OTH 91 356

LITERATURE SURVEY ONFATIGUE STRENGTHS OFLOAD-CARRYING FILLET

WELDED JOINTS FAILING IN THEWELD

Authors

T R Gurney and K MacDonald

The Welding InstituteAbington Hall, Abington

Cambridge CB1 6AL

HSE BOOKS

Health and Safety Executive - Offshore Technology Report

© Crown copyright 1995Applications for reproduction should be made to HMSO

First published 1995ISBN 0-7176-0977-4

This report is published by the Health and Safety Executive aspart of a series of reports of work which has been supported byfunds provided by the Executive. Neither the Executive, or thecontractors concerned assume any liability for the report nor dothey necessarily reflect the views or policy of the Executive.

Results, including detailed evaluation and, where relevant,recommendations stemming from their research projects arepublished in the OTH series of reports.

Background information and data arising from these researchprojects are published in the OTI series of reports.

CONTENTS

APPENDIX A

FIGURES

TABLES

29REFERENCES11.

27CONCLUSIONS AND RECOMMENDATIONS10.

25CONTINUOUS LONGITUDINAL FILLET WELDS9.

23LONGITUDINAL LOAD-CARRYING FILLET WELDS8.

21THICKNESS CORRECTION FACTOR7.

19JOINTS WITH PARTIAL PENETRATION WELDS6.

17CONVENTIONAL ANALYSIS OR RESULTS5.

13EXPERIMENTAL RESULTS4.

11FRACTURE MECHANICS ANALYSIS3.

9NOMENCLATURE2.

7INTRODUCTION1.

5SUMMARY

PAGE

3

4

SUMMARY

In load-carrying fillet welded joints, which may involve welds either transverse orparallel to the direction of the applied stress (or a combination of the two), there aretwo potential modes of fatigue cracking. One involves failure at the weld toe andthe other cracking from the root through the weld throat. However, if weld throatfailure occurs it is clear that the optimum strength of the joint has not beendeveloped, since a higher strength could be obtained by increasing the weld size.When toe failure supervenes no further strength increase is possible by increasingthe weld size. Hence in the current fatigue design rules (1) two sets of designstresses are given, one relating to plate failure (Class F2) and one to weld throatfailure (Class W), and the effect of using those design stresses should be to arrive atjoint dimensions close to the optimum.

However, there are several reasons why the existing Class W design stresses maybe incorrect so the available test data for transverse load carrying fillet weldssubjected to tensile loading have been re-analysed.

5

6

1. INTRODUCTION

In load-carrying fillet welded joints, which may involve welds either transverse orparallel to the direction of the applied stress (or a combination of the two), there aretwo potential modes of fatigue cracking. One involves failure at the weld toe andthe other cracking from the root through the weld throat. However, if weld throatfailure occurs it is clear that the optimum strength of the joint has not beendeveloped, since a higher strength could be obtained by increasing the weld size.Hence in the current fatigue design rules (1a) two sets of design stresses are given,one relating to plate failure (Class F2) and one to weld throat failure (Class W), andthe effect of using those design stresses should be to arrive at joint dimensions closeto the optimum.

However, there are several reasons why the existing Class W design stresses maybe incorrect, including the following:

a) The data for transverse load-carrying fillet welds used in the original analysiswere fairly limited, being derived from only seven investigations. There are nowmany more data available.

b) At the time that the Class W design curve was produced Class F2 did not exist,so that Class W was derived by comparison with Class F. However transverseload-carrying fillet welded joints are now included in Class F2 as far as platefailure is concerned so that the stresses given by Class W are now inconsistent.

c) The data which were used were all obtained using relatively thin specimens(typically 13 mm) made with double fillet welds and subjected to axial loading.Since that time a correction factor to allow for the influence of plate thicknesshas been introduced but has not been carried through into Class W. There istherefore a need to examine the implications of that amendment.

In view of the above, the available test data for transverse load carrying fillet weldssubjected to tensile loading have been re-analysed and the results are set out below.

7

8

2. NOMENCLATURE

‘Basic’ joint thicknesstB

Fatigue strength of a joint withthickness tB

SB

Fatigue strength of a joint withthickness t

S

Initial crack length (for fillet weld a1

= 0.5 Tp)a1

A specific value of mwIX

Generalised stress parameter definedby equation [8]

r&

Integral defined by equation [5]ICyclesN

Particular values of D (equations[10] and [11])

U,V

= 1CD

Constants in the Paris crack growthequation (equation [2])

C,m

Stress on weld throat, as defined byequation [18]

rw

Stress applied to platerp

Non-dimensional crack length aWa

Instantaneous crack lengthaA2)

Functions of (see equation [1] andHTp

Appendix A)A1)

Stress intensity factorK

)))

Assumed total length of half crack atfailure (=H + 0.5 TP)

W

)Fig. 1)

Weld leg lengthH

) See)

Thickness of central cross plateTc

))

Main plate thicknessTP,t

9

10

3. FRACTURE MECHANICS ANALYSIS

In general, the problem of specifying the optimum weld size for load carryingtransverse fillet welded joints is more complex than making a simple comparison ofweld and plate load-carrying areas. The fatigue strengths of joints that fail from theroot depend on many variables, chiefly the weld size, plate thickness and depth ofweld penetration, the last two defining the size of pre-existing defect at the root. Itwould be impractical to attempt to specify the fatigue strengths for weld failures interms of simple S-N curves for all geometric configurations that are likely to occur inpractice. Equally, one would expect the available test results, when plotted as anS-N curve on the basis of weld throat stress, to be relatively widely scattered, simplyon account of the side variation in the geometrical parameters of the various jointswhich have been tested. A better approach is to make use of the fact that fatiguecracks are thought to initiate early on in the lives of such joints, and to use a fracturemechanics crack propagation relationship to calculate the fatigue strength. In thatway the influence of variations in joint geometry can be eliminated, so that one mightreasonably expect the scatter in test results to be reduced.

As far as is known, the only solution for the stress intensity factor (k) for a crack atthe root of such a weld is that published by Frank (2). The geometry of the joint,with relevant dimensions, is shown in Fig.1. The analysis below will be limited tojoints in which the following conditions are met:

a) The weld angle, 0, = 45O Frank found that although 0 influenced the value ofK, it was not very significant. Also, in practice, it would usually be necessaryto assume an angle, and 45O is the value normally assumed in design.

b) The main plates and cross plate are of equal thickness, ie TP = Tc. The effectof varying the ratio TP/Tc was not investigated by Frank but it would not beexpected to be significant for typical joint dimensions used in practice.

For these conditions the stress intensity factor of a root crack of total length 2a canbe written as

K = f { , } [1]rP oa aw

HTP

Where W, H and TP are defined in Fig 1 and the function f{} is defined in AppendixA.

Now it is well known that, for pulsating tension loading, the rate of propagation of afatigue crack can, to a close approximation, be related to the range of stressda

dNintensity factor by the Paris relationship:(DK)

= C m [2]dadN (DK)

Where C and m are material constants. This can be integrated to predict thefatigue life of a crack. Thus, from equations [1] and [2],

11

da = C mN¶ a2 (DrP ) { [ , }m [3]a1

aw

HTP

oa

Alternatively, expressing a in non-dimensional form as = equation [3] can bea aW

rewritten as

- 1 m2

da = C m W N¶ a1 (DrP )

{f [ ] }m [4]a2 a, HtP oa

Where the crack propagates from an initial size = = to a size at failure a2a1a 1W a1

= in N cycles. Hence, assuming failure to occur when af = W, so that = 1.0,afW a2

and assuming also that m = 3.0, which is a typical value for structural steels,equation [4] becomes

1.0 da = C 3 W½N(DrP ) { f [ }3 [5]¶ a i a, H

TPoa

Thus, writing the integral in equation [5] as I, for brevity, we get

I = C 3 W½N [6](DaP )

Clearly, the value of I can be evaluated for any particular joint geometry underconsideration and some typical values are shown in Fig 2.

It will be seen that equation [6] can be rewritten as

{ ( ) }3 N = DDaP\/WI

13

Where D [7]h 1c

This represents an equation of an S-N curve with ‘stress’ replaced by thegeneralised stress parameter , wherea&

= ( ) Da& DaP\/WI

13

12

4. EXPERIMENTAL RESULTS

Transverse load-carrying fillet welds of the type under consideration have beentested by several investigators and in some instances (2, 4-22) the joint geometrieshave been such that failure occurred in the weld throat. All the available data havebeen analysed by plotting the results in terms of and they are shown in FigsDa&

3-22. These results only include specimens which were tested under tensile loadingand in which the welds appeared to have virtually zero penetration. The relevantjoint geometries are summarised in Table 1, which also shows the correspondingvalues of

= ( ) Da &

Da P

\/WI

13

For ease of comparison, Figs 3-22 also show the theoretical mean curve deducedfrom the fracture mechanics analysis assuming C = 1.83 x 10-13. This is the averagevalue m = 3, assuming that is expressed in mm/cycle and in N/mm-3/2 (3).da

dN DKMeanwhile, statistical analysis of the data (504 results), assuming a relationship ofthe form

( ) N = D DaP&

[9]

showed that the standard deviation of n(N) was 0.7925 and that the mean andfmean -2 standard deviations curves could be represented by D = 6.255 x 1012 and1.272 x 1012 respectively. On the same basis the mean of the theoretical scatterband corresponds to

D = = 5.46 x 101211.83x10 −13

The mean and lower limit experimental curves are also indicated in Figs 3-22

Thus, the joint behaviour when weld failure occurs can clearly be represented by acurve of the form

( )3 N = U DaP&

[10]

where U is some selected value of D.

Similarly, the strength of joints failing from the weld toe can, from BS 5400, bewritten as

( )3 N = V DaP&

[11]

where the value of V depends upon the selected probability of failure

13

Hence, in order for failure in the plate and in the weld throat to occur at the sametime, equations [10] and [11] must be satisfied simultaneously. Thus, rewriting [10]in terms of , using equation [8], we getDaP

( )3 N = U DaP& \/W

I[12]

so that, from [11] and [12]

= \/WI

UV

[13]

There are several possible strategies for selecting the appropriate values of U andV. The two most obvious possibilities are:

1 to use values corresponding to the mean -2 standard deviations experimentalcurve for weld throat failure (S3N = 1.272 x 1012) and the similar curve forClass F2 in BS 5400, since Class F2 includes this type of joint and relates tofailure from the weld toe (S3N = 4.32 x 1011). Inserting these values in equation[13] thus gives

= = 2.94 \/WI

1.272x10 12

4.32x10 11

[14] Conceptually, this value represents an equal chance of failure at the weld toe and atthe root at the ‘design’ level,

2 to use values corresponding to the mean curve for weld throat failure (6.255 x1012) and the mean curve for Class F2 (1.231 x 1012). Inserting these valuestherefore gives

= = 5.08 \/WI

6.255x10 12

1.231x10 12

[15] Given that a suitable value of (say x) can be selected, it is now necessary to\/W

Irelate that value to the corresponding stresses in the plate and the weld throat.

From Fig. 2 it will be seen that the value of I is related to the joint geometry via theparameters and , so that it is possible to define a unique relationship for filletH

TP

a iW

welds with zero penetration, as shown.

Replotting that relationship in terms of n (I) v n ( ) as shown in Fig. 23, it isn n a iW

apparent that, to a close approximation, we can write

( ) 5.37 I = 0.002 a iW

[16]

so that, eliminating I by introducing the value X

( ) 5.37 \/W = 0.002 Xa iW

14

or ( ) = { } a iW

0.002X\/W

15.37

[17]

Now, if the stress in the plate is , and the stress on the weld throat is , weaP aW

have, since the total weld throat area in a fillet welded joint with no penetration is (2X 0.7H),

Tp = (1.4H) aP aW

[18]

so what

= 1 a W

[19] 1.4 ( )aP

HTP

But from geometrical considerations (see Fig. 1)

= = 1- HTP

W−a i2a i

a iW

2a iW

[20]

Therefore, from equations [19] and [20],

= a Wa P

2 iaW

1.4 (1 - ) a iW

[21]

Then, eliminating ( ) from [21] with [17]a iW

= { = (0.002 X) a Wa P

0.002x\/W }

15.37

15.37

[22]

0.7 {1 - ) } 0.7 {\/W - (0.002 X }0.002X\/W

15.37

15.37

15.37

Which can be evaluated for the relevant values of W and X. Some typical valuesare shown in Table 2.

However, W is not directly proportional to the plate thickness ( ). FromTP

geometrical considerations (see Fig. 1)

W = H + 0.5 Tp [23]

and from equation [18]

H = 1 TP1.4

[24]

15

( )a Wa P

so that

Tp = W [25] {0.5 + 1 } 1.4

a Wa P

Given that the values of for given values of W and X are defined by equationa Wa P

[22] and set out in Table 2, it is easy to derive the corresponding values of Tp fromequation [25]. They are also shown in Table 2, and the resulting relationshipsbetween and Tp are shown in Fig. 24. From this diagram it will be seen that, at

a Wa P

the basic thickness of 13mm used for deriving the majority of the fatigue designcurves in BS 5400, the values of corresponding to the two possible values of X

a Wa P

specifically considered in this report (namely 2.94 and 5.08, see equations [14] and[15] are 0.58 and 0.68 respectively. Thus, knowing that the mean and mean - 2standard deviations fatigue strengths for Class F2 at 2 x 106 cycles are = 85 andaP

60 N/mm2, it follows that the corresponding strengths required for Class W are asshown in the following table. This assumes, of course, that the standard deviation ofLog N for Class W, in terms of rather than , is the same as for Class F2.aW a&

4357Existing stresses for Class W(for comparison)

41580.685.08 (eq. [15])35490.582.94 (eq. [14])Mean - 2 S.D.Mean

Required fatigue strength of Class W at 2 x 106

cyclesAt Tp = 13

a Wa PAssumed

values of x

These strengths may be compared with the existing strengths for Class W at 2 x 106

cycles, which are 57 and 43 N/mm2. It will be noted that these are very similar tothe values derived with X = 5.08 (ie equating the means of the scatter bands for toeand root failure).

16

5. CONVENTIONAL ANALYSIS OF RESULTS

In order to enable a comparison to be made between the available data and thevarious design stresses suggested above, the results have been replotted on the basisof the nominal weld throat stress in Figs 25 to 47, with the existing mean and meanminus 2 standard deviations S-N curves for Class W also shown in each Figure. Itwill be seen that, in total, 22 results lie below the mean - 2 S.D. Design line. This isslightly higher than the number which would strictly be anticipated on the basis of ablind application of statistics, namely 11, but 6 of the relevant failures related tospecimens which were either galvanised (series 1775) or coated with zinc epoxyprimer (series 1779 and 1780) prior to welding. It seems at least possible that thosespecimens may well have contained additional root defects prior to testing, but it hasnot been possible to check whether that was actually the case.

A summary of the 22 low results is shown in Fig 48, together with the current mean- 2SD S-N curve for Class W and the corresponding curve for X = 2.94 (equation[14]). The curve corresponding to X = 5.08 is not shown since it is virtually identicalto the Class W curve. It will be seen that for all practical purposes the curve for X= 2.94 forms a lower limit to all the results.

It would obviously be unrealistic to assess the scatter in life of the results, expressedin terms of weld throat stress, because of the wide variations in geometry of thevarious test series. An analysis has, however, been made of the results for thosejoint geometries for which 20 or more individual results existed. With best fit curvesfor each geometry forced to slope m = 3 (for consistency with Class F2), the resultswere as shown in Table 3. Compared with a quoted standard deviation of n(N)nfor Class F2 (toe failure) equal to 0.5248, a simple average of the 4 results shown inTable 3 is 0.5658 while a weighted average gives 0.5207. On the basis of theseresults it does not seem unrealistic to assume that the standards deviations for rootfailure (Class W) and toe failure (Class F2) are equal (ie 0.5248). This representsan increase in the degree of scatter implicit in the existing Class W design stresses,for which the corresponding standard deviation is 0.4251.

In passing, it should be noted that 3 of these 4 test series gave relatively lowstandard deviations of n(N), up to 0.579. The ‘average’ values were increasedfmainly by the fact that the fourth series had a high S.D. Of 0.729. It is interesting tospeculate that, since the high value was associated with 20mm thick specimenswhile the lower values related to 10-12mm thick specimens, the higher S.D. Wasdirectly related to the specimen thickness as a result, for example, of a more variableresidual stress distribution. Thus, although the residual stresses at a weld toe willalways be highly tensile, those at the root may well depend on how well the plate isrestrained and the number and size of weld passes. In a multipass fillet weld theroot stresses may be compressive and the larger the joint the greater is the possibilityof crack closure and an enhanced fatigue life. This also offers a possibleexplanation for a size effect that is less pronounced in practice than would beinferred from the theoretical fracture mechanics treatment (see Section 7).

17

18

6. JOINTS WITH PARTIAL PENETRATION WELDS

The proposed design curve should, of course, also apply to joints with partialpenetration but failing the weld throat. In order to check its validity in that situationsome readily available test results have been plotted, with the existing Class Wcurves for comparison, in Figs 49-52. It will be seen that, of the 56 results shown inFigs 49-51, only 3 are lower than (and one is on) the mean minus 2 standarddeviations curve. Of the results in Fig. 52, however, the majority are all below theproposed design curves, but all refer to weld throat stresses above 175 N/mm2 andhence to short life failures. Equally it will be noted that two of the three low resultsreferred to previously were also at high stress (265 N/mm2). Given that all these‘unsafe’ results were at stresses above those which would normally be used forweld metal, it seems reasonable to conclude that the test results for partialpenetration welds support the use of the proposed curve for the design of suchjoints.

19

20

7. THICKNESS CORRECTION FACTOR

As can be seen from Table 2 and from Fig, 24, the theoretical analysis shows that,as plate thickness (Tp) increases decreases. Indeed, it is easy to deduce from

a Wa P

Fig. 24 that, for a particular value of X,

( ) TP 0.15 = constant a Wa P

[26]

However, for failure from the weld toe in the plate, it has been found that fatiguestrength tends to decrease with increasing plate thickness and in the Guidance Notes[1a] it is stated that the design stress should be decreased with increasing platethickness such that

= SB ( ) 0.25 aPtBt

[27]

Where is the fatigue strength of a joint with thickness t and SB is the fatigueaP

strength of a joint of the basic thickness tB. For the type of joint under considerationtB is assumed to be 22mm, even though that is illogical.

Now, from equation [26] we have

( ) = ( ) ( ) 0.15 a Wa P

a Wa PB

tBt

[28]

so that, substituting from [27] into [28]

= ( ) SB ( ) 0.4 = ( )B( ) 0.4 aWa Wa PB

tBt aW

tBt

[29]

as the required thickness correction factor for Class W based simply on theoreticalconsiderations.

Unfortunately, it cannot really be said that the test results support that conclusion,but that may, in part, be due to the fact that extremely few of the results relate to‘thick’ specimens. As far as specimens thicker that the current ‘basic’ thicknesses(tB = 22mm) are concerned, Figs 32, 33, 36 and 42 show results for specimens withplate thicknesses of 25, 38, 25 and 32mm respectively and all those results are ‘safe’without the applications of any thickness correction factor. The only results for‘thick’ specimens which are currently unsafe are those shown in Fig 34 for 30mmthick specimens. If the correction factor derived in equation [29] were applied tothe Class W mean - 2 standard deviations curve it would be reduced from 116 and43 N/mm2 at 105 and 2 x 106 cycles respectively to 102 and 38 N/mm2. The relevantcurve is indicated on Fig. 34. It will be seen that only the results at high stress andless than 104 cycles are then ‘unsafe’.

In this context it is of interest to note that, as part of the UKOSRP-I testprogramme, an investigation was carried out with the objective of defining theoptimum weld size of some joints with load-carrying fillet and partial penetration

21

welds in specimens 25 and 38mm thick. The results are summarised in Table 4,which indicates that the recommended weld throat stresses varied from 0.94 to 1.13times the corresponding plate stresses. It will be noted that these are far higher thanthose corresponding either to the current Class W curves or to curves based on X =2.94 or 5.08.

It would clearly be useful to investigate the need for a thickness correction factor toClass W in more detail by carrying out some tests on a range of joints fabricatedfrom thicker material. In the meantime it is strongly suggested that, in view of thescanty evidence, the thickness correction for Class W should be no more severethan for Class F2.

22

8. LONGITUDINAL LOAD-CARRYING FILLET WELDS

No attempt has been made to carry out a similar analysis for joints with longitudinalfillet welds. Relatively few test results have been obtained for such joints and thegreat majority of those involved failures in the plate rather than in the weld throat.However, in a few specimens tested by Gurney (24) there were small subsidiarycracks in the weld throat near the weld ends. More recently, in tests carried out byChapeau and Plumier (15), it was reported that weld failures had been obtained, butunfortunately the specimen dimensions were not reported in sufficient detail toenable the relevant weld throat stress to be calculated and, in addition, it is not clearwhich specimens failed in the weld and which in the plate.

23

24

9. CONTINUOS LONGITUDINAL FILLET WELDS

Another form of load-carrying fillet welded joint is the continuous longitudinal (i.e.Parallel to the direction of stress) fillet weld, such as often exists at the web toflange joint of fabricated girders. As far as is known no failures of such welds in theweld throat along the length of the weld have ever been reported either in service orin laboratory tests on welded beams. Under normal circumstances, if failure occurs,it consists of cracking transverse to the weld direction initiated either at weld ripplesor weld defects and the strength is high.

Recently, however, some tests have been reported (25) on an equivalent joint. Thespecimens consisted of 108mm diameter tubes fillet welded to end plates andsubjected to alternating (R=-1) torsion. This did result in shear failure in the weld.The results, for specimens fabricated from two different (but no details given) steels,are shown in Figs 53 and 54. However, since there was little difference betweenthe two sets of results all were analysed together, ignoring the test results givinglives greater than 6 x 106 cycles, and the corresponding best-fit mean and mean-2standard deviations S-N curves are also shown in Figs 53 and 54. In addition thebest-fit curves with slope forced to m=3 are also shown.

For these particular tests it seems clear that the slope was much shallower than 3.0and there may well be a temptation to introduce a new Class to cover such joints.This should be resisted, at least for the moment, because

a) The tests were at R = -1 and it is known that stress ratio can affect the slope,depending to a large extent upon residual stresses;

b) As far as is known there are no confirmatory test results available from adifferent source;

c) The introduction of a new S-N curve with a different slope would tend tocomplicate design and cause confusion.

With the last point in mind it may be noted that the mean and mean -S.D. Strengthsat 2 x 106 cycles, assuming m=3, are 107 and 74 N/mm2. These lie midway (in bothcases) between Classes E and F. It is therefore proposed that this type of weldshould be classified as F.

25

26

10.CONCLUSIONS AND RECOMMENDATIONS

Some 500 test results for transverse load-carrying fillet welds failing in the weldthroat have been analysed. It has been shown that, theoretically, an equal chance offailure at the weld toe and at the weld root at the design level is obtained by makingthe basic mean and means-2SD strengths for Class W at 2 x 106 cycles equal to 49and 35N/mm2 respectively. Further analysis of the results showed that, in fact, sucha design curve effectively formed a lower bound to all the test results.

It is suggested that, in fact, the strengths at 2 x 106 cycles should be made equal to50 and 36 N/mm2 respectively, since that is a small change which would make thedesign curve coincide with the corresponding strength proposed for Eurocode No 3:Common Unified Rules for Steel Structures, and there is obviously some benefit inmoving towards a likely European Standard.

On this basis the design equation for the revised (mean-2SD) Class W curve wouldbe

Log N = 10.970 - 3 Log S

with a standard deviation (SD) equal to 0.214 where N is the predicted number ofcycles to failure under stress range S. Alternatively the equation can be expressedas

S3 N = 9.331 x 1010

Results from 56 tests on joints with partial penetration welds (failing in the weldthroat) showed that the proposed design curve would also be acceptable for thedesign of those joints.

Although there are theoretical reasons for believing that the Class W design stressesshould be subject to a thickness correction factor which is more severe than that fortoe failure, the available experimental results do not support that conclusion. It isrecommended that some experimental tests should be put in hand to clarify thesituation but, in the meantime, that any thickness correction introduced at this stageshould be no more severe than for Class F2.

As far as continuous longitudinal fillet welds (e.g. Web-to-flange welds) areconcerned, it is proposed that they should be included in Class F, the relevant stressbeing the range of weld throat shear stress. There is no evidence regarding thepossible existence of a thickness effect in such joints but there are no obvioustheoretical reasons for expecting one.

27

28

11.REFERENCES

Denny A K and Jubb J E M. ‘Fatigue crack propagation in submerged arc18.

Ouchida H. And Nishioka A. ‘A study of fatigue strength of fillet weldedjoints’, Schweisstechnik (Berlin) 16 (4) 1966 pp 150-157 (English translnIIW Doc XIII-338-65).

17.

Anon. NRIM (Japan) Data Sheet No 5, 1978.16.

Chapear W and Plumier A. ‘Critical size of fillet welds in high strengthsteel joints’, IABSE Colloquium, Lausanne 1982.

15.

Drayton P A. Weld Inst Contract Report C178/5/69 (Confidential)14.

Gregory E N. Weld Inst Contract Report C208/16/70 (Confidential)13.

Recho N and Brozzetti J. ‘Design fatigue life of welded cruciform joints’,IABSE Colloquium, Lausanne 1982.

12.

Lieurade H P. ‘Study of the behaviour in low cycle fatigue of cruciformwelded joints in high yield strength steels’, Soud Tech Conn, 32 (11/12)1978.

11.

Nagai K et al. ‘Fatigue strength of fillet welded cruciform joints in mildsteel under pulsating tension loading’, J Soc Nav Arch Jap, 148 Dec 1980.

10.

Booth G S. Weld Inst Contract Report 3463/4/77 (confidential).9.

Anon. ECSC Report F4.1/73, Apr. 1977.8.

Anon. ECSC Report EUR 5357, 1975.7.

Haibach E. ‘The fatigue strength of welded joints as depending on theload spectrum’, LBF Rep. FB 79 1968.

6.

Stallmeyer J E et al. ‘Fatigue strength of welds in low alloy structuralsteels’, Weld Res Suppl No 6 1956 pp 298-307s.

5.

Macfarlane D S and Harrison J D. ‘Some fatigue tests of load carryingtransverse fillet welds’, Br Weld J 12 (12) pp 613-623.

4.

Gurney TR. ‘Fatigue of welded structures’, Cambridge Univ. Press, 2ndEd, 1979.

3.

Frank KH. ‘The fatigue strength of fillet welded connections’, PhD Thesis,Lehigh Univ, Oct. 1971.

2.

HSE Guidance Notes.1a.

BS 5400, Part 10, and HSE Guidance Notes.1.

29

Seeger T and Olider R. ‘Ertragbare und zulässige SchubspannungenSchwingbeanspruchter Schweissverbindungen’, Stahlbau, Aug. 1987,p.231.

25.

Gurney T R. ‘The influence of artificially induced residual stresses on thefatigue strength of load carrying fillet welded joints in mild steel’, Brit weldJ 8 (11) 1961 pp 541-553.

24.

Tsuji I and Ogawa K. J Jap Weld Soc 44 (7) 1975.23.

Andrew R M. Weld Inst Report 345/1987 (Confidential to ResearchMembers).

22.

Ohta S and Eguchi Y. ‘Fatigue strength of 80 kg/mm2 (780 N/mm2) hightensile strength steel plates with transverse fillet welded attachments’,Trans Jap Weld Sec 3 (2) Sep. 1972 pp 247-254.

21.

Yamaguchi I, Terada Y And Nitta A. ‘On the fatigue strength of steelsfor ship structure’, IIW Doc XIII-425-66.

20.

Shingai K and Imanura N. ‘Fatigue crack propagation in cruciform points’,IIW-XIII-691-73A.

19.cruciform welded joints’, Weld Res Int 7 (1) 1977 pp 57-75.

30

TABLE 1 SUMMARY OF JOINT USED IN THE ANALYSIS

2.340.3101E+00.4010.74815.856.359.512.722354/872.630.2843E+00.4070.72727.011.016.022.03.430.1186E+00.4780.54523.011.012.022.02134511.710.7679E+00.3331.00015.05.010.010.02033934.510.4065E-10.5710.37514.08.06.016.01931673.800.7836E-10.5140.47418.59.59.019.03.170.1418E+00.4630.57920.59.511.019.02.530.2444E+00.4190.69215.56.69.013.02.100.4544E+00.3710.84617.56.511.013.03.980.6525E-10.5290.44417.09.08.018.02.820.1706E+00.4480.61514.56.58.013.01831372.620.3547E+00.3900.78141.016.025.032.04.180.7602E+00.5160.46931.016.015.032.01.660.1117E+00.3081.12526.08.018.016.01.850.7679E+00.3331.00024.08.016.016.02.870.1788E+00.4440.62518.08.010.016.04.510.4065E-10.5710.37514.08.06.016.01723382.460.2177E+00.4290.66710.54.56.09.01620791.570.1040E+00.3131.10016.05.011.010.01.710.7679E+00.3331.00015.05.010.010.01.890.5510E+00.3570.90014.05.09.010.02.110.3817E+00.3850.80013.05.08.010.02.390.2528E+00.4170.70012.05.07.010.01519942.250.4005E+00.3810.81321.08.013.016.0219072.290.3380E+00.3940.76916.56.510.013.02.290.3380E+00.3940.76916.56.510.013.01417782.290.3380E+00.3940.76916.56.510.013.002.290.3380E+00.3940.76916.56.510.013.002.290.3380E+00.3940.76916.56.510.013.02.290.3380E+00.3940.76916.56.510.013.01317712.250.5136E+00.3620.88034.512.522.025.02.580.3257E+00.3970.76031.512.519.025.02.100.4333E+00.3750.83316.06.010.012.01217013.940.5437E-10.5450.41711.06.05.012.01116252.360.4896E+00.3660.86741.015.026.030.01016143.170.2090E+00.4320.65844.019.025.038.03.560.1418E+00.4630.57941.019.022.038.03.880.1067E+00.4870.52639.019.020.038.02.460.3817E+00.3850.80032.512.520.025.02.720.2757E+00.4100.72030.512.518.025.03.030.1922E+00.4390.64028.512.516.025.03.030.1922E+00.4390.64028.512.516.025.03.290.1846E+00.4420.63243.019.024.038.02.350.4442E+00.3730.84033.512.521.025.02.250.5136E+00.3620.88034.512.522.025.02.580.3257E+00.3970.76031.512.519.025.0913432.440.3460E+00.3920.77525.510.015.520.02.100.4333E+00.3750.83316.06.010.012.088322.100.4333E+00.3750.83316.06.010.012.077902.110.3817E+00.3850.80013.05.08.010.066794.210.5599E-10.5430.42117.59.58.019.056501.790.7679E+00.5001.00019.56.513.013.02.820.1706E+00.4480.61514.56.58.013.04545

a&/aPIai/WH/TPWaiHTPRefCAR No

31

TABLE 2 VALUES OF AND Tp FOR A RANGE OFa wa p

VALUES OF W AND 3 POSSIBLE VALUES OF X

55.40.54850.10.47743.230.39410042.70.56838.50.49433.320.4087529.50.59926.70.52023.050.4280.435024.10.61721.80.53518.840.4404018.600.64216.800.55514.520.4563015.770.65814.240.56912.300.4662512.870.67811.630.58610.040.479209.920.7068.960.6087.740.497156.870.756.200.6425.360.5210

TPa Wa PTP

a Wa PTP

a Wa PW,mm

0.4250.3840.339(0.002 X)

15.37

5.082.941.34X

32

TABLE 3 SUMMARY OF RESULTS FOR INDIVIDUAL JOINTGEOMETRIES FOR WHICH 20 OR MORE SPECIMENS

HAVE BEEN TESTED

4311657154Current Class W (for comparison)3698531440.57912010103387531430.72912715.72043116601620.503788101252143711930.4511768102 x 1061052 x 106105

nNf Mean-2 SDMeanSD ofNo. of

resultsWeldleg

Platethick-

Fatigue strengths (weld throat)

Note: The above analyses are based upon the assumption that the S-N curves haveslope m=3

TABLE 4 SUMMARY OF RECOMMENDED WELD SIZESDETERMINED FOR 25 AND 38MM THICK AS-WELDED

SPECIMENS IN UKOSRP I

1.1324zero381.021150%250.9419zero25

a Wa P

CorrespondingRecommended leglength, mm

Degree of weldpenetration

Platethickness

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

APPENDIX A

STRESS INTENSITY FACTOR FOR TRANSVERSELOAD-CARRYING FILLET WELDS FAILING

FROM THE WELD ROOT

The value of K for this type of joint, assuming a root crack of total length 2a, hasbeen shown by Frank (2) to be approximately

K = \/( ) [A1 + A2 ( ) ] (sec ) ½ aP oa aW

oa2W

( 1 + )2HTP

where W, H and TP are defined in Fig. 1 and where the coefficients A1 and A2 are:

A1 = 0.528 + 3.287 ( ) - 4.361 ( )2 + 3.696 ( )3 - 1.874 ( )4 + 0.415 ( )5HTP

HTP

HTP

HTP

HTP

A2 = 0.218 + 2.717 ( ) - 10.171 ( )2 + 13.122 ( )3 - 7.755 ( )4 + 1.783 ( )5HTP

HTP

HTP

HTP

HTP

The expression [A1 + A2 ( )] is an approximation to a polynominal expression andaw

for 0.2 < ( ) < 1.2 the resulting solution for K is within 2% of the accurate solution.HTP

63