Project Overview
RUL estimation for critical spacecraft components using advanced survival analysis and deep learning, providing operators with a precise countdown to failure.
The Problem
Traditional Reliability block diagrams and MTBF (Mean Time Between Failures) metrics are static and do not account for the unique operational history or current degradation state of an individual satellite component.
Our Solution
A deep survival analysis engine that ingests high-frequency telemetry and outputs a probability density function representing the Remaining Useful Life (RUL) of specific components, such as gyroscopes and batteries.
System Architecture
Built on a microservices architecture using Go and Python. The models use Deep Survival Machines and Weibull Time-To-Event RNNs, exposed via a low-latency gRPC API to the mission control dashboard.
Impact & Results
Achieved a 94% confidence interval in predicting the end-of-life for lithium-ion power cells operating in highly elliptical orbits, completely eliminating unexpected power-loss anomalies for the client.