Data-driven health monitoring and fault detection for the RS-25 staged combustion liquid rocket rngine in EcosimPro/ESPSS framework

This paper presents a health monitoring system for open-set fault recognition applied to the RS-25 liquid rocket engine. A digital twin developed in the EcosimPro/ESPSS framework is used to simulate both nominal and faulty operation, covering 24 failure modes of varying intensity. The system extends a standard closed-set classifier with a distance-based rejection mechanism, allowing unknown faults — not seen during training — to be flagged rather than misclassified. Among the evaluated classifiers, K-Nearest Neighbors proves the most robust, minimizing open space risk while maintaining high accuracy on known failure patterns. A multi-objective optimization is then performed to identify the optimal sensor subset: results show that 15 sensors suffice to correctly classify over 94% of both known and unknown failure modes, with excessive dimensionality degrading performance. The proposed framework serves both as a sensor placement tool during engine design and as a reliable realtime diagnostic system.