Multidimensional Time-Series Shapelet-Based Real-Time Fault Detection and Localization on ISS Electrical Power Distribution System
Qin Liu, Weiran Chen, Venkata Dinavahi · IEEE Transactions on Instrumentation and Measurement · 2023
International Space Station (ISS) is a grand invention for human beings to have a chance at exploring the outer space. Its operation is completely dependent on the autonomous power distribution system which transforms energy by solar arrays from the sun. There is a high demand for a reliable monitoring system that can accurately and timely detect and localize faults in its power system for the special working environment of the ISS. In this article, a fault detection and localization (FDL) based on multidimensional time-series trend extracted shapelet (MTES) method was proposed. A fast shapelet discovery was created to accelerate the process of extracting shape features from time-series signals collected from the ISS electrical power distribution system (EPDS). Then, the techniques of randomization and information gain were exploited for further shapelet selection. Finally, multidimensional time-series classification (TSC) for FDL was solved by a designed random forest classifier. The real-time FDL measurement instrument was emulated on the Xilinx VCU128 FPGA board, while a hardware-in-the-loop (HIL) testing platform was established to verify the effectiveness, execution speed, and accuracy of the MTES method. Comparing with other state-of-the-art data-driven methods, higher accuracy (above 96%) and easier hardware implementation were achieved using MTES.