Researchers Alireza Rezaee, Niloofar Nobahari, Amin Asgarifar, and Farshid Hajati, affiliated with the Department of Electrical Engineering at the University of Tehran, have developed a novel method for detecting faults in the electrical power systems of nanosatellites operating in Low Earth Orbit (LEO). Their work, published in the journal Acta Astronautica, focuses on identifying and classifying faults without relying on an Attitude Determination Control Subsystem (ADCS), which can be beneficial for simplifying satellite design and reducing costs.
Nanosatellites, while offering advantages in terms of cost and launch flexibility, face significant challenges related to fault detection and management in their electrical power systems. The researchers identified common faults such as line-to-line faults and open circuits in the photovoltaic subsystem, short circuits and open circuits in the DC to DC converter’s Insulated Gate Bipolar Transistors (IGBTs), and regulator faults in the ground battery. These faults can arise due to pressure tolerance issues, launcher pressure, and environmental circumstances.
To address these challenges, the researchers developed a neural network-based system that simulates the electrical power system under normal operating conditions. The neural network uses solar radiation and solar panel surface temperature as input data to predict current and load outputs. By comparing the predicted outputs with actual measurements, the system can detect anomalies indicative of faults.
The researchers employed various machine learning techniques, including neural network classifiers, Principal Component Analysis (PCA) classification, decision trees, and K-Nearest Neighbors (KNN), to classify different types of faults based on their patterns. This approach allows for the identification of specific fault types, enabling targeted maintenance and repair actions.
The practical applications of this research for the energy sector, particularly in the space industry, are significant. By improving fault detection and classification in nanosatellite electrical power systems, the researchers’ method can enhance the reliability and longevity of these systems. This, in turn, can reduce the costs associated with satellite failures and improve the overall efficiency of space-based energy systems. Additionally, the machine learning techniques developed in this research could be adapted for use in other energy systems, such as terrestrial solar power plants, to improve fault detection and management.
This article is based on research available at arXiv.