High Cycle Fatigue (HCF) failure is a common failure type for many mechanical components. Traditional HCF design is based on the deterministic safety factor approach, typically used in conjunction with the point stress method. A current development is to explicitly model the uncertainty of the design set-up, and compute the probability of failure, pf. If pf can be computed in an appropriate way, the contributions to fatigue can be identified and managed. Probabilistic design gives improved control over safety, which helps to avoid overly conservative design.
One of the applications dealt with in this work is gas turbine compressor blades. For this type of component requirements on safety coincide with requirements on high efficiency, low weight, etc. In such case, methods for accurate fatigue assessment become extra important.
In order to perform an appropriate fatigue design, certain requirements must be fulfilled. For example, the fatigue model that is used must be accurate and the relevant material parameters must be accurately determined. Other requirements are that the mesh used in the FE-computations for the stress field is fine enough, a HCF post-processor that enables application of fatigue models to real components must be available and a method for computation of pf including all uncertainties should also be available.
In Paper A, it is shown that for a gas turbine compressor blade, it is the number of elements through the blade’s thickness that is the most important mesh property for convergence in .
In Paper B, it is investigated which test strategy that should be used in order to perform accurate estimations of material parameters in multiaxial HCF criteria by use of as few laboratory tests as possible when different types of scatter are present, and when the cost to perform the fatigue tests is taken into consideration. It is shown that performing tests on few stress ratios located far away from each other is the best strategy, and that for tests performed in a high quality laboratory, scatter in specimen misalignment has an insignificant influence on the parameter estimation.
In Papers C and D, the prediction accuracy for the probabilistic volume based Weakest Link (WL) model and the Volume method for the Probability of Fatigue (VPF) is investigated. A novel specimen design is suggested for investigation of the volume effect. Based on the results, the newly developed VPF is favoured for design purpose. In Paper D, the HCF post-processor AROMA-PF is also presented, and used for computation of pf for a real gas turbine compressor blade geometry. The behavior of the predicted fatigue probability curves is very different between WL and the VPF for low pf-values.
In Paper E, a new method for fatigue probability assessment is presented. The classification of aleatory uncertainty type 1 and type 2 is also introduced. The suggested method is applied to a bladed disk in a gas turbine for computation of pf. The results show that the epistemic uncertainty in the modeling of the aero-forcing gives the major contribution to uncertainty in pf.
Stockholm: KTH Royal Institute of Technology, 2016. , 45 p.