SLAM 시스템과의 연계를 통한 3D 선량률 예측 심층신경망 모델 개발
Development of 3D Dose Rate Prediction DNN Model in conjunction with SLAM System
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Based on the principle of ALARA (As Low As Reasonable Achievable), workers in radiation field ofshould check their effective dose using dosimeters in the workplace and ensure that it does not exceed 50mSv peryear, and 100 mSv for 5 years. However, the dosimeter, which has been mainly used such as a thermoluminescencedosimeter (TLD), is a method of evaluating the accumulated dose through post-processing and cannot be utilized toprevent additional exposure in real-time. This research team developed a Deep Neural Network (DNN) model thatgenerates a 3D dose rate distribution map based on a Simultaneous Localization And Mapping (SLAM) algorithmconsidering time and location by combining the dosimeter and LiDAR sensor. The developed SLAM algorithm notonly records the x, y, z coordinates and dose rates on the moving path, but also outputs a 3D point cloud data (PCD)that scans the entire space. Furthermore, to increase the prediction accuracy, d, which is the distance to the highestdose rate measurement location, was calculated through PCD and utilized in the DNN model. Finally, the DNNmodel was trained by setting the moving path coordinates and distance as input data and the dose rate as output data. The indoor/outdoor PCD and distance were used as inputs to the DNN model to predict the dose rate distribution forthe entire space. The predicted value was integrated into the PCD file by setting the color map in proportion to itsintensity. Finally, a system to predict the dose rates for the entire scan area and display them on the scanned 3D mapwas proposed. The generated 3D map can be used for follow-up management to minimize the exposure dose for theworkers by enabling the prediction of the risk area for the entire area.
Abstract
Based on the principle of ALARA (As Low As Reasonable Achievable), workers in radiation field ofshould check their effective dose using dosimeters in the workplace and ensure that it does not exceed 50mSv peryear, and 100 mSv for 5 years. However, the dosimeter, which has been mainly used such as a thermoluminescencedosimeter (TLD), is a method of evaluating the accumulated dose through post-processing and cannot be utilized toprevent additional exposure in real-time. This research team developed a Deep Neural Network (DNN) model thatgenerates a 3D dose rate distribution map based on a Simultaneous Localization And Mapping (SLAM) algorithmconsidering time and location by combining the dosimeter and LiDAR sensor. The developed SLAM algorithm notonly records the x, y, z coordinates and dose rates on the moving path, but also outputs a 3D point cloud data (PCD)that scans the entire space. Furthermore, to increase the prediction accuracy, d, which is the distance to the highestdose rate measurement location, was calculated through PCD and utilized in the DNN model. Finally, the DNNmodel was trained by setting the moving path coordinates and distance as input data and the dose rate as output data. The indoor/outdoor PCD and distance were used as inputs to the DNN model to predict the dose rate distribution forthe entire space. The predicted value was integrated into the PCD file by setting the color map in proportion to itsintensity. Finally, a system to predict the dose rates for the entire scan area and display them on the scanned 3D mapwas proposed. The generated 3D map can be used for follow-up management to minimize the exposure dose for theworkers by enabling the prediction of the risk area for the entire area.
참고문헌 (12)
- ICRP, "Recommendations of the ICRP, ICRP Publication 26, Ann. ICRP 1 (3)", ~ (1977)
- MacFarlane JW, "Lightweight Aerial Vehicles for Monitoring, Assessment and Mapping of Radiation Anomalies", Journal of Environmental Radioactivity, 136, pp.127~130 (2014)[DOI]
- Martin PG, "The Use of Unmanned Aerial Systems for the Mapping of Legacy Uranium Mines", Journal of Environmental Radioactivity, 143, pp.135~140 (2015)[DOI]
- Simons M, "Localization of hotspots via a lightweight system combining Compton imaging with a 3D lidar camera", Nuclear Engineering and Technology, 56(8), pp.3188~3198 (2024)[DOI]
- Katagiri H, "Development of an all-sky gamma-ray Compton camera based on scintillators for high-dose environments", Journal of Nuclear Science and Technology, 55(10), pp.1172~1179 (2018)[DOI]
- Sato Y, "Radiation imaging using a compact Compton camera inside the Fukushima Daiichi Nuclear Power Station building", Journal of Nuclear Science and Technology, 55(9), pp.965~970 (2018)[DOI]
- Sato Y, "Radiation imaging using an integrated Radiation Imaging System based on a compact Compton camera under unit 1", 2 exhaust stack of Fukushima Daiichi Nuclear Power Station, 2022, pp.59(6), 677~687
- Lee U, "Machine learning-based evaluation technology of 3D spatial distribution of residual radioactivity in large-scale radioactive structures", Nuclear Engineering and Technology, 56, pp.3199~3209 (2024)[DOI]
- Kwack D, "Development of Real-time Path-based 3D Dose Rate Mapping System", Nuclear Engineering and Technology under review
- "Understanding and Implementing SLAM with NVIDIA Jetson Nano" (2023)[URL]
- Xu W, "FAST-LIO : A Fast, Robust LiDAR-inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter", IEEE Robotics and Automation Letters, 6(2), pp.3317~3324 (2021)[DOI]
- Kingma DP, "Adam: A method for stochastic optimization", Proc. 3rd Int. Conf. on Learning Representations (2015)