Model-Based Spacecraft Fault Identification

Verus Research, in partnership with the University of Colorado Boulder, proposes to continue development of a model-based fault identification solution that utilizes runtime Bayesian multi-mode hypothesis estimation. We will further integrate a data contextualization layer that incorporates additional machine learning models, correlation analysis, and decision trees. This contextualization is necessary when the model-based method encounters unmodeled anomalies. Our team has spent several years developing machine learning models for fault detection, which are currently deployed in Sat-SAM, our ground service fault monitoring software. This STTR effort aims to supplement our existing detection algorithms with the proposed combined model-based and data-driven fault identification framework. By the end of Phase II of this STTR effort, we will deliver a prototype version of Sat-SAM as a complete ground software service for satellite operators and subject matter experts (SMEs) that both accurately detects and identifies the cause of faults and presents results to the operator/analyst with associated confidence levels, so that operators can immediately begin recovery plans to rapidly bring the satellite back online in minimal time.