(Georgia Institute of Technology, 2019-12)
Puranik, Tejas G.; Mavris, Dimitri N.
Quantification and improvement of safety is one of the most important objectives among the General Aviation community. In recent years, machine learning techniques have emerged as an important enabler in the data-driven safety enhancement of aviation operations with a number of techniques being applied to flight data to identify and isolate anomalous (and potentially unsafe) operations. Energy-based metrics provide measurable indications of the energy state of the aircraft and can be viewed as an objective currency to evaluate various safety-critical conditions across a heterogeneous fleet of aircraft and operations. In this paper, a novel method of identifying instantaneous anomalies for retrospective safety analysis in General Aviation using energy-based metrics is proposed. Each flight data record is processed by a sliding window across the multi-variate time series of evaluated metrics. A Gaussian Mixture Model using energy metrics and their variability within each window is fit in order to predict the probability of any instant during the flight being nominal. Instances during flights that deviate from the nominal are isolated to identify potential increased levels of risk. The identified anomalies are compared with traditional methods of safety assessment such as exceedance detection to highlight the benefits of the developed method. The methodology is demonstrated using flight data records from two representative aircraft for critical phases of flight.