Aerospace and Electronic Systems Magazine April 2017 - 4

Feature Article:

DOI. No. 10.1109/MAES.2017.150242

Airplane Flight Safety Using Error-Tolerant Data
Stream Processing
Shigeru Imai, Rensselaer Polytechnic Institute, Troy, NY, USA
Erik Blasch, Air Force Research Laboratory, Rome, NY, USA
Alessandro Galli, Wennan Zhu, Frederick Lee, Carlos A. Varela, Rensselaer Polytechnic
Institute, Troy, NY, USA

INTRODUCTION
Ubiquitous sensing is pervasive in society for such applications
as biometrics for health care, smart grids for power delivery, and
avionics for transportation safety [1]. As society continues to rely
ever more on sensors for various applications, there is a need to address the accuracy of sensor readings for health maintenance, signal identification, and control [2]. While there have been advances
in information fusion [3] for avionics control [4] and user warnings
[5], there is still a need for further research in methods that allow
for fault detection and recovery techniques to be easily realized
and implemented with minimal risk of software errors.

SENSOR DATA ERRORS
Aircraft sensor errors have been at the root of many flight accidents [6]. Sensor faults inducing erroneous airspeed, altitude,
and attitude data have led to pilot inability to correctly fly a
plane. In February 1996, Birgenair Flight 301's airspeed indicator failed due to a blocked pitot-static system that gave the indication that airspeed was increasing during climb-out, leading to
a stall and subsequent fatal crash [6]. In Dec. 1999, the Korean
Air Cargo Flight 8509 crashed after takeoff due to pilot error
induced by an unreliable attitude director indicator that resulted
in the pilot applying excessive bank angle [6]. In both cases, an
air data inertial reference unit could benefit from logical redundancy with independently-measured data. A third example involves a military B-2 plane which resulted in a loss of $1 billion
[7]. Caused by bad weather, the three pressure sensors (transducers) were improperly calibrated due to condensation inside
the devices. During flight, the sensor errors caused diesel fuel
to enter skin-flush air-data sensors that determine the airspeed,
altitude, and attitude. Investigation of the accident determined
Authors' current addresses: S. Imai, A. Galli, W. Zhu, F. Lee, C. A.
Varela, Rensselaer Polytechnic Institute, New York, 110 8th Street,
Troy, NY 12180, USA. E. Blasch, Air Force Research Lab, Information Directorate, 525 Brooks Road, Rome, NY 13441-4505, USA.
Manuscript received October 19, 2015, revised February 16,
2016, and ready for publication July 18, 2016.
Review handled by P. Willett.
0885/8985/17/$26.00 © 2017 IEEE
4

that computers calculated inaccurate aircraft angle of attack and
airspeed [8].
The 2009 crash of Air France Flight 447 [9] resulted from iced
pitot tubes that led to erroneous air speed data. In this article, we
advocate the use of logical redundancy; specifically, the ground
speed and estimated wind speed data-which were available from
onboard Global Positioning System (GPS) monitors and weather
forecasts-could have been used to catch erroneous air speed readings coming from the malfunctioning pitot tubes.
Figure 1 highlights the airspeed indicator components of the
Airbus 330 configurations, including the pitot tubes for dynamic
pressure measurement, the static port for static pressure measurement, and the instrumented diaphragm that determines airspeed by
measuring the difference between these two quantities. To maintain accurate airspeed measurements, it is important to heat the
pitot tube at high altitudes or in cases of low temperatures in which
moisture would freeze and obstruct airflow.

SENSOR FAULT DETECTION
Two important directions of research for safe air transportation are:
Data Stream Management Systems (DSMS) and Fault Detection,
Isolation, and Reconfiguration (FDIR).
DSMS are prevalent in transportation systems designs, but few
studies have been applied to aviation. Available systems such as
PLACE [10] and Microsoft StreamInsight [11], which are DSMSbased systems supporting spatio-temporal streams, combine data
stream processing and database management. These DSMS-based
spatio-temporal stream management systems support general continuous queries for multiple moving objects such as "Find all the
planes flying within a sphere diameter of X from a point Y in the
past Z time." [12]. Unlike these DSMS-based systems which handle
multiple spatio-temporal objects, the ProgrammIng Language for
spatiO-Temporal data Streaming applications (PILOTS) [13] assumes each application to be running on a single platform and tries
to extrapolate data that is relevant to its current location and time.
The PILOTS approach for spatio-temporal data stream filtering, and
error detection and correction, uses high-level abstractions, so that
users can more easily design error models to correct data in flight.
FDIR has been intensively studied in the control and aerospace
communities [14]. Mission critical systems, such as nuclear power
plants, flight control systems, and automotive systems, are key ap-

IEEE A&E SYSTEMS MAGAZINE

APRIL 2017

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine April 2017