Several (accumulative) lifetime models were developed to assess the lifetime consumption of high-temperature components of steam and gas turbine power plants during flexible operation modes. These accumulative methods have several drawbacks, e.g. that measured loading profiles cannot be used within accumulative lifetime methods without manual corrections, and cannot be combined directly to sophisticated probabilistic methods. Although these methods are widely accepted and used for years, the accumulative lifetime prediction procedures need improvement regarding the lifetime consumption of thermal power plants during flexible operation modes. Furthermore, previous investigations show that the main influencing factor from the materials perspective, the critical damage threshold, cannot be statistically estimated from typical creep-fatigue experiments due to massive experimental effort and a low amount of available data.
This paper seeks to investigate simple damage mechanics concepts applied to high-temperature components under creep-fatigue loading to demonstrate that these methods can overcome some drawbacks and use improvement potentials of traditional accumulative lifetime methods. Furthermore, damage mechanics models do not provide any reliability information, and the assessment of the resultant lifetime prediction is nearly impossible. At this point, probabilistic methods are used to quantify the missing information concerning failure probabilities and sensitivities and thus, the combination of both provides rigorous information for engineering judgment.
Nearly 50 low cycle fatigue experiments of a high chromium cast steel, including dwell times and service-type cycles, are used to investigate the model properties of a simple damage evolution equation using the strain equivalence hypothesis. Furthermore, different temperatures from 300 °C to 625 °C and different strain ranges from 0.35% to 2% were applied during the experiments. The determination of the specimen stiffness allows a quantification of the damage evolution during the experiment. The model parameters are determined by Nelder-Mead optimization procedure, and the dependencies of the model parameters concerning to different temperatures and strain ranges are investigated. In this paper, polynomial chaos expansion (PCE) is used for uncertainty propagation of the model uncertainties while using non-intrusive methods (regression techniques). In a further post-processing step, the computed PCE coefficients of the damage variable are used to determine the probability of failure as a function of cycles and evolution of the probability density function (pdf).
Except for the selected damage mechanics model which is considered simple, the advantages of using damage mechanics concepts combined with sophisticated probabilistic methods are presented in this paper.