Combining deep learning algorithm and a-star algorithm to increase the accuracy of tracking lost gamma source.

Area of Science:

• Nuclear Engineering
• Artificial Intelligence
• Robotics

Background:

• Locating lost gamma sources is critical for safety and efficiency in nuclear applications.
• Traditional search algorithms face challenges with complex environments and obstacles.

Purpose of the Study:

• To develop and evaluate an automatic search algorithm, A-star CNN-RNN (ACR), for locating lost gamma sources.
• To assess the performance of ACR compared to gradient search (GS) in various simulated environments.

Main Methods:

• A hybrid deep learning model combining Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) with the A-star algorithm.
• Input data generated using Monte Carlo N-Particle (MCNP) code simulating radiation dose rates.
• Performance evaluation based on average steps and prediction failure rates in rooms with and without obstacles.

Main Results:

• ACR demonstrated approximately half the average tracking steps compared to GS in obstacle-free environments.
• ACR achieved high accuracy (failure rates <5%) with obstacles, significantly outperforming GS (44% failure rate with an 8m wall).
• ACR maintained over 94% prediction accuracy in complex obstacle scenarios (parallel walls, L-shaped walls) without retraining.

Conclusions:

• The A-star CNN-RNN (ACR) algorithm is a robust and adaptable solution for lost gamma source detection.
• ACR offers superior performance and accuracy over traditional methods in challenging environments.
• ACR shows significant potential as a practical search algorithm for nuclear source localization.