Prediction of High-Temperature Creep Life of Austenitic Heat-Resistant Steels Based on Data Fusion

Limin Wei, Shuo Wang, Weixun Hao, Jingtao Huang, Nan Qu, Yong Liu, Jingchuan ZhuMetals 2023 · DOI

TL;DR

This paper introduces a machine learning approach to predict the high-temperature creep life of austenitic heat-resistant steels, crucial for the safe operation of power plant components. By fusing data from various steels and employing models like Gaussian Regression, they achieve significantly higher prediction accuracy and generalization than traditional methods, offering a more robust tool for material performance assessment.

What Problem Does This Paper Address?

High-temperature components in thermal power plants, often made of austenitic heat-resistant steels, are susceptible to creep damage. Predicting the remaining service life (creep life) of these components is vital for ensuring operational safety and preventing catastrophic failures. Traditional methods, such as the Larson-Miller (LM) parameter method, often rely on a limited number of input parameters (stress, temperature, time) and can struggle with the complex, non-linear relationships that govern creep behavior, especially when dealing with diverse material compositions and microstructures. This limits their generalizability and accuracy, particularly for predicting long-term creep life. The paper aims to bridge this gap by leveraging the power of machine learning to create more accurate and adaptable creep life prediction models.

Methods

The core of this research lies in applying and comparing various machine learning (ML) regression algorithms for creep life prediction. The authors explored:

  • Support Vector Machine (SVM): A powerful algorithm for classification and regression, known for its ability to handle non-linear relationships through kernel functions.
  • Ensemble Learning: Combines multiple base learners to improve generalization performance.
  • Gaussian Regression (Gaussian): A non-parametric probabilistic model that excels at quantifying prediction uncertainty.
  • Decision Tree (DT): A simple, interpretable model that partitions data based on feature values.
  • BP Neural Network (BPNN): A widely used neural network architecture for pattern recognition and prediction.

To handle the high dimensionality of material data, Principal Component Analysis (PCA) was employed to reduce the number of input features while retaining essential information, leading to the creation of three distinct datasets (based on 1, 2, or all principal components). The models were trained and evaluated using a 10-fold cross-validation approach, with performance assessed by prediction accuracy and the coefficient of determination (R²). The study also compared the ML models against the conventional Larson-Miller (LM) parametric method.

Datasets

The study utilized a dataset comprising creep life data from six typical austenitic heat-resistant steels: TP304H, TP321H, TP316H, TP347H, S30432, and TP310HCbN. A total of 1060 datasets were collected. The characteristic parameters included chemical composition, grain size, test temperature, and test stress. For comparative analysis, specific data for S30432 steel was used to directly compare ML predictions with the LM method. No new benchmark datasets were explicitly introduced, but the collected dataset represents a fusion of data from multiple steel grades.

Key Insights and Ideas

  • Data Fusion is Key: Combining creep life data from multiple austenitic steel grades significantly enhances the predictive power of machine learning models, allowing them to generalize better than models trained on single-grade data.
  • Model-Dataset Synergy: The "best" machine learning model is not universal; its performance is highly dependent on the dataset it's trained on. The Gaussian model, when trained on a dataset derived from the first two principal components (8 features), achieved the highest accuracy.
  • ML Outperforms Traditional Methods: Machine learning models, particularly Gaussian Regression, demonstrated superior prediction accuracy and generalization capabilities compared to the conventional Larson-Miller method, especially when dealing with a broader range of input features.
  • Input Feature Importance is Complex: While some elements (C, P, B) show a significant, non-linear impact on creep life, the influence of others (Cr, Ni, Al, S) is more nuanced and can even be detrimental at higher concentrations. Grain size also plays a crucial role, with larger grains generally leading to longer creep life.
  • Uncertainty Quantification Matters: Gaussian Regression's ability to quantify prediction uncertainty is a valuable asset, providing not just a prediction but also an indication of its reliability.

Key Results

  • Highest Prediction Accuracy: The Gaussian model, using the second dataset (derived from the first two principal components), achieved the highest prediction accuracy of 0.991.
  • SVM Performance: The SVM model showed the lowest prediction accuracy, ranging from 0.40 to 0.48, though it improved with more input features.
  • LM vs. ML: For S30432 steel, the conventional LM method had a prediction accuracy of 0.752, while the Gaussian model (using the second dataset) achieved 0.972, demonstrating a significant improvement.
  • Generalization: The ML models, especially Gaussian, showed better generalization ability across different austenitic steels compared to the LM method.
  • Input Feature Trends: The study identified specific trends for various elements and grain size influencing creep life, largely consistent with experimental observations and theoretical understanding. For instance, increasing grain size and amounts of C, B, and Nb generally improve creep life, while decreasing Cr, Ni, and Al is advisable.

Limitations and Future Work

  • Data Availability: The study notes the unavailability of data on precipitated phases, which are known to significantly influence creep properties. Future work could incorporate this data if available.
  • Model Specificity: While the Gaussian model with the second dataset performed best, the paper emphasizes that optimal model-dataset combinations are crucial. Further exploration of other ML architectures and feature selection techniques could yield even better results.
  • Experimental Validation: While the ML predictions align with theoretical understanding, direct experimental validation of the optimized compositions suggested by the ML models would be valuable.
  • Broader Steel Grades: Expanding the dataset to include a wider range of austenitic heat-resistant steels could further enhance the robustness and applicability of the developed ML models.

Summary updated

Abstract

The creep life prediction of austenitic heat-resistant steel is necessary to guarantee the safe operation of the high-temperature components in thermal power plants. This work presents a machine learning model that can be applied to predict the creep life of austenitic steels, offering a novel method and approach for such predictions. In this paper, creep life data from six typical austenitic heat-resistant steels are used to predict their creep life using various machine learning models. Moreover, the dissimilarities between the machine learning model and the conventional lifetime prediction method are compared. Finally, the influence of different input characteristics on creep life is discussed. The results demonstrate that the prediction accuracy of machine learning depends on both the model and the dataset used. The Gaussian model based on the second dataset achieves the highest level of prediction accuracy. Additionally, the accuracy and the generalization ability of the machine learning model prediction are significantly better than those of the traditional model. Lastly, the effect of the input characteristics on creep life is generally consistent with experimental observations and theoretical analyses.
BibTeX
@inproceedings{wei2023prediction,
  author = {Limin Wei and Shuo Wang and Weixun Hao and Jingtao Huang and Nan Qu and Yong Liu and Jingchuan Zhu},
  title = {{Prediction of High-Temperature Creep Life of Austenitic Heat-Resistant Steels Based on Data Fusion}},
  booktitle = {Metals},
  year = {2023},
  doi = {10.3390/met13091630},
}
Figure