Air Collision Risk Analysis
Client: Civil Aviation Authority
How can you sure that two planes will not collide?
The Civil Aviation Authority (CAA) was involved in an international
study looking at the possibility of reducing the standard for separation
of aircraft on international flights. Two changes were proposed:
- Move from a time-based to a distance-based standard.
- Reduce the minimum separation from (effectively) about 80 nautical
miles to 50 nautical miles.
CAA was interested in modelling collision risk. CSIRO provided advice
relating to the longitudinal separation of aircraft.
This involved answering several questions:
- Can a rigorous theoretical justification be provided for the notion
that, for any particular separation distance, the chance of a reported
separation more extreme than this distance is at least as large as the
chance of a true separation more extreme than this distance.
- Can the answers to Question 1 be satisfactorily applied to a
proposed time-based data collection process.
- What is the best way to estimate the route navigation (RNAV) error
distribution, as calculated from the triple-mix position displayed by
an aircraft navigation system?
To help answer Questions 1 & 2, CSIRO devised a simple model for
the position of an aircraft. A model for the separation of two aircraft
follows from it. Together with certain statistical assumptions, this led
to an important conclusion:
`Probabilities based on reported separations are conservative'.
That is, a true small separation is less likely than estimates based on
reported separations imply. This supported the supposition in Q 1.
However, the same model applied to the situation of Q 2 gave a rather
different picture. The assumptions behind the model had altered, and this
suggested strongly that it was now most unwise to rely on the same
conclusions; reported separation probabilities may no longer be
conservative. While this result may be inconvenient, it is a most
important factor in consideration of whether to use time-based separation
standards.
For Q 3, CSIRO showed that some form of modelling, or other additional
piece of information, was needed to estimate extreme RNAV errors; data
alone will never be adequate. Further, CSIRO recommended that modelling be
done directly on the whole triple-mix error distribution directly, rather
than examining each component separately.
Further Information
For further information, please contact Mark Westcott
|
To
top