Complex space systems exhibit non-linear dynamics that are unmodelled during
the design phase. This can cause detrimental effects on performance after launch. On-board identification of these dynamics can be used to maximise operational performance and mitigate risk. This paper proposes an algorithm that identifies the non-linear dynamics of fuel sloshing using only on-board measurements acquired during normal operation. No additional excitation
manoeuvres are used, thus conserving propellant and maintaining the mission timeline. The method uses an l1-regularised linear regression to determine the governing equations of the dynamics. The algorithm is demonstrated on a communication satellite with a dual-tank architecture and a chemical propulsion system. Real flight data, provided by Airbus Defence and Space, demonstrate the algorithm’s applicability to industry-grade systems.