Localizing Perturbations in Pressurized Water Reactors Using One-Dimensional Deep Convolutional Neural Networks.

Area of Science:

• Nuclear Engineering
• Artificial Intelligence

Background:

• Nuclear reactor core monitoring is crucial for safety and efficiency.
• Anomalies can affect reactor performance and require precise localization.
• Traditional methods may have limitations in real-time anomaly detection.

Purpose of the Study:

• To develop and validate a deep learning approach for localizing anomalies in nuclear reactor cores.
• To utilize neutron noise signals for perturbation diagnostics.
• To assess the methodology's effectiveness across different reactor types.

Main Methods:

• Simulating reactor core perturbation scenarios to generate datasets.
• Training deep, one-dimensional, convolutional neural networks (1D CNNs) on these datasets.
• Validating the trained models using actual plant measurements from VVER-1000 reactors.

Main Results:

• The deep learning models successfully identified and located simulated perturbations.
• The methodology was validated on hexagonal, pre-Konvoi, pressurized water, and VVER-1000 reactors.
• The approach demonstrated effectiveness in analyzing neutron noise signals for anomaly detection.

Conclusions:

• Deep learning, specifically 1D CNNs, offers a promising approach for anomaly localization in nuclear reactors.
• The integration of simulated data with real plant measurements enhances model robustness.
• This methodology contributes to advanced reactor monitoring and safety systems.