5:15 PM - 7:15 PM
[PEM10-P12] Development of an Automated Detection System for Satellite Surface Charging of Satellites Using DMSP Satellite ObservationData
Keywords:Surface Charging, Satellite, Machine Learning
Surface charging of spacecraft is a critical issue in space development, particularly for missions involving debris removal and on-orbit servicing. Discharge events caused by potential differences between the spacecraft and target objects pose a risks of damage to on-board systems and instruments. Therefore, it is essential to understanding the mechanisms of surface charging and developing techniques for its prediction and detection are essential.
The surface charging of Llow Earth Oorbit (LEO)satellites is mainly influenced primarily by two factors: auroral electrons and ionospheric plasma density. Previous studies have defined the following two conditions as indicators of charging events: (1) the integral number flux of electrons above 14 keV must be greater than 108 [eV cm-2 s-1 str-1 eV-1], (2) the ionospheric plasma density must be less than 104 [cm-3] (e.g., Gussenhoven, JGR, 1985). However, charging events have been observed even when these conditions are not met, and the specific plasma environments in which charging occurs remain unclear. (Masuno, SGEPSS, 2024).
To further investigate the plasma environment responsible for surface charging, it is necessary to detecting a large number of charging events is necessary. PriorPrevious research has identified a characteristic pattern, known as the **ion line structure**, in energy-time (ET) diagrams of ions during charging events (Anderson, 2012). This structure occurs when a spacecraft accumulates a surface potential of -ΦV[V], causing surrounding low-energy ions (<1 eV) to be accelerated to ΦeV[eV], where they pass through the sensor aperture. As a result, an enhanced ion flux at ΦeV[eV] is observed, forming the distinctivecharacteristic ion line structure. While efforts have been made to statistically detect this structure (Masuno, SGEPSS, 2024), accurate statistical detection remains challenging, and the process still relies heavily on manual identification.
In this research, we developed an automatic detection system for spacecraft surface charging detection system using convolutional neural networks (CNNs) and data from the DMSP-F16 satellite. The dataset consists of ion ET diagrams recorded from 2009 to 2019, including 3,081 cases of charging events with ion line structures and 2,819 cases of non-charging cases. Data augmentation through noise addition was performed, increasing the dataset to 8,599 charging cases and 7,898 non-charging cases for model training. For validation, 341 charging cases and 315 non-charging cases were used.
Our results demonstrate high detection accuracy, with the model achieving **99.6% accuracy** and a **loss of 0.015** on the training dataset, and **99.6% accuracy** andwith a **loss of 0.014** on the validation dataset. Furthermore, our CNN-based method successfully identified charging events that did not meet previously established charging conditions, suggesting the possibility of discovering additional contributing factors contributing to spacecraft surface charging.
The surface charging of Llow Earth Oorbit (LEO)satellites is mainly influenced primarily by two factors: auroral electrons and ionospheric plasma density. Previous studies have defined the following two conditions as indicators of charging events: (1) the integral number flux of electrons above 14 keV must be greater than 108 [eV cm-2 s-1 str-1 eV-1], (2) the ionospheric plasma density must be less than 104 [cm-3] (e.g., Gussenhoven, JGR, 1985). However, charging events have been observed even when these conditions are not met, and the specific plasma environments in which charging occurs remain unclear. (Masuno, SGEPSS, 2024).
To further investigate the plasma environment responsible for surface charging, it is necessary to detecting a large number of charging events is necessary. PriorPrevious research has identified a characteristic pattern, known as the **ion line structure**, in energy-time (ET) diagrams of ions during charging events (Anderson, 2012). This structure occurs when a spacecraft accumulates a surface potential of -ΦV[V], causing surrounding low-energy ions (<1 eV) to be accelerated to ΦeV[eV], where they pass through the sensor aperture. As a result, an enhanced ion flux at ΦeV[eV] is observed, forming the distinctivecharacteristic ion line structure. While efforts have been made to statistically detect this structure (Masuno, SGEPSS, 2024), accurate statistical detection remains challenging, and the process still relies heavily on manual identification.
In this research, we developed an automatic detection system for spacecraft surface charging detection system using convolutional neural networks (CNNs) and data from the DMSP-F16 satellite. The dataset consists of ion ET diagrams recorded from 2009 to 2019, including 3,081 cases of charging events with ion line structures and 2,819 cases of non-charging cases. Data augmentation through noise addition was performed, increasing the dataset to 8,599 charging cases and 7,898 non-charging cases for model training. For validation, 341 charging cases and 315 non-charging cases were used.
Our results demonstrate high detection accuracy, with the model achieving **99.6% accuracy** and a **loss of 0.015** on the training dataset, and **99.6% accuracy** andwith a **loss of 0.014** on the validation dataset. Furthermore, our CNN-based method successfully identified charging events that did not meet previously established charging conditions, suggesting the possibility of discovering additional contributing factors contributing to spacecraft surface charging.