Learning from Neutrons: Machine Learning and Monte Carlo Synergy in Shielding Design.

ICCESEN 2025, Antalya, Türkiye, 17 - 20 Ekim 2025, ss.33, (Özet Bildiri)

Yayın Türü: Bildiri / Özet Bildiri
Basıldığı Şehir: Antalya
Basıldığı Ülke: Türkiye
Sayfa Sayıları: ss.33
Manisa Celal Bayar Üniversitesi Adresli: Evet

Özet

Machine learning (ML), as a prominent discipline within artificial intelligence, has emerged as a key enabler of significant advancements in data-driven decision-making processes and is widely employed across diverse scientific domains for data analysis and predictive model development. In the field of radiation protection, ML offers substantial advantages over traditional approaches in terms of cost, computation time, and complexity, particularly in dose estimation and shielding design processes.

In this study, the applicability of ML algorithms for predicting radiation dose values at various distances within a tunnel system designed to attenuate secondary neutrons generated under abnormal operating conditions of proton accelerators at 250 and 1000 MeV energy regions was investigated. The performance of the developed ML models was evaluated against reference results obtained from FLUKA Monte Carlo simulations. Linear Regression, Gradient Boosting Regressor, KNN Regressor and Random Forest algorithms were employed, and the results indicated that these methods achieved high prediction accuracy (R² ≈ 0.85–0.98) for dose distributions while significantly reducing computation time compared to Monte Carlo simulations. The findings demonstrate that ML-based models can provide reliable predictions across different material, environment and energy combinations, thereby accelerating the design process and offering cost-effective solutions that contribute to enhancing safety levels in nuclear facilities.