Aerospace and Electronic Systems Magazine April 2017 - 56
DOI. No. 10.1109/MAES.2017.160185
Bayesian Methods in the Search for MH3701
Samuel Davey, Neil Gordon, Ian Holland, Mark Rutten, and Jason Williams
Springer, New York, NY, USA,
Review by Lawrence Stone, Metron, Inc., Reston, VA, USA
his slim paper-bound volume (111 pages) in the Springer Briefs in Electrical and Computer Engineering series
presents a detailed description of the Bayesian search
analysis performed by the authors to produce a probability distribution for the location of the crash of flight
MH370. They performed this analysis under the direction
of the Australian Transport Safety Bureau (ATSB) as members of the MH370 Flight Path Reconstruction group.
Malaysian Airlines Flight MH370, a Boeing 777 with 239 passengers and crew aboard, left Kuala Lumpur at 16:42 UTC12 on
March 7, 2014 bound for Beijing. The last radio transmission from
MH370 was at 17:19. Contact with the plane was lost during the
handover from Malaysian to Vietnamese air controllers. The initial
search was in the Gulf of Thailand centered around the last reported position from the transponder on the aircraft at 17:21.
After several days of intensive air and surface search, the Malaysian military revealed that their radar had tracked the plane after
the last air traffic control contact. This track showed that almost
immediately after 17:21, the plane headed southwest, recrossed the
Malay Peninsula, turned northwest near Penang Island, and continued in this direction until continuous radar contact was lost at
18:02. A single and final radar position was obtained at 18:22. After this the only information about the plane's location came from
two failed ground-to-air phone calls and eight communications between the satellite terminal aboard the aircraft and it's associated
ground station via one of the Inmarsat satellites used by the aircraft
for data and voice services to both the cockpit and the cabin.
If there has been no satellite communication information from
an aircraft for an hour, the ground station automatically initiates a
handshake communication to confirm the presence of the aircraft.
If the aircraft receives the query, then an automatic response is
sent indicating the aircraft terminal is still logged on to the satellite
communications network. (Subsequent to this loss, the 1 hr time
On January 17, 2017 Australian officials announced that the search for
MH370 has been officially called off.
All times are given in Coordinated Universal Time (UTC).
period has been reduced to
15 min to prevent long gaps
in aircraft reporting times.)
The aircraft terminal initiated a logon to the Inmarsat
system at 18:25 as well as an
access request at 18:28. At
19:41, 20:41, 21:41, 22:41,
and 00:11 handshake messages were sent to the aircraft and the terminal on the
aircraft responded. At 00:19
the aircraft satellite terminal
initiated a logon request. Due
to the timing of this logon request, less than 9 min after the last handshake, the analysis team
concluded that the aircraft had run out fuel, lost power, and upon
subsequent activation of the auxiliary power unit, which is powered by airflow, the aircraft terminal had powered up and initiated
a logon request. There was no response to the ground-station initiated handshake request at 01:16. From this they concluded that the
plane must have crashed shortly after the 00:19 logon request. In
addition to the satellite messages, there were two unanswered attempts to call the aircraft, at 18:22 and 23:16.
The satellite messages contain two types of information, burst
time offset (BTO) and burst frequency offset (BFO) information that
can be converted to range and Doppler measurements relative to the
satellite relaying the messages to and from the ground-station. The
unanswered phone calls were analyzed to provide Doppler measurements relative to the satellite. Using the BTO and BFO information,
Inmarsat was able to determine that MH370 must have headed south
into the Indian Ocean . But where in the Indian Ocean had the plane
crashed? This is the question that this book answers by using Bayesian
methods to compute a probability distribution for the crash location.
The approach taken by the authors was to produce the prior distribution on the crash location by developing a particle filter model
IEEE A&E SYSTEMS MAGAZINE