A novel unsupervised time series discord detection algorithm in aircraft engine gearbox

Abstract

Aircraft engine discord detection is an important way to ensure flight safety. Unsupervised algorithms will be relatively effective due to the lack of models and tagged discord data. For the aircraft time series data collected from sensors, this paper proposes a Trend Featured Dynamic Time Wrapping for J Distance Discord Discovery algorithm based on the J-Distance Discord anomaly definition which combined with the trend information of the data. The experiments on aircraft engine gearbox data show that the TFDTW for JDD Discovery algorithm is better than the normal J-Distance Discord Discovery algorithm and also better than some other classic time series data discord detection algorithms.