Ph.D. Proposal
Ruby Al Fawares
(Advisor: Prof. Prof. Dimitri Mavris)
"A Predictive Maintenance Framework for Mega Satellite Constellations Using Clustering and Synthetic Data"
Monday, February 17
10:00 a.m.
Weber Space Science and Technology Building (SST II)
Collaborative Design Environment (CoVE)
and
Abstract
The rapid deployment of mega satellite constellations in low Earth orbit (LEO) introduces significant challenges in maintaining operational reliability and sustainment. Predictive maintenance offers a proactive approach to minimizing satellite failures and optimizing resource allocation. However, the proprietary nature of telemetry data and the complexity of large-scale constellations hinder the development of effective predictive models. This research presents a novel framework integrating synthetic data generation, life-cycle-based clustering, and decision-tree-based predictive maintenance to address these challenges.
The framework consists of three key components. First, synthetic telemetry data generation substitutes for inaccessible real-world data by simulating critical performance metrics, such as coverage availability and communication latency, across a satellite’s operational life cycle. By incorporating failure modes and phased deployment scenarios, the synthetic datasets enable robust model development and testing, ensuring adaptability to the evolving constellation landscape. A ten-year simulation with 100 satellites ensures comprehensive coverage of life-cycle phases, strengthening clustering and predictive maintenance analyses.
Second, a life-cycle-based clustering method is introduced to categorize satellites into operational phases—early-life, mid-life, and late-life—using clustering algorithms such as K-Means and DBSCAN. This approach facilitates the aggregation of performance data, enabling scalable predictive maintenance for large constellations. The clustering framework dynamically updates as satellites transition between life-cycle phases or as new satellites are deployed, ensuring adaptability in a constantly evolving system.
Finally, a decision-tree-based predictive model is developed to transition from reactive to proactive maintenance. By leveraging aggregated cluster-level data, the model predicts failures based on key performance thresholds. This predictive approach allows for optimized maintenance decisions, balancing proactive interventions with resource constraints. The framework ensures that maintenance strategies align with operational scenarios, adapting to life-cycle phases and satellite health metrics.
This research evaluates the effectiveness of synthetic data in predictive maintenance, compares clustering techniques for satellite grouping, and assesses the decision-tree model’s capability in shifting maintenance strategies. The results demonstrate the framework’s scalability, adaptability, and predictive accuracy. By bridging synthetic data modeling with real-world operational challenges, this work contributes to the advancement of predictive maintenance in space systems, with broader implications for autonomous asset management in aerospace, maritime, and industrial applications.
Committee
- Prof. Dimitri Mavris– School of Aerospace Engineering (advisor)
- Prof. Brian German– School of Aerospace Engineering
- Prof. Graeme Kennedy– School of Aerospace Engineering
- Prof. Jenna Jordan– Sam Nunn School of International Affairs
- Dr. Adam Cox – School of Aerospace Engineering (Research Faculty)