Detecting Aircraft Performance Anomalies from Cruise Flight Data

We propose an approach to detecting anomalies from aircraft cruise flight data. The detection is based on a model learned from the historical data of a fleet of aircraft. For a variety of cruise flight conditions with and without turbulence, we validate the approach using a FOQA dataset generated by a NASA flight simulator. We identify a regression model that maps the flight conditions and aircraft control inputs into accelerations (linear and rotational). Anomalies are detected as outliers that exceed the scatter caused by turbulence and the modeling error. The detection method is related to multivariable statistical process control. Sensor offset anomalies that are a fraction of a degree in flight surface position and a small percentage of the experienced sensor range are reliably detected from the data collected under light turbulence conditions.