2021 |
Wachulec, Małgorzata; Luckner, Marcin Fault detection of jet engine heat sensor Journal Article Procedia Computer Science, 192 , pp. 844–852, 2021, ISSN: 18770509. Abstract | Links | BibTeX | Tags: Anomaly detection, Oil temperature sensor, Outlier detection, Sister engines, Time series @article{Wachulec2021, title = {Fault detection of jet engine heat sensor}, author = {Małgorzata Wachulec and Marcin Luckner}, url = {https://doi.org/10.1016/j.procs.2021.08.087}, doi = {10.1016/j.procs.2021.08.087}, issn = {18770509}, year = {2021}, date = {2021-01-01}, journal = {Procedia Computer Science}, volume = {192}, pages = {844--852}, publisher = {Elsevier B.V.}, abstract = {This paper presents an algorithm predicting oil level and temperature sensor (OLTS) failure to replace it before it carries serious costs. OLTS sensor showing too high oil temperature cockpit indications is a driver of significant air turnback events and commanded in-flight shutdown (IFSD). A prediction of sensor malfunction is possible, but an operator requires at least 11 months of historical data. The developed algorithm automates the process of identifying potential failures using a data-driven, dissimilarity based model. It calculates the rolling mean of the oil temperature difference between sister engines for short-term and long-term periods (counted in flights). If the difference between the short-term and long-term means is greater than a set threshold at least confirmation window times, it sets an alert. The proposed model requires less than three months of data to detect the malfunction, with the final F1 score measured on the test set equal to 0.71.}, keywords = {Anomaly detection, Oil temperature sensor, Outlier detection, Sister engines, Time series}, pubstate = {published}, tppubtype = {article} } This paper presents an algorithm predicting oil level and temperature sensor (OLTS) failure to replace it before it carries serious costs. OLTS sensor showing too high oil temperature cockpit indications is a driver of significant air turnback events and commanded in-flight shutdown (IFSD). A prediction of sensor malfunction is possible, but an operator requires at least 11 months of historical data. The developed algorithm automates the process of identifying potential failures using a data-driven, dissimilarity based model. It calculates the rolling mean of the oil temperature difference between sister engines for short-term and long-term periods (counted in flights). If the difference between the short-term and long-term means is greater than a set threshold at least confirmation window times, it sets an alert. The proposed model requires less than three months of data to detect the malfunction, with the final F1 score measured on the test set equal to 0.71. |
Publications
2021 |
Fault detection of jet engine heat sensor Journal Article Procedia Computer Science, 192 , pp. 844–852, 2021, ISSN: 18770509. |