You are on page 1of 10

20th International Conference on Structural Mechanics in Reactor Technology (SMiRT 20) Espoo, Finland, August 9-14, 2009 SMiRT

20-Division 1, Paper 3135

Fatigue curve and stress strain response for stainless steel

Jussi Solina, Gerhard Nagelb and Wolfgang Mayingerb

VTT, Finland, , e-mail: E.ON Kernkraft GmbH, Tresckowstrasse 5, Hannover, Germany

Keywords: Fatigue, Stainless steel, Design curve.


Niobium stabilized austenitic stainless steel (X6CrNiNb1810 mod) was studied in strain controlled fatigue tests ranging from low cycle (LCF) to high cycle (HCF) regime in RT air. An experimental strain life curve was determined as a base line for component specific evaluations and for comparison with the Langer and Chopra curves, which are the basis of the ASME III and NRC RG 1.207 design criteria. Stress strain responses were carefully measured to clarify the strain amplitude dependent fatigue behaviour of this steel. In the LCF regime our data lie within a common scatter band between the Langer and Chopra curves, but an endurance limit was observed at a 0,19 %. This is due to notable secondary hardening at low strain amplitudes. In HCF regime the Chopra curve and NRC RG 1.207 become highly conservative for this material. Relevance of all proposed design curves should be carefully considered.


Fatigue design curves given in the ASME Code Section III are derived from reference mean curves proposed by Langer . They are based on strain controlled low cycle fatigue tests in room temperature. Arbitrary design margins, 20 against life and 2 against strain, were considered appropriate for ensuring transferability of the data to plant components, as by ASME (1972). Similar curves are included also in the German and French design codes KTA and RCC-M. The ASME III design by analysis philosophy and local strain approach assume that the designer has relevant material data available. Although generalized design curves have been included in the codes to reduce need for material testing, choice and applicability of the code curve or an experimental curve for the particular application remains the responsibility of the designer. Consideration of operation environment is a good example on this. The code itself does not give specific curves or quantitative factors for adopting influence of reactor coolant to fatigue calculation. Moderate environmental effects are accounted for through the design curve definition (within the margin of 20 in life), but the responsibility of considering eventual environmental effects was left to the designer as clearly stated in the ASME (1972) Criteria Document for the ASME III Design by analysis procedure as follows: protection against environmental conditions such as corrosion and radiation effects are the responsibility of the designer It is obvious that the designer may choose to use an appropriately determined and more relevant experimental curve. This is literally recommended by STUK (2002) in the Finnish YVL guide for ensuring strength of NPP pressure devices. According to YVL guide 3.5, fatigue assessment shall be based on S-N curves applicable to each material and conditions. The development of the ASME code was primarily aiming to prevent catastrophic fractures of pressure vessels and the fatigue assessment was focusing on severe but rare thermal transients that can cause notable low cycle fatigue damage in heavy equipment. Later on, higher numbers of small stress cycles acting in the piping have been addressed, in particular for small bore pipes. Simultaneously fatigue tests have been conducted to longer lives. Encouraged by new experimental data and the proposal by Chopra and Shack (2007), the NRC (2007) endorsed a new air curve for stainless steels as part of a Regulatory Guide for new designs un USA. Data and regression curves in line with the ANL/NRC reference curve (often referred as Chopra curve) have been published by Jaske (1977), Higuchi (2004), Solomon (2004), Faidy (2008) et al.

It should be noted that in the LCF regime all air data for austenitic stainless steels lie practically within a common scatter band and the later proposed curves do not much differ from the Langer curve. The difference grows to an order of magnitude in life in the HCF regime (Nf > 105 cycles), where the Chopra curve becomes more conservative. In other words, the ASME code stainless steel air curve has become a subject of debate because of gradually expanding scope of fatigue assessments to HCF region. But HCF testing of stainless steels is not simple and valid data is still rare. Aim of the current study is to test applicability of the existing and proposed design criteria. For that purpose we selected a material batch, which would have been ready for plant use, e.g., in a PWR surge line. An experimental strain life fatigue curve was determined to be used as a base line for component specific evaluations and for comparison with the Langer and Chopra curves, which are the basis of the ASME III and NRC RG 1.207 design curves. Monotonic and cyclic stress strain responses were carefully analyzed to clarify the deformation and fatigue mechanisms for this steel.


Solution annealed Niobium stabilized austenitic stainless steel (X6CrNiNb1810 mod) was received as a 360x32 mm pipe, which fulfils all KTA material requirements for primary components in BWR and PWR. Chemical composition of the test material is given in Table 1. The grain size in this pipe varies so that the material report classified 50 % to ASTM 0-1 and 50 % to ASTM 2-3. Smooth round bar specimens were turned and polished from samples extracted all round the circumference of the pipe. Tensile and CSSC test specimens were selected from each quarter segment and LCF specimens were randomly picked round the pipe. The LCF specimen dimensions are shown in Fig. 1. Table 1. Composition of test material (wt %).
C 0,031 N 0,021 Si 0,235 Mn 1,885 Cr 17,30
R2 0

Ni 10,29
! 8

Mo 0,405

Nb 0,357

P 0,030

S 0,004


! 15

20 113

Figure 1. Fatigue specimen and dimensions in mm. Tensile and fatigue testes were both performed in a MTS 100 kN rig with precision alignment grips and digital control unit. Alignment of load train was adjusted with the help of strain gauged specimens according to the ASTM E 1012-05 procedure. MTS extensometers with gage lengths of 50 mm (tensile tests) and 8 mm (fatigue) were used for measuring strains. The gauge section of the tensile specimens was 8 mm x 58 mm. Five tensile tests were performed with nominal strain rates varying between 210-5 and 210-3. Partial unloadings were introduced to measure change of apparent elastic modulus as function of strain. Until onset of necking true stress ( = S(1+e) and strain ( = ln(1+e) were obtained from the nominal values (S and e). Beyond instability the local strains were deduced from periodic optical measurements of necking. Strain controlled low cycle fatigue tests were performed according to the ASTM E 606 procedure using sinusoidal waveform. An average strain rate of 0,02 1/s was used for constant amplitude tests (e.g. 1 Hz for a = 0,5 %). For longest HCF tests, the frequency was increased to 6 Hz during the late secondary hardening phase with decreasing plastic strain. For determination of cyclic stress strain curves (CSSC) a spectrum straining method developed by Solin (1986) was used. The frequency of sinusoidal ramps was varied to keep the average strain rate constant at 0,01 1/s.

A simplified definition of fatigue life (Nf,25) as the number of cycles to 25 % drop of peak stress from its absolute maximum was adopted to avoid practical problems with variable cyclic softening and hardening behaviour of stainless steels. However, all final load drop phases were so short that the obtained lives are insensitive to the selected failure criterion.



Stress strain response tensile properties and hysteresis loops

The results of tensile tests together with the material specification report results for the particular pipe and averages for the melt are summarized in Table 2. The elongation (A5) varied from 57 % to 74 %, the largest values obtained by the lowest applied strain rates. In spite of the well known correlation of instantaneous flow stress and pulling rate, an inverse correlation was seen in ultimate tensile strength, because of longer range of strain hardening with low rate. Partial unloadings seemed to have minimal effect on the tensile curves. The nominal and true stress strain curves are shown in Fig. 2. Table 2. Tensile test results in current study and in the material specification report.
data source minimum of 5 tests maximum of 5 tests average of 5 tests material report / pipe material report / melt E 195 GPa 201 GPa 197 GPa Rp0,2 224 MPa 249 MPa 238 MPa 239 MPa 251 MPa UTS 535 MPa 559 MPa 544 MPa 548 MPa 544 MPa


nominal / true stress MPa

1750 1500

true stress strain curve

1250 1000 750 X6CrNiNb1810 mod monotonic pull intermediate rate 250 unloads, slow rate 500

nominal stress strain curves

0 0 20 40 60 80 100 120

true stress strain 140 160 180

nominal / true strain %

Figure 2. Monotonic nominal and true stress strain curves. As the strain amplitude is fixed in fatigue tests, stress amplitude and mean stress depend on the material response. The stress strain responses during constant amplitude tests were studied through hysteresis loops and hardening softening curves. Hysteresis loops at half-life (N = N25 / 2) are shown in Fig. 3a. The general tradition of reporting the half-life data is applied here, although this practice may be questioned for stainless steels, because the cyclic response doesnt stabilize. A minimum stress response is found at 0,2 % strain amplitude due to secondary hardening beginning before half-life for smaller amplitudes. The hysteresis loops in Fig. 3a are positioned to reveal different strain hardening paths in tensile direction. According to Bayerlein et al. (1987) non-Masing behaviour (not coinciding loops) indicates that the dislocation microstructures and/or cyclic deformation mechanisms are amplitude dependent. But for this material the deformation mechanisms change even during a single test. Fig. 3b shows evolution of hysteresis 3

loops during a test at or just below the endurance limit. Inelastic strains are notable even in this HCF test, which lasted over 12 million cycles. More than one third of the total strain ( a = 0,185 % ) was inelastic before secondary hardening eliminated most of the hysteresis. See chapter 4.3 for further description of cyclic softening and hardening.
MPa 700








1,00% 0,80%


0,50% 0,40% 0,30% 0,22%


!a = 0,185 % N25 = 12 10 6
begin 1-2 hard 28 soft 80k sec.h. 1M EOL 10M


0,20% 0,195% 0,185%


0 0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6 1,8 % 2,0








0,15 % 0,20

Figure 3. Hysteresis loops at half-lives of different tests (a) and at different stages of a single test (b). 4.2 Apparent elastic modulus

The elastic modulus was determined in tensile tests to 197 GPa, but unloading reloading modulus is not constant as shown in Fig. 4. The apparent elastic modulus reaches a maximum just when yielding starts at about 0,1 % total strain and then gradually decreases by about 5 % until 1 % strain and by about 15 % until 10 % strain. Note that the reduction of specimen cross section was accounted for. The modules based on nominal stresses decrease even more. At high strains, changes of grain orientations may contribute, but at low strains the changes must be mainly due to internal stresses. Austenitic stainless steels are known for strain hardening and ability to generate large internal stresses by plastic strain.
210 210 X6CrNiNb1810 mod ( cross section corrected ) repeated unloads




X6CrNiNb1810 mod ( cross section corrected ) first pull 20-70 MPa


repeated unloads single unloads

160 0,0 0,2 0,4 0,6 0,8 1,0

Apparent module GPa




130 0 10 20

strain %


Figure 4. The measured apparent elastic modules as function of applied tensile strain. Evolution of microstructure and internal stresses cause changes in hysteresis loop shape. KuhlmannWilsdorf and Laird (1979) divided the shear stresses acting on the slip planes into thermal (rate dependent) 4

friction stress and athermal (elastic type) back stress rising from short distance dislocation motion. They used friction stress and back stress components for modelling hysteresis loops for stainless steels. We assume that together with the real elastic modulus microscopic or mesoscopic internal (back) stresses play a role in evolution of modulus (linear part of hysteresis loops) during cyclic deformation. A drop of modulus was systematically observed at all strain amplitude levels and during the whole fatigue life as demonstrated in Fig. 5. Reduction of modulus scaled off or even turned to increase very late during the longest tests, but it is worth of noting that even there the reducing trend continued also during the secondary hardening phase. The modulus change was not in phase with softening or hardening.
200 GPa 195




up 0,185% up 0,200% up 0,220% up 0,300% up 0,500% up 0,800% up 1,000% 10 1,E+01 100 1,E+02 1000 1,E+03 104 1,E+04 10 5 1,E+05 10 6 1,E+06

dwn 0,185% dwn 0,200% dwn 0,220% dwn 0,300% dwn 0,500% dwn 0,800% dwn 1,000% 10 7 N 1,E+08 10 8 1,E+07


170 1 1,E+00

Figure 5. Change of modulus in increasing and decreasing ramps during cyclic straining. 4.3 Cyclic softening and hardening

Change of cyclic stress response during a test is best visualized by softening / hardening curves. They show stress responses (Sa = S / 2) as function of number of cycles during constant strain amplitude tests, Fig. 6.
380 MPa 360 340 320 300 280 260 240 220 200 1 1,E+00 10 1,E+01 100 1,E+02 1000 1,E+03 104 1,E+04 105 1,E+05 106 N 1,E+07 107 1,E+06

1,0 0,8% 0,5 0,4 0,3% 0,22 0,2% 0,195% 0,190% 0,185%

Figure 6. Cyclic hardening and softening curves. Consistent and well repeatable cyclic stress strain responses including initial hardening, softening and secondary hardening were observed. Dislocation density and internal stresses are increased during the first 5

10 to 20 hardening cycles. Subsequent dislocation activity utilizes the driving force and repeated opportunities to optimize the dislocation structures. This is seen as cyclic softening, which lasts for 2000 to 100 000 cycles depending on the strain amplitude. During softening the stress response may be lowered below that for the soft annealed material. Softening is followed by secondary hardening. It is not clear, how onset of secondary hardening is determined after up to 105 softening cycles, but it is astonishingly well predictable as a function of amplitude. Secondary hardening is soon interrupted by crack growth and fracture at strain amplitudes above 0,2 %, but below this threshold, most of the fatigue life is spent in secondary hardening and the stress response will notably increase. As shown in Figs. 3b and 6 (see lowest curve in Fig. 6), the secondary hardening phase began after 80 000 cycles and occupied 99 % of the 12 million cycles endurance of a specimen tested at a = 0,185 %. 4.4 Cyclic stress strain curves

Stress strain data can also be presented as cyclic stress strain curves (CSSC), which give the stress amplitude as function of strain amplitude. Cyclic stress strain curves (CSSC) are traditionally defined as half life response of the material in constant amplitude as shown in Fig. 7a. The half life is assumed to represent stabilized response occupying most of the fatigue life. But such stabilisation doesnt occur for austenitic stainless steels. A peak is observed below 0,2 % strain amplitude, where secondary hardening is prominent. Solin (1989) demonstrated applicability of spectrum straining to determine CSSC for stainless steels, which do not stabilize in cyclic deformation. By spectrum straining method with a single specimen determined CSSC represents a certain microstructure (dislocation density and configuration), which depends mostly on the largest strains in spectrum. Therefore, realistic amplitude scales should be applied. Results with different amplitude scales are shown together with the constant amplitude data in Fig. 7a. By the spectrum straining method cyclic hardening or softening can be modelled as evolution of the CSSC in different phases of the test as demonstrated in Fig. 7b. The six first cycles are in line with the monotonic tensile curve. A reasonably stabilized CSSC for the early test condition is reached already within 40 first cycles, but cyclic hardening gradually increases the curve until crack growth takes over.
350 350 MPa 300 250 200 150 1,00% half-life 3650 100 50 0 0,0 0,2 0,4 0,6 0,8 % 1,0 0,50% half-life 18000 0,25% half-life 125000 pl. half-life 0 0,0 0,2 0,4 0,6 0,8 % 1,0 100 50 250 200 150 end-of-life (N=7300) hardest (N=5750) half-life (N=3650) early phase (N=500) first block (N=39..100) first cycles (N=21..38) first cycles (N=7..20) first cycles (N=1..6) m onotonic tensile test MPa 300

Figure 7. Summary of half-life CSSCs (a) and evolution of CSSC in a spectrum straining test (b). 4.5 Fatigue lives

The obtained fatigue lives are summarized in Figs. 8-9. The individual data points are shown in Fig. 8 together with the data set used by Langer to derive the stainless steel reference curve and the ASME design curve. Our data is within the scatter band of Langers data set and well above the ASME design curve.


Langer data Raw data N(25%) Interrupted test Mean curve Langer curve

0,1 1,E+03 10 3 1,E+04 10 4 1,E+05 10 5 1,E+06 10 6 1,E+07 107 N N 1,E+08 10 8

Figure 8. Strain life data compared to ASME 3 design curve and air curve in US NRC Reg.Guide 1.207.
1,0 0,7 0,5

strain amplitude %

ASME design curve

% 0,2

N / 12

a / 2
0,1 0,07 0,05 103 10 4 [ NUREG 6909 ] [ NRC RG 1.207 ] [ Current data & Chopra margins ]

10 5


10 7

10 8

Figure 9. Strain life data compared to ASME 3 design curve and air curve in US NRC Reg.Guide 1.207. The scatter in obtained fatigue lives (and endurance limit) is small, except at the secondary hardening transition regime ( a 0,2 % ; N25 105 ). At 0,2 % strain amplitude the secondary hardening begins at 25 to 50 % of N25. As notable fraction of LCF tests is usually spent in crack growth, we may assume that cracks had initiated before the hardening was effective. This may explain, why the fatigue lives are shorter at this level. Furthermore, three specimens of five tested at 0,2 % broke at or near the altitude of a extensometer 7

knife edge. Two other specimens broke just outside of the strain measurement gauge section (at 0,4 % and 0,185 %). These five specimens were marked with validity concerns, because such crack locations may lead to early acceleration of crack growth and reduce life. However, to stay in conservative side, even those data points are included in the geometric averages and for determining the mean curve. The in this study determined mean curve is shown in Fig. 9 together with the mean curve model developed by Chopra (2007) for 304, 316 and other grades of stainless steels (310, 347, 348, ). Our data is within the same scatter band in LCF regime, but scales off in HCF. A remarkable difference is seen for long lives ( N25 > 105 ). For comparison with the design curve endorsed by NRC (2007) for new designs in US, we determined an experimental design curve with the design margins proposed by Chopra (2007). We can see in Fig. 9 that the NRC design curve would be very conservative for this particular material batch in HCF.



Endurance limit

The obtained data in room temperature air demonstrates a typical endurance limit behaviour for this material. Note that all data below 0,195 % (at amplitudes 0,17 %; 0,185 %; 0,19 %) were run-outs against a definition of endurance limit for 107 cycles. Two tests were interrupted and one lasted for 12,3 million cycles, before fracturing outside of gauge section. Three tests at a = 0,195 % resulted to less than one million cycles, but the test at a = 0,190 % was interrupted after ten million cycles. This specimen was actually tested till failure at a = 0,22 % and it lasted more than double the longest life received at that amplitude clearly demonstrating that the specimen was not near its end of life when interrupted, and also that the preload secondary hardening was effective even at the higher amplitude. It is naturally possible that several repeats could bring finite lives at a = 0,190 %, but all three tests below a = 0,195 % showed remarkable secondary hardening and we have good reason to assume that there is an effective endurance limit below, but not much below a = 0,195 %. The endurance limit behaviour is enhanced by secondary hardening in a synergistic manner. As the specimen doesnt fail, secondary hardening gets a chance to decrease the damaging inelastic strains, thus effectively ensuring that the specimen endures even longer. Further details and discussion on the secondary hardening phenomena go beyond this presentation, but it is worth of noting that it may occur also in operation temperatures. The GE/EDF programme actually resulted to extreme secondary hardening in 300 C as convincingly reported by Solomon et al. (2004). Chopra (2007) also observed secondary hardening in elevated temperature. 5.2 Strain life curves and different stainless grades

The Langer and Chopra curves are not much different for LCF and our data lie within a common scatter band in between. The curves deviate in HCF regime, where Chopra curve becomes more conservative. Our current data for the stabilized steel has lower slope and extends even above the Langer curve in high cycle region, Fig. 8. In addition to secondary hardening, good HCF properties for this steel can be explained by a hardening effect of Niobium carbides and in general terms with the classic models correlating material strength and strain life curve slope. In recent decades, mitigation of susceptibility to stress corrosion cracking has been a major challenge for material scientists in the nuclear industry. Sensitization has been prevented mainly by reducing carbon content of the steels. As a side effect, high cycle fatigue strength of the nuclear grade stainless steels have probably decreased. If this is true, it might explain part of the differences between recent experimental data and the Langer curve. On the other hand, our current results provide supporting arguments for use of stabilized stainless steels in components, where high cycle fatigue is a concern. Together with parallel results by Solin (2006, 2009) this study clearly demonstrated that different grades of stainless steels exhibit different fatigue performance, and that the new air curve endorsed by the NRC is directly applicable only to part of the stainless steels. Based on general experience, it seems probable, that many soft grades developed particularly to exclude sensitivity to stress corrosion cracking will comply with the Chopra curve, but many grades used in the existing plants will not. So, as the NRC also states, the Regulatory Guide 1.207 as such is applicable to new (US) designs only. Furthermore, one might consider 8

determining more relevant experimental curves for material batches to be used in building of new nuclear power plants and use code curves, when relevant experimental data is not available.


Strain controlled fatigue tests for an austenitic Niobium stabilized (X6CrNiNb1810 mod) stainless steel in room temperature air led to the following conclusions:
The data demonstrates an endurance limit behaviour for this material. Fatigue failures were not

obtained within ten million cycles at strain amplitudes below 0,195 %.

The endurance limit coincides with a marked transition in the cyclic stress response. Secondary hardening is pronounced at strain amplitudes below 0,195 %. Secondary hardening

reduces inelastic strains and thus eliminates fatigue damage.

Determination of cyclic stress strain curve (CSSC) for this material is complicated by continuous

change of stress response through initial hardening, softening and secondary hardening. The data for CSSC should be based on largest strain amplitudes relevant in plant operation.
The apparent elastic modulus decreases as function of the applied strain. This was consistently

observed after partial unloads during tensile tests and also in cyclic hysteresis loops. Our LCF data for Niobium stabilized austenitic stainless steel extends to very long lives, where it is in better agreement with the Langer curve than some newly proposed air curves. This means that the original basis of ASME 3 and KTA design criteria is valid for this material batch, which is completely relevant for primary loops in certain operating PWRs. Applicability of the new air curve given in NRC RG 1.207, which is based on different fatigue data, should be questioned also for new designs utilizing this kind of material.

Acknowledgements. The reported work is part of the Technical Programme of E.ON Case on Thermal Transients funded by E.ON Kernkraft GmbH. The experiments were carried out at VTT by Mr. J. Alhainen and E. Arilahti. The analysis of results was partly funded by the Academy of Finland decision 117700.

ASME 1972. Criteria of the ASME Boiler and Pressure Vessel Code for design by analysis in sections III and VIII division 2. Pressure Vessels and Piping: Design and Analysis, A Decade of Progress, Volume One. P. 61-83. Bayerlein, M., Christ, H-J., Mughrabi, H., 1987. A critical evaluation of the incremental step test. In: Rie, KT. (ed.), Low Cycle Fatigue and Elasto-plastic Behaviour of Materials. Essex, Elsevier. pp. 149-154. Chopra, O., Shack, W., 2007. Effect of LWR Coolant Environments on the Fatigue Life of Reactor Materials, Final Report. NUREG/CR-6909, ANL-06/08, Argonne National Laboratory. 118 p. Faidy, C., 2008. Status of French fatigue analysis procedure. Proceedings of PVP 2008-ICPVT11 2008 ASME Pressure Vessel and Piping Division Conference, July 27-31, 2008, Chigago, USA. Paper PVP200861911. 9 p. Higuchi, M., 2004. Japanese program overview. 3rd International Conf. on Fatigue of Reactor Components, 3-6. 10. 2004, Seville, Spain, EPRI/OECD, 15 p. Jaske, C.E., ODonnell, W.J., 1977. Fatigue Design Criteria for Pressure Vessel Alloys, Trans. ASME J. Pressure Vessel Technol. 99, pp. 584-592. Kuhlmann-Wilsdorf, D., Laird, C., 1979. Dislocation behavior in fatigue II: friction stress and back stress as inferred from an analysis of hysteresis loops. Materials Science and Engineering 37, ss. 111-120. NRC, 2007. U.S. Nuclear Regulatory Commission Regulatory Guide 1.207, 2007. Guidelines for evaluating fatigue analyses incorporating the life reduction of metal components due to the effects of the light-water reactor environment for new reactors. 7 p.

Solin, J., 1986. A Spectrum straining method for low-cycle fatigue resistance determination (C 245/86). Fatigue of Engineering Materials and Structures, IMechE, 1986, vol 1, pp. 273-280. Solin, J., 1989. Cyclic stress strain behavior of an austenitic stainless steel Polarit 778 under variable amplitude loading. VTT Research reports 647, Espoo 1989, 42 p. Solin, J., 2006. Fatigue of stabilized SS and 316 NG alloy in PWR environment. Proceedings of PVP 2006ICPVT11 2006 ASME Pressure Vessel and Piping Division Conference, July 23-27, 2006, Vancouver, BC, Canada. Paper PVP2006-ICPVT11-93833. Solin, J., Alhainen, J., Karlsen, W., 2009. Fatigue of primary circuit components (Fate-Safir) cyclic behaviour and fatigue design criteria for stainless steel. SAFIR2010 - The Finnish Research Programme on Nuclear Power Plant Safety 2007-2010, Interim Seminar, 12-13. 3. 2009. 13 p.
( see )

Solomon, H., DeLair, R.E., Vallee, A.J., Amzallag, C., 2004. 3rd Int. Conf. on Fatigue of Reactor Components, 3-6. 10. 2004, Seville, EPRI/OECD, 22 p. STUK, 2002. YVL-guide 3.5, Ensuring the strength of nuclear power plant pressure devices, issue 5.4.2002. (in Finnish, but translations exist)