Combining predictions in a simplified 4-Θ methodology for creep life assessment: An application to 2Cr—1Mo steel

Abstract

A simplification of the 4-Θ methodology is proposed and then used to predict creep properties under non-accelerated test conditions using data on 2Cr–1Mo steel. In the proposed simplification, creep rates at very low strains were first used to predict the minimum creep rates which are then used in the Monkman–Grant relation to predict times to failure. Because various strains can be used in this simplification, predictions using creep rates at various strains were combined in an optimal manner. These predictions were then compared with those obtained through the traditional application of the 4-Θ methodology. These predictions were assessed using the mean absolute percentage error (MAPE) and mean square error (MSE) which were further decomposed into systematic and random components.

When considering the accuracy of the extrapolations for both minimum creep rates and times to failure, most of the simplified 4-Θ models produced lower MSE and MAPE compared with the traditional implementation of the 4-Θ methodology. All models also had very similar random components of the MSE. A simplified 4-Θ model that used creep rates measured at higher strain usually produced better extrapolations (as measured by the MAPE) than those obtained by combing the extrapolations from a number of simplified 4-Θ models. However, combing predictions in this way tended to increase the random component of the MSE.