Validation of statistical models for creep rupture by parametric analysis

Statistical analysis is an efficient method for the optimisation of any candidate mathematical model of creep rupture data, and for the comparative ranking of competing models. However, when a series of candidate models has been examined and the best of the series has been identified, there is no statistical criterion to determine whether a yet more accurate model might be devised. Hence there remains some uncertainty that the best of any series examined is sufficiently accurate to be considered reliable as a basis for extrapolation.This paper proposes that models should be validated primarily by parametric graphical comparison to rupture data and rupture gradient data. It proposes that no mathematical model should be considered reliable for extrapolation unless the visible divergence between model and data is so small as to leave no apparent scope for further reduction.This study is based on the data for a 12% Cr alloy steel used in BS PD6605:1998 to exemplify its recommended statistical analysis procedure. The models considered in this paper include a) a relatively simple model, b) the PD6605 recommended model and c) a more accurate model of somewhat greater complexity.

► The paper discusses the validation of creep rupture models derived from statistical analysis.
► It demonstrates that models can be satisfactorily validated by a visual-graphic comparison of models to data.
► The method proposed utilises test data both as conventional rupture stress and as rupture stress gradient.
► The approach is shown to be more reliable than a well-established and widely used method (BS PD6605).