DSC-86 Predicting RAF flight safety from time series data

Use accident reports to predict the future. Identify trends which deviate from the prediction and use this to instigate investigations/corrective actions.

Team members

  • Gareth Clews
  • Ian Grimstead
  • Flt Lt Peter Kennedy (RAF)

The need

  • Improve MOD safety culture from the bottom up
  • Although limited flight safety analysis exists, action taken to prevent accidents is often reactive to unforseen, disparate combinations of contributory factors, after an event has happened
  • Identify when large scale avoidable incidents may occur which may prompt investigations and focused training to try to prevent them

Impact

  • Potential to improve training on reporting, understanding reporting culture, human behaviour and more
  • Detecting anomalous squadron behaviour
  • Allowing a more generative approach to flight safety
  • How will we share this - published work including code-base, presentations, training courses etc.
  • Use cases for MOD, Military Aviation, Civil Aviation, ‘Total Safety’ and other areas where there are reports (logistics/shipping)

Data science

  • Potential for innovative means of feature extraction, Named Entity Recognition applications on report text, LSTMs and beyond for time series prediction
  • What is the data science stack ? Baleen/spaCy, keras: autoencoder, LSTM/GRU, diffusion manifold methods (whatever they live in - LAPACK+BLAS?), python but with opportunities for others (C, Haskell, Fortran)
  • Does it use ONS infrastructure or expertise, or extend capacity in some way?
    • ONS: no, only DSC
    • Capacity: involvement from MOD, RAF, CAA + MAA who have analysts untrained in these things who will contribute and learn, prototype systems for use

Stakeholders

  • Who are the Partners / stakeholders? MOD, RAF, MAA, CAA, Flight Safety Analytics
  • Duty holders at all levels, Delivery Duty Holder (e.g. Station Commander), Operational Duty Holder (e.g. Air Officer Commanding), Senior Duty Holder (e.g. Chief of the Air Staff)
  • Who are we working with? The partners/stakeholders mentioned above

Code and outputs

  • What are the outputs? Repo, paper
  • Links to (public) Github repositories. As and when they are made?
  • Links to related research, other groups (inside + outside gov): Accelerator project presentation will be made available soon
  • Related + similar projects: NLP projects (optimus, patent_explorer, keywords, people survey), LSTM+extra (patent explorer, data project?)

Delivery

  • Project started: November 2018?
  • A milestone
  • Another milestone
  • Estimated delivery

Further information

Please contact datasciencecampus@ons.gov.uk for more information.

Updates

  • No updates yet.

Notes

This page has been automatically generated. Click here to download this project description as a pdf or click here to download as a word document.


Updated