Suitably Accounting for Mean Stress Effects in Pseudo-Stress-Based Design Fatigue Curves

Abstract Design fatigue curves in nuclear pressure vessel codes, such as those presented in Appendix I of Section III of the ASME Boiler and Pressure Vessel Code (BPVC), are typically based on fully-reversed, constant-amplitude strain-controlled testing, with nominally zero mean stress. It has long been recognised that the presence of a mean stress has an effect on fatigue life, with cycling about a mean tensile stress being more damaging than cycling of the same amplitude in the presence of zero mean stress. A number of different mean stress correction approaches have been developed, each with their own strengths, with the majority of ASME fatigue curves utilising the Modified Goodman correction. Sources of mean stress include residual stresses from manufacture and mean stresses set up by prior loading. During the design phase it is not typically possible to know what mean stresses will be present in conjunction with cyclic loading, particularly because mean stresses set up by prior loading will be a function of the order of those operational loadings, something which is not typically known or is difficult to predict the resulting stress response of. Therefore, in the ASME BPVC and various other codes, a bounding mean stress correction is applied within the design fatigue curve construction. Existing design fatigue curves within the ASME code present what are really strain-vs-life curves as pseudo-stress-vs-life curves, for ease of integration with elastic assessments. These curves generally utilise the method presented by B. F. Langer in 1962 to account for this bounding mean stress effect. Transference factors are also applied on both stress and cycles, to account for data scatter and other perceived differences between real components and the small, polished, constant-amplitude specimen tests, upon which the mean fatigue curves are based. The lower of the two shifted curves at any point defines the design curve. The approach to mean stress correction makes an assumption of elastic-perfectly-plastic material behaviour, which leads to an over-estimate of mean stress shakedown behaviour. Furthermore, the incorrect assumption of linear-elastic behaviour leads to the misuse of stress-life based mean stress correction methods, which are non-conservative when applied based on pseudo-stress. Depending on the conservatism of the transference factors, and whether bounding mean stresses are actually present, there is a danger that the resulting design fatigue curve could be non-conservative. A large number of studies in recent years have indicated there is substantial conservatism in the existing transference factors. However, removal of such conservatism will require a more robust mean stress correction approach. This paper presents recent variable amplitude fatigue initiation testing of stainless steel in air and simulated PWR environment conducted under the INCEFA-SCALE test program specifically designed to set up and periodically reinforce tensile or compressive mean stresses for a range of loading conditions. This testing verifies the discussed non-conservatisms, both the shortfall in the stress range over which mean stresses can persist, and the magnitude of required mean stress correction. The paper goes on to present two new design fatigue curve derivation methods that address these issues in a pragmatic and proportionate way; a detailed approach from first principals, and a simplified approach that is more suited to code implementation.