MARS: A Machine Learning-Based Flight Condition Recognition and Maneuver Reporting Framework for Helicopter Predictive Maintenance
2026 2nd International Symposium on AI-Driven Engineering Systems (ISADES), Mbale, Uganda, 19 - 20 Haziran 2026, ss.1-4, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/isades69945.2026.11608067
- Basıldığı Şehir: Mbale
- Basıldığı Ülke: Uganda
- Sayfa Sayıları: ss.1-4
- Kocaeli Üniversitesi Adresli: Evet
Özet
Helicopter Health and Usage Monitoring Systems (HUMS) play a vital role in enhancing flight safety and supporting maintenance planning. However, many operational implementations still rely on event-based or rule-driven Flight Condition Recognition (FCR) methods. While these approaches effectively identify threshold exceedances, hard landings, and high-load events, they may inadequately capture unsteady maneuvers and recurrent sub-threshold aggressive operations that contribute to cumulative fatigue damage. To address this limitation, this paper presents MARS, a software-oriented Maneuver Analysis and Reporting System designed to recognize helicopter flight conditions from multivariate time-series sensor data and transform maneuver classifications into maintenance-oriented analytical insights. The proposed framework integrates robust preprocessing, sensor synchronization, sliding-window segmentation, time-series data augmentation, supervised time-series classification, a backend data management layer, and web-based reporting capabilities. Random Forest, Long Short-Term Memory (LSTM), and Temporal Convolutional Network (TCN) models are employed as comparative classifiers for maneuver recognition. The system is designed to enhance existing Flight Data Recorder (FDR) and HUMS data streams without requiring modifications to onboard hardware, thereby supporting usage-based maintenance strategies and providing a foundation for future Remaining Useful Life (RUL) estimation.