Satellite Telemetry Feature Extraction and Parameter Identification using Supervised Learning Algorithm

Robert P

摘要

Indian Space Research Organisation (ISRO) is nodal in building, launching amp; maintaining LEO, GEO, NAV amp; Interplanetary Indian space resources meeting the nation's requirements. The ever increasing on-orbit satellites yield gigabytes of spacecraft telemetry data archived over the years. This presents an excellent opportunity for analysing telemetry data using the emergent cutting-edge techniques in big data, data mining, machine learning (ML) and artificial intelligence (AI). Manual Telemetry monitoring is a humungous activity at spacecraft control centre (SCC) for deciphering amp; monitoring the health and activities of the spacecraft. Considering the number of telemetry parameters pertaining to various subsystems, it's a mammoth effort which only explodes in complexity given the sheer number of critical parameters to keep track. The rote and tedious nature of this task may lead to undetected anomaly signatures, causing missed opportunities for timely apriori response or early mitigation post contingency. ML based solution to circumvent this is attempted. Identification amp; classification of ground received telemetry parameters is attempted before developing automated ground procedures related to spacecraft health. A neuralnet (NN) based machine learning methodology is developed for identifying the telemetry parameter from the stream. The complexity of unique feature extraction manifolds with the way telemetry data is sampled and stored in onboard storage using various storage ratios suiting operation simplicity. The telemetry feature metamorphosises and manifests differently depending on the storage ratio and sampling time, which is easily distinguishable for a human operator but not for a machine. The challenge is to extract and present the myriad features belonging to the same parameter for supervised learning. The aim is to realize a robustly trained NN which is resilient and can be used for identifying features and can further be extended for automation. This paper elucidates a methodology to handle telemetry data, extract unique features, train a neural net and identify the telemetry stream in Python Machine learning eco system.

Satellite Telemetry Feature Extraction and Parameter Identification using Supervised Learning Algorithm

摘要

著录项

相关主题

期刊订阅