Sunteți pe pagina 1din 6

Quantitative reliability analysis as a technique for assessing the fatigue performance of welded structures

G. K. Cole and B. F. Ronalds

An increasing trend in the offshore industry is to use quantitative reliability analysis to predict the fatigue performance of welded structures. This approach takes account of the uncertainties involved in fatigue assessment explicitly to arrive at estimates of the probability of fatigue failure. It may be used as the foundation for risk based inspection programs and is also a useful tool in the rational optimisation of engineering designs on the basis of minimised total through-life costs. This review paper describes the sources of uncertainty in conventional fatigue analysis before introducing a technique for reliability based fatigue analysis. An example of its use is provided, and the elements required for its application to other industries are then discussed.

application of reliability based fatigue life analysis to other industries is considered.

Sources of uncertainty in fatigue analysis


Conventional fatigue analyses are routinely applied to many different types of welded engineering structures which are subjected to uctuating loads. These include: ships, railway rolling stock, offshore structures, bridges and pipelines. The components of conventional fatigue analysis have been well documented 1,2,3 and will not be discussed in detail here. However, it is worthwhile at this point to summarise the main steps in estimating the fatigue life of a joint in a welded structure: 1. Cyclic fatigue loads acting on the joint are estimated and classied into some form of table or loading spectrum. 2. An appropriate S-N curve or fatigue crack growth model which describes the fatigue behaviour of the welded joint is selected. 3. The cumulative fatigue damage associated with the cyclic loads in the fatigue spectrum is calculated. The time period associated with the spectrum can then be used to predict the fatigue life of the joint. In practice, there are uncertainties associated with each of these steps. These can conveniently be divided into two broad categories. Type I uncertainties reflect natural or inherent variability in a physical quantity or process. Type II uncertainties are associated with the modelling of these physical processes.

Keywords
Fatigue, quantitative reliability analysis, welded structure.

Introduction
The potential for engineering structures to fail by fatigue under the action of uctuating loads is well understood. In particular, welded joints have been identied as being prone to fatigue cracking. Fatigue design is therefore an important aspect of welded structures in many industries. Conventional fatigue assessment as applied to welded structures in many industries is deterministic, that is, it results in a single estimate of fatigue life. If this estimate is quoted with precision it can give the misleading impression that fatigue analysis is an exact science. In fact, there is considerable variability in the factors which contribute to the fatigue performance of welded structures, and in most cases this leads to substantial uncertainty surrounding fatigue life predictions. In a practical sense this uncertainty is dealt with by applying a high degree of conservatism in the design or analysis. Factors of safety between 2 and 5 are typically applied on estimates of fatigue life. An increasing trend in the offshore industry is to use quantitative reliability analysis to predict the fatigue behaviour of welded structures. The differentiating features of this technique are that uncertainties associated with fatigue are accounted for explicitly, and the result is a probability of failure of the welded joint after a certain time period. It is also used to provide a basis for quantitative risk based structural inspection programs. The purpose of this paper is to introduce the technique of quantitative reliability analysis as a method of assessing the fatigue performance of welded engineering structures. The uncertainties surrounding conventional fatigue analysis are rstly discussed, and the procedure of quantitative fatigue reliability analysis as it is applied to offshore platforms is then summarised. A case study of a simple platform located on the North West Shelf of Australia is presented, before the Centre for Oil & Gas Engineering, University of Western Australia
WELDING RESEARCH SUPPLEMENT

Fatigue loads
The estimation of fatigue loads may have both Type I and Type II uncertainties associated with it. For example, there is inherent variability in loading caused by natural phenomena such as waves and wind. The translation of these actions into stresses at the welded joint involves some structural modelling. This modelling will involve assumptions and simplications which will lead to uncertainties. For example, Figure 1 shows the measured total force (or base shear) recorded on an instrumented offshore platform in response to wave loading compared with the base shear modelled or calculated for the same wave height using standard analytical techniques. If the modelling were perfect, all the results would fall on a single, 45-degree line. However, there is a bias toward the calculated result, that is, calculation of the forces on the structure is conservative. Moreover, there is signicant scatter in the results. This illustrates two important concepts in reliability analysis, namely bias and variance. Bias is dened as the ratio of the actual value to the predicted value, that is: Actual Value Bias = (1) Predicted Value Variance is apparent as scatter in the results and is an example of the uncertainty involved in making a prediction of the wave forces.
AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

33

105 90 Calculated Base Shear (KIPS) 75 60 45 30 15 0


Winter 76/77 Hurricane Anita Winter 77/78 Conductors Only

1000
T mean curve for 16 mm joints (m=3)

Hot spot stress range (MPa.)

100

T design curve for 16 mm joints (m=3)

16 mm experimental results

10 104

105 106 107 Endurance (N3) for 16 mm tubular joints

108

15

30

45

60

75

90

105

Measured Base Shear (KIPS) Figure 1. Measured total force plotted against calculated total force on an offshore platform in California waters 4 . 20 16 Predicted SCF 12 8 4 0

Figure 2. Typical scatter in fatigue test data; test results for welded tubular joints 5.

0.35 0.3 Frequency of Occurrence 0.25 0.2 0.15 0.1 0.05 0

8 12 16 Experimental SCF Figure 3. Experimentally measured Stress Concentration Factor (SCF) versus SCF predicted by parametric equations7.

20

4 5 6 Miners Sum

Figure 4. Histogram showing the values of Miners sum calculated for a series of variable amplitude fatigue tests on welded joints 5.

S-N Data
Constant amplitude fatigue test data for welded joints are characterised by considerable inherent variability. Important factors contributing to this variability include joint alignment, as well as weld shape and quality. Figure 2 illustrates the scatter in fatigue data for welded tubular joints which are typical for an offshore platform structure 5. Test data at a given stress range can span almost an order of magnitude on fatigue life. Additionally, signicant stress concentrations exist in the vicinity of tubular welded joints. These are dependent upon the joint conguration, its geometry, and the type of loading applied. Various sets of parametric Stress Concentration Factors (SCFs) have been proposed for estimating the localised, or Hot Spot Stresses that occur at various locations around the welded joint6,7. These are, of course, subject to some modelling uncertainty. An indicative comparison between predicted SCFs and those recorded experimentally is shown in Figure 3.

Cumulative fatigue damage


Once the hot spot stresses corresponding to each loading cycle in the fatigue spectrum are quantied, the cumulative fatigue damage may be calculated. The Palmgren-Miner linear cumulative damage hypothesis (Miners rule) remains the most widely used method for predicting the fatigue lives of welded joints subjected to variable amplitude (spectrum) loading. However, the variability associated with Miners rule is well documented8, 9 and represents a Type II uncertainty. The Miners sum calculated from a series of variable amplitude fatigue tests on welded joints is shown in Figure 4. This demonstrates clearly that the value of Miners sum at failure is not always 1.0 as the hypothesis suggests. In fact, it is a variable that may be represented by a suitable statistical distribution. A separate shortcoming of Miners rule is that the fatigue damage summation is purely notional, that is, it has no physical meaning. This leads to an important alternative procedure where
WELDING RESEARCH SUPPLEMENT

34

AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

the fatigue damage is correctly characterised as a growing crack, and the fatigue analysis is carried out using fracture mechanics principles10,11. It should be emphasised that while the fracture mechanics method more correctly describes the physics of the fatigue phenomenon, it is also characterised by its own set of uncertainties surrounding initial crack lengths, crack growth models, etc. The conventional method of dealing with the uncertainties associated with fatigue test data is to incorporate considerable conservatism into fatigue design. For example, design S-N curves are typically set at two standard deviations below the mean line representing test data (see Figure 2). When all the uncertainties associated with conventional fatigue life estimation are taken into consideration, it becomes clear that fatigue life prediction is an inexact science. In fact, a low calculated fatigue life simply indicates that the welded joint has an increased likelihood of fatigue failure. Recognition of this has led to a reliability based approach for quantifying the probability (or likelihood) of fatigue failure. This is described in the following section.

where m K B ~ is the slope of the S-N curve, = N Sm , a random variable representing the S-N curve, is a random variable representing the bias in fatigue stress estimates, is a random variable denoting the value of Miners sum at failure, tilde, denoting the median value of the above variables,

is called the stress parameter, which is a function of the magnitude and distribution of fatigue stress ranges. If the fatigue stresses have been categorised into a number of discrete stress ranges (with a number of stress cycles applicable to each range), the value of may be calculated as: (6) where f0 is the average frequency of the stress cycles, Si is the i th stress range, and i is the fraction of the total stress ranges that Si is acting. The next step in the reliability analysis is the calculation of the fatigue reliability index, F, using the following equation: (7) where is the intended service life of the structure, and

Reliability based fatigue life analysis


The important, and distinguishing, characteristics of reliability based fatigue life analysis are that it uses best estimates of fatigue performance, and takes account of uncertainties explicitly to arrive at estimates of the probability of fatigue failure for a given service life. In doing so, the variables associated with the fatigue process are described statistically. The methodology of reliability based fatigue life analysis using safe-life principles has been described by Wirsching12. An important underlying assumption is that: the fatigue stress, the S-N curve, and the value of Miners sum at failure, are all variables which can be described by a lognormal distribution, that is, the natural logarithms of the values are normally distributed. Expressed mathematically, if the variable is denoted as X , then Z is normally distributed, where:

is a measure of the combined uncertainty of the variables associated with the fatigue life estimate and is given by: (8) VK , VB and V represent the coefcients of variation of the variables K, B and respectively. Evaluation of the fatigue reliability index then allows the calculation of the probability of fatigue failure, PF , from (9) where is the standardised cumulative normal distribution. The translation from fatigue reliability index to the probability of fatigue failure may be found from statistical tables of the cumulative normal distribution, or by using the appropriate function in software such as Microsoft Excel. An approximation is also given by the equation: (10) The reliability is dened simply as Reliability = 1 PF (11)

Z = ln(X )

(2)

The properties of the lognormal distribution are described by Thoft-Christensen and Baker13. Important properties from the viewpoint of reliability analysis are the median and the coefcient of variation, which are described by the following expressions: ~ Median: X = exp( Z ) (3) Coefcient of Variation: V X = exp( Z 2 ) 1 Z (4)

where is the arithmetic mean of the Z values and Z is their standard deviation. An example is illustrated in Figure 4 where Miners sum is shown to conform to a lognormal distribution. The probability of fatigue failure is calculated using the following procedure. The median value of the time to fatigue failure (denoted ) is calculated and is equal to: (5)
WELDING RESEARCH SUPPLEMENT

It may be seen from the foregoing discussion that once the magnitude and distribution of fatigue loading cycles making up the fatigue load spectrum have been determined, it is reasonably straightforward to calculate the fatigue reliability provided that the estimates of K, B and , and their coefcients of variation are known. In practice, these estimates are likely to depend on a mix of statistical analysis and engineering judgement. Deriving such estimates will be illustrated in the following case study, and discussed in greater detail in a later section.
AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

35

Although not discussed in this paper, It should also be noted that an analogous method for calculating fatigue reliability based on fracture mechanics has been described by Kirkemo14.

Case study: fatigue reliability of a minimum North West Shelf offshore structure
The following section presents a case study where the method of calculating fatigue reliability is applied to a minimum offshore oil and gas platform similar to that shown in Figure 5. The idealised structure is shown in Figure 6. This idealised braced monotower structure has key features which are representative of several of the minimum offshore facilities located in shallow water on the North West Shelf of Australia. These include the height of the brace to caisson joint above the mudline, and the brace to caisson angle (31 degrees). Previous work has shown that the joint between the

braces and vertical caisson is critical in these structures. This is shown in Figure 6. The mean water depth was 41m and the central caisson of the platform was 1.8 m in diameter. The topsides weight was 200 tonnes. The structural members were designed in accordance with the American Petroleum Institutes Recommended Practice for Planning, Designing and Constructing Fixed Offshore Platforms, API RP2A WSD 20th Ed. While strength design is dominated by the joint action of extreme wind, wave and current loads, typical of those experienced by the structure during a tropical cyclone, fatigue loading is dominated by cyclic wave loading alone. Figure 7 shows a typical North West Shelf wave height exceedance curve. The lled points represent extreme events and are used for strength design. The open points represent the wave spectrum applied to the structure in the fatigue analysis. Further details of the analysis, which involved an evaluation of the hydrodynamic forces on the structure and calculation of the resulting hot spot stresses at critical locations around the welded tubular joint, are provided elsewhere 15. The fatigue reliability of the critical joint was calculated using the method described in the previous section. Revised estimates of uncertainties associated with predicting the fatigue lives of tubular joints have recently become available 16. These estimates have been interpreted by the current authors to develop a new data set for reliability analysis of offshore structures in Australia. These are provided in Table 1 below: Table 1. Fatigue reliability analysis data for a tubular joint in an offshore structure. Variable Value for tubular joint in offshore structure 3.82x1012 m 3.00 0.95 1.3

Figure 5. Typical braced monotower minimum offshore oil & gas platform.

VK VB V

0.67 0.54 0.50

By varying the value of Ts in equation (7) it is possible to derive the fatigue reliability index, F as a function of time, Figure 8. Similarly, using equation (9) the probability of fatigue
Significant Wave height, H sig (m)
16.0 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 10-2 10-1 100 101 102 103 104 105 106 107 108

Critical joint

Extreme Storm Design Criteria

Fatigue Wave Climate

Number of Waves Exceeding H sig per Year

Figure 6. Idealised structural model of braced monotower platform.

Figure 7. Typical North West Shelf wave height exceedance curve showing the cumulative number of waves exceeding a particular wave height.
WELDING RESEARCH SUPPLEMENT

36

AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

failure, PF can be plotted also as a function of time, and this is shown in Figure 9. As expected, the probability of failure increases with time, that is, the longer the joint is exposed to the fatigue environment, the greater the likelihood of fatigue failure. In this instance fatigue failure is dened as through thickness cracking of the critical member. Clearly, for Figure 9 to be of use it must be compared with some target probability of failure. Therefore some understanding of the acceptable risk of failure is required. This will depend on both the exposure to hazards and the consequence of failure. There are a number of considerations which may play a role in setting this value including: 1. Socially acceptable risks. 2. Historical failure rates. 3. Cost/benet analysis. A detailed discussion of the setting of appropriate target probabilities of failure is beyond the scope of this paper. Interested readers are referred to the work of Bea 4 and Melchers 17.

RISKEX may be evaluated as the probability of failure multiplied by the cost of that failure. All potential failure modes and their associated probabilities may be included in this cost, although the challenges of estimating both the probability and cost associated with caTastrophic failure of large structures have been highlighted in the wake of recent terrorist attacks. In the case of welded joints it may simply be the probability of fatigue failure, PF, multiplied by the cost of weld repair. However, other costs such as deferred production, and intangibles such as loss of good will may also be incorporated. A Net Present Value (NPV) function may be incorporated into the formulation of OPEX and RISKEX to account for the present worth of future costs. Commonly, as CAPEX increases, OPEX and RISKEX decrease, and a balanced solution may be found with a minimum total cost. The probability of failure associated with this solution may be regarded as the Target PF mentioned above, provided that regulatory and health, safety and environmental (HSE) requirements can be met.

Advantages of reliability based fatigue life analysis


Figure 9 immediately illustrates one of the uses of reliability based fatigue analysis, namely, it can be used to provide a rational estimate of inspection frequency. If the acceptable probability of fatigue failure is 10-4, the appropriate time to rst inspect this joint is approximately 10 years after installation. If no crack is found during this inspection, Bayesian updating techniques can be used to review the probability of failure curve. This is the basis of quantitative Risk Based Inspection (RBI) programs 18. A second use of reliability based fatigue analysis is in optimising design solutions on the basis of total life-cycle cost, where this is dened as: Total Cost = CAPEX + OPEX + RISKEX (12)

Application of reliability based fatigue life analysis to other industries


From the previous discussion, it can be seen that there is no impediment to using reliability based fatigue life analysis in other industries. Indeed, the current status of fatigue reliability reassessment in railroad and highway bridges has been described by Byers et al.20,21. However, the quality of fatigue reliability assessments is critically dependent on the availability of appropriate data describing the uncertainties in the fatigue assessment process. The important specic requirements include: 1. Estimates of the bias and uncertainty associated with fatigue stress prediction, which are necessary to evaluate the parameters B and VB in the reliability assessment. This is likely to require some experimental validation of the structural design model combined with engineering judgement. A well planned strain gauging investigation is one practical method of obtaining this information. The accuracy of fatigue stress prediction is an important consideration as the calculated probability of fatigue failure is quite sensitive to the fatigue stress bias parameter B. 2. Estimates of the mean S-N curve and uncertainty in fatigue test data for the welded joints under consideration. It is
10-2 Probability of Fatigue Failure, PF 10-3 10-4 10-5 10-6 10-7 0 5 10 15 Service Life (years) 20

where CAPEX is the capital expenditure associated with fabricating and installing the structure, OPEX is the cost of inspecting joints during the service life, and RISKEX is the probabilistic cost associated with structural failure 19. Calculation of the probability of fatigue failure has a direct input into the evaluation of OPEX by setting the frequency and type of inspection required to maintain structural integrity.
5.5 Fatigue Reliability Index, F 5.0 4.5 4.0 3.5 3.0 2.5 0 5 15 10 Service Life (years) 20

Figure 8. Variation of the fatigue reliability index, F , of the monotower critical joint with time.
WELDING RESEARCH SUPPLEMENT

Figure 9. Variation of the probability of fatigue failure, PF , of the monotower critical joint with time.
AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

37

fortunate that BS 7608 3 provides mean S-N curves as well as design S-N curves for a wide variety of weld details. The mean curves and the associated standard deviations described in this standard are useful starting points for estimation of the parameters, K, m, and VK. 3. Estimates of the median value and uncertainty in Miners sum which results in failure for the weld detail and fatigue spectrum under consideration. This is the basis for the parameters and V. This information can only be derived from appropriate spectrum fatigue testing programs and may be difcult to obtain. Some results have been provided by Gurney 9 and Blom 22. In the absence of such test results, the literature suggests that Miners sum at failure commonly has a median value of 1.0. Finally, it must be emphasised that while the reliability approach has the advantage of taking the uncertainties associated with fatigue life estimation into account explicitly, the results will depend upon the quality of the techniques used and the experience of the analyst. Like all fatigue life assessments, the results must be approached with insight and caution, and viewed within the context of a broader structural integrity management program.

References
1. T.R. Gurney, Fatigue of Welded Structures, Cambridge University Press, 2nd Edition, 1979. 2. D. Radaj, Design and Analysis of Fatigue Resistant Welded Structures, Abington Publishing, 1990. 3. BS 7608:1993 Fatigue Design and Assessment of Steel Structures, British Standards Institute. 4. R. G. Bea, Reliability based design criteria for coastal and ocean structures, The Institution of Engineers, Australia, 1990. 5. HSE, Background to new fatigue guidance for steel joints and connections in offshore structures, Offshore Technology Report OTH 92 390, 1992, Health and Safety Executive, UK. 6. HSE, Stress concentration factors for simple tubular joints, Offshore Technology Report OTH 354, 1997, Health and Safety Executive, UK. 7. A. K. Hellier, M. P. Connolly and W. D. Dover, Stress concentration factors for tubular Y- and T- joints, Int. J. Fatigue, 1990, vol 12, no. 1, pp. 13-23. 8. W. Schutz, The prediction of fatigue life in the crack initiation and propagation stages a state of the art review, Eng. Fracture Mech., 1979, vol. 11, pp. 405-421. 9. T. R. Gurney, Cumulative damage of welded joints, Joining and Materials, 1989, vol. 2, no. 7, pp. 320-323 & no. 8, pp. 390-395. 10. D. Broek, The Practical Use of Fracture Mechanics, Kluwer Academic Publishers, 1988. 11. BS 7910:1999 Guide on methods for assessing the acceptability of aws in metallic structures, British Standards Institute. 12. P. H. Wirsching Fatigue reliability for offshore structures, J. Struct. Eng., 1984, vol. 110, no. 10, pp. 2340-2356. 13. P. Thoft-Christensen and M. J. Baker, Structural Reliability Theory and its Applications, Springer-Verlag, 1982. 14. F. Kirkemo, Applications of probabilistic fracture mechanics to offshore structures, Appl. Mech. Review, 1988, vol. 41, no. 2, pp. 61-84. 15. G. K. Cole, B. F. Ronalds and E. Fakas, An assessment of the interaction between strength and fatigue reliability for shallow water platforms, Proceedings of the 20th International Conference on Offshore Mechanics and Arctic Engineering, 2001, ASME. 16. T. Xu, R.G. Bea, R. Ramos, O. Valle and V. Valdes, Uncertainties in the fatigue lives of tubular joints, OTC 10849, Proceedings of the 1999 Offshore Technology Conference. 17. R. E. Melchers, Structural Reliability Analysis and Prediction, Ellis Horwood, 1987. 18. MTD, Probability-based fatigue inspection planning, MTD Publication 92/100, Marine Technology Directorate, 1992. 19. G. K. Cole, R. Pinna, B. F. Ronalds and P. Romagnolo, Investigation of a simplied fatigue reliability model for optimising the design of tubular joints, Proceedings of the 21st International Conference on Offshore Mechanics and Arctic Engineering, 2002, ASME. 20. W. G. Byers, M. J. Marley, J. Mohammadi, R.J. Nielsen and S. Sarkani, Fatigue reliability reassessment procedures: state-of-the-art paper, J. Struct. Eng., 1997, vol. 123, no. 3, pp. 271-276. 21. W. G. Byers, M. J. Marley, J. Mohammadi, R. J. Nielsen and S. Sarkani, Fatigue reliability reassessment applications: state-of-the-art paper, J. Struct. Eng., 1997, vol. 123, no. 3, pp. 277-285. 22. A. Blom, Spectrum fatigue behaviour of welded joints, Int. J. Fatigue, 1995, vol. 17, no. 7, pp. 485-491.

Conclusions
1. There is considerable inherent variability in the fatigue performance of welded joints. This, along with variability associated with load estimates and Miners sum, contribute to uncertainty in estimating the fatigue lives of welded structures. This uncertainty is dealt with in conventional deterministic fatigue design by simply incorporating a very conservative factor of safety. 2. Reliability based fatigue life assessment offers a method of accounting for uncertainties explicitly and results in the calculation of a likelihood, or probability of fatigue failure. 3. The probability of fatigue failure may be used as a basis for risk based inspection programs, and also as a method for rationally optimising design on the basis of minimised through-life costs. 4. Despite the advantages of reliability based fatigue life analysis, the results depend upon the quality of the input data and techniques used, and the experience of the analyst. As with all fatigue analysis, the results should be approached with understanding.

Acknowledgements
The case study described in this paper was undertaken as part of a research project within the Cooperative Research Centre for Welded Structures (CRC-WS). The CRC-WS was established and is supported under the Australian Governments Cooperative Research Centres Program. Support in the form of a University of Western Australia Small Grant is also gratefully acknowledged. The authors extend their thanks to Mr. Steve Buchan of WNI Science and Engineering for valuable discussions regarding environmental data on the North West Shelf.

38

AUSTRALASIAN WELDING JOURNAL VOLUME 47, THIRD QUARTER 2002

WELDING RESEARCH SUPPLEMENT

S-ar putea să vă placă și