A framework for probabilistic prediction of remaining useful life in structural materials

TL;DR

This paper presents a probabilistic framework for predicting the remaining useful life of high-temperature structural materials under creep conditions, incorporating uncertainty quantification and model selection techniques.

Contribution

It introduces a comprehensive probabilistic approach combining robust regression, sensitivity analysis, and Monte Carlo simulations for more accurate RUL predictions.

Findings

Framework quantifies uncertainties in creep rupture time predictions.

Sensitivity analysis identifies key contributors to model uncertainty.

Model selection based on statistical criteria improves prediction reliability.

Abstract

Accurate prediction of remaining useful life under creep conditions is essential for the structural reliability of high-temperature components in critical engineering systems. Traditional approaches based on deterministic parametric models often overlook the substantial variability inherent in experimental data, compromising the accuracy and robustness of long-term predictions. This study introduces a probabilistic framework to quantify uncertainties in creep rupture time prediction. Robust regression techniques are first applied to mitigate the influence of outliers and enhance the stability of model estimates. Global sensitivity analysis using Sobol indices is then employed to identify the dominant contributors to model uncertainty, followed by Monte Carlo simulations to propagate these uncertainties and estimate the distribution of the remaining useful life. Finally, model selection is guided by statistical criteria, including the Akaike and Bayesian information criteria, to identify the most reliable predictive model. The proposed framework not only enables the definition of safe operational limits with quantifiable confidence levels but is also general and extensible to other time-dependent degradation phenomena, such as fatigue and creep-fatigue interaction.