Combining CNN and Q-learning for increasing the accuracy of lost gamma source finding

Area of Science:

• Nuclear Engineering
• Robotics and Automation
• Artificial Intelligence

Background:

• Nuclear technology, particularly gamma sources, is vital in industry and medicine.
• Locating lost gamma sources in shielded irradiation rooms is a significant safety challenge.
• Existing methods may be inefficient or inaccurate due to radiation shielding barriers.

Purpose of the Study:

• To develop an efficient and robust method for simultaneously locating and routing lost gamma sources.
• To address the challenges posed by radiation blocking barriers in gamma irradiation rooms.
• To improve the safety and efficiency of operations involving gamma sources.

Main Methods:

• Utilized a combination of convolutional neural network (CNN) and Q-learning algorithms.
• Developed a simulated environment using Geant4 for gamma source scenarios.
• Tested the combined CNN and Q-learning approach in geometries with radiation blocking barriers.

Main Results:

• Achieved 90% accuracy in locating lost gamma sources.
• Demonstrated the capability to perform simultaneous locating and routing.
• The combined method significantly reduced search time and increased accuracy, even with thick barriers.

Conclusions:

• The integration of CNN and Q-learning offers a powerful solution for lost gamma source retrieval.
• This hybrid approach overcomes limitations of traditional methods in complex shielded environments.
• The findings contribute to enhanced safety protocols in nuclear technology applications.