To improve pilot operations, airlines have begun an extensive monitoring of flight operations, thanks to an abundance of data recorded by digital flight data recorders, Quick Access Recorder (QAR). However, traditional data analytics methods are becoming obsolete for proactive safety management. For instance, Exceedance Detection, widely used by the airline industry, can only detect hazardous behaviors from a pre-defined list comprised of “known issues of safety concerns”; it cannot respond to emerging, previously unidentified issues.
This study proposes a new approach, cluster-based anomaly detection, to detect abnormal flights, which can support domain experts in detecting anomalies and associated risks from routine airline operations. The new approach, enabled by data from QAR, a heterogonous dataset collected by ubiquitous sensors on aircraft, applies clustering techniques, i.e. DBSCAN, GMM, to detect abnormal flights of unique data patterns. These flights may indicate an increased level of risks under the assumption that normal flights share common patterns, while anomalies do not. Safety experts can then review these flights in detail to identify risks, if any. Expert reviews and case studies are conducted to validate the proposed methods. Results show that the new breed of data driven approach can identify ‘common patterns’ as well as anomalies, allowing airlines to examine the consistency of current operations while focusing efforts on investigating unusual behaviors to look for latent risks.