) were also determined. Table 2 summarises the resulting differences between the FAT 90 mean and design curves and the fitted S-N curves. They are compared on the basis of the ratio of the fatigue strength given by the data to that obtained from the FAT90 curve at the same life. Where possible, this is shown for the experimental data expressed in terms of both the IIW hot-spot stress range and the stress 0.5t from the weld toe. A ratio of less than 1.0 means that the experimental value is below FAT 90, indicating that FAT 90 would not be safe.

Table 2 Comparison of published hot-spot stress S-N data and FAT90 mean and design curves

Reference	Figure	Detail	Hot-spot type (see Fig.1)	Fatigue strength +
				FAT90 fatigue strength
				FAT90 mean		FAT90 design
				IIW	0.5t	IIW	0.5t
Yagi, 1992	4(v)	Corner gusset	a	1.01	0.73	1.03	0.73
Yagi, 1992	4(ii)	Plate with gussets	a	1.21	1.04	1.28	1.09
Yagi, 1992	4(iv)	Plate with doubler	c	1.08	0.95	1.23	1.11
Kawono, 1993	4(vi)	Corner gusset	a	0.87	0.77	0.98	0.86
Dexter, 1993	4(v)	Angle to plate	c	1.25	-	0.99	-
Koskimaki, 1995	4(ii)	Plate with gussets	a	1.17	-	1.16	-
Marquis, 1995	4(ix & ii) 4(ix)	Box beam corner and plate with gussets	a	1.14	-	1.08	-
		Box beam corner	a	-	0.83	-	0.66
Huther, 1996	4(ii)	Plate with gussets	a	1.29	1.23	1.39	1.35
Miki et al, 1997	-	Cope holes	a	1.28*	-	0.99*	-
Andrews, 1997	4(iii)	Cruciform joint	c	0.94	0.93	0.97	0.96
Dahle, 1997	4(vii)	Box beam connection	c	1.07	-	0.81	-
Kim, 1999	4(ii)	Plate with gussets	a	1.05	0.96	1.19	1.09
	4(iv)	Plate with doubler	c	0.95	0.88	1.18	1.09
	4(viii)	Hopper corner structure	c	1.53	1.21	1.42	1.14
Maddox, 1993	4(i,ii,iii & iv)	Various in structural components	a and c	1.08	0.94	1.04	0.93
+Values shown bold are cases for which FAT90 would not be safe *Hot-spot stress from formula based in linear extrapolation from positions 4 and 10mm (0.25t and 0.625t) from the weld toe.

It will be seen that almost every set of data supports FAT 90 used in conjunction with the IIW hot-spot stress range. One exception, the corner gussets tested by Kawono (1993), involved virtually the same test specimen as Yagi (1992), which did satisfy FAT 90. The small difference may reflect the difficulty of estimating the IIW hot-spot stress from the information provided. The other case where the FAT 90 design curve was unconservative, Dahle (1997), was one in which the lower 95% confidence limit was particularly low because of the few very high fatigue lives (which were obtained under constant amplitude loading). Thus, again it would seem reasonable to overlook these results. In some cases FAT 90 appears to be over-conservative, but it will be noted that this was not the case for the structural components, apart from the hopper corner structure tested by Kim (1999). However, this case was somewhat surprising in that the test results were consistent with published data for failure from a transverse weld toe when expressed in terms of nominal stresses, which were well defined. A possible explanation is that the hot-spot stress has been overestimated. Certainly there was some doubt about the validity of the strain measurements. The results used to determine the hot-spot stress in the present evaluation were those which were closest to the FEA stress distribution. However, much lower stresses were also measured and they would have led to lower hot-spot stress estimates and much closer agreement with FAT 90. In view of the importance of the hopper corner detail to FPSOs, it is planned to investigate these results further in Phase II of the JIP.

Turning to the use of the hot-spot stress 0.5t from the weld toe, it will be seen that more results indicate that FAT 90 is un-conservative. Thus, FAT 80 may be the better choice for use with this hot-spot stress. The main problem was associated with the structural corner connections. A feature was that the stress gradient approaching the weld toe was relatively high, with the result that the stress 0.5t from the weld was significantly lower that the IIW hot-spot stress. Examination of the stress distributions provided in the papers by Yagi (1992), Kawono (1993) and Marquis (1995) indicates that a better estimate of the IIW hot-spot stress, for use with the FAT 90 design curve, would be the stress 0.3t from the weld toe.

Finally, from the few cases in which the hot-spot stress could be estimated using the '0.5t/1.5t' extrapolation method, it was up to 5% lower than the IIW value for plate specimens, but up to 12% lower for structural corner joints, reflecting the high stress gradient referred to above. Assuming that these differences were typical, it would seem that FAT 90 is still suitable for simple details, but FAT 80, 12% lower fatigue strength, would be required for structural components incorporating high stress gradients. However, note that it was only possible to estimate the '0.5t/1.5t' hot-spot stress in four of the investigations considered and so this conclusion is tentative.

Thus, the published data highlight the problem of estimating the hot-spot stress for situations in which there is a high stress gradient approaching the weld, as in most of the structural components. It seems that the IIW '0.4t/1.0t' extrapolation method produces hot-spot stresses that satisfy FAT 90, but FAT 80 may be more appropriate if the hot-spot stress is assumed to be the stress 0.5t from the weld toe. Few data were available to check the '0.5t/1.5t' extrapolation method, but those available also favoured a lower design curve than FAT 90 for situations of high stress gradient.

In practice, it would be unrealistic to relate the choice of stress analysis method and design curve to the stress gradient since it is not known in advance. A better approach is to relate the S-N curve to the method of stress analysis regardless of the stress gradient. On the basis of the data considered here, this would be reasonable for the IIW (FAT 90) and 0.5t (FAT 80) definitions of the hot-spot stress. However, the distinction is less clear for the '0.5t/1.5t' extrapolation method and downgrading to FAT 80 for all cases could be over-conservative. Further consideration of the data in terms of the two hot-spot types (a and c) helps to resolve this issue.

Analysis of available S-N data

Type (a) hot-spots

All but one set of the available test results for specimens and components failing from type (a) hot-spots, that is the ends of longitudinal attachments or gussets (see Fig.1), are plotted in Figs.5(a) and (b), in terms of the IIW hot-spot stress range and that 0.5t from the weld toe respectively. Since the results for beams with cope holes were expressed in terms of a different hot-spot stress, they are not included. The previous assessment (see Table 2) showed that they were consistent with FAT 90 based on another definition of the hot-spot stress, obtained by extrapolation from 4 and 10mm from the weld toe.

Figures 5(a) and (b) include the FAT 90 and FAT 80 mean and design S-N curves respectively, and the mean and 95% confidence intervals fitted to the data (neglecting results for unfailed specimens) by regression analysis. In both cases, the results from plate specimens, all of which consisted of plates with longitudinal fillet welded attachments, are in close agreement. The results obtained from structural components are more widely scattered, especially when expressed using the hot-spot stress 0.5t from the weld toe ( Fig.5(b)), but generally agree with those for plate specimens. Based on the fitted mean S-N curve and 95% confidence intervals, the results in Fig.5(a), in terms of the IIW hot-spot stress range, show that FAT 90 is conservative and would actually support FAT 100. This margin is sufficient to encompass the differences between hot-spot stresses estimated by the IIW and '0.5t/1.5t' extrapolation methods. Thus, FAT 90 should still be applicable for both extrapolation methods. In contrast, due to the low estimates of the hot-spot stress in the structural corner joints using 0.5t, as discussed earlier, FAT 80 is appropriate. Indeed, the lower 95% confidence interval actually coincides with the FAT 80 design curve.

Thus, the total database for type (a) hot-spots supports the adoption of FAT 90 for hot-spot stresses estimated by either of the two extrapolation methods, and FAT 80 for hot-spot stresses 0.5t from the weld toe.

Fig.5a) Fatigue test results for type (a) hot-spots, expressed in terms of the IIW hot-spot stress

Fig.5b) Fatigue test results for type (a) hot-spots, expressed in terms of the hot-spot stress 0.5t from the weld toe

Type (c) hot-spots

The available test results for type (c) hot-spots, corresponding to fatigue failure from the toe of a transverse fillet or butt weld (see Fig.1), cover a wider variety of weld details. Five sets of results refer to members with doubler pads. The remaining data refer to members with transverse fillet welded attachments, cruciform or T-joints, and to intersecting tubes. A preliminary review of the results showed that although there was reasonable correlation between most data sets, three sets stood out as being significantly higher or more widely scattered than the others (Dexter 1993, Dahle 1997 and the hopper corner structure tested by Kim 1999). Therefore, they were excluded in the first compilation of the results for comparison with the design curves.

The remaining results are presented in the same way as the type (a) hot-spot data in Figs.6(a) and (b). It will be seen that there is close agreement between all the results, with little scatter. The mean S-N curve fitted to the results expressed in terms of the IIW hot-spot stress range ( Fig.6(a)) is slightly lower than the FAT 90 mean, while the lower 95% confidence interval is just above the FAT 90 design curve, due to the low scatter. Thus, the results provide reasonable support for FAT 90. The results in Fig.6(b), expressed in terms of the hot-spot stress range 0.5t from the weld toe, are compared with FAT 80. In this case, the mean S-N curve virtually coincides with FAT 80 mean, but the lower 95% confidence limit is significantly higher than FAT 80 design. In fact, it lies just above FAT 90 design. However, results below the confidence limit lie close to or on the FAT 80 design curve and therefore FAT 80 seems to be appropriate.

Fig.6a) Fatigue test results for type (c) hot-spots, expressed in terms of the IIW hot-spot stress

Fig.6b) Fatigue test results for type (c) hot-spots, expressed in terms of the hot-spot stress 0.5t from the weld toe

The three sets of results which did not agree with those in Fig.6 are plotted in Fig.7. Also shown are the 95% confidence intervals enclosing the results in Fig.6. The results obtained by Dahle (1997) are only available in terms of the IIW hot-spot stress and therefore they are only shown in Fig.7(a). Similarly, those from Dexter (1993) are only available in terms of the hot-spot stress 0.5t from the weld toe and they are only included in Fig.7(b). However, the overall picture is the same in both figures, with the extra sets of data generally lying above the confidence intervals enclosing the remaining hot-spot type (c) results. Higher results might be expected from Dahle, since the stress ratio was -1 and the fatigue crack grew in a shell bending stress field. Similarly, the fatigue cracks in Dexter's specimens grew in shell bending stress fields and therefore higher fatigue lives are to be expected. This might also be an explanation for the hopper corner structure data from Kim. Thus, in every case there are aspects of the results that could explain why they were higher than expected on the basis of the results given in Fig.6. From the practical viewpoint, the proposed FAT 90 for use with the IIW hot-spot stress or FAT 80 with the 0.5t hot-spot stress are conservative for these extra data.

Fig.7a) Exceptionally high fatigue test results for type (c) hot-spots, expressed in terms of the IIW hot-spot stress

Fig.7b) Exceptionally high fatigue test results for type (c) hot-spots, expressed in terms of the hot-spot stress 0.5t from the weld toe

Hot-spot stresses determined by extrapolation from 0.5t and 1.5t lie between the other two. However, since both lower 95% confidence limits in Fig.6 lie above the FAT 90 design curve, it will probably be reasonable to assume FAT 90 for use with this hot-spot stress. Against this argument, it will be recalled that the use of hot-spot S-N curves based on measured stresses may be unconservative when used for design with calculated stresses. Therefore, caution is needed before accepting the conclusions drawn. The lack of calculated hot-spot stresses for the published data provides an important limitation on their use for defining hot-spot stress design curves. Thus, their value would increase if, in future, the specimen geometries could be analysed using the FEA procedures recommended in Fricke (2001). Relevant work would form part of Phase II of the JIP.

Conclusions

Available fatigue data for weld details relevant to FPSOs were evaluated as the basis of hot-spot stress design S-N curves. Three methods of determining the hot-spot stress were considered: by extrapolation from stresses 0.4t and 1.0t (IIW) or 0.5t and 1.5t (Classification Societies) from the weld toe, and as the stress 0.5t from the weld toe. They were applied to weld ends on plate surfaces (type (a) hot-spots) and to weld toes on plate surfaces (type(c)). The following conclusions were drawn:

• Individual sets of fatigue data support FAT 90 as the hot-spot stress design S-N curve when used in conjunction with the IIW definition of the hot-spot stress. The same is true for the other two definitions of the hot-spot stress considered for simple weld details, but FAT 80 would be more appropriate for structural components incorporating high stress gradients approaching the weld toe.
• However, there was limited support for the more useful solution of adopting FAT 90 for hot-spot stresses obtained by either of the extrapolation methods and FAT 80 for hot-spot stresses 0.5t from the weld toe.
• Evaluation of the results was hindered by the wide scatter in some cases and the lack of data expressed in terms of all three hot-spot stress definitions. However, support for the more practical solution comes by combining the data and considering them in terms of hot-spot type (a) or (c).
• Fatigue data from beams with cope-holes also support the use of FAT 90 but only in conjunction with a different definition of the hot-spot stress (extrapolation from 4 and 10mm from weld toe).
• The stress 0.3t from the weld toe proved to be a better estimate of the IIW hot-spot stress than 0.5t.
• Further work is needed to analyse the results in terms of hot-spot stresses calculated using FEA, and to address type (b) hot-spots, welds on plate edges.

Acknowledgements

The author is grateful to sponsors of the FPSO and TWI/EWI JIPs and to Hyundai Heavy Engineering for permission to include their test data.

References

Andrews, R. M (1996): The effect of misalignment on the fatigue strength of welded cruciform joints. Fatigue Fract. Engng.Mater. Struct. Vol.19. No.6, pp755-768

BSI (1993): BS 7608:1993 Code of Practice for Fatigue Design and Assessment of Steel Structures. BSI, London.

Dahle T (1997): 'Fatigue behaviour of a railway component - application of new design criteria and TIG remelting techniques', in 'Welded High-Strength Steel Structures', Proc. 1st North European Engineering and Science Conference (NESCO I), EMAS Publishing, Cradley Heath, West Midlands, p.519-526.

Dexter R J, Tarquinio J E and Fisher J W (1994): 'Application of hot-spot stress fatigue analysis to attachments on flexible plate', Proc. 13 Intl. Conf. 'Offshore Mechanics and Arctic Engineering', ASME, New York, p85-92.

DNV(1998): Classification Note No.: 30.7. Fatigue Assessment of Ship Structures. Det Norske Veritas, Oslo.

DNV (2000): Recommended Practice RP203. Fatigue Assessment of Offshore Structures. Det Norske Veritas, Oslo.

Fricke W (2001): 'Recommended hot-spot analysis procedure for structural details of FPSOs and ships based on round-robin FE analysis', ISOPE, Stavanger, June.

Hobbacher, A.(1996): Fatigue Design of Welded Joints and Components. IIW, Abington Publishing, Abington, Cambridge

Huther I, Lieurade H P, Sayki N and Buisson R (1996): 'Fatigue strength of longitudinal non-load carrying welded joints. Use of the hot-spot stress approach', in ' Fatigue of Welded Components and Structures', Proc. 3rd Intl. Spring Meeting, Les Editions de Physique, Les Ulis Cedex A, France, p.57-64

Kawono H and Inoue K (1993): 'A load approach for the fatigue strength evaluation of ship structure', in ' Fatigue Design', Proc. Intl Symposium, Mechanical Engineering Publications Limited, Bury St Edmunds, p.95-108

Kim, W. S (1999): 'Fatigue Tests of Typical Welded Joints'. Hyundai Heavy Industries Co., Ltd. Ulsan (unpublished)

Koskimaki M and Niemi E (1995): 'Fatigue strength of welded joints in three types of stainless steel', IIW Doc. XIII-1603-95.

Lotsberg, I 'Background and overview of the FPSO-Fatigue capacity JIP', to be presented at OMAE, Rio, June.

Maddox S J (1997): 'Developments in fatigue design codes and fitness-for-service assessment methods', IIW International Conference on Performance of Dynamically-loaded Welded Structures, WRC, New York, pp22-42.

Maddox S J and Manteghi S (1993): 'The application of fatigue design rules to large welded structures', TWI Report No. 5587/22/93, Dec.1993 (unpublished).

Marquis G B and Kahonen A (1995): 'Fatigue testing and analysis using the hot-spot method', VTT Publication No. 240 'High Cycle Spectrum Fatigue of Welded Components', P O Box 2000, Fin-02044 VTT, Finland.

Miki C and Tateishi K (1997): 'Fatigue design of cope hole details in steel bridges', Int. J. Fatigue, Vol.19, No.6, p.445-455.

HSE (1995): 'Offshore Installations: Guidance on Design, Construction and Certification'. HSE. London.

Partenen T and Niemi E (1995): 'Collection of hot-spot S-N curves based on tests of small arc-welded steel specimens', IIW Document XIII-1602-99.

Yagi J and Tomita Y (1992): 'Definition of hot-spot stress in welded plate type structure for fatigue assessment (Report 1),' J. Society of Naval Architects of Japan, Vol.169, June, p311-318 (IIW Doc XIII-1414-91).