Aging grade estimation of Z3CN20-09M steel from nuclear power plants using LIBS based on fiber laser ablation combined with mutual information-random forest

Abstract

Estimating the aging state of the main pipeline steel Z3CN20-09M during its service life is critical for the safe operation of nuclear power plants. This study proposes an innovative approach combining fiber laser-based laser-induced breakdown spectroscopy (FL-LIBS) with mutual information-random forest (MI-RF) to estimate the aging grade of Z3CN20-09M steel. The spectral characteristics corresponding to ten distinct aging grades of the steel were analyzed. Considering the surface elemental inhomogeneity of alloy steels, the impact of various spectral characterization scales (SCS) on the classification accuracy of the RF-based model was investigated. The model's performance was further optimized through the application of the MI feature extraction method. Finally, the robustness of the model was evaluated under conditions of limited training data. The results demonstrate significant inhomogeneity in the distribution of elemental concentrations across the surface of the Z3CN20-09M samples. The RF model achieved optimal performance at an SCS of 1.2 mm. By extracting the first 1476 high-scoring features via mutual information, the classification accuracy of the prediction set rose to 99.0%, with notable enhancements in both precision and recall. Finally, the robustness of the MI-RF model was verified even when the number of samples obtained was insufficient. These findings indicate that the combination of FL-LIBS with MI-RF provides a promising approach for the in situ, rapid estimation of the aging state of essential metal components in nuclear facilities.