camillecochener / anomaly-detection-of-accelerometers-data

This repository contains the analysis I performed during the advanced machine learning data challenge (MDI341) at Télécom Paris.

Challenge Large Scale Machine Learning

Theme: anomaly detection for accelerometer data

Flight engineers attach a large number of sensors to the test helicopters to capture every nuance of their behaviour. To improve the detection of early warning signals in this vast amount of data, Airbus is encouraging research into a new way to pinpoint potential problems, including outliers. A multi-disciplinary team of specialist engineers accompanies each flight to study this mass of observations - a major investment for every flight made.

The aim of the challenge is thus to develop a method for using unmonitored AI to detect anomalies in the accelerometer data of helicopters pre-certified by Airbus.

The dataset provided by Airbus consists in measurements of accelerometers of helicopters during 1 minute at frequency 1024 Hertz, which yields time series measured at in total 60 * 1024 = 61440 equidistant time points.

This challenge is inspired by the Airbus AI Challenge that tooks place in december 2019. 140 teams participated. The first prize in the "Airbus AI Gym" competition was awarded to Fujitsu.