Uncertainty quantification on the prediction of creep remaining useful life

Victor Maudonet,Carlos Frederico Trotta Matt,Americo Cunha Jr

2024-09-29

Abstract:Accurate prediction of remaining useful life (RUL) under creep conditions is crucial for the design and maintenance of industrial equipment operating at high temperatures. Traditional deterministic methods often overlook significant variability in experimental data, leading to unreliable predictions. This study introduces a probabilistic framework to address uncertainties in predicting creep rupture time. We utilize robust regression methods to minimize the influence of outliers and enhance model estimates. Sobol indices-based global sensitivity analysis identifies the most influential parameters, followed by Monte Carlo simulations to determine the probability distribution of the material's RUL. Model selection techniques, including the Akaike and Bayesian information criteria, ensure the optimal predictive model. This probabilistic approach allows for the delineation of safe operational limits with quantifiable confidence levels, thereby improving the reliability and safety of high-temperature applications. The framework's versatility also allows integration with various mathematical models, offering a comprehensive understanding of creep behavior.

Computational Engineering, Finance, and Science

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to accurately predict the remaining useful life (RUL) of materials under high - temperature conditions. Traditional deterministic methods have limitations when dealing with significant variability in experimental data and model inaccuracies, resulting in unreliable prediction results. Therefore, this paper proposes a probabilistic framework to quantify and handle uncertainties in creep rupture time prediction.

Specifically, the research aims to:

1. **Introduce a probabilistic framework**: Improve the accuracy of model estimation by using robust regression methods to reduce the influence of outliers.
2. **Global sensitivity analysis**: Use Sobol indices to identify the most influential parameters.
3. **Monte Carlo simulation**: Determine the probability distribution of the material's RUL.
4. **Model selection techniques**: Include Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to ensure the selection of the optimal prediction model.

Through these methods, this research hopes to provide a more reliable, safe, and quantifiable confidence - level method to evaluate material behavior in high - temperature applications, thereby improving the design and maintenance strategies of industrial equipment.

### Summary of Key Formulas

- **Larson - Miller Parameter Relationship**:
\[
P_{\text{LM}}(\sigma)=T(C_{\text{LM}}+\log t_r)
\]
where \(P_{\text{LM}}\) is the Larson - Miller parameter, \(C_{\text{LM}}\) is the Larson - Miller constant, \(T\) is the temperature, and \(t_r\) is the rupture time.

- **Orr - Sherby - Dorn Parameter Relationship**:
\[
P_{\text{OSD}}(\sigma)=\log t_r-\frac{C_{\text{OSD}}}{T}
\]
where \(P_{\text{OSD}}\) is the Orr - Sherby - Dorn parameter, and \(C_{\text{OSD}}\) is the Orr - Sherby - Dorn constant.

- **Manson - Succop Parameter Relationship**:
\[
P_{\text{MS}}(\sigma)=\log(t_r)+C_{\text{MS}}T
\]
where \(P_{\text{MS}}\) is the Manson - Succop parameter, and \(C_{\text{MS}}\) is the Manson - Succop constant.

- **Polynomial Regression**:
\[
P(\sigma)=a_0 + a_1\sigma+\cdots+a_{n - 1}\sigma^{n - 1}
\]

- **STLS Optimization Problem**:
\[
x^*=\arg\min_{x\in\mathbb{R}^m}\|y - [A]x\|_2^2+\lambda\|x\|_0
\]

- **Root Mean Square Error (RMSE)**:
\[
\text{RMSE}=\sqrt{\frac{1}{k}\sum_{i = 1}^k e_i^2}
\]

- **Sobol Index**:
\[
S_i=\frac{V(E(Y|X_i))}{V(Y)}
\]

Through these formulas and methods, the paper provides a systematic framework to quantify and handle uncertainties in creep life prediction, thereby improving the reliability and accuracy of prediction.