Aerospace and Electronic Systems Magazine April 2017 - 11
Imai et al.
error signatures vector shown in Table 3
(for more details on how to create error
signatures vectors, please refer to ).
Speed Check PILOTS Program
A PILOTS program called SpeedCheck
implementing the error signatures vector of Table 3 is presented in Figure 8.
SpeedCheck checks if the wind speed,
airspeed, and ground speed are consistent with each other or not, and computes
a crab angle, which is used to adjust the
direction of the aircraft to keep a desired
ground track based on the estimated values. If a "GPS fault" is detected by the
error signature S2, the Error Analyzer
estimates the ground speed from the
airspeed and wind speed Equation (7). Figure 8.
Similarly, if a "Pitot tube fault" is detect- PILOTS program to detect and correct for speed data faults.
ed by S1, a similar data correction can
be applied to estimate the airspeed. On the other hand, if a "Both
the flight data recorder, and so it was computed from recorded
pitot tube and GPS faults" condition is detected by S3, PILOTS
Mach (M) and static air temperature (SAT) data. The airspeed was
cannot estimate any values due to not having sufficient redundancy
obtained by using the relationship: va = a0M SAT / T0 , where a0 is
in the data. With the inability to estimate speeds, an alert or warnthe speed of sound at standard sea level (661.47 knots) and T0 is
ing should be sent to the aircrew. It is assumed that with aircraft
the temperature at standard sea level (288.15 Kelvin). Independent
inspections, the likelihood of redundant fault-independent sensors
wind speed information was not recorded either. According to the
producing data errors would be limited to situations of aircraft disdescription from page 47 of the final report: "(From the weather
tress, such as structural failure.
forecast) the wind and temperature charts show that the average
For the SpeedCheck PILOTS program to be applicable to the
effective wind along the route can be estimated at approximately ten knots tail-wind." We followed this description and created
AF447 flight, we use a cruise speed of 470 knots as va, the cruise
the wind speed data stream as ten knots tail wind. As shown in
speed of the AF447 flight available from the flight data recorder
Figure 2b, weather forecast wind speed was used for checking the
information before the accident. Note that the angle signs are recorrectness of air speed and ground speed. If the SpeedCheck PIversed in the PILOTS implementation as opposed to the equations
presented in the previous sections. In trigonometry, angles increase
LOTS program detects a pitot tube failure (S1) or a GPS failure
counter-clockwise (with 0° representing East) while in aviation,
(S2), accurate wind speed from the last normal mode is used for
angles increase clockwise (with 0° representing North).
correction of air speed (S1) or ground speed (S2).
Experimental settings: According to the final report ,
speed data was provided from 2:09:00 UTC on June 1st, 2009 and
it became invalid after 2:11:42 UTC on the same day. Thus, we
Flight data: The ground speed and airspeed were collected
examine the valid 162 seconds of speed data including a period of
based on Appendix 3 in the final accident report of Air France
pitot tube failure which occurred from 2:10:03 to 2:10:36 UTC. We
Flight 447 . Note that the (true) airspeed was not recorded in
used the SpeedCheck PILOTS program shown in Figure 8.
Results: With ω = 1 and τ = 0.8, Figure 9 shows
the PILOTS results. From Figure 9a, the airspeed
error is successfully corrected at 69 seconds, which
Error Signatures Vector for Speed Data
is 5 seconds after the start of the fault, and seamlessly transitions to the normal airspeed when pitot
tubes recover at 98 seconds. From Figure 9b, we
can see that the mode spikes twice to the both faults
condition at around 63 seconds. This is due to the
gradual change in the error; however, the pitot tube
−ava k ava
fault condition is successfully detected in the time
Pilot tube fault
(1 - 2a - bh)va k (1 + a − |a - bl|)va
it would have taken a human expert (e.g., cockpit
crew) given the same data. The accuracy of the preGPS fault
−(a + 1)va k −|a - 1|va
diction is 96.3% for the entire test period, with 3.7%
−(a + bh)va k −|a - bl|va
false negative rate, and 0% false positive rate.
IEEE A&E SYSTEMS MAGAZINE