/5bc834b8ba7add0027f3ac5a

  • Airbus Artificial Intelligence Challenges
    AI Gym
    https://aigym-v.airbus.com/contest/5bc834b8ba7add0027f3ac5a

    Open: 18 Oct 2018 | Closed: 01 Jun 2019
    2 months ago

    Interested parties ranging from established companies, start-ups, research labs, schools or individuals, can express their
    interest to register to the challenge by email to timeserieschallenge.request@airbus.com anytime till end of 2018.

    CONTEXT
    Technologies at the intersection of #Artificial_Intelligence and #Internet_of_Things / #Big_Data are pushing the boundaries of the state of the art in #Time_Series_Analysis and #Predictive_Maintenance.

    #AIRBUS is launching this scientific challenge on anomaly detection in time series data in order to:
    ● scout for top players in the field of Time Series Analysis
    ● encourage the research community to tackle the specific issues of related to the data generated by the aerospace industry during tests and in operations.

    OVERVIEW
    Data collected from our platforms is mostly considered normal. Due to the high quality of our products and of aerospace context, the occurrence of faults and failures is very rare, and we cannot afford to wait for reaching hundreds of new fault types to be able to identify and anticipate them. We are interested in unexpected changes in the behavior of the systems we monitor and have a rapid reaction time in analyzing suspect behavior.

    TECHNICAL SCOPE
    We set up a three stage challenge to benchmark unsupervised detection algorithms, based on three use cases:

    1) Business Domain : Helicopters // number of input sensors : 1 // Sampling Frequency : 1000Hz // expected output : classify sequence as OK / KO

    2) Business Domain : Satellites // number of input sensors : 30 // Sampling Frequency : 1000Hz // expected output : classify sequence as OK / KO

    3) Business Domain : Commercial Aircraft // number of context sensors: 81 // number of sensors for anomaly detection: 9 // Sampling Frequency: 8Hz // expected output : identify anomalous time windows on sensors of interest

    We welcome all and every working technical approaches, ranging from statistics (eg. SCP) to more
    established machine learning techniques (eg. Isolation Forest) to modern AI (eg. Deep Learning LSTM).

    TIMELINE
    The challenge will officially start beginning 2019 with a first training phase on Q1 2019. The second phase will be a shorter evaluation on Q2 2019. A restitution workshop is going to be organised in June 2019.

    #IA
    #AI #IoT