Swelling of Steels Subjected to Fast Neutrons’ Irradiation using Optimized Ensemble Regressions
- 1Department of Materials and Construction Engineering, Hebei Institute of Mechanical and Electrical Technology, Xingtai City, Hebei Province, 054000, China.Hebei Technology Innovation Centre for New Energy Vehicle Integrated Die Casting, Xingtai City, Hebei Province, 054000, China.
- 2Department of Materials and Construction Engineering, Hebei Institute of Mechanical and Electrical Technology, Xingtai City, Hebei Province, 054000, China.Hebei Science Popularization Base for Additive Manufacturing, Xingtai City, Hebei Province, 054000, China.Hebei Technology Innovation Centre for New Energy Vehicle Integrated Die Casting, Xingtai City, Hebei Province, 054000, China.
Abstract
Radiation-induced swelling (swe) of structural steels under fast neutron irradiation is a critical degradation phenomenon that limits the service lifetime and safety of nuclear reactor components. Conventional experimental approaches for evaluating swelling behavior are costly, time-consuming, and difficult to apply under high-dose irradiation conditions. This study develops an optimized machine learning framework for predicting the swelling behavior of austenitic steels exposed to fast neutron irradiation. A dataset containing 444 experimental samples collected from published literature was established, including twelve input variables related to chemical composition (Fe, Cr, Ni, Mo, Mn, C, Ti, Si, and P), cold work degree, irradiation dose, and temperature. Two ensemble regression models, Adaptive Boosting Regression (ABR) and Extra Tree Regression (ETR), were optimized using the Tyrannosaurus Rex Optimization (TRO) algorithm to improve hyperparameter selection and prediction performance. The dataset was randomly divided into 75% training data (333 samples) and 25% testing data (111 samples), with preprocessing methods including normalization, randomization, interquartile range (IQR)-based outlier treatment, and four-fold cross-validation applied to enhance model reliability. The optimized Extra Tree Regression model (ExtraT-T) achieved superior predictive performance, obtaining coefficients of determination (R²) of 0.9908 and 0.9563 for the training and testing datasets, respectively, compared with 0.9679 and 0.8991 for the optimized ABR model (AdaB-T). Feature analysis revealed that irradiation dose and cold work degree were the dominant factors affecting swelling behavior. The proposed optimized machine learning framework provides an efficient approach for predicting radiation-induced swelling, reducing dependence on irradiation experiments, and supporting the design of radiation-resistant nuclear materials.
Related
- Altermagnetism revealed by polarized neutrons in MnF 2
- Large deviations for linear regressions
- Neutron Optical Constraints on Exotic Parity-odd Interactions between Neutrons and Nucleons
- Fault-Class-Matched Test Oracles for Output-Invisible Quantum Transpiler Regressions
- Urban climate regressions require representative monitoring networks
BibTeX
@article{ren2026swelling,
title = {{Swelling of Steels Subjected to Fast Neutrons’ Irradiation using Optimized Ensemble Regressions}},
author = {Ziqi Ren and Yaping Zhao and Baoxiu Li and Zixuan Zhang},
journal = {Journal of Applied Science and Engineering},
year = {2026},
month = oct,
doi = {10.6180/jase.202701_36.046},
url = {https://doi.org/10.6180/jase.202701_36.046},
note = {CC BY},
}
RIS
TY - JOUR
TI - Swelling of Steels Subjected to Fast Neutrons’ Irradiation using Optimized Ensemble Regressions
AU - Ziqi Ren
AU - Yaping Zhao
AU - Baoxiu Li
AU - Zixuan Zhang
PY - 2026
DA - 2026/10/31
JO - Journal of Applied Science and Engineering
DO - 10.6180/jase.202701_36.046
UR - https://doi.org/10.6180/jase.202701_36.046
AB - Radiation-induced swelling (swe) of structural steels under fast neutron irradiation is a critical degradation phenomenon that limits the service lifetime and safety of nuclear reactor components. Conventional experimental approaches for evaluating swelling behavior are costly, time-consuming, and difficult to apply under high-dose irradiation conditions. This study develops an optimized machine learning framework for predicting the swelling behavior of austenitic steels exposed to fast neutron irradiation. A dataset containing 444 experimental samples collected from published literature was established, including twelve input variables related to chemical composition (Fe, Cr, Ni, Mo, Mn, C, Ti, Si, and P), cold work degree, irradiation dose, and temperature. Two ensemble regression models, Adaptive Boosting Regression (ABR) and Extra Tree Regression (ETR), were optimized using the Tyrannosaurus Rex Optimization (TRO) algorithm to improve hyperparameter selection and prediction performance. The dataset was randomly divided into 75% training data (333 samples) and 25% testing data (111 samples), with preprocessing methods including normalization, randomization, interquartile range (IQR)-based outlier treatment, and four-fold cross-validation applied to enhance model reliability. The optimized Extra Tree Regression model (ExtraT-T) achieved superior predictive performance, obtaining coefficients of determination (R²) of 0.9908 and 0.9563 for the training and testing datasets, respectively, compared with 0.9679 and 0.8991 for the optimized ABR model (AdaB-T). Feature analysis revealed that irradiation dose and cold work degree were the dominant factors affecting swelling behavior. The proposed optimized machine learning framework provides an efficient approach for predicting radiation-induced swelling, reducing dependence on irradiation experiments, and supporting the design of radiation-resistant nuclear materials.
SN - 2708-9967
SN - 2708-9975
ER -