The constant amplitude database

2 The constant amplitude database

2.1

Introduction

As noted in the previous chapter, it is the results that have been obtained in tests under constant amplitude loading that necessarily form the database from which the basic design S-N curves for the various types of joint have been derived. Design calculations relating to variable amplitude loading are then based on these curves. It is therefore of some interest to summarise how those design curves were derived and the type of information that now exists. It was noted previously that the use of welding really only ‘took off’ at the time of the Second World War. It was only after the war that it was recognised that there was a need to elucidate the fatigue performance of welded joints, particularly so that design rules could be compiled. At that stage the objective was therefore to define S-N curves, typically over the range of endurance from about 105 to 2 × 106 cycles, for a typical range of ‘basic’ joints. It was only considerably later that it was realised that geometry variations in a particular type of joint, such as changes in plate thickness or width and changes in weld size and shape, might have anything other than a relatively minor influence, so tests tended to be concentrated on specimens of fairly standard dimensions. Primarily as a result of what the available machines were capable of testing, both in terms of the maximum thickness that could be gripped in the machine jaws and also in terms of the maximum loads that could be applied, there was a great tendency for specimen dimensions to lie in the range 10–15 mm thick and 100–150 mm wide. By testing beams in bending it was possible to obtain test results for a slightly extended range of joint types, but the flange dimensions tended to be very similar to those of axially loaded specimens. Given that published data exist for several thousand test series it was clearly necessary to apply some fairly arbitrary restrictions to the data that were actually used in the analysis. The most recent review of the data, carried out to check the validity of the British fatigue design rules, was performed in 1987, and since none of the results obtained since then (except possibly for welded tubular joints) have 19

20

Cumulative damage of welded joints

given any very different strengths, it is convenient to base the results presented here on those used in that analysis. In other words the data considered were almost all obtained in the period 1959–1986. That effectively eliminates the very early test results, since it has to be recognised that considerable improvements in welding processes have occurred since then and it is therefore unlikely that the early results were obtained using joints representative of current welding practice. The data used were also subjected to other constraints. In the first place, since it has been shown in numerous investigations that the high cycle fatigue strength of welded joints in structural steel seems to be independent of the static strength of the parent material, that variable was ignored and results were accepted for joints in any steels with yield strengths in the range from approximately 225–850 N/mm2, although the great majority of the results related to normal structural steels. Secondly, the analyses were restricted to data emanating from specimens tested under uniaxial tensile loading or from tests on beams in bending, where the stress conditions in the flange are, in fact, similar to those in an axially loaded specimen. The stress ratio was required to be approximately zero (actually between –0.2 and +0.3, although few results exist for small negative stress ratios). Although many data exist for tests carried out at (larger) negative stress ratios (e.g. R = –1) it was felt that it might be unrealistic to make use of them. This was because, in a relatively small specimen, there might only have been small residual stresses, so that the stresses applied to the joint might actually have been partially compressive. In a real structure, high tensile residual stresses would be more likely to occur so that, in that case, the actual stress range under the same nominal loading would be fully tensile, and therefore more damaging. Hence, there is a danger that the test results obtained under compressive stress ratios may be somewhat optimistic. By the same token it would, theoretically, be safe to make use of the data obtained at higher (tensile) stress ratios, but simply for consistency they were also ignored. The main implication of the restriction on the range of stress ratios that were used, is that the proposed design stresses derived from this work refer primarily to as-welded (i.e. not stress-relieved) structures. However, for fully tensile loading the design stresses for stress-relieved joints would be expected to be very similar to those for as-welded joints. For stresses varying between tension and compression it is probable that an analysis of the relevant test results would suggest that there could be a considerable increase in design stress for stress-relieved structures. A brief summary of the influence of mean stress on the fatigue strength of welded joints under constant amplitude loading is included at the end of the chapter; residual stresses are considered in Chapter 3.

The constant amplitude database

2.2

21

Method of analysis and joint design classification

For each type of joint the individual test results were all assumed to form part of a single population. They were therefore plotted and analysed statistically, assuming that the whole set of results (often consisting of several individual test series) could be represented by a straight line on a plot of log stress (S) against log endurance (N), i.e. log N = log C – m log S

[2.1]

where m and C are constants for each individual population (i.e. joint type). This analysis was restricted to joints failing within the endurance range 5 × 104–5 × 106 cycles. In fact the analyses were carried out in two ways, firstly with the slope of the curve (m) not pre-defined, so as to obtain the actual best fit curve, and secondly with m pre-defined as (usually) 3.0, which (as discussed in Chapter 1) is a reasonable average value for structural steels and which is the value normally used for design purposes. For the higher strength ‘joints’, notably steel in the as-rolled condition and some types of continuous longitudinal welds, it was in fact found more realistic to use higher values of m (3.5 or 4.0) instead of 3.0. However, such joints are rarely critical because of the almost inevitable presence of lower strength joints, so it is not strictly relevant to pursue the matter here. For each set of data (i.e. each type of joint) the regression line and the standard deviation of log N (both for m = 3.0) were then calculated and the mean life and scatter band (±2 standard deviations) were plotted on the graph of the data. A summary of the results, showing the mean life and mean life minus two standard deviations of log N, corresponding to an applied stress of 124 N/mm2 is shown in Fig. 2.1. This includes only the lower strength joints, since it is those that are usually relevant in the design process. In general it would not have been practical to regard the lower limits of the various scatter bands as the design curves, since that would have meant that there would have been a different design curve for every joint, which would have been very unwieldy. It was therefore decided, since many joints have quite similar fatigue strengths, to limit the number of design S-N curves by considering all joints with similar strengths as belonging to a single ‘class’ and then to provide design S-N curves for each class. This is in fact the principle that has also been followed in design rules emanating from other countries, and although most are similar to the British rules there are, almost inevitably, some minor differences in the various joint classification systems that have been used. For convenience, at least in the British design rules, the various classes were designated by letters, with Classes A–G referring essentially to joints

22

Cumulative damage of welded joints Stiffener-to-web joints Submerged arc butt welds Class E Butt weld on backing bar K butt welds (cruciform) Intermittent longitudinal fillets Longitudinal NLC fillets Transverse NLC fillets Stud shear connectors Class F Transverse LC fillets (cruciform) Transverse LC fillets (lap joints) Longitudinal LC fillets (weld ends not on plate edge) Class F2 Gussets welded to plate edge Longitudinal LC fillets (weld end on plate edge) Beams with Ends not welded welded

Ends welded

cover plates

Wide plates, ends not welded

Class G 105

106 Endurance (cycles)

2

2.1 Comparison of mean life and mean life –2 standard deviations for various joints and Classes at 124 N/mm2 (note: the right-hand end of each bar represents the mean life and the left hand end the mean –2SD).

involving failure from the weld toe or end, while one (Class W) refers to failure through the weld throat from the weld root in load-carrying fillet welds. In fact one of the classes (Class A) is not really applicable in the context of welded joints, since it refers to parent material of uniform crosssection and with polished surfaces and thus represents an unattainable ideal; for that reason no design stresses are given for Class A. As far as Classes B, C and D are concerned the allocation of joints to classes was fairly obvious; for the lower classes (E, F, F2 and G) it became so when the results were plotted in the form shown in Fig. 2.1, although it proved necessary to make a few minor adjustments. Having decided upon the make-up of each class the results were re-analysed

The constant amplitude database

23

300

200

Stress range (N/mm2)

100 80

B C

60 50 D 40 E F 30 F2 G 20 W

10

105

106 107 Endurance (cycles)

108

2.2 Mean S-N curves for British joint classes.

so as to give the mean S-N curves and standard deviations for the classes. The results, expressed in the same terms as for the individual joints, are also shown in Fig. 2.1, while the S-N curves themselves are shown in Figs 2.2 and 2.3. The co-ordinates of these curves are given in Table 2.1. Since they are linear on a log stress against log N basis they can obviously be expressed as log N = log a – dσ – m log s, where σ is the standard deviation of log N and the curve is relevant to d standard deviations below the mean. The values of the relevant constants are given in Table 2.2. Obviously for any particular value of ‘a’ the equation of the S-N curve can be written as: Sm N = C

[2.2]

where C is a constant (different for each curve). For the set of mean curves C = a, while the relevant values of Sm N for the set of mean minus two standard deviation curves are also included in Table 2.2 for convenience.

24

Cumulative damage of welded joints 300

200

Stress range (N/mm2)

100 80 B 60 50

C

40 30

D E F

20 F2 G W

10

105

106 107 Endurance (cycles)

108

2.3 Mean –2 standard deviations S-N curves.

Reverting to the joint classification system (Fig. 2.1), by comparing the results for the classes and the results for individual joints it is easy to see that three joints appear to be in the wrong class. The reasons for these apparent anomalies are as follows. As far as transverse submerged arc butt welds are concerned the results suggest that they should be in Class F but have been placed in Class E. This was because most of the low test results are relatively old, and it is known that submerged arc butt welds made nowadays normally have a better weld shape and tend to give better strengths. The other two modifications both involved an apparent downgrading of class. The reason for putting stud shear connections in Class F rather than Class E was twofold; first the majority of the available test results refer to unloaded studs, and it seems possible that, with load applied to the studs, the fatigue strength may be lower; and secondly studs will normally be encased in concrete and therefore be uninspectable, so that it seemed prudent to apply a larger than normal factor of safety. Finally, the reason why transverse load-carrying fillet welds of the lap-plate type were downgraded from Class F to Class F2

The constant amplitude database

25

Table 2.1 Co-ordinates of S-N curves (N/mm2) Endurance, N (cycles) Curve

105

106

2 × 106

107(a)

2 × 107(b)

108

Class B Mean Mean – 1 S.D. Mean – 2 S.D.

391 352 317

220 198 178

185 167 150

124 111 100

104 94 84

70 63 56

Class C Mean Mean – 1 S.D. Mean – 2 S.D.

381 333 291

197 173 151

162 142 124

102 89 78

84 73 64

53 46 40

Class D Mean Mean – 1 S.D Mean – 2 S.D.

342 291 248

159 135 115

126 107 91

74 63 53

58 50 42

34 29 25

Class E Mean Mean – 1 S.D Mean – 2 S.D.

320 264 218

149 123 101

118 97 80

69 57 47

55 45 37

32 26 22

Class F Mean Mean – 1 S.D. Mean – 2 S.D.

258 219 185

120 101 86

95 80 68

56 47 40

44 37 32

26 22 18

Class F2 Mean Mean – 1 S.D. Mean – 2 S.D.

231 194 163

107 90 75

85 71 60

50 42 35

39 33 28

23 19 16

Class G Mean Mean – 1 S.D. Mean – 2 S.D.

178 155 135

83 72 63

66 57 50

38 33 29

30 27 23

18 15 13

Class W Mean Mean – 1 S.D. Mean – 2 S.D.

154 134 116

72 62 54

57 49 43

33 29 25

26 23 20

15 13 12

(a) Initial non-propagation stress S0. (b) Cut-off stress for cumulative damage calculations, but see further comments in Chapter 9.

 for joints in air, as given   in BS5400 and BS7608.

was that joints of the simple type which have normally been tested in the laboratory, (i.e. with a transverse weld alone) are virtually never used in practice; such a joint would always have welds along the sides as well and be more akin to a cover plate. A summary of the resulting joint classifications used in the British design rules, at least for the simple ‘basic’ joints, is shown in Table 2.3. As noted

12.6007

1.082 × 1014

3.988 × 1012

3.289 × 10

1.726 × 1012

1.231 × 10

0.566 × 1012

0.368 × 10

C

D

E

F

F2

G

W

12

12

11.5662

11.7525

12.0900

12.2370

12.5169

15.3697 14.0342

2.343 × 1015

B

12

log10a

a

Class

Table 2.2 Values of the relevant constants

26.6324

27.0614

27.8387

28.1770

28.8216

29.0144

32.3153

35.3900

logea

3.0

3.0

3.0

3.0

3.0

3.0

3.5

4.0

m

0.1846

0.1793

0.2279

0.2183

0.2509

0.2095

0.2041

0.1822

log10

0.4251

0.4129

0.5248

0.5027

0.5777

0.4824

0.4700

0.4194

loge

Standard deviation, σ

1.575 × 1011

2.50 × 1011

4.325 × 1011

6.31 × 1011

1.024 × 1012

1.516 × 1012

4.20 × 1013

Mean – 2SD

SmN

38

Cumulative damage of welded joints

previously there are some minor differences in other sets of design rules. At this stage it may be helpful to comment in a little more detail on the results for some of the individual joint types. Continuous longitudinal welds One of the more frequent uses of continuous longitudinal welds occurs in the web to flange joints of fabricated I beams and in general that is the type of specimen that has been used in obtaining fatigue test data. There are three main variants: 1. Joints made with full penetration (of the web) welds; 2. Fillet welds made automatically; 3. Fillet welds made manually. Given that the major stress concentration in welded joints is normally at the weld toe, or in some types of joint at the weld root, but that in continuous longitudinal welds both of those are parallel to the direction of stress, it would be anticipated that all these types of joint would give a relatively high fatigue strength, which has been shown to be the case. In most instances the mode of failure involves crack initiation from weld surface ripples or, in the case of fillet welds, from roughness at the weld root. Weld stop-start positions are particularly likely to be the source of failure since the resulting crater at that position tends to produce the most pronounced change in longitudinal profile. In service, however, this type of joint rarely leads to fatigue cracking since nearly all structures also contain details with much lower fatigue strengths, such as fillet welded stiffeners, etc. Intermittent longitudinal fillet welds In contrast to continuous longitudinal welds, intermittent welds inevitably result in multiple weld ends with every weld end acting as a stress concentration. As a result the fatigue strength is low, with a mean strength at 2 × 106 cycles of about 112 N/mm2. This is equivalent to a reduction in strength of about 35% compared with beams fabricated with continuous manual web-to-flange fillet welds. In other words, intermittent welds should not be used under fatigue loading conditions. Transverse butt welds In Chapter 1 it was noted that the behaviour of a structure under conditions likely to result in fatigue failure was determined primarily by the severity of the stress concentrations which it contained. With this in mind, the problem of joining together two plates may now be considered. Figure 2.4 shows four

The constant amplitude database

39

2.4 Various methods of making welded joints between two plates and the corresponding lines of stress flow.

possible methods by which such a joint could be made, one fabricated with a transverse butt weld, and three with fillet welds. The diagram also shows the influence of the form of the joint on the stress flow between the connected plates. It is quite obvious that the least disturbance to the stress flow occurs with the butt weld and it is therefore to be expected that this form of joint will give a better fatigue performance than would any of the fillet-welded joints. While this expectation is usually fulfilled, the fatigue strength of transverse butt welds can still vary between wide limits. In the absence of weld defects, the major stress concentration in a specimen containing a transverse butt weld with the weld reinforcement left in the aswelded condition occurs at the weld toes. In such a specimen it is therefore from the weld toe, either of the top or of the backing run, that fatigue failure invariably occurs (Fig. 2.5). Thus the propagating crack is initially located either just in weld metal or in the heat affected zone of the parent material, but subsequently it may spread either into the parent metal or into weld metal, depending on the type of joint. The latter mode of failure is associated in particular with butt welds made with a single-sided preparation failing from the toe of the penetration bead, as shown in Fig. 2.5b. The effect of weld defects is outside the scope of this book. However, even with so-called good-quality butt welds (i.e. welds without defects) a

40

Cumulative damage of welded joints (a)

HAZ

Fracture

(b)

Fracture

(c)

Fracture

2.5 Typical modes of fracture in specimens containing transverse butt welds: (a) in heat affected zone material and parent plate initiated at weld toe; (b) and (c) in weld metal initiated at the edge of the weld root.

large number of factors have at one time or another been thought to influence fatigue strength. These include, in addition to the normal variables such as type of loading and the applied stress ratio: • • • • • • • •

composition and mechanical properties of the parent material specimen width and thickness type and method of weld preparation welding process and type of electrode welding position weld shape post-weld machining post-weld heat treatment.

In reality, however, it appears that weld shape is the overriding factor determining the fatigue strength of transverse butt welds, and the influence of many of the other factors listed above depends upon their effects on the shape at the weld toe. Plate thickness also has an important influence, but since that is a factor in all types of transverse welds it is convenient to consider its influence separately later in this chapter. Influence of weld shape As far as is known the first investigators to appreciate the possible influence of weld profile on fatigue strength were Wilson et al. (1941) They found reasonable correlation between fatigue strength and a qualitative assessment of shape based upon the height of the reinforcement and the sharpness of the angle at its edge, with specimens in which the reinforcement flowed by a relatively smooth concave surface into the parent metal giving a higher fatigue strength than specimens in which the reinforcement was high and rough and which had a sharply re-entrant angle at the weld toe. Subsequently, Becker and Rieger (1954) also noted the importance of the shape at the weld toe;

The constant amplitude database

41

they noted that electrodes giving spray transfer produced a smooth gradual junction and higher fatigue strengths than electrodes giving globular transfer, for which the junctions were more abrupt. A more extensive study of the problem was carried out some years later by Newman and Gurney (1959). They tested several types of butt welds, made both by manual and automatic welding, and obtained a wide range of fatigue strengths, which varied from 100–178 N/mm2 at 2 × 106 cycles under pulsating tension loading. As a quantitative measure of reinforcement shape the (obtuse) angle θ between the plate surface and the tangent to the reinforcement at its point of contact with the plate surface was used. Examination of the specimens revealed that this ‘reinforcement angle’ varied along the length of a weld – particularly in manually welded joints – but that failure usually originated at the point of minimum angle. In order to try to make the measurements critical, a few specimens of each test series were selected from those which gave fatigue test results lying close to the relevant S-N curve; these were then sectioned at the point of crack initiation and the angle was measured with the aid of a projection microscope. The measured angles were then plotted against the fatigue strength at 2 × 106 cycles of the particular test series from which the specimen originated, as shown in Fig. 2.6. For the manually welded series the scatter was about 15°, but for the automatic welds it was somewhat less; but it can be seen that all the experimental points lie within a scatter band which can conveniently be located at its upper end by the strengths of plain plate with and without millscale. Figure 2.6 also contains, for comparison, comparable results obtained by other investigators, using reinforcement angles deduced from macrosections of the joints illustrated in the relevant reports. It will be seen that all these results lie within the scatter band. It is not suggested that this rough relationship adequately defines the fatigue strength of transverse butt welds. However, the fact that any relationship exists proves that reinforcement shape is important. It also shows that it may be possible partly to define ‘good’ and ‘bad’ shapes. The difference between the two, as found by Newman and Gurney (1959) is clearly demonstrated by the macrosections of four of the joints which were tested and which are shown in Fig. 2.7. Differences in reinforcement shape are almost certainly responsible for the different forms of fracture which are normally obtained in tests on manual and automatic welds respectively. Automatically welded joints tend to vary comparatively little in profile along their length, and fracture initiation has frequently been noted to occur over a considerable width of the specimen (Fig. 2.8(a)). This necessarily leads to the formation of a ‘continuous’ crack at the weld toe. In contrast, with manual welds, cracks are usually initiated only at isolated points along the toe of the weld (Fig. 2.8(b)) as might be expected in view of the greater variation in reinforcement angle. As a result the failure mode normally involves

42

Cumulative damage of welded joints 300

Fatigue strength at 2 × 106 cycles (N/mm2)

θ°

Plain plate (machined)

Fatigue crack Plain plate (with millscale)

200

100

0 100

110

120 130 140 150 160 Reinforcement angle, θ (deg.)

170

180

2.6 Relation between reinforcement angle and fatigue strength of transverse butt welds.

the propagation of one or more semi-elliptical cracks at the weld toe. It is the combination of the sharper reinforcement angle and the existence of a continuous toe crack that ultimately leads to the much lower fatigue strength of automatic as compared to manual welds. This was subsequently demonstrated by fracture mechanics methods combined with finite element stress analyses of a range of idealised joints. A comparison between the results of fatigue tests on manual and submerged arc welds is shown in Fig. 2.9 and this clearly demonstrates the difference in fatigue strength. It must, however, be noted that it is possible, with care, to adjust the welding conditions in submerged arc welding to give a much more favourable weld shape and consequently a higher strength. The obvious conclusion is, therefore, that if automatic transverse butt welds are to be used in a structure subjected to fatigue loading, then welding trials should be carried out to determine suitable conditions to produce favourable weld profiles. Transverse butt weld on a permanent backing bar From the point of view of fatigue strength, a transverse butt weld made on a permanent backing bar has a worse effective shape than is provided by a

The constant amplitude database

43

(a)

(b)

(c)

(d)

2.7 Macrosections of some of the joints tested by Newman and Gurney (1964). Note particularly the differences in reinforcement angle between the various specimens: (a) manual weld made with rutile electrodes giving a pulsating tension fatigue strength at 2 × 106 cycles of 154 N/mm2; (b) submerged arc (automatic) weld (100 N/ mm2); (c) close square butt manual weld made with deep penetration electrodes (108 N/mm2); (d) submerged arc weld made under controlled conditions to give good reinforcement shape (170 N/mm2).

normal weld reinforcement. Such welds invariably fail from the notch at the junction between the weld metal and the backing bar (Fig. 2.5(c)), and not from the toe of the reinforcement on the opposite side. The strength of such a joint is about 100 N/mm2 at 2 × 106 cycles under pulsating tension loading (Newman and Gurney, 1964) and is therefore about the same as the worst form of butt weld made without a backing bar. Cruciform joints In many respects cruciform joints can be regarded as another form of transverse butt weld. In all cases the form of the specimen that has been used to investigate the strength of this type of joint has been as shown in Fig. 2.10, in which the

44

Cumulative damage of welded joints

(a)

(b)

2.8 Typical fracture surfaces of transverse butt weld specimens, (a) automatic weld showing crack initiation over the major part of the specimen width; (b) manual weld with a few discrete points of initiation growing as semi-elliptical cracks.

load has been transferred from one longitudinal plate to another through a transverse plate by way of two welds placed directly opposite each other. Joints of this general type may in fact be made with full penetration welds, as shown in Fig. 2.10, with partial penetration welds or with fillet welds, and the external appearance may be very similar in each case. The behaviour under fatigue conditions may, however, be different, particularly in so far as the mode of failure of fillet welds and partial penetration butt welds may involve fatigue cracking through the weld, while full penetration welds seem to have resulted invariably in plate failure initiated at the weld toe. For this reason, the discussion here will be confined to full penetration welds; partial penetration welds will be considered in conjunction with fillet welds later in the chapter.

The constant amplitude database

45

99.8 99 98 95 90

Probability (%)

80 Manual welds

70 60 50 40 30 Submerged arc welds

20 10 5 2 1 0.5 0.2 60

80

100

120 140 160 180 200 220 Fatigue strength (N/mm2) at 2 × 106 cycles

240

260

2.9 Comparison of the fatigue strength at 2 × 106 cycles under pulsating tension loading of transverse butt welds made manually and automatically.

2.10 Type of specimen used to investigate the fatigue behaviour of K butt welds.

For this type of joint test results are most numerous for the alternating load condition. At 2 × 106 cycles strengths varying from ±69 to ±138 N/mm2 have been recorded. However, if one ignores the very high strengths obtained in Germany (Kloppel and Wiehermuller, 1957; Wintergest and Ruckerl, 1957), it seems more reasonable to accept a strength within the range of ±69 N/mm2 obtained by Ferguson (1942) to ±97 N/mm2 recorded by Höisveen and Persson (1963). Such figures are certainly more in accord with the corresponding pulsating tension fatigue strength of 100–108 N/mm2 found by Neumann (1960) and 117–128 N/mm2 found by Yamaguchi et al. (1966), Inoue (1974) and Donato et al. (1972).

46

Cumulative damage of welded joints

2.11 K butt weld with load causing bending of the transverse plate.

It is important to appreciate the type of specimen used in these investigations. In many cases in practice the load will be taken out of the longitudinal plate not by another longitudinal plate positioned immediately opposite the first but by bending of the transverse plate as shown in Fig. 2.11. Such a loading condition may easily produce a difference in fatigue strength, even with failure still occurring in the longitudinal plate from the weld toe, because of the more severe local stress conditions. Once again this may be visualised by considering the distribution of the hypothetical lines of stress flow, which must inevitably be more crowded in the region of the stress concentration than they would be if they were able to flow straight through the transverse plate. Few test results exist, however, for specimens loaded in this way. Stud shear connectors In composite construction there is an obvious need for attachments to be welded to the flange to act as shear connectors, and these often take the form of welded studs. In most applications the majority of the studs will be welded to the compression flange of the beam, so that even if the residual stresses induced by welding enable fatigue cracks to initiate and grow to an appreciable size, it is most unlikely that catastrophic failure would occur. However, the use of ‘continuous’ structures necessitates the welding of shear connectors to the tension flange in the regions where the beam passes over the supports. In these regions it would certainly be possible for a fatigue crack to propagate through the entire flange. What is more, such a crack would not be visible since it would be buried in concrete. Although few investigations have been made of the effect on fatigue strength of studs welded to a member, there is excellent agreement between the results that do exist. For the most part the specimens used have consisted of a flat plate, with one or more studs welded to it, subjected to axial loading. The results obtained with such specimens have varied from about 93–120 N/ mm2 at 2 × 106 cycles for arc welded studs tested at R = 0 and, in a single investigation (Selby et al., 1963), the strength at R = –1 was found to be 112

The constant amplitude database

47

N/mm2 (range). Friction welded studs gave 93 N/mm2 at R = 0. These strengths are very similar to those for non-load-carrying fillet-welded joints, which the stud closely resembles. Grinding the transition region of arc welded studs has given improved strengths ranging from 116–137 N/mm2. Unfortunately, this type of specimen does not fully represent a beam flange with stud shear connectors, since the stud is not subjected to a shear force at the same time as the member is stressed, as occurs in actual composite construction. A few comparative tests have, however, been carried out on beam specimens with and without a concrete slab (Selby et al., 1963). These showed some indication that the additional shear force imposed by the concrete did tend to reduce fatigue strength. Fillet welded joints Fillet welded joints can conveniently be divided into two types, those in which the welds are continuous and loaded parallel to their length and those in which the welds may be considered as discontinuous. The former class is typified by the fillet welds used to make web-to-flange joints in the construction of beams and the effect of such welds under conditions of repeated loading was considered above. This section will be restricted to a consideration of the effect of so-called ‘discontinuous’ welds, i.e., those that do not extend over the whole length of a stressed member. This group may be further sub-divided into welds that are either transverse or parallel to the direction of the applied stress, and in either of these cases the welds may be defined as load-carrying or non-load-carrying. In this context ‘load-carrying’ means that a major part of the load is transferred through the weld in shear while a ‘non-load-carrying’ weld may be defined as any weld on a stressed member, which does not carry the load, or an appreciable part of it, into the member. It should be realised, however, that a non-load-carrying weld, as defined here, is not entirely load free since it will be strained by the load in the stressed member on which it is laid. Typical examples of these four types of joint are shown in Figs 2.12 and 2.13, which also show, qualitatively, the stress distributions associated with each under axial tensile loading. Figure 2.12 refers to welds which are transverse, and Fig. 2.13 to welds that are parallel, to the direction of the applied stress. In the former case the stress distribution, although uniform along the length of the weld (i.e. across the width of the specimen) varies through the specimen thickness since the toes of the welds form stress concentrations. On the other hand, with longitudinal fillet welds it is the weld ends that form the stress concentrations, and the stress distribution is non-uniform both over the specimen width and through its thickness. In the case of the load-carrying welds it should also be noted that the stress distribution in the welds is not uniform. The transverse weld has a stress concentration

48

Cumulative damage of welded joints

(a) Non-load-carrying joint.

(b) Load-carrying joint.

2.12 Qualitative stress distributions in transverse fillet welded joints.

Shear stress in the weld

(a) Non-load-carrying joint.

(b) Load-carrying joint.

2.13 Qualitative stress distributions in longitudinal fillet welded joints.

at the weld root, while the longitudinal weld tends to transmit the majority of the load near its two ends and a lesser amount at the centre of its length. From a consideration of these stress distributions the possible modes of failure to be expected under repeated loading are obvious. If failure occurs in the plate, as will invariably happen with non-load-carrying fillet welds, the fatigue crack will be initiated at one of the weld ends in the case of a longitudinal weld or at the weld toe with a transverse weld. But in loadcarrying fillet-welded joints failure may either occur in the plate, with the cracks being initiated at the same points as in the case of non-load-carrying welds, or in the welds themselves. In the latter event failure will be initiated

The constant amplitude database

1

1

2

1

49

1

2

3

2.14 Modes of failure of fillet welded joints under fatigue loading. The black spots indicate points of crack initiation.

(a) Specimen with gussets welded on plate surfaces.

2.15 Typical specimens used to study the fatigue behaviour of joints involving longitudinal non-load-carrying fillet welds.

at the weld ends of a longitudinal weld or at the weld root of a transverse weld. All these modes of failure are shown diagrammatically in Fig. 2.14. Longitudinal non-load-carrying welds on the plate surface In the investigation of the fatigue strength of joints involving non-loadcarrying fillet welds on the plate surface the type of specimen which has usually been employed is as shown in Fig. 2.15. There have, however, been several variations of a comparatively minor nature, particularly with respect to the number of attachments on each surface. Nevertheless, the majority of tests have involved specimens with two gussets, one on each surface opposite each other, in order to maintain symmetry. A theoretical fracture mechanics study of this type of joint, Smith and Gurney (1986) suggested that, at least for specimens 13 mm thick and 152 mm wide, attachment length has a significant influence on fatigue life, particularly for attachments less than 150 mm long. It showed that fatigue life tended to increase as attachment length decreased, with a theoretical

50

Cumulative damage of welded joints

increase in life of about 70% on reducing length from 150 to 50 mm. For longer attachments the effect was much smaller with a reduction in length from 300 to 150 mm only giving a theoretical increase in life of about 10%. In general terms these findings are in agreement with available experimental results, which are summarised in Fig. 2.16. In the same study (Smith and Gurney, 1986) attachment height and thickness and weld leg length were all found only to have a relatively minor influence on fatigue life. However, rather surprisingly, an increase in main plate thickness was found to give a theoretical increase in fatigue life. This is in distinct contrast to the well proven finding for transverse attachments where an increase in thickness leads to a decrease in fatigue strength. (Part of the increase derives from the larger value of af in the larger joints, but the effect is still apparent even if a constant value of af is used.) Further work is still required to confirm this finding experimentally. Transverse non-load-carrying welds The type of specimen most often used in the study of transverse non-loadcarrying welds is shown in Figs 2.17(a) and 2.17(b) and consists merely of a gusset or gussets welded transversely to either one or both surfaces of the main plate. These types of specimen have been tested by a very large number of investigators and, in comparison with joints with longitudinal attachments, the results have been much more scattered. Based on a statistical analysis of the results of 454 individual tests under pulsating tension loading the mean fatigue strengths at 105 and 2 × 106 cycles were 315 and 105 N/mm2 respectively, with the limits of the scatter band, as represented by the mean ±2 standard deviations of log N, being 165 and 66 N/mm2 at 2 × 106 cycles. This may be compared with a scatter range from 114 to 68 N/mm2 at 2 × 106 cycles for specimens with longitudinal attachments. The influence of joint geometry on the fatigue strength of this type of joint has been studied in considerable detail, both theoretically, by means of fracture mechanics, and experimentally. As a result it has been found that the main plate thickness, the attachment thickness and the weld size can all have an important influence on fatigue strength. The influence of main plate thickness is considered in more detail later, since it is relevant to all types of joints involving welds transverse to the direction of stress (e.g. transverse butt welds, K butt welds, etc.). However, it may be helpful to summarise here the work relating to the influence of attachment thickness and weld size. It has in fact been found that these two variables can, to a very large extent, be considered as one, since the critical dimension, as far as attachment size is concerned, seems to be the overall distance from weld toe to weld toe, namely L (see Fig. 2.18). In general it is convenient to think in terms of the ratio L /T, where T is the main plate thickness.

Stress range, Nmm–2

40 104

50

60

70

80

90

100

150

200

300

350

400

3 4 5

6 7 8 9 105 2 3 4

5 6 7 8 9 106

Endurance, cycles

2

3

4

2.16 Results for longitudinal attachments, showing influence of length of stiffener.

2

50 mm long attachments 150 mm long attachments 305 mm long atttachments

305 mm long atttachments

5 6 7 8 9 107

150 mm long attachments

50 mm long attachments

52

Cumulative damage of welded joints

(a) (b) Specimens with welded gussets

(c) Specimens with welded pad plates.

2.17 Typical specimens used to study the fatigue behaviour of joints involving transverse non-load-carrying fillet welds. L

45°

a T

b

w

2.18 Joint geometry used in theoretical analysis of transverse nonload-carrying fillet welded joints.

By means of fracture mechanics analysis it has been found that the relationship between the fatigue strength for a particular plate thickness and ratio L /T is of the form shown in Fig. 2.19. This shows the results both for a semi-elliptical toe crack and a continuous toe crack. It should be noted that the two relationships are for slightly different thicknesses, namely 20 and 22 mm respectively, because they are the thicknesses at which the rate of decrease of fatigue strength of geometrically similar joints (constant L/T) changes; the reason for this change is not yet understood. Although the two relationships are, in principle, very similar to each other, the strength of the joint with a continuous toe crack is significantly lower than that with a semi-elliptical crack. It will be seen that, in both cases, fatigue strength seems to reach a minimum value at L / T = 2–2.7. In other words small attachments (i.e. those having L / T <2.0) give higher strengths. In this context it may be noted that

The constant amplitude database 140

53

m=3 T = 20

130 S20 = 101.64 ( LT ) –0.232 120

Fatigue strength at 2 × 106 cycles N/mm2

110

S20 = 98.75 ( LT ) –0.111

100

S20 = 89.2 90 Semi-elliptical cracks

80

Continuous toe cracks

70

60 0.1

1

10

L T

2.19 Theoretical relationships between fatigue strength and L /T for semi-elliptical and continuous cracks in transverse non-load-carrying fillet welds (contact thickness).

the ‘basic’ type of joint which has usually been tested has tended to have L /T ≈ 2.0 with T = 13 mm and L = 26 mm (t = 13 mm and w = 6.5 mm, or t = 10 and w = 8). A set of test results obtained with this ‘basic’ joint geometry is shown in Fig. 2.20. Given these results for several thicknesses it is easy to deduce the theoretical relationship between fatigue strength and plate thickness; obviously this results in a family of curves depending upon the value of L /T (Fig. 2.21). Although this figure relates to semi-elliptical cracks, a very similar (but lower) set of curves can be derived for continuous cracks. Alternatively, Fig. 2.22 shows the relationship between fatigue strength and thickness for a range of values of L (instead of L/T). Again, the results for continuous

54

Cumulative damage of welded joints

260 220

Stress range, N/mm2

180

140

100

10 Leg = 8

80

13

60 104 1.5 2

3

4 5

Ref. This project Ref. 17

105

1.5 2

3

4

5

106 1.5 2

3

4 5

107

Endurance, cycles

2.20 Summary of experimental results for ‘basic’ joint geometry, aswelded specimens (T = 13, L = T + 2W = 26 mm).

m = 3.0

Fatigue strength of 2 × 106 cycles, N/mm2

150 140 130 120

L T

110

0.29

100

0.45

90

0.70 1.00 1.32 2.00

80

70

60 10

12.5

15

20

25 40 Thickness, mm

50

60

80

100

2.21 Theoretical fatigue strength of geometrically similar joints with semi-elliptic toe cracks under axial loading.

The constant amplitude database

55

150 m=3

Fatigue strength of 2 × 106 cycles, N/mm2

140 130

L 13

120 110

20

100

29 41

90

58 80 100 250

70

60 10

12.5

15

20

25 40 Thickness, mm

50

60

80

100

2.22 Influence of attachment size, and plate thickness on the fatigue strength of joints with semi-elliptic toe cracks under axial loading (m = 3).

cracks are similar, but lower. Effectively, therefore, these results confirm (as discussed later) that fatigue strength would be expected to decrease with increasing plate thickness. However, they also show that this ‘thickness effect’ can be mitigated by reducing the attachment thickness (L). Transverse load-carrying fillet welds The investigation of the fatigue strength of this joint has involved two distinct types of specimen, and these are shown in Fig. 2.23. The main difference between them is that, in the specimen forming a lap joint (Fig. 2.23(a)) the load is transferred through the weld into the cover plate, whereas in the cruciform joint (Fig. 2.23(b)) the load distribution in the centre plate will be less uniform. So far as failure from the weld toe is concerned, however, there is no significant difference between the two and the strengths that have been measured experimentally have been very much the same. Typically the mean fatigue strength at 2 × 106 cycles has been found to be in the range 85–99 N/ mm2. As might be expected, this is somewhat lower than the comparable strength for K butt welds; although the external appearance may be much the same it is obvious that the existence of the lack of penetration in the fillet welded joint requires a much greater deviation in the lines of stress flow through the joint, and hence produces a larger stress concentration. This has been confirmed by finite element analysis (Gurney, 1976).

56

Cumulative damage of welded joints

(a) Lap joint.

(b) Cruciform joint.

2.23 Typical specimens with transverse load-carrying fillet welds.

In designing joints of these two types it is necessary to consider carefully the question of weld size. The critical size is that which gives an equal chance of failure in the weld or in the plate, and for design purposes the ideal weld size is the one that will just ensure plate failure. At that stage a further increase in weld size will not increase the fatigue strength, while a decrease could lead to weld failure and a lower-than-optimum strength. In the case of cruciform joints, which occur much more commonly than lap plate joints, it is possible to use fracture mechanics methods to derive design stresses for weld throat failure that are consistent with the corresponding stresses for weld toe failure; indeed this provides a good example of the use of the generalised stress parameter ∆σ* (see eqn [1.13]). Maddox (1974) and, subsequently, Gurney and MacDonald (1988) have analysed the results in this way, making use of the K calibration for weld failure put forward by Frank (1971) and assuming the slope m = 3.0. Frank’s K calibration is:

∆K =

πa  A 1 + A 2 (a / W) ∆ σ ( π a)  see 2W (1 + 2S/T) 

1/2

[2.3]

where ∆σ is the stress in the plate, S, T and W are as given in Fig. 2.24 and the coefficients A1 and A2 are:

( ) ( ) + 3.696 ( TS ) – 1.874 ( S ) + 0.415 ( S ) T T S A = 0.218 + 2.717 ( S ) – 10.171 ( ) + 13.122 ( S ) T T T – 7.755 ( S ) + 1.785 ( S ) T T

A1 = 0.528 + 3.287 S – 4.361 S T T 4

2

3

5

2

3

2

4

5

Thus in eqn [1.13] we have: I=

∫

1.0

a i /W

(

 S  1+ 2 T 

) ( ) 3

d a W

( )

3/2 3  A + A a   π a sec π a   1 2  W  W 2W   

[2.4]

The constant amplitude database

57

S θ

W ai

σ

σ

T

2.24 Load-carrying transverse fillet welds as considered in the fracture mechanics analysis of Frank (1971) and Maddox (1975).

and

∆σ* = ∆σ (√W/I)1/3.

[2.5]

Now, based on an analysis of 504 results for joints with virtually zero weld penetration subjected to tensile loading and failing in the weld, it was found that the mean and mean –2 standard deviation curves could be represented by: (∆σ*)3 · N = 6.255 × 1012 and 1.272 × 1012 respecively.

[2.6]

For convenience this can be written as: (∆σ*)3 · N = U

[2.7]

where U is the value relating to the particular curve under consideration. Using eqn [2.5] to eliminate ∆σ* this becomes: ( ∆σ ) 3

W ⋅N=U I

[2.8]

Equally, for failure from the weld toe we can write, (from eqn [2.2]) (∆σ)3 · N = V

[2.9]

where V depends upon the chosen probability of failure and can be derived from the relevant design standard (BS 5400 or BS 7608). Thus, eliminating ∆σ between eqns [2.8] and [2.9] V⋅

W =U I

[2.10]

Hence, if we insert the values of U and V corresponding to the relevant mean

58

Cumulative damage of welded joints

–2 standard deviation curves, we have U = 1.272 × 1012 (from eqn [2.6]) and V = 4.32 × 1011 for curve F2 in BS 5400 or BS 7608, so that, from eqn [2.10]: W 1.272 × 10 12 = = 2.94 I 4.32 × 10 11

[2.11]

Alternatively, using the corresponding mean values: 12 W = U 6.255 × 1012 = 5.08 I V 1.231 × 10

[2.12]

W (say X) can be selected, it becomes Given that a suitable value of I possible to relate the corresponding stresses in the plate and the weld throat. From Fig. 2.25 it will be seen that, since the value of I is related to the joint geometry, it becomes possible to define a unique relationship for fillet welds with zero penetration, and replotting that relationship in terms of ln (I) versus ln (ai/W) (see Fig. 2.26) it is apparent that, to a close approximation:

0.9 H Tp

0.8 1.0 0.7 0.8 0.6 0.6

0.5

Fillet welds

ai W

0.4 0.4

0.2

0.3

0.2

0.1

0.01

0.1

1.0

10

I

2.25 Values of I as a function of initial crack size and joint geometry.

The constant amplitude database

59

0.6 0.5 0.4

ai W

0.3

0.2

0.01 1.5 2

3

4 5

0.1 1.5

2

3 4 5

1.0 1.5 2

3

4 5

10.0

Value of I

2.26 Values of I for fillet welds with zero root penetration.

 ai   W

5.37

⋅ I = 0.002

[2.13]

so that, eliminating I by introducing the value X ai   = 0.002 X  W  W 

1/5.37

[2.14]

Now, if the stress in the plate is σ and the stress on the weld throat is σw we have, since the weld throat area of two fillet welds is (2 × 0.7S) σ · T = σw (1.4S)

[2.15]

and since T = 2ai and S = W–ai

σw = σ

2a i W a 1.4  1 – i   W

[2.16]

so that, eliminating ai/W with eqn [2.14]

(0.002 X)1/5.37 σw = σp 0.7{( W )1/5.37 – (0.002 X)1/5.37}

[2.17]

60

Cumulative damage of welded joints

 which can be evaluated for the relevant values of W and X  = 

W . I 

It must be noted, however, that W is not directly proportional to T since (see Fig. 2.24) W = S + 0.5T

[2.18]

which gives, using eqn [2.15] to eliminate S T=

W     1  0.5 +  σ  1.4  w    σ   

[2.19]

Given that the values of (σw /σ) for given values of W and X are defined by eqn [2.17] it is easy to derive the corresponding values of T from eqn [2.19]. The resulting relationships between σw /σp and T are shown in Fig. 2.27.

1.0 0.9 0.8

0.7 Value of X 0.6 σw σp

7.78 5.08 0.5 2.94

0.4 1.34

0.3 3

4

2.27 Values of

5

10 13 2 3 4 Plate thickness (Tp) mm

5

σW as a function of plate thickness and X. σP

100

The constant amplitude database

61

Table 2.4 Fatigue strengths required for the weld throat in fillet welds Assumed value of X=

2.94 5.08

W I

σW σP

Required value of σw at 2 × 106 cycles, N/mm2

at T = 13 mm

Mean

Mean –2 standard deviation

0.58 0.68

49 58

35 41

From this it will be seen that, at the basic thickness of 13 mm used for deriving most of the British fatigue design curves, the values of σw/σp corresponding to the two values of X considered here (namely 2.94 (eqn [2.11]) and 5.08 (eqn [2.12])) are 0.58 and 0.68 respectively. Thus, knowing that the mean and mean –2 standard deviation fatigue strengths at 2 × 106 cycles for Class F2 in the British design rules (relating to weld toe failure in this type of joint) are 85 and 60 N/mm2 respectively, it is easy to show that the corresponding strengths required for Class W (weld throat failure), expressed in terms of stress on the weld throat (σw) become those given in Table 2.4. It will be seen that the existing strengths for Class W in the British design rules, namely 57 and 43 N/mm2 at 2 × 106 cycles, are very similar to the values derived for X = 5.08 (i.e. equating the means of the scatter bands for toe and root failure). However, it might be more logical to reduce them to the values corresponding to the mean –2 standard deviation design curve, namely 49 and 35 N/mm2, or, for consistency, to the values proposed for the Eurocode, which are 50 and 36 N/mm2. A check on a small database of some 56 results for partial penetration welds, rather than fillet welds with no penetration, showed that these proposed design stresses would be equally acceptable for the design of those joints. Longitudinal load-carrying welds Work on longitudinal load-carrying fillet-welded joints has been confined almost exclusively to specimens of the general form shown in Fig. 2.28(a), although there have been a large number of minor variants. In particular the cover plates, shown in the diagram as flat plates, have sometimes consisted of rolled steel sections such as channels or angles and there has been a large variation in the relative dimensions of the ‘main’ and ‘cover’ plates. The alternative type of specimen, which has occasionally been used, simulates what is usually referred to as egg-box construction and is shown in Fig. 2.28(b). Either the two gripped plates, or the central plate, or all three, may have slots machined in them to receive the adjacent plates. Most usually it has been all three plates that have been slotted so that the ends of the slots

62

Cumulative damage of welded joints

(a) Cover plate type

(b) Egg-box type

2.28 Typical specimens with longitudinal load-carrying fillet welds.

are finally positioned in the middle of the weld length; otherwise the slot ends coincide with weld ends and form an enhanced stress concentration. These types of joints have low fatigue strengths and, in general, are rarely used, certainly in the form shown in Fig. 2.28. When they do have to be used the joint normally includes a transverse fillet weld across the end of the lap plate. Beams with welded flange cover plates One of the simplest and most economical methods of increasing the section modulus of a girder, particularly if such increase is only required over a comparatively short length, is to add flange cover plates. However, in exactly the same way as other fillet-welded attachments, flange cover plates produce severe local changes of geometry and the resulting stress concentrations can exert large effects on fatigue performance. The main stress concentration produced by a cover plate occurs, of course, at its ends. Hence, if the cover plate extends over the full length of the beam it might reasonably be expected that it would not lower the fatigue strength very much and this has been confirmed by Wilson (1948). In such beams, with the cover plate attached by continuous manual fillet welds, fatigue cracking initiated at the weld stop-start positions, with much the same fatigue strength as that found for continuous manual web-to-flange fillet welds. In the same investigation Wilson also showed that attaching the cover plates with intermittent welds resulted in a large decrease in fatigue strength even though the cover plates themselves extended over the full length of the beam. This is because each weld end forms a stress concentration, incidentally on a plate edge, and fatigue cracks initiate at the weld ends. The actual fatigue strength obtained was 100 N/mm2, this being the stress at the plane of the weld. This is in good agreement with the strength found for plain beams with intermittent web-to-flange welds. However, the reduction in fatigue strength which results from the use of partial length cover plates is even more drastic. Wilson demonstrated this fact quite clearly when, in the same investigation he tested three series having partial length cover plates attached to the beam with intermittent welds along their length and a continuous fillet weld across the ends. The nominal stress at the ends of the cover plates was arranged to be the same as that at mid-

The constant amplitude database

63

(a)

(b)

2.29 (a) Type of specimen used to investigate the effect of discontinuous longitudinal butt welds (b) specimen with gussets welded on plate edges.

span but all these specimens failed at the ends of the cover plates. The fatigue strengths which he obtained with these and other test series, which also had partial length cover plates but which were welded with continuous fillet welds, were all within the range 43–75 N/mm2 at 2 × 106 cycles under pulsating tension loading. Extensive work carried out subsequently, mainly in the USA (Frank and Fisher, 1969; Heins and Murad, 1971; Murad and Heins, 1972; Schilling et al., 1975) has more-or-less confirmed the validity of Wilson’s earlier results. Longitudinal gusset on a plate edge Work on this form of joint has involved two main types of specimen, one with a gusset or gussets butt welded to the edge of the main plate (Fig. 2.29(a)) and one with a pair of gussets fillet welded to the edge (Fig. 2.29(b)). In both instances a feature of the test results has been a remarkable lack of scatter and a relatively low fatigue strength. It will be noted that this type of joint has distinct similarities to a joint with transverse non-load-carrying fillet welds but with the plate width (W) being equivalent to the plate thickness in the transverse fillet welded joints. This suggests that fatigue strength would be expected to decrease as plate width increases, and in a limited test programme on specimens of four different widths, all with 150 mm long attachments, that has been shown to be the situation (Fig. 2.30). In passing it may be noted that the low fatigue classification (Class G) in the British, and other, fatigue design rules for ‘a plate with edge attachments’ is probably a direct result of the fact that the classification is based upon results for relatively wide specimens (typically 125–150 mm).

64

Cumulative damage of welded joints 300 260 150 220

Stress range, N/mm2

W 180

Glass G mean curve (BS 5400)

140 120 W = 40 100

i

W = 200 80

60

Plate width, W 40 mm 80 mm 200 mm 105

1.5

W = 80 W = 125

2

3 4 5 106 Endurance, cycles

1.5

2

3

4

5

2.30 Test results for plates with attachments welded to the plate edge (stress relieved, R = 0).

The relationship between fatigue strength and plate width found in these tests is summarised in Fig. 2.31. Assuming that the behaviour is similar to that predicted for transverse attachments (Fig. 2.19), and also that the result for 200 mm wide specimens is correct, it is interesting to speculate that the lower limit of strength for this type of joint is reached when the attachment length is equal to the plate width. Common sense would certainly suggest that attachment length, or more accurately the ratio between attachment length and plate width, must be an important parameter. In any case, however, it is obvious that the reduction in fatigue strength with increasing width found in these tests is not covered by the existing thickness correction rules. They would require the design stress to be reduced as the plate thickness increased, yet in this instance one would have expected the plate thickness to be largely irrelevant. These results therefore suggest that the current requirements for a ‘thickness correction’ on fatigue strength are, at best, a misnomer and, at worst, wrong. What is probably needed is a correction to take account of size ‘in the direction of primary crack propagation’. In the case of welds transverse to the direction of stress that is obviously the member thickness, but with welds parallel to the stress direction (i.e. with a localised crack initiation site) it may well be the width of the member. At this stage it is important to recognise

The constant amplitude database

65

120

Fatigue strength at 2 × 106 cycles, N/mm2

110 100 90

80

Possible relationship for L = 150 (?)

70

Slope = – 0.23

60

50 40

50

60 100 Main plate width, mm

150

200

2.31 Influence of plate width on the fatigue strength of plates with edge attachments (see 2.30).

that all the data used in the formulation of the original ‘thickness correction’ rules (discussed later, see Fig. 2.33) and most of the check data produced subsequently, have involved transverse welds.

2.3

Influence of plate thickness

It has been known for a long time that, so far as fatigue strength is concerned, plate thickness was likely to be a relevant variable under bending stresses. This is due to the fact that the stress gradient through the thickness of a ‘thin’ specimen will be steeper, and therefore less damaging, than in a ‘thick’ specimen. However, it is only in the relatively recent past that many tests have been carried out on welded joints in bending, but they have indeed confirmed that thickness matters. By way of example Fig. 2.32 shows a set of results for transverse non-load-carrying fillet welded joints of a range of thicknesses tested in bending. However, such a situation does not occur in very many structures, so that the comparatively limited testing resources available have tended to be concentrated on tests under axial loading. It has always been assumed, for design purposes, that the use of fatigue strengths obtained under axial loading would be safe, even if conservative, under bending stresses. In consequence fatigue design rules for welded joints have traditionally been based upon results obtained under axial loading.

66

Cumulative damage of welded joints

300 260 T = 6.4

Stress range, N/mm2

220 180 t = 13 140

T = 38

w T=8

T 100

80

T w 6.4 8 8 8 13 10 38 10

60 104 1.5 2

3

L Ref. 29 30 29 30 33 27 33 27

4 5

105 1.5 2 3 4 5 Endurance, cycles

106 1.5

2

3

4

5

107

2.32 Results for transverse non-load-carrying fillet welds with t = 13 mm tested in bending at R = 0.

It has also long been known (for example Phillips and Heywood (1951)) that the fatigue strength of notched, but unwelded, specimens is size-dependent. However, it was not recognised that the same might also be true of welded joints, and it was not until early in 1977 that it first became apparent, on the basis of theoretical fracture mechanics calculations, that the fatigue strength of welded joints could be affected by plate thickness, even when they were subjected to axial loading (Gurney, 1977). In simple terms the reason for this is that the rate of fatigue crack growth is predominantly a function of the crack depth (a), while the stresses acting on the crack are a function of a/B, where B is the plate thickness. Consequently, in thick plates the stress for a given crack depth is larger than for the same sized crack in a thinner plate, so that the rate of crack growth is higher. Hence, making the reasonable assumption that the size of the initial defect will be independent of thickness, the fatigue strength will obviously be lower for the thicker joint. In the light of these calculations some fatigue tests under axial loading were subsequently carried out by Johnston (1978) on specimens with transverse non-load-carrying fillet welds fabricated from plates of various thicknesses. In these specimens the thickness of the attachments was the same as that of the main plate, but the weld size was varied. In these tests it was found that there was a general tendency for fatigue strength to decrease as plate thickness increased. In view of the theoretical analysis and of these test results it therefore came as no particular surprise that several investigators (Booth,

The constant amplitude database

67

2.4 (R = –1)

Relative fatigue strength, N/mm2

2.0

Tubular T joints

1.8

Wildschut – data normalised to 40 mm thickness, instead of 32 mm Morgan – data normalised to 38 mm thickness, instead of 32 mm

1.6

1.4

Sl 1.2

op

e

=

–

1 /4

1.0

0.8

0.6 5

10

20 Thickness, mm

50

100

2.33 Summary of results used in deriving the thickness correction factor in the British Department of Energy Guidance Notes.

1978; Haibach et al., 1978; Dijkstra and Hartog, 1978) in the combined European Offshore fatigue programme also found a tendency for fatigue strength to decrease as thickness increased. This was found to be the case for transverse K butt welds (Booth, 1978) and transverse non-load-carrying fillet welds (Haibach et al., 1978) tested in bending, and also for tubular T joints (Dijkstra and Hartog, 1978) under axial brace loading. Effectively, apart from some limited tests by Morgan (1983) the work outlined above represented the ‘state of play’ at the time that the Department of Energy Guidance Notes were being drafted, these being the first official design rules to incorporate thickness effect requirements. The results, expressed in terms of relative fatigue strength normalised to a thickness of 32 mm, are summarised in Fig. 2.33. It was on the basis of those data that it was decided to introduce a thickness correction factor of the form:

68

Cumulative damage of welded joints

t 1/4 S = SB  B   t  where S is the t is the SB is the tB is the

[2.20]

fatigue strength of the joint under consideration thickness fatigue strength of the joint using the basic S-N curve thickness corresponding to the basic S-N curve.

For design purposes it was assumed that for tubular nodal joints tB = 32 mm but that for other types of joint tB = 22 mm. In passing it may be noted that this is consistent with the data which were actually used for tubular joints, but for other types of joint the chosen value of tB is very much an illogical approximation; most of the data for such joints were derived using approximately 13 mm thick specimens. Following the ‘discovery’ of the thickness effect in welded joints, research on the problem, including both theoretical fracture mechanics and experimental work, was carried out by several workers. It has been adequately summarised elsewhere (De Back et al., 1989) and there appears now to be general agreement that the thickness effect is ‘real’, although there is still some argument about its severity. However, as far as transverse fillet welded joints are concerned, an important finding has been, as noted previously, that fatigue strength is affected not only by the thickness of the main plate but also by the size of the attachment as represented, in particular, by its length (L) in the direction of the applied stress (see Fig. 2.19). In view of this situation it seems reasonable to suggest that the thickness correction rule expressed by eqn [2.20] should be re-written in terms of an ‘apparent thickness’ (T′), where T′ is dependent both on the plate thickness (T) and on the length (L). The rule then becomes: t k S = SB  B   T′ 

[2.21]

where, for axial loading if

L ≤ 2T, T′ = 0.5 L

if

L ≥ 2R T′ = T

and for bending if

L ≤ 0.65 T, T′ = 1.55 L

if

L ≥ 0.65 T, T′ = T

This proposed rule does not yet exist in any official set of design rules, but it does seem both simple to apply and a good approximation to the facts. By way of example Fig. 2.34 shows the results obtained in a single investigation

The constant amplitude database

69

140 T 13 25 38 50 100

Fatigue strength at 2 × 106 cycles, N/mm2

120

100 90 k = 0.25 80

70 k = 0.29 60

50 5

10

20 50 Apparent thickness (T I ), mm

100

2.34 Relationship between fatigue strength at 2 × 106 cycles and apparent thickness for as-welded specimens, axial loading (from Gurney, 1991).

plotted in this way and it is obvious that it leads to a consistent interpretation of the data. The slope of the curve shown is k = 0.29, but it relies very heavily on a single data point at T′ = 100 mm. The dashed curve with k = 0.25 gives just as good a fit to the results in the more usual range of medium ‘apparent thickness’, so the use of the ‘simple’ value k = 0.25 commends itself. Although this study was concerned solely with transverse non-load-carrying fillet welds, it would seem entirely logical to assume that the results, and the above recommendations, would be equally applicable to transverse full penetration K butt welds. The overall external shape of such joints is, after all, virtually identical to a transverse fillet weld. Similarly, it would be surprising if the results were not also applicable to transverse butt welds. The majority of such joints will tend to have an overall width of weld reinforcement less than 2T, which implies that the thickness correction should be based upon the reinforcement width rather than the plate thickness (under axial loading). While it would clearly be desirable to check this prediction experimentally, it seems likely that transverse butt welds are probably critical only on rare occasions. Hence the introduction of a revised rule based on intuition is probably justifiable.

70

2.4

Cumulative damage of welded joints

Influence of mean stress

At the beginning of this chapter it was pointed out that the constant amplitude S-N curve appearing in design Standards is based upon test results obtained under essentially pulsating tension loading; such results are, after all, far more numerous than those relating to any other mean stress or stress ratio. Nevertheless, since variable amplitude loading necessarily involves stress cycles with a wide variety of mean stresses and stress ratios it may be helpful briefly to summarise their influence. For all practical purposes the relevant information has been obtained primarily for transverse butt welds and for discontinuous fillet welded joints and it is convenient to start by considering the results for these two types of joint separately. Transverse butt welds Because of the large number of variables which may affect the fatigue strength of transverse butt welds it is apparent that, as in the study of the other parameters, the only way to determine the effect of mean stress is to compare results obtained from test series which were, in all other respects, nominally similar. The comparison is shown in Fig. 2.35 where the ratios of the fatigue strength range for alternating and half tensile loading to the fatigue strength for pulsating tension loading, all at 2 × 106 cycles, have been plotted in the form of a normal probability diagram. In this plot the probability of a particular ratio occurring was computed as P = [n – (m-1/2)]/n where m is the order number of the result under consideration and n is the total number of results. Some of the results included in this diagram for alternating loading, and particularly those obtained in one of Wilson’s investigations (Wilson et al., 1943), are based on a very small number of test specimens. However, the mean of his results is the same as the mean of all the results, so that it does not seem necessary to reject them. Apart from the ratio of 1.79 obtained for alternating loading by Baron and Brine (1963) it will be seen that the results can be represented reasonably satisfactorily by straight lines, indicating that they can be considered as normally distributed. For half tensile loading the mean value of the ratio is 0.815 with 95% probability limits of 0.735 and 0.895, while for alternating loading the mean value is 1.235 with 95% probability of it lying between 1.045 and 1.425. Hence, if we consider the mean values, the fatigue strengths corresponding to a pulsating tension fatigue strength at 2 × 106 cycles of σ1 would be: ±1/2(1.235)σ1 = 0.617σ1 (i.e. stress range = 1.235σ1)

The constant amplitude database

71

99.5 99 98 FS under alternating loading 95

FS under pulsating tension loading

90

Probability (%)

80 70 60 50 40 30 20 10 5 2 1 0.5

FS under half tensile loading FS under pulsating tension loading 0.6

0.8

1.0 1.2 Value of ratio

1.4

1.6

1.8

2.35 Effect of stress ratio on the fatigue strength at 2 × 106 cycles of transverse butt welds.

and 0.815σ1 to (2 × 0.815)σ1 = 0.815σ1 to 1.630σ1 (i.e. stress range = 0.815σ1) These fatigue strengths refer, of course, to results obtained by using relatively small specimens with (probably) insignificant residual stresses. In the presence of high tensile residual stresses the stress ranges under half-tensile, pulsating tension and alternating loading would be expected to be more nearly equal, as explained in Chapter 3. For shorter endurances there is very much less information available on the effect of mean stress. However, from Wilson’s results, it appears that for alternating loading of the ratio of the fatigue strength range at 105 cycles to the corresponding pulsating tension fatigue strength tends to be rather greater than at 2 × 106 cycles. Fillet welded joints For fillet welded joints a similar situation exists with by far the largest number of series having been tested at R = 0 and relatively few investigations

Type of specimen R=0 R = 0.5

81

42

171

154

242 204

211 246 196 105 97 76 128 177 113 140

139 148 147 68 56 50 85 114 69 93 216

185

– – – – – –

– –

Transverse non-load-carrying fillet welds

108 150 154

63 120 100

Longitudinal non-load-carrying fillet welds

R = –1

Fatigue strength at 2 × 106 cycles (N/mm2)

Table 2.5 Effect of stress ratio on the fatigue strength of fillet-welded joints

1.32

1.23

1.50 1.29 1.15 1.32 1.33 1.29

1.31 1.21

1.05

1.60 1.30

1.17

At R = –1 At R = 0

0.77

0.82

– – – – – –

– –

0.95

0.81 0.66

0.79

At R = 0.5 At R = 0

Ratio of fatigue strength ranges

–

Failed from root

– – Tested in bending – – –

– –

–

– –

–

Remarks

Type of specimen

Table 2.5 Continued

R=0 R = 0.5

65 54 54

34 28 34

103 113 103 232 96 162

71 79 54 128 59 93

Transverse load-carrying fillet welds

66

38

167 –

–

– – –

–

– –

113

Longitudinal load-carrying fillet welds

R = –1

Fatigue strength at 2 × 106 cycles (N/mm2)

1.23 1.14

1.11

1.37 1.40 1.05

1.25

1.05 1.03

1.16

At R = –1 At R = 0

0.87 –

–

– – –

–

– –

0.85

At R = 0.5 At R = 0

Ratio of fatigue strength ranges

–

Plate failure

– – Failure in weld

–

Failed in lap plates Failed in lap plates

–

Remarks

74

Cumulative damage of welded joints

having also included tests at R = 0.5 and R = –1. A summary of the results obtained under these two types of loading is shown in Table 2.5. The range of fatigue strengths that have been obtained under pulsating tension loading for fillet welded joints is quite large and there is no reason to believe that the corresponding ranges for other load cycles would be any smaller. However, since insufficient results are available to specify these ranges the table also shows the ratios of the fatigue strengths under alternating and half tensile loading to those obtained under pulsating tension loads in the various investigations. These ratios are measures of the effect of stress ratio alone with all other variables held constant. It is very important to remember, however, that nearly all the results shown in Table 2.5 were obtained in tests on relatively small specimens which probably did not contain high tensile residual stresses. They therefore need to be applied with care to large aswelded structures and this problem will be considered in more detail in Chapter 3. For the time being attention is drawn particularly to the result Gregory (1972) obtained for transverse non-load-carrying fillet welds which were spot heated to induce high residual stresses artificially; it will be seen that in the situation there was no difference in stress range between the tests at R = 0 and R = –1. It is evident from Table 2.5 that the ratios show quite a large variation, although there is a well-defined tendency for stress range to decrease as mean stress increases. Over the whole range of fillet-welded joints, but ignoring the results for the spot-heated specimens referred to above, the average values of the ratios are 1.25 for alternating stresses and 0.82 for half tensile; in order words the results are remarkably similar to the values found for butt welds (see above). It is tempting, therefore to regard these values as typical for all (effectively stress relieved) welded joints. In a large as-welded structure containing high tensile residual stresses one would usually expect there to be no difference in the total stress range at a given life, regardless of the stress ratio. In other words, the tendency for the range to increase as the stress ratio decreases is really relevant only in the case of small specimens, which are not capable of holding high residual stresses, or in stress-relieved structures. This is discussed in more detail in Chapter 3 but at this stage, in addition to the results for spot heated specimens referred to above, it is worth noting the slight tendency (see Table 2.5) for longitudinal welds (which would be expected to contain higher residual stresses than transverse welds) to give a fatigue strength at R = –1 which was closer to that for R = 0 than was the case with the other joints.

Table 2.3 Joint classifications Type 1. Material free from welding Notes on potential modes of failure In plain steel, fatigue cracks initiate at the surface, usually either at surface irregularities

or at corners of the cross-section. In welded construction, fatigue failure will rarely occur in a region of plain material since the fatigue strength of the welded joints will usually be

much lower. In steel with rivet or bolt holes or other stress concentrations arising from the shape of the member, failure will usually initiate at the stress concentration.

Type number, description and notes on mode of failure

Class Explanatory comments

Examples, including failure modes

1.1 Plain steel (a) With all surfaces machined and polished, and of uniform or uniformly varying crosssection.

A

(b) In the as-rolled condition, or with cleaned surfaces but with no flame-cut edges or reentrant corners

B

Beware of using class B for a member which may acquire stress concentrations during its life, e.g., as a result of rust pitting. In such an event class C would be more appropriate

(c) As (b) but with any flame-cut edges subsequently ground or machined to remove all visible sign of the drag lines

B

Any re-entrant corners in flame-cut edges should have a radius greater than the plate thickness

(d) As (b) but with the edges machine flamecut by a controlled procedure to ensure that the cut surface is free from cracks

C

Note, however, that the presence of a re-entrant corner implies the existence of a stress concentration so that the design stress should be taken as the net stress multiplied by the relevant stress concentration factor

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

Type 2. Continuous welds essentially parallel Notes on potential modes of failure With the excess weld metal dressed flush, fatigue cracks would be expected to initiate at weld defect locations. In the as-welded condition, cracks might initiate at stop-start positions or, if these are not present, at weld surface ripples.

to the direction of applied stress continuous, and (b) if they are attached by welding those welds must also comply with the relevant class requirements (note particularly that tack welds, unless subsequently ground out or covered by a continuous weld, reduced the joint to class F, see joint type 6.5.

General comments 1. Backing strips. If backing strips are used in making these joints: (a) they must be

2. Edge distance. An edge distance criterion exists to limit the possibility of local stress concentrations occurring at unwelded edges

2.1 Full or partial penetration butt welds, or fillet welds Parent or weld metal in members, without attachments, built up of plates or sections, and joined by continuous welds. (a). Full penetration butt welds with the weld overfill dressed flush with the surface and finish-machined in the direction of stress, and with the weld proved free from significant defects by non-destructive examination. (b) Butt or fillet welds with the welds made by an automatic submerged or open arc process and with no stop–start positions within the length.

Examples, including failure modes

as a result, for example, of undercut, weld spatter, or accidental overweave in manual fillet welding (see also notes on joint type 4). Although an edge distance can be specified only for the ‘width’ direction of an element, it is equally important to ensure that no accidental undercutting occurs on the unwelded corners of, for example, cover plates or box girder flanges. If it does occur it should subsequently be ground smooth.

Applied stress

B

The significance of defects should be determined with the aid of specialist advice and/or by the use of fracture mechanics analysis. The NDT technique must be selected with a view to ensuring the detection of such significant defects.

C

If an accidental stop–start occurs in a region where class C is required remedial action should be taken so that the finished weld has a similar surface and root profile to that intended.

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

(c) As (b) but with the weld containing stopstart positions within the length. This class includes welds made manually.

D

Examples, including failure modes

For situation at the ends of flange cover plates see joint type 6.4. Edge distance from weld toe to edge of flange > 10 mm

Type 3. Transverse butt welds (i.e. essentially perpendicular to the direction of applied stress) Notes on potential modes of failure conditions), at the weld root. Unless made approximate method of allowing for eccentricity With the weld ends machined flush with the on a permanent backing (type 3.3) welds made in the thickness direction is to multiply the plate edges, fatigue cracks in the as-welded entirely from one side are not classified for nominal stress by (1 + 3e/t), where e is the condition normally initiate at the weld toe, so fatigue purposes, since adequate control distance between centres of thickness of the that the fatigue strength depends largely upon cannot be exercised over the profile of the two abutting members (if one of the members the shape of the weld overfill. If this is dressed root bead which is where fatigue cracks would is tapered, the centre of the untapered thickness flush the stress concentration caused by it is be likely to initiate. must be used); and t is the thickness of the removed and failure is then associated with thinner member. With connections which are weld defects. In welds made on a permanent Design stresses supported laterally, e.g. flanges of a beam which backing strip, fatigue cracks initiate at the weld In the design of butt welds of types 3.1 or are supported by the web, eccentricity may be metal/strip junction, and in partial penetration 3.2 which are not aligned the stresses must neglected. welds (which should not be used under fatigue include the effect of any eccentricity. An 3.1 Parent metal adjacent to, or weld metal in, full penetration butt joints welded from both sides between plates of equal width and thickness or where differences in width and thickness are machined to a smooth transition not steeper than 1 in 4.

Note that this includes butt welds which do not completely traverse the member, such as circular welds used for inserting infilling plates into temporary holes.

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

(a) With the weld overfill dressed flush with the surface and with the weld proved free from significant defects by nondestructive examination.

C

The significance of defects should be determined with the aid of specialist advice and/or by the use of fracture mechanics analysis. The nondestructive testing technique must be selected with a view to ensuring the detection of such significant defects.

(b) With the welds made in the shop, either manually or by an automatic process other than submerged arc, provided all runs are made in the flat position.

D

In general welds made positionally, or on site, or by the submerged arc process tend to have a poor reinforcement shape, from the point of view of fatigue strength. Hence such welds are downgraded from D to E.

(c) Welds made other than in (a) or (b).

E

In both (b) and (c) the corners of the cross-section of the stressed element at the weld toes should be dressed to a smooth profile. Note that step changes in thickness are, in general, not permitted under fatigue conditions, but that where the thickness of the thicker member is not greater than 1.15 × the thickness of the thinner member, the change can be accommodated in the weld profile without any machining. Step changes in width lead to large reductions in strength (see joint type 3.3).

Examples, including failure modes

t

e

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

3.2 Parent metal adjacent to, or weld metal in, full penetration butt joints made on a permanent backing strip between plates of equal width and thickness or with differences in width and thickness machined to a smooth transition not steeper than 1 in 4.

F

Examples, including failure modes

Note that if the backing strip is fillet welded or tack welded to the member the joint could be reduced to class G (joint type 4.2).

No tack welds

3.3 Parent metal adjacent to, or weld metal in, full penetration butt-welded joints made from both sides between plates of unequal width, with the weld ends ground to a radius not less than 1.25 × the thickness t.

F2

Step changes in width can often be avoided by the use of shaped transition plates, arranged so as to enable butt welds to be made between plates of equal width. Note that for this detail the stress concentration has been taken into account in the joint classification.

Type 4. Welded attachments on the surface or edge of a stressed member Notes on potential modes of failure attachments involving a single as opposed When the weld is parallel to the direction of to a double weld, cracks may also initiate at the applied stress fatigue cracks normally the weld root. The cracks then propagate into initiate at the weld ends, but when it is the stressed member. When the welds are transverse to the direction of stressing they on or adjacent to the edge of the stressed usually initiate at the weld toe; for member the stress concentration is increased

t

r ≥ 1.25t

and the fatigue strength is reduced; this is the reason for specifying an ‘edge distance’ in some of these joints (see also note on edge distance in joint type 2)

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

4.1 Parent metal (of the stressed member) adjacent to toes or ends of bevel-butt or filletwelded attachments, regardless of the orientation of the weld to the direction of applied stress, and whether or not the welds are continuous round the attachment.

Butt-welded joints should be made with an additional reinforcing fillet so as to provide a toe profile similar to that which would exist in a fillet-welded joint.

(a) With attachment length (parallel to the direction of the applied stress) ≤150 mm and with edge distance ≥10 mm.

F

(b) With attachment length (parallel to the direction of the applied stress) >150 mm and with edge distance ≥10 mm.

F2

4.2 Parent metal (of the stressed member) at the toes or the ends of butt- or fillet-welded attachments on or within 10 mm of the edges or corners of a stressed member, and regardless of the shape of the attachment.

G

The decrease in fatigue strength with increasing attachment length is because more load is transferred into the longer gusset, giving an increase in stress concentration.

Note that the classification aplies to all sizes of attachment. It would therefore include, for example, the junction of two flanges at right angles. In such situations a low fatigue classification can often be avoided by the use of a transition plate (see also joint type 3.3).

Examples, including failure modes

Edge distance

Edge distance

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

4.3 Parent metal (of the stressed member) at the toe of a butt weld connecting the stressed member to another member slotted through it.

Note that this classification does not apply to fillet-welded joints (see joint type 5.1b). However, it does apply to loading in either direction (L or T in the sketch).

(a) With the length of the slotted-through member parallel to the direction of the applied stress ≤150 mm and with edge distance ≥10 mm.

F

(b) With the length of the slotted-through member parallel to the direction of the applied stress >150 mm and with edge distance ≥10 mm.

F2

(c) With edge distance <10 mm.

G

Type 5. Load-carrying fillet and T butt welds Notes on potential modes of failure Failure in cruciform or T joints with full penetration welds will normally initiate at the weld toe, but in joints made with loadcarrying fillet or partial penetration butt welds cracking may initiate either at the weld toe

and propagate into the plate or at the weld root and propagate through the weld. In welds parallel to the direction of the applied stress, however, weld failure is uncommon; cracks normally initiate at the weld end and propagate into the plate perpendicular to the

Examples, including failure modes

T

L

L T

direction of applied stress. The stress concentration is increased, and the fatigue strength is therefore reduced, if the weld end is located on or adjacent to the edge of a stressed member rather than on its surface.

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

5.1 Parent metal adjacent to cruciform joints or T joints (member marked X in sketches).

Member Y can be regarded as one with a non-load-carrying weld (see joint type 4.1). Note that in this instance the edge distance limitation applies.

(a) Joint made with full penetration welds and with any undercutting at the corners of the member dressed out by local grinding.

Examples, including failure modes

X

Y

F X

X

Y

X Y

(b) Joint made with partial penetration or fillet welds with any undercutting at the corners of the member dressed out by local grinding.

F2

X

In this type of joint, failure is likely to occur in the weld throat unless the weld is made sufficiently large (see joint type 5.4). Y

X

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

5.2 Parent metal adjacent to the toe of loadcarrying fillet welds which are essentially transverse to the direction of applied stress (member X in sketch).

The relevant stress in member X should be calculated on the assumption that its effective width is the same as the width of member Y.

(a) Edge distance ≥10 mm.

F2

(b) Edge distance <10 mm.

G

5.3 Parent metal at the ends of load-carrying fillet welds which are essentially parallel to the direction of applied stress, with the weld end on plate edge (member Y in sketch).

G

These classifications also apply to joints with longitudinal welds only.

Examples, including failure modes

Edge distance Y

X

Edge distance

X

5.4 Weld metal in load-carrying joints made with fillet or partial penetration welds, with the welds either transverse or parallel to the direction of applied stress (based on nominal shear stress on the minimum weld throat area).

W

This includes joints in which a pulsating load may be carried in bearing, such as the connection of bearing stiffeners to flanges. In such examples the welds should be designed on the assumption that none of the load is carried in bearing.

Y

Table 2.3 Continued Type number, description and notes on mode of failure Type 6. Details in welded girders Notes on potential modes of failure Fatigue cracks generally initiate at weld toes and are especially associated with local stress concentrations at weld ends, short lengths of return welds, and changes of weld direction. Concentrations are enhanced when

Class Explanatory comments

Examples, including failure modes

these features occur at or near an edge of a part (see notes on joint type 4).

shown, in a more general form, in joint type 4; they are included here for convenience as being the joints which occur most frequently in welded girders.

General comment Most of the joints in this section are also

6.1 Parent metal at the toe of a weld connecting a stiffener, diaphragm, etc., to a girder flange.

Edge distance refers to distance from a free, i.e., unwelded, edge. In this example, therefore, it is not relevant as far as the (welded) edge of the web plate is concerned. For reason for edge distance see note on joint type 2.

(a) Edge distance ≥10 mm (see joint type 4.2).

F

(b) Edge distance <10 mm.

G

6.2 Parent metal at the end of a weld connecting a stiffener, diaphragm, etc., to a girder web in a region of combined bending and shear.

E

Edge distance

This classification includes attachments to girder webs.

all

Table 2.3 Continued Type number, description and notes on mode of failure

Class Explanatory comments

Examples, including failure modes

6.3 Parent metal adjacent to welded shear connectors. (a) Edge distance ≥10 mm.

F

(b) Edge distance <10 mm (see type 4.2).

G

6.4 Parent metal at the end of a partial length welded cover plate, regardless of whether the plate has square or tapered ends and whether or not there are welds across the ends.

G

This class includes cover plates which are wider than the flange. However, such a detail is not recommended because it will almost inevitably result in undercutting of the flange edge where the transverse weld crosses it, as well as involving a longitudinal weld terminating on the flange edge and causing a high stress concentration.

6.5 Parent metal adjacent to the ends of discontinuous welds, e.g., intermittent webto-flange welds, or tack welds unless subsequently buried in continuous runs.

E

As above, but adjacent to cope holes.

F

This also includes tack welds which are not subsequently buried in a continuous weld. This may be particularly relevant in tack-welded backing strips. Note that the existence of the cope hole is allowed for in the joint classification; it should not be regarded as an additional stress concentration as far as the web-to-flange welds are concerned.

Edge distance