Coupling principal component analysis-based sensor data reduction techniques and multi-net systems for simultaneous prediction of multi-component degradation levels in hybrid electric rotorcraft engines

Hybrid Electric Power Systems (HEPS) have gained popularity as a more efficient and eco-friendly alternative. However, with increasing system complexity, fault potential rises. The work aims on implementing a diagnostic system for rotorcraft engine health within a hybrid-electric system. Health monitoring tools are still understudied for HEPS, so this work can represent a valid contribution in the literature. The main goal is assessing degradation and monitoring multi-component simultaneous degradation. Various machine learning techniques for Engine Health Monitoring (EHM) have been compared, varying in network architecture and data reduction. A dynamic model of the entire HEPS generated a dataset including fault information. This dataset trained FFNNs to predict performance parameters (PPs) of degraded components from sensor data. The proposed EHM system's efficacy was evaluated by comparing neural network predictions to dynamic model data. Results show that the Multi-net architecture, with distinct networks for each PP, works more effectively reducing training time.