Embedding machine-learned anomaly detection into relevance-driven mission operations workflows

01 / Abstract

Machine-learned anomaly detection has reached the point where it reliably identifies departures from a spacecraft’s own normal behaviour. Operationally, that is not yet useful on its own: a detector that flags every deviation produces a stream of escalations that operators cannot triage, and the cost of that noise is paid by the rare event that actually matters.

This paper describes how detection output is embedded into relevance-driven mission operations workflows, so that a flagged deviation is qualified against mission context, spacecraft state and scheduled activity before it reaches a person. The result is a decision-ready event carrying context, history and a recommended procedure, rather than a score.

Presented at SmallSat Europe 2026 with Telespazio Germany.

02 / Get the presentation deck

The full presentation deck is available on request. Use the button above and we will send it over.