Autonomous Real-Time Fault Detection in Commercial Spacecraft Systems

A fault detection on systems in commercial spacecrafts requires autonomy in real time in order to be able to safely, reliably and effectively see the missions through. The dynamic nature and multifaceted ness of space vehicles also leads to the large volumes of telemetry created as the space vehicles are operating through the interaction of different sub systems. Rigorous fault detection techniques such as model-based and classical machine learning-based have very low scalability and cross-subsystem dependency modelling problems and are not sensitive to the changes in the type of faults. It is on such weaknesses that they cannot be feasible in real-time applications in the environment in the onboard setting that will require fast and precise fault prognostic responses to outlaw cascade-like failures and the reality that autonomous remedial measures can be executed without the assistance before ground before ground. This system suggest of such a solution to such issues a Hybrid Adaptive Transformer-Graph Neural Network model (HAT-GNN) that is capable of identifying and localising faults in commercial spacecraft systems in real-time by itself. We have a research advantage which consists of an advantage called Interdependencies are expressly represented in our solution which means that it considers the spacecraft subsystems and sensors as a grab base system with dynamic edge connection implying dependencies. The trends in technical sensor telemetry amongst timeseries are detected timely by a lightweight Transformer encoder with the ability to identify temporal abnormalities. The Graph neural Network module condenses and disperses explicitly the information amongst the adjacent subsystem nodes and thereby allows fault localization and inter-subsystem diagnosis to be perfect.

Hali tarjima qilinmagan