Smart fault detection in a nanosatellite electrical power system

This research‑based resource looks at a single residual framework that can find and isolate faults in a nanosatellite’s electrical power system (EPS). The diagnostic setup covers fault cases in the key parts of the EPS: the photovoltaic (PV) array, the maximum‑power‑point‑tracking (MPPT) converter, the regulator converter, and the battery‑to‑ground interface. The method was built in MATLAB R2021a. It uses healthy‑state and fault‑specific multilayer perceptron (MLP) identifiers to create residual vectors. Those residuals are then sorted by several machine‑learning classifiers: MLP, k‑nearest neighbors (KNN), decision trees (DT), and principal component analysis (PCA) with a nearest‑centroid rule. Testing was done with orbit‑disjoint data. A total of 145 orbit simulations—each with 69 samples—were

split into 87 training orbits, 29 validation orbits, and 29 sealed test orbits to avoid any data leakage. In ideal test conditions, the MLP classifier reached 100 % accuracy (ten‑seed mean = 100.000 % ± 0.000 %). KNN, DT, and PCA achieved accuracies of 97.311 %, 97.201 %, and 79.620 % respectively. The framework offers a solid base for building smart, onboard diagnostic systems for small satellite missions.

Message: Hi FE, I need help with “Smart fault detection in a nanosatellite electrical power system” in “Electrical Engineering”