Review of Fatigue Assessment Procedures for Welded Aluminium Structures
S J Maddox, TWI Ltd
Paper published in International Journal of Fatigue, Paper JIJF1005. vol. 25, no. 12, December 2003, pp.1359-1378 - available here
This paper presents a review of methods and corresponding Codes and Standards for the fatigue assessment of welded aluminium alloy structures. Methods for the fatigue evaluation of welded aluminium structures are assessed from the viewpoints of original design and estimation of the residual life of existing structures. Based partly on a literature search, but also reference to data used in the formulation of recent fatigue design Standards, it goes on to review the information available for such assessments in design or guidance specifications in the light of relevant fatigue data. With regard to design specifications, particular attention is focussed on recent fatigue data obtained from structural components representative of actual structures. Recommendations are made for future research.
There is growing interest in the structural use of aluminium alloys, for such applications as automotive and railway vehicles, bridges, offshore structure topsides and high-speed ships. In all cases, welding is the primary joining method and fatigue is a major design criterion. However, as is well known, welded joints can exhibit poor fatigue properties. Thus, clear design guidelines are needed to ensure that fatigue failures are avoided in welded aluminium alloy structures. Apart from basic design of new structures, there is also increasing interest in methods for assessing the remaining fatigue lives of existing structures.
Prompted by difficulties experienced in reaching a consensus on fatigue design rules, extensive testing and analysis of the fatigue performance of welded aluminium alloys has been undertaken over the past 20 years. A measure of the research effort is the series of International Aluminium Conferences (INALCO) which has produced seven volumes of papers since 1981. The fatigue research work culminated in the production of new design specifications, notably BS 8118  , Eurocode 9  , the International Institute of Welding (IIW)  recommendations and specifications from the Aluminum Association  in the USA and the Canadian Standards Association.  In relation to ships, DNV issued supplementary guidance  to their 'Rules for the Classification of High Speed and Light Craft' based on the ECCS recommendations  , the forerunner to Eurocode 9.
Thus, it is an opportune time to review the various fatigue design procedures for welded aluminium alloy structures. This paper summarises methods for design and remaining-life assessments of fatigue-loaded aluminium alloy welded structures and compares and contrasts the information contained in the various recent fatigue design specifications. This includes assessment of the corresponding design curves in the light of fatigue test results, chiefly obtained in recent research programmes where a major aim was to reproduce the fatigue performance of full-scale welded structures, particularly with respect to the effect of high tensile residual stresses. Finally, areas still requiring further research are identified.
2. Fatigue assessment methods
Broadly speaking, any fatigue assessment involves comparison of the actions which the component or structure under consideration will be required to sustain during its design life (e.g. fatigue loading history, resulting stresses and the number of times they occur, any environmental influences, etc) with its resistance to fatigue. Clearly, the resistance must be sufficient to withstand the actions without failure occurring. The form and source of the resistance data depend on the type of assessment being performed.
A useful summary of the four main methods for assessing the fatigue lives of welded joints is contained in the new IIW fatigue design recommendations  . They are:
- S-N curves for specific welded joints used in conjunction with nominal stresses.
- S-N curves for welds used in conjunction with hot spot stresses.
- S-N curves for materials used in conjunction with local notch stresses.
- The fracture mechanics approach, whereby fatigue crack growth data are used in conjunction with the stress intensity factor to calculate the progress of a known flaw.
The first three are intended for application at the design stage and are described in detail in Section 2.2. The fourth is not generally used for design but for assessing known or assumed flaws. Thus, it would be applicable in an assessment of residual fatigue life, as described in Section 2.3.
2.2 Design of a new structure
Fatigue resistance data for design are usually expressed in terms of S-N curves, relating nominal applied cyclic stress range S and the corresponding number of cycles N needed to cause failure. In the simplest situation, the designer would ensure that the number of applied load fluctuations, n, in the design life that resulted in stress range S did not exceed N. In the more general case of a detail which will experience a spectrum of applied loads, the cumulative damage due to individual load cycles would need to be determined. The usual method is to apply Miner's rule, which assumes that the fatigue damage due to n i cycles of stress S i is directly proportional to n i / N i . An important step in the assessment is estimation of the stress history that will be experienced by the detail under consideration. In general, this involves identification of the loading history, conversion from loads to stresses (e.g. by finite element analysis (FEA) or strain gauge measurements) and, finally, extraction of recognisable stress cycles from the stress spectrum (the process of cycle counting) to provide input to Miner's rule. The full procedure is illustrated in Fig.1.
Fig.1a) Miner's linear cumulative damage rule for estimating fatigue lives under variable amplitude loading
Fig.1b) Analysis of fatigue loading for cumulative damage calculations
2.2.2 S-N curves used with nominal stresses
The S-N curves used in fatigue design depend on the procedure being used. Referring to those mentioned earlier, by far the most common approach is to use S-N curves obtained from fatigue tests on specimens containing the weld detail of interest ( Fig.2). Such S-N curves appear in many codes and standards, including some that apply to welded aluminium alloys. The design curve is usually some statistical lower bound to published experimental data, typically mean -2 standard deviations of log N. Since the S-N curves refer to particular weld details, there is no need for the user to attempt to quantify the local stress concentration effect of the weld detail itself. Thus, the curves are used in conjunction with the nominal stress range near the detail. In codes and standards, the curves are identified by arbitrary letters or, increasingly, by the fatigue strength at a particular life, usually 2x10 6 cycles. The current status of fatigue design rules for welded aluminium alloys is discussed in more detail later.
Fig.2. Examples of design S-N curves for welded joints (from IIW Recommendations for aluminium).
3.2.3 Hot-spot stress approach
The hot-spot stress method is an extension of the S-N curve approach in that it makes use of S-N curves obtained from tests on actual welded joints. However, the S-N curve is based on the hot-spot stress range rather than the nominal. Nominal stress is easy to define in simple laboratory specimens. However, in real structures the presence of gross structural discontinuities, non-uniform stress distributions and through-thickness stress gradients can be so complex that the nominal stress is no longer obvious. Experimental (e.g. strain gauges) and numerical (e.g. FEA) stress analysis methods are capable of providing detailed information about the stresses arising near welded joints. In such circumstances, the structural stress, which includes the effect of all sources of stress concentration except the weld itself, can be used. The hot-spot stress, which is discussed in more detail by Niemi  , is the structural stress at the weld toe. It is usually necessary to determine it by extrapolation from the stress distribution approaching the weld ( Fig.3). However, parametric formulae exist for calculating the hot-spot stress in some tubular joints  , and more such formulae are likely to be developed as the hot-spot stress method becomes more widely used. A practical limitation is that the hot-spot stress method is only suitable for assessing weld details from the point of view of potential failure from the weld toe.
Fig.3. Stress distribution approaching a welded joint and the definition of the hot-spot stress
Apart from tubular joints, there are no established S-N curves for use with the hot-spot stress. The S-N curves for use with the nominal stress are not generally suitable because they include some influence of the stress concentration effect of the welded joint. Thus, for example, the S-N curve for a fillet welded cover plate is below that for a simple fillet welded stiffener because of the greater stress concentration effect of the cover plate. An obvious candidate for a hot-spot stress S-N curve is that for transverse butt welds, since there is essentially no stress concentration effect due to the joint (provided it is perfectly aligned), only the weld bead. Indeed this is the case for tubular joints.
2.2.4 Notch stress approach
While the notch stress approach applies only to assessments of potential failure from the weld toe or root, the method attempts to include all sources of stress concentration, including the weld itself, in the stress used with the design S-N curve. Thus, in principle, a single S-N curve is sufficient for a given type of material. A practical problem is that the local geometry of the toe or root of a weld is highly variable and, at the design stage, not known. In recognition of this problem, the weld geometry is normally idealised as having a particular shape and weld toe or root radius. The local stress is then calculated by numerical analysis. Alternatively, parametric formulae are available for a range of joint geometries. 
Until very recently [6,11] , the notch stress method did not appear in any fatigue design specifications. Indeed, one of its protagonists  only recommends it for carrying out comparative studies of the fatigue performance of different welded joint options. Furthermore, the method has not been developed to any extent for aluminium alloys. Consequently, it is not considered further in this report.
2.3 Remaining life of existing structures
Broadly three approaches can be envisaged for the fatigue assessment of existing structures which have already experienced some service duty. The approach used will depend on the circumstances, particularly whether or not the structure was designed for fatigue loading, the time in service and what measures will be taken to assess its current condition with respect to potential fatigue damage already introduced during previous service.
Three assessment methods are described below. Examples of their application or reference to their development may be found in Refs. [12-14]
2.3.1 Fatigue design assessment
This method follows the procedure outlined in Section 2.2.1 for original design. If the structure was designed for fatigue loading, the same actions can be assumed, after any modifications to allow for changes such as reduced severity of the stress history from reinforcement or a change in the operating conditions. Fatigue resistance is still represented by the design S-N curves. If repairs are introduced, the design curves may still be applicable, but a safety factor could be introduced if the repair was of uncertain quality. Post-weld improvement of repair welds, for example by toe grinding (to be discussed in more detail later), would justify a higher S-N curve, which may be included in a design specification or obtained from appropriate published information. Finally, assuming Miner's rule (see Section 3.3.5 regarding validity of this assumption), it is used to calculate the fatigue damage introduced before and after the time of the assessment, on the basis that
( Σ/N) before + ( Σn/N) after < 1.
Then, the remaining life is given by:
Remaining life (e.g. in years) = life so far (years) + 
2.3.2 Fatigue design review
The aim of a design review would be to improve the accuracy of the original design process to provide a better estimate of the proportions of fatigue life used and remaining at the time of the assessment. A fatigue design process involves many assumptions and hence potential inaccuracies. When assessing an existing structure, there may be scope for improving the accuracy of some of those assumptions. For example, records of the service operation or even measurements made on the structure may enable a more precise definition of the stress history. Knowledge of the actual structural arrangement and weld details used, including their quality (e.g. alignment), coupled with appropriate stress analysis may allow the more precise hot spot stress approach to be used instead of the normal S-N curves. A further refinement might be the characterisation of actions and/or resistance data in statistical terms to enable reliability methods to be used to assess the risk associated with a particular estimate of remaining life. Some progress has been made in such an approach in the context of steel bridges. 
2.3.3 Fracture mechanics approach
The third method specifically addresses circumstances in which it has been found, or it must be assumed, that flaws (e.g. fatigue cracks) have been introduced during the service life endured so far. Such flaws would be those detected or measured by non-destructive testing (NDT), or assumed flaws corresponding to the limit of detection of the NDT methods used.
A fracture mechanics assessment  utilises the same actions as those determined for design calculations. However, fatigue resistance is represented by fatigue crack growth rate data for the material under consideration, expressed in terms of the fracture mechanics stress intensity factor parameter Δ K. Δ K is a function of applied stress range ( S), and crack size ( a), such that:
is a function of geometry and loading. The use of Δ K
ensures that the relationship between crack growth rate and Δ K
can be regarded as a law applicable to any geometry of the same material. The crack growth law approximates to a linear relationship (usually referred to as the Paris law):
with deviations as Δ K approaches a threshold value below which crack growth is insignificant ( Δ K o) and as K max approaches the critical value for fracture, as illustrated in Fig.4. In practice, the gradual transitions from the Paris law may be modelled more accurately by defining several linear relationships. For a flaw size a i and a critical fatigue crack size of a f , the remaining fatigue life N under stress range S is obtained by integrating Eq.:
For variable amplitude loading the integration will be performed for each individual cycle or block of equal stress cycles, to give:
Fig.4. Fracture mechanics fatigue crack growth relationship
3. Fatigue design data
3.1 Design specifications
As noted earlier, there is a wide choice of fatigue design specifications for welded aluminium alloys. The main ones, in chronological order, are as follows:
BS 8118:1991: 'Structural use of aluminium - Part 1 Code of practice for design', BSI, London 1991.
ECCS: 'European recommendations for aluminium alloy structures, fatigue design', European Convention for Constructional Steelwork, Document No.68, 1992.
Canadian Standards Association CAN/CSA-S157-M92: 'Strength design in aluminium', 1993.
The Aluminum Association: 'Specifications for aluminium structures', Washington DC, 1994.
DNV: 'Class note: fatigue assessment of aluminium structures', Technical Report No.LIB-J-000010, 1995.
International Institute of Welding: 'Fatigue design of welded joints and components', Abington Publishing, 1996.
Eurocode 9: 'Design of aluminium structures: Part 2: Structures susceptible to fatigue', ENV, 1999-2: 1998, CEN, Brussels, 1998.
Apart from the DNV document, these all provide a selection of design S-N curves expressed in terms of nominal stress ranges. In the DNV note, attention is confined to the use of the hot-spot stress range, a method that is also referred to in the IIW Recommendations and Eurocode 9, but specific design data are not provided. There are significant differences between the S-N curves in the rules and how they are used, and hence the different specifications will lead to different estimates of fatigue life. In order to provide a basis for judging their applicability to welded aluminium structures, key features are compared and where possible assessed in the light of relevant published data.
3.2 Historical developments
In order to appreciate why so many fatigue design specifications have been produced in recent years, it is useful to review the developments over the past 20 years, mainly in Europe, which have influenced them.
In 1979 it was decided that the British Standard design specification for aluminium, CP 118, which was the most comprehensive Standard for aluminium at the time and used throughout the world, should be revised as a limit state Standard. This followed the publication of new steel bridge design rules, BS 5400, on the same basis. With regard to fatigue, as a starting point the possibility was examined that the new rules for steels could be simply factored in accordance with the difference in Young's modulus between steel and aluminium to provide the aluminium fatigue rules.  This approach stemmed largely from the good correlation between fatigue crack growth data for the two materials on the basis of Δ K/E and the assumption that the fatigue lives of welded joints are dominated by fatigue crack growth.  Thus, the fatigue design stresses for welded aluminium alloys would be obtained simply by dividing those for steel by 3. A review of published data tended to support this approach and was adopted for the Draft for Public Comment of the Standard that would replace BS CP 118, BS 8118, in 1985. However, meanwhile some industries had drawn attention to the fact that the new rules were considerably more conservative than those in BS CP 118 in the high-cycle regime, while others felt that the steel/3 approach was too simplistic and effectively penalised aluminium alloys as compared with steel.
The initial review of fatigue data for welded joints in aluminium alloys had drawn attention to the wide scatter in published data and the fact that most data referred to small-scale specimens of variable, unspecified quality. Another important characteristic of small-scale specimens, particularly those incorporating transverse welds, is that they will contain much lower tensile residual stresses than would be expected to be present in a real structure. It was felt that more realistic fatigue data relevant to actual welded structures were required. These arguments influence the newly-formed ECCS Committee charged with the task of drafting a European Standard. Consequently, they placed particular reliance on data obtained from realistic structural specimens, mainly beams. A large database was available from one source (Alusuisse) and this was made available to the Committee. In addition, a number of new European projects provided additional data that were taken into consideration. To some extent, the same results were used to review the BS 8118 Draft for Public Comment and the fatigue rules were revised slightly as a result.
The resulting ECCS  and BS 8118  fatigue rules were finally considered together as the basis of the new Eurocode 9  in the early 1990s. Even more large-scale specimen data were available by then and so the final form of Eurocode 9 is different from both BS 8118 and the ECCS specification.
Other significant developments were the drafting of fatigue design rules in the USA  and Canada  , both of which are known to have been influenced by the European activities. 
3.3 Summary of design rules
3.3.1 Design S-N curves
All the fatigue design specifications for welded aluminium alloys present a series of S-N curves for particular weld details, with a classification scheme linking a description of the welded joint with the appropriate design curve. Examples of the S-N curves provided are shown in Fig.5. The classification usually depends on the joint type, geometry, loading direction and mode of fatigue failure, as illustrated for one group of joint types in Eurocode 9 in Fig.2. Rather less comprehensive guidance is given by the Aluminum Association  which only refers to joint types and loading direction. The S-N curves are derived from linear regression analysis of log S versus log N fatigue data to establish mean curves and statistical lower bound, usually mean - 2 standard deviations of log N. The S-N curves have the form:
S m N=A, where A and m are constants.
Fig.5a) Aluminum Association 
Fig.5b) IIW 
Fig.5c) Eurocode 9 
Fig.5. Examples of fatigue design S-N curves for aluminium alloys in recent specifications
Fig.6. Eurocode 9  allowance for reduction in fatigue strength due to marine corrosion
The curves are assumed to extend up to stress levels corresponding to the static design limit for the material, and down to a fatigue endurance limit. A constant amplitude fatigue endurance limit is introduced at an endurance of 5x10 6 cycles in all the aluminium specifications. Two basic approaches have been used to define the design S-N curves:
(a) An arbitrary grid of S-N curves, usually equally spaced on log-log scales, is defined and the curve closest to the selected lower bound to experimental data of a particular detail is allocated to that detail. The IIW Recommendations  are based on this approach with the S-N curves defined in terms of the stress range in N/mm 2 at 2x10 6 cycles (See Fig.7(a)).
(b) The design S-N curves are derived directly from experimental data. In some cases, the fatigue lives of different details are so similar that the experimental data can be combined to produce a single class for all of them. The resulting S-N curves, which are not usually equally spaced, may be described in terms of the fatigue strength at 2x10 6 cycles, and possibly the slope m of the S-N curve as in draft Eurocode 9  , or by arbitrary letters such as Class A, B, C, etc, in the Aluminum Association's specification  (see Fig.5(b) and (c)).
Fig.7a) Comparison of fatigue test results obtained from continuous longitudinal butt and fillet welds without stop/starts and design curves
Fig.7b) Comparison of fatigue test results obtained from continuous longitudinal fillet welds containing stop/starts and design curves
In most cases, the S-N curves get progressively steeper as the fatigue strength of the detail decreases. The steepest curve usually has a slope which is consistent with crack growth data (i.e. m = n, typically 3-4), reflecting the fact that the lives of the low fatigue strength details are dominated by crack growth. [17,18]
Apart from the implied need to utilise mechanised welding to achieve continuous welds without stop/starts, no distinction is drawn between different welding processes. The bulk of the test data upon which the design curves are based have been obtained from arc welds. However, as other processes became more viable for welding aluminium, notably friction stir welding  (see later) which seems to offer advantages from the fatigue viewpoint as well as production, it may become necessary to introduce new process-related categories.
3.3.2 Effect of residual stress and mean stress
Welding introduces tensile residual stresses, which modify the mean stress experienced by the welded joint under fatigue loading. Long-range, or reaction, residual stresses will also be introduced when welded sub-assemblies are connected together, due to imperfect fit-up. It is generally assumed that tensile residual stresses up to the proof strength of the material will be present in a welded structure. As a result, its fatigue life will be independent of mean stress and depend only on the applied stress range, even if this is compressive.  Consequently, all the fatigue design specifications are based on the use of full stress range regardless of whether it is tensile or compressive.
A common feature of all the specifications is that no distinction is drawn between different aluminium alloys when welded, unless they are exposed to a corrosive environment. This reflects the fact that fatigue crack growth rates are not significantly different in different alloys  and that fatigue crack growth dominates the fatigue lives of welded joints. Consequently, precise details of alloys used to produce welded test specimens discussed later are only given if they are significant.
For unwelded material, some of the specifications [2,3] provide higher design stresses for high strength 7000 series aluminium alloys as compared with all the other alloy types.
3.3.4 Effect of plate thickness
It is generally acknowledged that the fatigue strengths of welded joints failing from the weld toe can decrease with increase in plate thickness.  This has led to thickness effect penalties, applied to the fatigue strength obtained from the S-N curve, in many fatigue design rules for welded steel of the form ( t ref/t) p , where t = thickness, t ref = reference thickness (usually around 25mm) and the exponent p = 0.25. Recent work showed that the thickness effect also depended on the overall proportions of the welded joint. [24,25] These influences are incorporated in the fatigue rules in Eurocode 9. A further refinement in the IIW Recommendations  is to modify the thickness correction exponent p for different weld details. Values range from 0.3 to 0.1, reflecting the fact that the thickness correction also depends on the level of stress concentration introduced by the welded joint. In contrast, the Aluminum Association take the view that the database used to establish the design S-N curves covered the full range of thicknesses of aluminium alloy likely to be used in practice.  Hence, there is no requirement to apply a thickness effect correction. This assumption may be reasonable for some applications (e.g. automotive or railways vehicles where plate thickness is unlikely to exceed 25mm) but not for large structures such as bridges or LNG tankers where plate thickness may be 100mm or more.
3.3.5 Cumulative damage
Miner's rule is universally adopted as the method for predicting fatigue lives under variable amplitude loading using the constant amplitude design S-N curves. However, the accuracy of the rule has been called into question in recent years as more and more fatigue tests obtained under random loading conditions have produced failures in shorter lives than those predicted by Miner's rule. [26,27] It is thought that part of the reason for this is that the crack closure conditions for a given stress fluctuation are different under constant and variable amplitude loading, with the result that a stress range may be more damaging in a variable amplitude sequence than it was under constant amplitude loading.  However, a second problem concerns the damaging effect of stress ranges below the constant amplitude fatigue limit. Some specifications take account of such stresses by assuming that the S-N curve extends below the constant amplitude fatigue limit at a shallower slope. For an S-N curve of the form S mN=A, the extrapolated curve would be of the form S m+2N=A, (see Fig.5(c)). However, on the basis of fatigue tests on large-scale welded beams (to be discussed later), the Aluminum Association  take the view that the S-N curve should be extrapolated indefinitely below the constant amplitude fatigue limit without a slope change. The extent to which these modifications to the S-N curve are successful will be considered later in the light of new experimental data.
3.3.6 Hot-spot stress approach
Only the DNV note  gives specific guidance on the use of the hot-spot stress fatigue design procedure. That guidance is related to four S-N curves from the ECCS  recommendations, one for unwelded material, two for welded connections and the fourth for welds exposed to a corrosive environment (presumably sea-water). The corresponding ECCS design curves are as follows:
|DNV Class||Material||ECCS S-N curve|
||Unwelded high strength 7020 alloy
||Flush ground butt welds
||As-welded transverse butt welds with good profile
||Welded, in corrosive environment
||Not included: 25% reduction in design stress from Class III curve
Thus, for as-welded joints it is assumed that the butt weld design S-N curve is applicable to both butt and fillet welds if the hot-spot stress is used. The reason for the choice of the S-N curve for the highest strength of unwelded aluminium alloy (lower curves were provided for other aluminium alloys) is not known. The corresponding S-N curves in Eurocode 9 are lower than those in the ECCS rules.
The curves are used in conjunction with specified stress concentration factors, K, by which stress ranges obtained from the specified design curve are divided, for a variety of typical welded connections used in ships. The range is not totally comprehensive and, as will be seen later, no guidance is offered for some relevant details.
3.3.7 Effect of environment
As already mentioned, the DNV note provides guidance on the effect of immersion in a marine environment by adopting an S-N curve which is 25% on stress below the lowest curve for welded joints. That same curve would be used for any detail, welded or unwelded. The reduction in allowable stresses for as-welded joints of 25% is in-line with the results of an extensive series of corrosion fatigue tests conducted in Norway in the early 1980s. 
Eurocode 9 also provides guidance on the influence of environment, industrial and marine. The basic approach is to reduce the detail classification, by up to three categories in the case of immersion in seawater, and to reduce the fatigue endurance limit, as illustrated in Fig.6. The design penalty is most severe for 7000 series alloys, which are also susceptible to stress corrosion cracking, while no reduction in design category is required for 3000 series and aluminium-magnesium 5000 series alloys, although the fatigue endurance limit is still reduced. In general, the extent of the reduction in fatigue strength due to environment depends on the detail and endurance since the S-N curves are not parallel. It is not clear how the category-reduction approach should be applied to the lower category details, for which the required reduction would take them below the design categories provided.
4. Comparison of design proposals and recent fatigue data
All the recent design S-N curves for welded joints in aluminium alloys are claimed to have been derived from experimental data by linear regression analysis such that they represent approximately 97.7% probability of survival. [19, 29-31] However, it is not always clear how this has been achieved. It is evident that in most cases some judgement has been applied and even some assumptions, like the slope of the design S-N curve, have been imposed. It is also claimed that special attention was paid to the provision of data relevant to real structures, particularly with respect to the influence of tensile residual stresses. Thus, wherever possible the main basis of the design curves has been experimental results obtained from full-scale welded specimens, usually beams, or from specimens tested under high tensile mean stress conditions to simulate the effect of high tensile residual stresses. In view of this, the design S-N curves should be consistent with such data, including data generated since the curves were published. In order to provide a basis for judging the validity of the proposed design curves, relevant published data have been assembled and they are presented in comparison with some of the design curves. In this exercise, whenever possible, attention has been focused on test results obtained from specimens made from material 8-16mm in thickness, since plate thickness is known to influence fatigue performance. Furthermore, only design curves from the four most recent specifications, namely the Aluminum Association specification, Eurocode 9, the IIW recommendations and the DNV design note, are considered. The majority of the experimental data were obtained under constant amplitude loading and presented in terms of nominal stress range. Hence, these will be used to assess the design S-N curves intended for use with nominal stress range. Many of the results for beams were obtained from the compilation of data in Ref.30, which does not always give full details of the source (which may have been an internal company report). Reference is made to the relevant series in that reference.
Limited data presented in terms of hot-spot stress range are also available and these will be used to assess the proposed hot-spot stress S-N curves. Finally, some data have been obtained under variable amplitude loading. These will be used to assess the validity of Miner's rule and the methods proposed to modify the S-N curve below the constant amplitude fatigue limit to take account of the damaging effect of low stresses. The variable amplitude fatigue data will be considered in terms of the equivalent constant amplitude stress range, calculated on the basis that Miner's rule is correct using the constant amplitude S-N data obtained in the same investigation. This equivalent stress range is as follows:
Seq = m 
where n i cycles were applied at stress range S i before failure, Σn i is the total number of cycles to failure and m = the slope of the S-N curve. In those cases where a significant number of applied stress cycles were below the constant amplitude fatigue limit, the S-N curve was assumed to be extrapolated below this limit at a slope of m+2, as proposed in most of the specifications.
4.2 Continuous longitudinal welds
The stress concentrations introduced by continuous longitudinal butt and fillet welds, weld ripples or lumps if stop/starts are present, are relatively minor. Their corresponding fatigue performance is relatively good. The severity of a stop/start would be intensified if crater cracking occurred, which is certainly a possibility when welding aluminium alloys, but these would normally be repaired if found.
In spite of the relatively good fatigue performance of continuous longitudinally loaded welds, they are important details, particularly in welded aluminium alloy structures. Such details may be the governing ones in well designed structures in which poorer transverse welds have been avoided or located in low stress regions. The potential for doing this is enhanced in the case of aluminium alloys because of the enormous scope for producing special extrusions, for example of sufficient rigidity in relevant parts to avoid the need for welded stiffeners.
Continuous longitudinal welds are not explicitly included in the DNV note. Of the other specifications, both the IIW and Eurocode 9 distinguish between welds with and without stop/starts. The IIW recommendations also draw a distinction between butt and fillet welds. The Aluminum Association gives only one design category for both butt and fillet welds with and without stop/starts.
Recent fatigue data obtained from structural components [30 (series C2), 30 (D1 and D2), 32, 33] are confined to I-section beams in which the test detail is either a continuous butt weld in the web or, in the case of fabricated beams, the web to flange weld. Data are available for 5000, 6000 and 7000 series alloys in thicknesses from 6-15mm. They are shown together with relevant design S-N curves in Fig.7(a) for welds without stop/starts and Fig.7(b) for welds with stop/starts.
Referring first to the results for welds without stop/starts ( Fig.7(a)), it will be seen that they are most consistent with the slope of the IIW design curves (i.e. m=3), although only the Category 40 design curve is safe for all the results. The Eurocode 9 and Aluminum Association design curves appear to be too shallow. This may be a situation in which the slope of the design curve has been imposed. Both the Eurocode 9 and Aluminum Association specifications provide S-N curves which become gradually steeper as the fatigue strength of the detail decreases and the adoption of a slope of m = 3 for such a high fatigue performance detail would introduce an anomaly into such a scheme. However, compared with actual data, the result is that both the Eurocode 9 and Aluminum Association design curves are particularly conservative in the high stress/low fatigue life regime.
The case for a shallow S-N curve is better for welds containing stop/starts positions, as seen in Fig.7(b). This situation arises mainly because of the wide scatter associated with the fillet weld results. There are some fatigue data below all the design curves, the IIW curve being particularly non-conservative and the Eurocode 9 curve being the most suitable. Further investigation of the reason for the low  results would be worthwhile. By and large, it may be noted that the database does not provide a strong indication that distinction should be drawn between butt and fillet welds in design specifications, but the results do support the distinction between welds with and without stop/starts.
4.3 Transverse butt welds
This section is concerned with transverse butt welds made from one or both sides, with the condition that they should be full penetration welds. A number of factors will influence the fatigue performance of transverse butt welds and some of them influence the design curves. In particular, a distinction may be drawn between welds made from one or both sides and welds with different profiles (expressed in terms of the weld toe angle). Further conditions might be that the weld should be proved free from significant defects (i.e. those which might replace the weld toe as the site for crack initiation and lead to a lower fatigue life) by appropriate inspection, and that the effect of misalignment as a source of secondary bending stress should be taken into consideration when calculating the stress experienced by the weld.
A reasonable database from structural specimens containing transverse butt welds is available, mainly from I-section beams. [33,30 (series B7, B8, B10, B11), 34] These include specimens fabricated or extruded from 5000, 6000 and 7000 series alloys in thicknesses ranging from 8-15mm. In some cases, the weld toe angle is reported. The data are shown in comparison with relevant design curves inFig.8.
Fig.8. Comparison of fatigue test results obtained from transverse butt welded 5000, 6000 and 7000 series aluminium alloy beams and design curves
There is some indication of an influence of weld toe angle in that the highest results were obtained from welds with a toe angle not exceeding 30°, while the lowest were from welds with angles up to 60°. However, some good profile welds also gave lives near the lower bound and overall the results do not indicate a strong correlation between weld angle and fatigue strength. Similarly, the results do not provide support for distinguishing between one- and two-sided full-penetration welds.
The DNV note distinguishes between one- and two-sided welds and welds with different profiles. Default stress concentration factor values (by which stresses obtained from the DNV III curve are divided) of K = 1.7 and 1.3 respectively are given for weld toe angles up to 50°. The IIW recommendation distinguishes between one-and two-sided welds and different weld profiles, Eurocode 9 only distinguishes between one- and two-sided welds, but the Aluminum Association provides only one design curve for any transverse butt weld. Referring to Fig.8, it will be seen that the Aluminum Association Category C curve and the DNV curve for welds made from both sides are very similar and provide reasonable lower bounds to the data. The Eurocode 9 curves are unaccountably low while the IIW curves appear to be too steep and, apart from FAT 28, too high. However, it is interesting to note that regression analysis of all the experimental data together results in a mean S-N curve with slope of m=2.95, very similar to the assumed slope of m = 3 in the IIW Recommendations.
4.4 Transverse butt welds made on permanent backing
One technique for ensuring full penetration for butt welds from one side only is to use a permanent backing bar or, in the case of aluminium alloys, backing lip included in the extrusion. For joints in steel, the fatigue strength of the resulting joint is lower than that obtained from butt welds made from both sides, due to the severe stress concentration introduced at the weld root between the main plate and backing bar.  Since this is a geometric effect, it would be expected that the same would be found from aluminium alloys. To some extent this is the case, but the database is surprisingly limited in view of the potential for extruding aluminium sections incorporating backing lips. In fact, only one reference to tests on structural components  could be found. In view of this, data obtained from specimens are also considered. The data found in the literature search are given in Fig.9 together with the appropriate design S-N curves. These refer to plate specimens in 6005 and 7020 alloys, [30 (series B4)] extruded bridge deck panels in 6005 alloy and specimens extracted from such panels.  In fact, these specimens were reported to be severely misaligned (angular distortion) with the result that secondary bending occurred at the weld. The corresponding stress magnification factor K m was estimated by the authors and the results are presented in terms of K m x nominal stress range in Fig.9.
Fig.9. Comparison of fatigue test results obtained from transverse butt welds made on permanent backing and design curves
It will be seen that most of the data lie above the DNV and Aluminum Association design curves, which are shallower than the IIW and Eurocode 9 curves. The data tend to follow the slope of the shallower curves, but with such a limited database confined to a very limited range of relatively low endurances this may be a misleading impression. Certainly, in the light of experience of joints in steel, the slopes of the IIW and Eurocode design curves seem to be more appropriate, but further experimental data are needed to confirm this.
4.5 Transverse cruciform joints
Fatigue data are available for I-section beams incorporating cruciform joints [30(series F1), 33] in 15mm thick 5083 alloy and 8mm thick 6005 aluminium alloy.  In all cases, fillet and full penetration joints, failure was by fatigue cracking from the weld toe. Also, some data have been obtained from a model of a structural connection used in an aluminium alloy ship.  This was essentially the joint between the hull, T-section longitudinal stiffeners and a transverse bulkhead (see Fig.10). The specimens were made in 6.4mm thick 5086 H116 alloy. In all cases, the fillet weld sizes were sufficient to avoid failure in the weld throat in preference to failure from the weld toe.
Fig.10. Structural detail representing intersection of hull, longitudinal stiffener and transverse bulkhead fatigue tested by Beach et al 
All the results are plotted in Fig.11 in comparison with the appropriate design S-N curves. The detail is not explicitly covered in the Aluminum Association specification but it has been assumed that the design curve for transverse fillet welded stiffeners is appropriate. The Eurocode 9 design curve depends on plate thickness and joint proportions and the curve shown is applicable to the sizes of specimens used to generate the data presented. As will be seen, the data are widely scattered. However, both the IIW and Eurocode 9 curves appear to be representative of the slope of the data and close to the lower bound. The DNV and Aluminum Association curves are similar, but both seem to be too shallow.
Fig.11. Comparison of fatigue test results obtained from structural specimens incorporating cruciform joints and design curves
4.6 Transverse fillet welded attachments and stiffeners
Transverse non-load carrying fillet welded attachments and stiffeners are very common in actual structures. Like transverse butt welds, small-scale specimens are unlikely to contain high tensile residual stresses and hence be representative of real structures from this viewpoint. Therefore, fatigue test results obtained from structural specimens are particularly valuable.
A reasonable database now exists, as shown in Fig.12. Most of the results were obtained from beams with full or partial depth web stiffeners in 11-15mm thick 5000, 6000 and 7000 series alloys. [30 (series E1), 37] In all cases, fatigue failure was from the weld toe in the flange. In addition, a few results were obtained from I-section beams in 12mm thick 6061-T6  or 15mm thick 7020 alloys [30 (series E8)] with simple transverse attachments on the tension flange.
Fig.12. Comparison of fatigue test results obtained from beams with transverse fillet welded attachments or web stiffeners, or plates with transverse fillet welded attachments, and design curves
Figure 12 also includes the relevant design curves from the four specifications being considered. As will be seen, the data strongly support the slopes of the Eurocode 9 and Aluminum Association S-N curves, between m=3.2 and 3.6, and indeed those design curves are close to the lower bound to the data. The DNV curve appears to be too shallow, with a result that it is unduly conservative in the short life regime. The IIW curve, on the other hand, seems to be too steep and over-conservative in the long life regime. However, introducing a set of results  obtained from small-scale specimens makes the slope of the IIW curve seem more reasonable. These specimens were tested with the maximum stress held constant at a value close to proof strength, to simulate the presence of high tensile residual stresses. As seen, they are rather similar to the results for beams with simple attachments. Clearly, more data for the long life regime are needed to clarify the slope issue.
4.7 Longitudinal non-load carrying fillet welded attachments
Specimens incorporating longitudinal non-load carrying fillet welded attachments offer the advantage that high tensile residual stresses exist even in small-scale specimens.  The detail itself is not particularly common in real structures, except perhaps as gussets to stiffen corners. Fatigue failure occurs by crack growth from the weld toe at the end of the attachment, the fatigue life being similar whether or not the weld is continued around the ends of the attachment. However, the fatigue life varies with attachment size, the stress concentration effect at the end of the attachment increasing with attachment length.
Fatigue data are available for beams, extruded and fabricated, with attachments fillet welded to the tension flange. [30 (series E6,E7), 34, 39] These have been obtained using specimens made from 5000, 6000 and 7000  series alloys in thicknesses of between 11 and 15mm. Some of these were tested under variable amplitude loading.  The loading spectrum used was based on strain measurements on a railway freight wagon. Finally, partly to extend the range of endurances, but also because they include further fatigue test results obtained under variable amplitude loading, recent results obtained from 10mm thick 7019 alloy specimens can be considered. 
All these data are plotted in Fig.13, together with the appropriate design S-N curves, including the curves extrapolated below the constant amplitude fatigue limit for use when performing cumulative damage calculations. As will be seen, the S-N curves are rather similar above the constant amplitude fatigue limit, with slopes that are consistent with the experimental data. The design curves also lie close to the lower bound to the data. Thus, any of the design curves could be supported by the database.
Fig.13. Comparison of fatigue test results obtained from beams and plates with longitudinal fillet welded attachments and design curves
Comparing the data in Fig.13 with those in Fig.7, which are representative of beams without stiffeners or other attachments, provides a striking illustration of the detrimental effect of attachments on fatigue performance. In practice, they should be avoided in highly-fatigue-loaded areas.
The variable amplitude fatigue test results for both beams and specimens are consistent with the respective constant amplitude data, supporting the validity of Miner's rule for the spectra used. Furthermore, the beam data extend well below the constant amplitude fatigue limit and hence provide a useful check on the validity of the extrapolated S-N curve. Again, the data could be used to support any of the proposals, but the fact that they appear to be consistent with the same S-N curve as the constant amplitude data provides support for the Aluminum Association's approach for extrapolating the S-N curve without any slope change. Further results in the high-cycle regime are needed to check this further, particularly for other loading spectra.
It will be noted that the results obtained from small-scale specimens are entirely consistent with those obtained from beams, confirming that the two types of specimen incorporated similar high tensile residual stresses. This adds confidence to the use of small-scale specimens for investigating the fatigue performance of this particular detail.
4.8 Beams with cover plates
Large cover plates welded to beam flanges represent extremely high stress concentrations and consequently result in the lowest fatigue performance for welds failing from the weld toe. Consequently, beams with cover plates have been widely investigated for providing design data. This has resulted in a large database for beams in aluminium alloys, chiefly from fabricated beams in 10-15mm thick 5000, 6000 and 7000 series alloys. [30 (series F3), 37] A large number of results have also been obtained from smaller beams, [40,41] in 4mm thick 6261-T6 aluminium alloy. All these results are plotted together in Fig.14, together with the relevant design S-N curves from the four specifications considered.
Fig.14. Comparison of fatigue test results obtained from beams with cover plates and design curves
It will be seen that the IIW and Eurocode 9 design curves are very similar and consistent with the database in terms of slope and position. The Aluminum Association curve is slightly lower but of similar slope, while the DNV curve appears to be too shallow with a result that it is excessively conservative in the short life regime.
This detail is one in which the thickness effect would be expected to apply, and indeed is evident from a comparison of the results obtained from 4mm thick specimens with the remainder. This thickness effect is incorporated in both the IIW Recommendations and Eurocode 9 and higher design curves would be used for the 4mm thick specimens.
4.9 Fatigue data expressed in terms of the hot-spot stress range
Although guidance exists for the determination of the hot-spot stress, notably the IIW recommendations,  there is still the need for corresponding S-N curves. Preliminary proposals for weldable aluminium alloys have been made by Partanen and Niemi  on the basis of fatigue data generated at their university over a period of years. The data were obtained from a variety of butt- and fillet-welded specimens in 5000 and 6000 series alloys, including a model of the structural connection between the deck, longitudinal stiffeners and a transverse bulkhead in a naval ship. The range of plate thicknesses was 3-6mm. In all cases, hot-spot stresses were determined from FEA or strain gauges using the procedures in the IIW recommendations. The results ( Fig.15) were in reasonable agreement and the authors proposed that the IIW FAT40 design curve for transverse butt welds, which is included in Fig.15, was suitable as a hot-spot stress S-N curve for both butt and fillet welded joints in plate thicknesses up to 6mm.
Fig.15. Fatigue data presented by Niemi and Partanen  as a basis for the hot-spot stress S-N curve for thickness up to 6mm
This reference to thickness is important because the thickness effect discussed earlier will still exist even if fatigue strength is expressed in terms of the hot spot stress. This is evident from another set of results for a wider range of thicknesses, 3-24mm, also expressed in terms of the hot-spot stress range.  The test specimens were all 6061-T6 aluminium alloy plates with transverse fillet welded attachments. The results are shown in Fig.16. Since the lowest S-N curve, for 24mm thick specimens, happens to correspond exactly with the FAT40 design curve, it is tempting to conclude that these data support Partanen and Niemi's proposal. However, the results clearly show a thickness effect that justifies different hot-spot stress S-N curves for different thicknesses. A much larger database is needed to establish such S-N curves.
Fig.16. Fatigue test results obtained from 6061-T6 beams with fillet welded attachments expressed in terms of the hot-spot stress range which illustrate a thickness effect 
Other variables that might need to be considered further are the influence of the joint type and the through-thickness stress gradient. Referring to the data from Partanen and Niemi, there is a tendency for the higher results to be for butt welds and the lower ones for fillet welds, suggesting that there may be a need for different hot spot stress S-N curves for the two types of joint. A notable exception is the single lap joint that gave the highest results. This probably reflects the influence of bending stress gradient, which would have been particularly high in these joints due to their inherent misalignment. The effect, which is particularly significant in thin sections, is to increase fatigue resistance. Thus, this is another 'thickness effect' that needs to be considered when selecting data from which hot-spot stress design curves could be deduced. In general, data for high stress gradients should be excluded unless the hot-spot stress curve will only be used for similar conditions.
Finally, in order to assess the DNV hot-spot stress S-N curves  mentioned earlier, they are included in Fig.15. As will be seen, both curves look reasonable. However, it should be mentioned that many of the results in Fig.16, virtually all those for 24mm thick specimens, lie below curve III, suggesting that the thickness effect correction should be introduced at a lower value than that currently specified (i.e. 25mm) in the DNV note.
4.10 Effect of marine environment
Marine corrosion fatigue (full immersion and a saline atmosphere) of welded aluminium alloys was studied in some depth in Norway  in the early 1980s. Fatigue tests were performed on transverse butt- and fillet-welded joints in 8-12mm thick 5052, 5083, 6351 and 7004 aluminium alloys. The tests on specimens in saline atmosphere or immersed in seawater were carried out at the low frequency of 1Hz to allow time for the corrosion reaction to take place. The tests were conducted in bending which meant that relatively high fatigue lives were obtained, as compared with those expected for axial loading. Based on comparison of the fatigue performance in air and seawater, the results were consistent with earlier studies by Sanders and McDowell  on 5000 series alloys. However, the effect of the environment varied with alloy type, 5052 and 7004 alloys being more susceptible to environment than the others. In general, immersion in seawater produced fatigue lives approximately one third of those obtained in air, corresponding to a 25% reduction in fatigue strength. A saline atmosphere was generally less harmful, but not always. It produced a similar reduction in fatigue life to full immersion in seawater in the 5052 alloy, while it produced an order of magnitude reduction in fatigue life in the case of butt welds in 7004 alloy. The effect was less severe in fillet welds. The influences of environment and alloy type seen in this study are reflected in Eurocode 9 (see Fig.6).
Fatigue crack growth studies in Russian AlMg5 and AlZnMg alloys immersed in 3% sodium chloride solution  showed rather similar effects of environment for both alloy types. Crack growth rate was increased, but only significantly, by up to seven times, at relatively high crack growth rates, with little effect of environment near the threshold. The threshold itself was effectively independent of environment. These results suggest that the effect of environment on S-N data referred to earlier may have been largely associated with crack initiation, which might also explain why butt welded 7004 alloy was more susceptible to environment than fillet welds.
5. Friction-stir welding
All the data presented so far were obtained from arc welded specimens. A new welding process that offers considerably better fatigue performance is friction-stir. Friction Stir Welding (FSW) was invented at TWI, and the first patent application was filed in December 1991. The process is an entirely new method of making continuous welds in several configurations using a solid state process. The concept of FSW is illustrated in Figure 17(a). This shows a rotating tool that consists of a shoulder and a pin. The former is pressed against the surface of the materials being welded, while the pin is forced between the two components by a downward force. The rotation of the tool under this force generates frictional heat which softens the work-piece, and the movement of the rotating tool along the joint line causes softened material to flow from the region ahead of the tool to the region behind, consolidating to form a solid phase weld. The process uses no filler, and for most materials a shielding gas is not required. As the process does not melt the materials being joined, materials such as series 2000 and 7000 aluminium alloy, which are often difficult to weld by fusion processes due to solidification problems, are readily weldable. Experience has shown that as the process is fully mechanised, high levels of consistency can be obtained in weld quality. In aluminium, it is possible to make full penetration single pass butt welds in thicknesses of less than 1mm to over 50mm.
Fig.17. Friction stir welding:
a) FSW process
b) FSW joint in aluminium sheet
The process is used commercially by an ever-growing list of companies in the aerospace, shipbuilding, railway and automotive sectors. Almost all of the current commercial usage involves aluminium alloys, although some copper and magnesium alloys are also being welded. The joining of other materials, including titanium alloys, steel and nickel alloys, is under development.
FSW of aluminium alloys produces joints of high quality with static mechanical properties that equal, or generally exceed, those of competing processes, but with lower scatter. An example of a weld is shown in Fig.17(b). In view of the favourable profile, it is not surprising to find that, under similar conditions, the fatigue properties of friction stir welds in aluminium alloys compare very favourably with those for welds made by MIG, the normal alternative. There are several examples in the literature, all relating to 6000 alloys (since 2000 and 7000 alloys cannot be welded easily by the MIG process), although they are mainly confined to tests on relatively thin specimens. [45-47] A typical example, for 5mm thick 6082 alloy tested under the relatively severe loading condition of R=0.5  is shown in Fig.18. In the same investigation FSW joints in 6005 alloy gave fatigue test results close to those obtained from the unwelded material, which is deliberately low to allow for possible weld root flaws.
Fig.18. Comparison of fatigue data for 5mm thick 6082 alloy butt welded by MIG or FSW 
In principle, it is possible to make any weld design which does not require the addition of a filler material by FSW, although most experience to-date relates to butt- and lap-joints. A particular benefit is that full penetration butt-welds are readily achieved in joints made from one side only, whereas such joints are highly vulnerable to root flaws in welds made by other processes. Fig.18 includes the current Eurocode 9 design curves for transverse butt welds made from one or two sides, for comparison with the test data. It will be clear that friction stir welds made from one side can achieve considerably better fatigue lives than those indicated by the design curve. Root flaws can still arise in FSWs, but they can be relatively large before they affect the fatigue performance of the joint.  Unfortunately, FSW cannot be used to make fillet welds, as no filler is used. Therefore comparative fatigue data for fillet welds do not exist.
6. Fatigue life improvement methods
Fatigue life improvement techniques play an important part in achieving higher design stresses when the fatigue lives of structures are restricted by the presence of low fatigue strength details like fillet welded attachments. They may also be needed to ensure that a weld repair of fatigue damage will survive longer than the original detail. Thus, they are relevant to both the original design and life extension of existing structure.
There are two main principles behind the various improvement techniques  :
- Reduction of the stress concentration due to weld geometry, e.g. dressing by machining, grinding or TIG remelting;
- Introduction of compressive residual stresses, e.g. peening (hammer, needle, shot and brush), ultrasonic impact treatment.
In general, both types of improvement technique are only readily applicable to surface stress concentrations, notably weld toes. Improvement techniques have been widely studied in the context of welded steel, but less so in aluminium alloys. However, the general principles should be applicable to aluminium, as confirmed in a recent review by Hobbacher.  However, the review showed that most published data referred to butt welds, whereas in practice fillet welds present the greater potential fatigue problem. It was concluded that a fatigue strength improvement at 2 x 10 6 cycles of around 1.4 or more was justified for all the joint types reviewed, leading to the recommendations summarised in the following table:
|Structural Detail||Treatment Method||Fatigue Strength Improvement Factor|
|Transverse butt welds
|Cruciform joint fillet welds
|Longitudinal fillet welded stiffener
|Transverse fillet welded stiffener
A practical problem with aluminium that can arise in the case of dressing techniques concerns porosity. Flush-grinding of butt welds or TIG dressing of butt or fillet welds can result in previously embedded pores being exposed. TIG dressing can actually cause surface-breaking pores.  They then act as crack initiation sites and can actually reduce the fatigue life of the weld detail.
Although some design codes refer to the use of improvement techniques, none provides recommendations on the improvement in fatigue life to be achieved. This partly reflects uncertainty about the correct application of the techniques and their effect on the fatigue performance of real structures. In relation to the last point, it is known that the fatigue performance of welds treated by the residual stress techniques are significantly affected by mean stress, any benefit disappearing at stresses approaching yield.  Thus the techniques may not be suitable for large structures containing high tensile residual stresses or there may be doubts about their benefits in situations where the mean stress is not known. The IIW is currently addressing both the provision of specifications for the application of improvement techniques and corresponding benefits in terms of revised S-N curves,  including preliminary recommendations for welded aluminium alloys.
7. Fracture mechanics assessment of fatigue
Fracture mechanics offers the ability to assess the fatigue performance of aluminium structures containing known or assumed flaws. In the context of the original structure, these could be manufacturing flaws, when the assessment might be required if the flaw exceeds the manufacturing quality standard being worked to. In the context of the assessment of existing structures, they are likely to be cracks formed during previous service by fatigue or other mechanisms.
Eurocode 9 contains an Appendix with guidance on the use of fracture mechanics for assessing welded aluminium alloys. This includes recommended fatigue crack growth laws based on a large database produced by Alusuisse.  These are expressed as a series of Paris laws, to model the data more accurately than a single Paris law, as illustrated for one alloy in Fig.19. The same data are referred to in BS 7910  and the corresponding IIW recommendations.  These also contain very detailed guidance on the use of fracture mechanics for assessing welded structures, including a comprehensive set of stress intensity factor solutions for the types of crack and welded joint geometry which are likely to be encountered. One advantage of the procedure in the IIW recommendations is that it is linked with the IIW fatigue design curves for welded aluminium alloys to facilitate direct comparison of the fatigue performance of flaws and that of basic design details in the same structure. However, BS 7910, which relates the fracture mechanics assessment to British Standard design S-N curves, is more up-to-date.
Fig.19. Example of multi-stage fatigue crack growth relationships for aluminium alloys propsed by Jaccard 
The multi-stage Paris law crack growth data from Eurocode 9 have also been presented as polynomials,  in order to deduce better estimates of the probability of failure associated with upper-bound curves. However, regardless of the method of presentation, little has been done to confirm that the same complex, multi-stage crack growth relationship is a general law, applicable to any cracks in real structures.
8. Future research
A number of aspects of both the design specifications and residual life assessment methods considered in this review would be improved by further research. The following are suggested as being the most important:
- Provision of fatigue data for non-arc welding processes, particularly friction stir but also laser welding, and their incorporation in design specifications.
- Study of cumulative damage under realistic stress spectra, with particular emphasis on the high-cycle regime and the damaging effect of stresses below the constant amplitude fatigue limit.
- Further fatigue tests and FEA of structural details to establish hot-spot stress S-N curves and guidance on the practical application of the approach.
- Identification of potential fatigue design improvements that could be achieved by better use of special extrusions, and generation of appropriate fatigue data. Data are also required for transverse butt welds made on the backing provided by an extruded lip.
- Establishment of specifications for applying improvement techniques (of particular relevance for life extension) to welded aluminium and experimental confirmation of their value under realistic loading conditions (e.g. mean stress,loading spectrum).
- As far as the use of fracture mechanics for estimating residual fatigue life is concerned, the information incorporated in BS 7910 probably represents the current state of the art. However, it does place particular emphasis on steel and experimental work to decide on the choice of fatigue crack growth relationships appropriate for aluminium alloys and experimental validation of the fracture mechanics approach in general would be useful.
Based on a review of published information on fatigue assessment procedures for welded aluminium structures and supporting experimental data, the following conclusions can be drawn:
- Of the three fatigue design assessment procedures described, that using the nominal stress S-N curves is the most developed and standardised, but the hot-spot stress approach will probably prove to be the most valuable forstructural design in future.
- Several national and international fatigue design specifications have been published in recent years. Eurocode 9 and the IIW recommendations are the most comprehensive.
- Even so, most design data refer to arc welded joints and there is a need for corresponding data for other welding processes (e.g. friction-stir).
- There are significant differences between all the proposed design S-N curves and the fatigue test database available for large-scale structural specimens, mainly due to the choice of S-N curve slope. Thus, some specifications areunduly conservative in the low endurance regime and others in the high cycle regime. Eurocode 9 curves were generally the most consistent with experimental data.
- There is a scarcity of data for weld details incorporating special extrusion shapes, (e.g. built in backing lip) and little effort seems to have been made to investigate improved fatigue performance of welded aluminium structuresby better use of extrusions.
The fatigue performance of commonly welded aluminium alloys is not greatly influenced by immersion in a marine environment. The recommendations in Eurocode 9 are consistent with relevant published data.
There is little information in the literature on remaining life assessment procedures. The most appropriate are the use of design S-N curves or, if allowance must be made for damage sustained during previous service, fracture mechanics. Comprehensive guidance and recommended input data for the application of fracture mechanics are contained in BS 7910:1999.
This work described in this paper was funded partly by the Australian Maritime Engineering CRC Ltd and partly by Industrial Members of TWI Ltd. The author is also grateful to his colleague Dr P L Threadgill for his assistance with the section on friction-stir welding.
- BS 8118:1991: 'Structural use of aluminium - Part 1 Code of practice for design', BSI, London 1991.
- Eurocode 9: 'Design of aluminium structures: Part 2: Structures susceptible to fatigue', ENV, 1999-2: 1998, CEN, Brussels, 1998.
- International Institute of Welding: 'Fatigue design of welded joints and components', Abington Publishing, Abington, Cambridge, 1996.
- The Aluminum Association: 'Specifications for aluminium structures', Washington DC, 1994.
- Canadian Standards Association CAN/CSA-S157-M92: 'Strength design in aluminium', 1993.
- DNV: 'Class note: fatigue assessment of aluminium structures', Technical Report No.LIB-J-000010, 1995.
- ECCS: 'European recommendations for aluminium alloy structures, fatigue design', European Convention for Constructional Steelwork, Document No.68, ECCS, Brussels, 1992.
- Niemi E: 'Stress determination for fatigue analysis of welded components', International Institute of Welding, Abington Publishing, Cambridge, 1995.
- Lloyds Register of Shipping: 'Stress concentration factors for simple tubular joints, assessment of existing and development of new parametric formulae', HSE Report No. 0TH91 354, 1991.
- Radaj D: 'Design and analysis of fatigue resistant welded structures', Abington Publishing, Abington, Cambridge, 1990.
- 'Fatigue strength of welded ship structures', Bureau Veritas Document NI 393 DSM ROIE, July 1998
- IABSE Workshop, Lausanne 1990: 'Remaining fatigue life of steel structures', IABSE Report, Vol.59, IABSE, Zurich, 1990.
- Hadley I and Manteghi S: 'Remnant life of semi-submersible rigs', Proc. Seminar 'Remnant Life Prediction', IMechE, November 1997.
- Maddox S J and Kenzie B W: 'Fatigue assessment of ageing ship structures', Proc. 11th Intl. Maritime and Shipping Conf. 'Ships, The Ageing Process', Institute of Marine Engineers, London, 1997.
- Moses F: 'Bridge load models for fatigue', Proc. IABSE Workshop 'Remaining Fatigue Life of Steel Structures', IABSE Report, vol.59, IABSE, Zurich, 1990.
- BS 7910:1999: 'Guide on methods for assessing the acceptability of flaws in metallic structures', BSI, London, 1999.
- Maddox S J: 'Fatigue design of welded aluminium alloy structures', Proc. 2nd Intl. Conf. of Aluminium Weldments, Aluminium-Verlag, Dusseldorf, 1982.
- Maddox S J and Webber D: 'Fatigue crack propagation in AlZnMg alloy fillet welded joints' in 'Fatigue Testing of Weldments', ASTM STP648, ASTM, Philadelphia, 1977.
- Menzemer C and Fisher J: 'Revisions to the Aluminum Association fatigue design specifications', Proc. 6th Intl Conf. on Aluminium Weldments, AWS, 1995, 41-45.
- Dawes C J and Thomas W M: 'Friction stir process welds aluminium alloys' Welding Journal, vol.75, no.3. March, 1986.
- Gurney T R: 'Fatigue of welded structures', Cambridge University Press, 2nd Edition, 1978.
- Jaccard R: 'Fatigue crack propagation in aluminium', IIW Doc.XIII-1377-90, 1990.
- Gurney T R:' The influence of thickness on the fatigue strength of welded joints', Proc. 2nd Intl. Conf. 'Behaviour of Offshore Structures', London 1979.
- Maddox S J: 'The effect of plate thickness on the fatigue strength of fillet welded joints', Abington Publishing, Abington, Cambridge, 1987.
- Maddox S J: 'Scale effect in fatigue of fillet welded aluminium alloys', Proc. 6th Intl Conf. on Aluminium Weldments', AWS, 1995.
- Gurney T R and Maddox S J: 'An alternative to Miners' rule for cumulative damage calculations?' Proc. IABSE Workshop ' Remaining Fatigue Life of Steel Structures', vol.59, IABSE, Zurich, 1990.
- Niemi E: 'Random loading behaviour of welded components', Proc. IIW Intl Conf. on 'Performance of Dynamically Loaded Welded Structures', Welding Research Council, New York, 1997.
- Engh B, Solli O, Arbo S and Paauw A J: 'The influence of a corrosive environment on the fatigue life of aluminium weldments', Proc. 3rd Intl Conf. on Aluminium Weldments, Aluminium-Verlag, Dussedorf,1985.
- Ogle M H and Maddox S J: 'Design rules for fatigue of aluminium structures according to BS8118 Part 1', Proc. 5th Intl Conf. on Aluminium Weldments, Aluminium-Zentrale/Technical University of Munich,1992.
- Jaccard R, Kosteas D and Ondra R: 'Background document to fatigue design curves for welded aluminium components', IIW Doc. No.XIII-1588-95, 1995.
- Hobbacher A: 'The development of the new IIW fatigue recommendations', Proc. IIW Intl. Conf. on 'Performance of Dynamically Loaded Welded Structures', Welding Research Council, New York, 1997.
- Mazzolani F and Grillo M: 'Fatigue strength of longitudinally welded aluminium alloy structures,' Proc.6th Intl. Conf. on Aluminium Weldments, AWS, 1995.
- Orjasæter O: 'Fatigue of large weldments in AlMgSi1 and AlMg4.5Mn aluminium', Proc. 5th Intl Conf. on Aluminium Weldments, Aluminium-Zentrale/Technical University of Munich, 1992.
- Soetens F, van Straalen I J and Dijkstra O: 'European research on fatigue of aluminium structures', Proc. 6th Intl Conf. of Aluminium Weldments, AWS, 1995.
- Haagensen P J, Raines M, Kluken A O and Kvale I: 'Fatigue performance of welded aluminium deck structures', Proc. 15th Intl Offshore Mechanics and Arctic Engineering (OMAE) Conf., Vol.3 - MaterialsEngineering, ASME, 1996.
- Beach J E, Johnson R E and Koehler F S: 'Fatigue of 5086 aluminium weldments', Proc. 2nd Intl. Conf. on Aluminium Weldments, Aluminium-Verlag, Dusseldorf, 1982.
- Menzemer C C: 'Fatigue behaviour of welded aluminium structures', PhD Thesis, Lehigh University, Bethlehem, PA, 1992.
- Webber D and Gurney T R: 'The effect of some programme variables on fillet weld cumulative fatigue damaged test results', Proc. 5th Intl. Conf. on Aluminium Weldments, Aluminium-Zentrale/TechnicalUniversity of Munich, 1992.
- Hirt M A, Kimberley G J and Smith I F C: 'Fatigue behaviour of aluminium beams with welded attachments', Proc. 5th Intl. Conf. on Aluminium Weldments, Aluminium-Zentrale/Technical University of Munich,1992.
- James M N, Lambrecht H O and Patterson A E: 'Fatigue strength of welded cover plates on 6261 aluminium alloy I-beams', Intl. J. of Fatigue, Vol.15, No.6, November 1993, 519-524.
- James M N, Patterson A E and Sutcliffe N: 'Constant and variable amplitude loading of 6261 aluminium alloy I-beams with welded cover plates - influence of weld quality and stress relief', Intl. J of Fatigue, Vol.19, No.2, February1997, 125-133.
- Partanen T and Niemi E: 'Hot spot S-N curves based on fatigue tests of small MIG welded aluminium specimens', Welding in the World, Vol.43, No. 1, Jan/Feb.1999, p.16-22.
- Sanders W W and McDowell K A: 'Fatigue behaviour of 5000 series aluminium alloy weldments in marine environments', Welding Research Council Bulletin, No.242, October 1978.
- Olik A P: 'Cyclic cracking resistance of ship hull plate aluminium alloys', Fiziko-Khimisheskaya Mekhanika Materialov, Vol.26, No.4, 1990 (translated as Soviet Materials Science), 60-64.
- Ranes M, Kluken A O and Midling O T: 'Fatigue properties of as-welded AA6005 and AA6082 aluminium alloys in T1 and T5 temper condition'. Proc. 4th Intl. Conf. on 'Trends in Welding Research', GatlinburgTN, 5-8 June 1995, p.639-644.
- Haagensen P J, Midling O T and Ranes M: 'Fatigue performance of friction stir butt welds in a 6000 series aluminium alloy', Proc. 2nd Intl. Conf. on 'Surface Treatment', Milan, 7-9 June 1995.
- Kawasaki T, Makino T, Todori S, Takai H, Ezumi M and Ina Y: 'Application of friction stir welding to the manufacturing of next generation A-Train type rolling stock', Proc. 2nd Intl. Symp. on 'Friction Stir Welding', Gothenburg,June 2000.
- Dickerson T L and Przydatek J: 'Fatigue of friction stir welds in aluminium alloys that contain root flaws', Intl. J Fatigue, this issue.
- Hobbacher A: 'The benefit of fatigue improvement techniques at welded aluminium joints in context of scatter of non-improved welds', Proc. 5th Intl. Conf. on Aluminium Weldments,Aluminium-Zentrale/Technical University of Munich, 1992.
- Calcraft R C, Viano D M, Schumann G O, Phillips R H and Ahmed N U: 'Fatigue of marine grade aluminium', Proc. 7th Intl. Conf. on 'Joints in Aluminium', Abington Publishing, Abington, Cambridge, 1998.
- Maddox S J: 'Improving the fatigue strength of welded joints by peening', Metal Construction, Vol.17, No.4, 1985, 220-224.
- Haagensen P J: 'IIW's Round Robin and design recommendations for weld improvement methods'. Proc. IIW Intl. Conf. on 'Performance of Dynamically Loaded Welded Structures', Welding Research Council, NewYork, 1997.
- IIW: 'Guidance on assessment of the fitness-for-purpose of welded structures', IIW Doc. No. SST-1157-90, International Institute of Welding, Paris, 1990.
- Ondra R and Kosteas D: 'Practice oriented design values for the crack propagation of welded aluminium joints', Proc. 7th Intl. Conf. on 'Joints in Aluminium', Abington Publishing, Abington, Cambridge,1998.
||constant in equation for S-N curve
||constant in fatigue crack growth law
||stress concentration factor
||stress magnification factor due to misalignment
||slope of S-N curve
||exponent in fatigue crack growth law
||number of applied load cycles at stress range S i (i = 1, 2, 3 . . . .)
||fatigue life in cycles
||fatigue life at stress range S i
||exponent in plate thickness correction
||equivalent constant amplitude stress range
||correction factor in formula for stress intensity factor
||stress intensity factor range