Machine Learning Aided Aerial Radiation Mapping
In recent years, radiation mapping has attracted widespread research interest along with increasing public concerns on environmental monitoring. However, due to the complex mechanisms of gaseous radionuclide dispersion, radiation-matter interaction, and the current limitation of dose rate data colle...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2022
|
| Online Access: | https://hdl.handle.net/1721.1/139890 https://orcid.org/0000-0003-2127-3414 |
| _version_ | 1826196725668249600 |
|---|---|
| author | Xue, Shangjie |
| author2 | Hu, Lin-wen |
| author_facet | Hu, Lin-wen Xue, Shangjie |
| author_sort | Xue, Shangjie |
| collection | MIT |
| description | In recent years, radiation mapping has attracted widespread research interest along with increasing public concerns on environmental monitoring. However, due to the complex mechanisms of gaseous radionuclide dispersion, radiation-matter interaction, and the current limitation of dose rate data collection, radiation mapping is considered to be a challenging task. In this study, a general framework for radiation mapping is proposed in static and dynamic scenarios using machine learning techniques. The proposed method enables rapid radiation mapping, as well as trajectory planning for measurements.
Firstly, a novel directional radiation detection algorithm is presented. Single-pad radiation detector arrays and attenuation materials are proposed to be used for radiation detection. This thesis presents a deep neural network model to estimate the angular distribution of the incident radiation. Wasserstein distance is applied as a loss function to train the neural network for accurate prediction. Furthermore, radiation mapping could be enabled by performing directional measurements at different positions. In particular, optimization-based approaches are presented to fuse the directional measurement results for source localization and radiation mapping in static scenarios.
Secondly, this thesis presents a model for tracking dynamic radionuclide atmospheric dispersion using probabilistic graphical models. Kalman Filter is applied for incremental estimation of atmospheric concentration and ground release simultaneously, as well as the prediction of concentration evolution. Moreover, a path planning algorithm by maximizing the information gathered from the measurements is also presented. The presented method enables joint estimation of the concentration and release source distributions in dynamic scenarios. Such method is also able to plan for future measurements in order to obtain more accurate estimations from the environments, given the previous measurement results. Simulation results for an environment similar to MIT research reactor showed that for radiation release during an accident, the proposed aerial radiation mapping algorithm is able to achieve relative error <10% within a few minutes using three or more agents, and hence showing potential advantages over conventional approaches, which require manual survey at selected locations, and take about an hour with the existing emergency procedure.
This work provides an algorithmic basis for radiation mapping problem, and shows potential to enable autonomous radiation surveys. |
| first_indexed | 2024-09-23T10:36:57Z |
| format | Thesis |
| id | mit-1721.1/139890 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T10:36:57Z |
| publishDate | 2022 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1398902022-02-08T03:49:33Z Machine Learning Aided Aerial Radiation Mapping Xue, Shangjie Hu, Lin-wen Li, Mingda Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology. Department of Nuclear Science and Engineering In recent years, radiation mapping has attracted widespread research interest along with increasing public concerns on environmental monitoring. However, due to the complex mechanisms of gaseous radionuclide dispersion, radiation-matter interaction, and the current limitation of dose rate data collection, radiation mapping is considered to be a challenging task. In this study, a general framework for radiation mapping is proposed in static and dynamic scenarios using machine learning techniques. The proposed method enables rapid radiation mapping, as well as trajectory planning for measurements. Firstly, a novel directional radiation detection algorithm is presented. Single-pad radiation detector arrays and attenuation materials are proposed to be used for radiation detection. This thesis presents a deep neural network model to estimate the angular distribution of the incident radiation. Wasserstein distance is applied as a loss function to train the neural network for accurate prediction. Furthermore, radiation mapping could be enabled by performing directional measurements at different positions. In particular, optimization-based approaches are presented to fuse the directional measurement results for source localization and radiation mapping in static scenarios. Secondly, this thesis presents a model for tracking dynamic radionuclide atmospheric dispersion using probabilistic graphical models. Kalman Filter is applied for incremental estimation of atmospheric concentration and ground release simultaneously, as well as the prediction of concentration evolution. Moreover, a path planning algorithm by maximizing the information gathered from the measurements is also presented. The presented method enables joint estimation of the concentration and release source distributions in dynamic scenarios. Such method is also able to plan for future measurements in order to obtain more accurate estimations from the environments, given the previous measurement results. Simulation results for an environment similar to MIT research reactor showed that for radiation release during an accident, the proposed aerial radiation mapping algorithm is able to achieve relative error <10% within a few minutes using three or more agents, and hence showing potential advantages over conventional approaches, which require manual survey at selected locations, and take about an hour with the existing emergency procedure. This work provides an algorithmic basis for radiation mapping problem, and shows potential to enable autonomous radiation surveys. S.M. S.M. 2022-02-07T15:11:03Z 2022-02-07T15:11:03Z 2021-09 2021-10-08T14:40:53.632Z Thesis https://hdl.handle.net/1721.1/139890 https://orcid.org/0000-0003-2127-3414 In Copyright - Educational Use Permitted Copyright MIT http://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Xue, Shangjie Machine Learning Aided Aerial Radiation Mapping |
| title | Machine Learning Aided Aerial Radiation Mapping |
| title_full | Machine Learning Aided Aerial Radiation Mapping |
| title_fullStr | Machine Learning Aided Aerial Radiation Mapping |
| title_full_unstemmed | Machine Learning Aided Aerial Radiation Mapping |
| title_short | Machine Learning Aided Aerial Radiation Mapping |
| title_sort | machine learning aided aerial radiation mapping |
| url | https://hdl.handle.net/1721.1/139890 https://orcid.org/0000-0003-2127-3414 |
| work_keys_str_mv | AT xueshangjie machinelearningaidedaerialradiationmapping |