Revolutionizing nanosatellites' data integrity with SEEnet: A real-time ensemble learning approach for Single-Event...

Area of Science:

• Spacecraft engineering
• Radiation effects in electronics
• Machine learning for space applications

Background:

• Nanosatellites offer affordable space access but face data integrity challenges due to radiation.
• Commercial off-the-shelf (COTS) components in nanosatellites are susceptible to Single-Event Effects (SEEs).
• Real-time prediction of SEEs is crucial for mission success and data protection.

Purpose of the Study:

• To develop a lightweight, real-time framework for predicting SEE occurrence in nanosatellites.
• To improve the reliability of onboard data protection in resource-constrained environments.
• To provide an efficient solution for early SEE risk assessment.

Main Methods:

• Proposed SEEnet, an ensemble learning framework combining decision-tree classifiers with soft-voting.
• Implemented a soft-voting strategy to enhance prediction reliability and maintain low computational complexity.
• Evaluated SEEnet using a public dataset of satellite spatial characteristics and SEE events.

Main Results:

• SEEnet achieved 77% classification accuracy, outperforming baseline models like SVM, Random Forests, and Gradient Boosting.
• Demonstrated balanced precision-recall performance.
• Provided bootstrap-based uncertainty estimates to increase prediction confidence.

Conclusions:

• SEEnet offers an effective and computationally efficient solution for real-time SEE risk assessment in nanosatellites.
• The framework supports proactive fault mitigation strategies.
• Enhances data reliability for modern low-cost space missions.