Anomaly detection via data mining techniques for aircraft engine operation monitoring

Today, system monitoring as a key function of condition based maintenance (CBM) is provided by the development of sensor applications. Data mining is a practical tool that can be used to extract latent information fiom data generated by sensors. This study examines data mining anomaly detection algorithms for identifying abnormal events in aircraft engine operations. The algorithms under investigation are density-based spatial clustering of application with noise (DBSCAN), expectation maximization (EM) and K-means as well-known clustering algorithms as well as one class support vector machine (OCSVM) as one class classification algorithm. No need for predefined criteria or domain knowledge and capability of determining unknown anomalies are the advantages of the proposed algorithms. The dataset used in this study was recorded from flight data obtained from 791 flights of a twin engine commercial aircraft (i.e. Fokker 100) at the cruise phase. The results show the effectiveness of the DBSCAN and OCSVM in identifying anomalous operations.

Anomaly detection via data mining techniques for aircraft engine operation monitoring Keywords:

Aircraft engine operation anomaly detection , Data mining , Cluster based anomaly detection , One class classification