Answers are grounded in this paper and cite their sources.
Flight-Control Electro-Mechanical Actuators (EMAs) are characterized by complex temporal dynamics developing on various time-scales and large variability of operating conditions, which make field data scarce. In the present work, we address the technical problem of anomaly detection and develop a novel signal reconstruction method combining Transfer Learning (TL) and Physics-Informed Long Short-Term Memory (TL-PILSTM) networks to estimate the position of the actuator motor expected in normal conditions. The model architecture comprises physics-informed cells for the solution of the differential equations governing the EMA operation and data-driven Long Short-Term Memory (LSTM) cells, which receive in input the output of the physics-informed cells and reconstruct the motor position expected in normal conditions. Given the scarcity of field data, the signal reconstruction model is pre-trained using data simulated by a high-fidelity model and fine-tuned with the scarce field data available. The Sequential Probability Ratio Test (SPRT) is, then, applied to the difference between the obtained signal reconstructions and the physical measurements to detect the occurrence of anomalies using an experimental test bench. The TL-PILSTM is demonstrated able to provide accurate and physics-consistent reconstructions of the motor position of an EMA; it outperforms state-of-the-art anomaly detection methods, yielding a 22.8% average increase in the Area Under Receiver Operating Characteristic (AUC-ROC).
No takes yet. Share an insight, caveat, or question.
Lai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: