A Novel Method of Overrun Risk Measurement and Assessment Using Large Scale QAR Data

Abstract

Currently, all of China's major commercial aircrafts have used QAR (Quick Access Recorder) to collect flight data. Considering that the large scale of QAR data both in the amount and high feature dimension, QAR data can be applied only 10% of its potential value. Moreover, how to measure the overrun risk for flight overrun issue is rarely involved. In this paper, we proposed a novel approach to evaluate the overrun risk using large scale QAR data. The main process of our approach is to find out the most important factors affecting the flight safety in the landing phase using extracted 76 variables from 2000+ factors. Then 6,395 flight cases were divided into two classes using overrun dangerous line. Finally, three machine-learning models were used to examine the links between overrun risk index and these flight parameters. Results indicate that the most risky factors for overrun incidents in landing phase are long touchdown distances and releasing brake too early in deceleration stage according to our experiments. We also give several suggestions to avoid flight overrun.