On the Scatter of Creep Data: Methods to Increase Modelling Accuracy Accounting for Batch-to-Batch Dispersion

Abstract

Gas turbine components and many industrial high temperature components suffer from creep, a viscous effect of the material that induce irreversible deformation, microstructural damage, and eventually failure. Creep strain (where the creep strain εcr is a function of time, stress, temperature) and creep rupture models (where the rupture time texp is a function of stress and temperature) are fitted to the results of expensive and time-consuming experimental tests, which can last for several years (e.g. test duration up to 100kh – 200kh). At longer times, in the range of components expected life target, when it is more likely to observe creep damage, the accuracy of creep models is required to be as high as possible. It is therefore crucial to optimize the model fitting process in order to minimize the error and reduce the number of tests required. To achieve such results the experimental result dispersion needs to be properly addressed. In particular, the differences between the different heats of the same material are known to be a dominant source of uncertainty in the experimental results. The differences are mainly linked to small variations in the fabrication process or chemical composition (even within the allowed variations of the purchase specification or standard recommendations), which can generate different microstructures and mechanical behavior. This is known as batch-to-batch dispersion and this phenomenon is responsible for significant creep strength differences between heats. It is essential for the model reliability to gain the best possible insight of how the model itself can be influenced by the peculiarities and homogeneity of the available data. In order to achieve such goal, many analyses can be performed: quantitative identification strong/weak batches, analysis of the dataset inhomogeneity (i.e. a predominance of weak batches at a certain temperature or times), identification of correlations (e.g. tensile strength and chemistry, etc.), identification of creep mechanisms transitions that affect the applicability range of the model. A statistical analysis of the test results is conducted in order to enable a non-deterministic modelling of creep rupture and strength, separately accounting for in-batch and batch-to-batch sources of dispersion. The soundness of the proposed probabilistic framework is validated via Monte Carlo simulation. The paper is intended to provide an overview of the most recent proposals and progresses of the existing methods to deal with the problem and propose additional original methods to improve the analysis and the fitting procedure.