Unsupervised Time Series Fault Detection on Marine Dual Fuel Engines Exhaust System using LSTM-AE*
Uložené v:
| Názov: | Unsupervised Time Series Fault Detection on Marine Dual Fuel Engines Exhaust System using LSTM-AE* |
|---|---|
| Autori: | Youssef, Ayah, Noura, Hassan, Dabaja, Hassan, El-Adel, El-Mostafa, Ouladsine, Mustapha |
| Prispievatelia: | Youssef, Ayah |
| Zdroj: | 2025 33rd Mediterranean Conference on Control and Automation (MED). :72-77 |
| Informácie o vydavateľovi: | IEEE, 2025. |
| Rok vydania: | 2025 |
| Predmety: | [INFO.INFO-LG] Computer Science [cs]/Machine Learning [cs.LG] |
| Popis: | Fault diagnosis in marine dual fuel systems is essential due to the complex operating conditions and the potential for significant safety and environmental impacts. This paper uses Long Short-Term Memory Autoencoder (LSTMAE) for fault detection in the exhaust systems of marine dual fuel engines. The LSTM-AE model is designed to detect anomalies in exhaust valve closing dead time (ECDT) using real historical data. The methodology involves creating input sequences, encoding and decoding them using LSTM layers, and detecting faults based on reconstruction errors. The model is trained on healthy data from Cylinder 3 and tested on other cylinders, achieving a precision of 1.00, recall of 0.82, and F1-score of 0.90. Results showed that the model successfully detects both known and unreported faults, demonstrating its potential for real-time fault monitoring in marine propulsion systems. |
| Druh dokumentu: | Article Conference object |
| Popis súboru: | application/pdf |
| DOI: | 10.1109/med64031.2025.11073222 |
| Prístupová URL adresa: | https://amu.hal.science/hal-05065731v1 https://doi.org/10.1109/med64031.2025.11073222 https://amu.hal.science/hal-05065731v1/document |
| Rights: | STM Policy #29 |
| Prístupové číslo: | edsair.doi.dedup.....cedf44aa4dc02115bfa9d0ebb29c7ae3 |
| Databáza: | OpenAIRE |
Nájsť tento článok vo Web of Science | Abstrakt: | Fault diagnosis in marine dual fuel systems is essential due to the complex operating conditions and the potential for significant safety and environmental impacts. This paper uses Long Short-Term Memory Autoencoder (LSTMAE) for fault detection in the exhaust systems of marine dual fuel engines. The LSTM-AE model is designed to detect anomalies in exhaust valve closing dead time (ECDT) using real historical data. The methodology involves creating input sequences, encoding and decoding them using LSTM layers, and detecting faults based on reconstruction errors. The model is trained on healthy data from Cylinder 3 and tested on other cylinders, achieving a precision of 1.00, recall of 0.82, and F1-score of 0.90. Results showed that the model successfully detects both known and unreported faults, demonstrating its potential for real-time fault monitoring in marine propulsion systems. |
|---|---|
| DOI: | 10.1109/med64031.2025.11073222 |