INTRODUCTION
Real-Time Decision Aiding:
Aircraft Guidance for Wind
Shear Avoidance
L). ALEXANDER STRATTON
ROBERT STENGEL, Fellow, IEEE
Princeton University
Modern estimation theory and artificial intelligence
technology are applied to the Wind Shear Safety Advisor, a
conceptual airborne advisory system to help flight crew avoid
or survive encounter with hazardous low-altitude wind shear.
Numerical and symbolic processes of the system fuse diverse,
time-varying data from ground-based and airborne measurements.
Simulated wind-shear-encounterscenarios illustrate the need
to consider a variety of factors for optimal decision reliability.
The wind-shear-encountersimulations show the potential of the
Wind Shear Safety Advisor for effectively integrating the available
information, highlighting the benefits of the compulational
techniques employed.
Air crews are faced with critical decisions in an
increasingly information-rich environment. While
automation handles a greater percentage of procedural
operations, the crew retains responsibility for vigilant
monitoring, situational awareness, and efficient
planning. Strategic decisions that significantly affect the
course of the mission must be made under uncertainty
using diverse and conflicting evidence as well as
background knowledge. A pilot's decision to continue
a takeoff or landing threatened by low-altitude wind
shear is an example in which time-critical strategic
decisions must be made from incomplete and possibly
conflicting data.
Low-altitude wind shear is a spatial and
temporal variation of winds near the ground. Severe,
low-altitude wind shear has caused several air carrier
accidents with great loss in life [l].Microbursts
(small (1-2 mi.), short-lived (5-30 min.)convective
downdrafts that form strong horizontal outflow
winds upon reaching the ground) can cause sudden,
disastrous aircraft deviations from approach or
departure flight paths, as depicted in Fig. 1. I f a
microburst is encountered, the pilot must react within
seconds to prevent an unacceptable loss of altitude.
Studies of microburst wind shear have provided an
understanding of phenomenology and structure (e.g.,
[2, 3]), detection technology (e.g., [4, 5]), flight crew
training programs [6], and flight guidance strategies
for recovery (e.g., [6, 71). Improving the flight crew's
capability to avoid wind shear is an issue of continuing
interest.
Microburst
4
,
Fig. 1. Wind shear avoidance. penetration, recovery
Manuscript received March 3, 199
IEEE Log No. T-AES/31/1/08009.
This work was supported by the Federal Aviation Administration and
NASA Langley Research Center under NASA Grant NAG-1.834.
Based on a presentation at the AIAA Aerospace Sciences Meeting,
Jan. 1992.
Authors' current addresses: D. A. Stratton, Parker Hannifin
Corp., Gull Electronics Div., Smithtown, NY; R. Stengel, Dept. of
Mechanical and Aerospace Engineering, School of Engineering and
Applied Sciences, Princeton University, EO. Box CNS263, Princeton,
NJ 08544-5263.
OOl8-9251/9S/$4.00@ 1995 IEEE
Flight safety is best achieved by avoiding hazardous
wind shear, but the crew cannot always predict its
presence along the flight path far enough in advance
to divert from the area. While microbursts often
occur in conjunction with locally severe convective
weather, conventional indicators of weather such
as radar reflectivity do not correlate reliably with
wind shear intensity. New ground-based systcms have
been developed for microburst detection at airports,
including Terminal Doppler Weather Radar (TDWR)
and Low-Level Wind Shear Alert System (LLWAS).
With LLWAS installations at approximately 100 U.S.
airports, and with TDWR installations planned for
47 airports, aircraft require on-board wind shear alert
systems not only to supplement ground-based systems
1EI.E TKANSAC'I'IONS O N AEROSPACE AND ELECTRONIC SYSTEMS
VOL. 31, NO. 1 JANUARY 1995
117
LLWAS
TDWR
Fareerits
Wrather d a u
Futum products
Wnihn n d . r
Fig. 2. Schematic diagram of wind shear safety advisor.
but to increase safety when landing at unequipped
airports. Airborne forward-looking sensors warn
several seconds in advance of encounter, enabling a
last-minute avoidance or recovery maneuver. Reactive
wind shear detectors that monitor aircraft motion
sensors can alert the flight crew only as wind shear
is encountered.
The variety of information sources complicates
time-critical decision-making, so it is important to
define ways for aiding the crew. Our earlier papers
present rule-based techniques for aiding wind shear
avoidance [8], Bayesian networks for assessing the
risk of encountering wind shear [9], and multivariable
estimators for predicting wind shear intensity [lo,
111. This paper describes computational logic that
integrates the methods presented in [S-111 for the
Wind Shear Safety Advisor (WSSA), a system to
help pilots avoid and escape hazardous wind shear.
The WSSA, depicted in Fig. 2, can fuse information
from on-board and ground-based sources to provide
advance warnings and guidance. Simulations of the
WSSA illustrate its ability to process time-varying
data from a number of information sources. The logic
development techniques used in the WSSA provide
a general framework that could be applied to many
other systems for real-time decision-making involving
uncertainty.
COMPUTATIONAL STRUCTURES FOR W I N D SHEAR
DECISION AIDING
The computational structure of the WSSA
integrates rule-based and Bayesian processes with
multivariable estimation (Fig. 3). The logic includes
234 rules, 150 conditional probability assignments, and
40 estimation algorithms. It incorporates statistical
information that spans several years of meteorological
study [2, 12-17], implementing a model of flight crew
actions described in a comprehensive flight crew
training program, the FAA Windshear Training Aid
[6].Rule-based processes that control the execution of
Bayesian reasoning and estimation contain declarative
statements that prescribe the conditions under which
all decision-aiding functions are executed. Bayesian
logic explicitly represents a model of uncertain
cause-effect relationships involving wind shear with
probability assignments. Quantitative stochastic models
are used to design multivariable estimators that
provide data for rule-based and Bayesian processing
118
Fig. 3. Wind shear safety advisor top-level structure with arrows
pointing to controlling process.
Fig. 4. Data flow through WSSA.
(Fig. 4). Rule-based, Bayesian, and multivariable
estimation processes of the WSSA are described
below.
RULE-BASED REPRESENTATION FOR DECISION
AIDING
WSSA primary functions are described with
rules-"if-then'' statements coded in an elementary
symbolic language. Implementing decision-aiding
functions in an explicit symbolic format simplifies
system development and facilitates execution-time
debugging and validation. The rule base can be
automatically translated into a procedural format for
real-time operation, following the methods of (181.
Primary WSSA functions are divided into separate
rule bases flor monitoring, assessment, planning, and
guidance, as described below.
The monitoring rule base compiles the most recent
reports from available ground-based and airborne
information sources by processing any available
measurements at each cycle. On-board state elements
determined from multivariable estimators include airand ground -referenced vehicle motions and predicted
wind shear intensity from forward-looking wind
sensors; location and intensity of radar reflectivity and
lightning flash rate also are obtained. Reports from
ground-based sources such as TDWR, LLWAS, and
the Automated Terminal Information System (ATIS)
are processed. The WSSA informs the flight crew
whenever the state of a system changes significantly.
The strategic decision to continue or abort a
takeoff or landing is based on the risk of encountering
IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 31. NO. 1 JANUARY 1995
weather hazards. The assessment rules control the
computation of the probabilities of encountering
severe wind shear and heavy precipitation given
the reported information, using a Bayesian network
described below. A risk level of low, medium, or
high is assigned as a function of these probabilities.
An avoidance strategy, such as a delay in takeoff or
a missed approach, is determined by the planning
rule base if the risk level is high. If the risk level is
medium, then precautionary procedures that increase
the performance of the aircraft in inadvertently
encountered wind shear are recommended.
The guidance rules produce various advisories
or alerts depending on severity and time-criticality
of the situation, informing the flight crew of the
relevant information leading to the advisory or
alert. The highest state of alert, a Time-Critical
Executive-Level Alert, occurs only when aircraft
motion sensors indicate the aircraft is in hazardous
performance-decreasing wind shear. If this occurs, the
WSSA recommends recovery procedures that depend
on the airspeed of the aircraft, groundspeed, runway
remaining, and proximity to terrain. Following a wind
shear encounter, the WSSA reminds the flight crew to
issue a pilot report.
BAYESIAN NETWORK UNCERTAINTY MANAG EMENT
Strategic decisions associated with avoiding
microburst wind shear must account for the
unccrtainty surrounding its forecasting and detection.
Observational study of wind shear has provided
useful statistical data on microburst frequency
and characteristics. For example, the frequency of
microbursts has been established as a function of
thunderstorm frequency [12]; air mass and supercell
thunderstorms, with accompanying precipitation,
lightning, and turbulence, are the most typical sources
of microburst [2]. In arid geographical regions,
microburst frequency is greater due to outflow
from high-altitude storms in conjunction with dry
surface conditions. Only about 1 of 20 storm cells is
capable of producing an outflow hazardous enough
to be classified as a microburst [6]. Distributions of
microburst intensity, size, and duration have been
obtained from dual Doppler radar measurements [15].
Statistical performance measures for ground-based
warning systems, such as probabilities of correct
warning and false warning, have been derived from
operational testing [16, 171.
A qualitative model of wind shear meteorology
was developed as a basis for computing the risk of
encountering hazardous weather. The probabilistic
model incorporates available statistical data from
[2, 12-17] and is based on Bayesian network
representation techniques [19]. Discrete-valued random
variables are represented as square nodes, and arrows
point from cause to effect in Fig. 5. Absence of a
Fig. 5. Bayesian network model for nsk assessment,
direct link between nodes indicates that they are
assumed to be independently generated; for example,
lightning and wind shear are modeled as independent
manifestations of convective weather, and one does
not cause the other. A vector of prior probabilities
is associated with each node; a matrix of conditional
probabilities is associated with each arrow of the
network, relating the likelihood that a given cause
will produce a given effect. Bayes's Theorem is used
to update the network through these conditional
relationships each time new evidence is obtained.
When new evidence is obtained, Bayesian
reasoning is initiated with probabilities related to the
spatial and temporal relevance of the new information.
For example, Fig. 6 is a histogram of microburst
duration derived from [13] with a Rayleigh probability
distribution function fit to the data. As this figure
shows, the relevance of ground-based wind shear
alerts and pilot reports (PIREPs) decays in a period
of minutes. When alerts are successively broadcast, the
more recent information renders the older information
obsolete; revised probabilities reflect the increased
relevance of the more recent information. The network
variables of Fig. 5 are functions of space and time;
whenever the place or time of' terminal operations is
altered, the Bayesian network is reinitialized to reflect
the change.
The risk level is determined from the probability of
< 0.002),
wind shear encounter, Pw,,to be low (Pws
medium (0.002 5 Pws< 0.01), or high (Pws
2 0.01).
The risk level also is assigned a value of high if the
probability of heavy precipitation exceeds 0.9. Ranges
for Pws were assigned so that the WSSKs risk levels
correspond with those designated by the Windshear
Training Aid authors for twelve Windshear Training
Aid Weather Evaluation Exercises [9]. For example,
STRATTON & STENGEL REAL-TIME DECISION AIDING: AIRCRAFT GUIDANCE FOR WIND SHEAR AVOIDANCE
119
0
2
4
6
8
IO 12
14 16 18 20 22 24 26 28
30
Duration time, min
t
\
R d u l H'md
Cornranenis
Fig. 6. Histogram of microburst duration with Rayleigh
distribution function, c = 0.0114 min-' (solid line).
observation of convective weather near the flight path,
unaccompanied by direct evidence of wind shear,
results in medium risk.
I
'
+
D~rw:e
MULTIVARIABLE STATE ESTIMATION A N D HAZARD
PREDICTION
Estimation of the aircraft state and predictions of
wind shear intensity must be made in an uncertain
environment from measurements corrupted by
error. Optimal estimation theory [20, 211 provides
a basis for developing multivariable estimators to
mitigate noise and disturbances, given a stochastic
model of the aircraft and disturbance environment.
State and hazard estimates are used in rules and to
assign likelihood vectors for updating the Bayesian
network. For example, an algorithm for predicting
wind shear hazard from forward-looking sensors
has been developed and refined for the WSSA
[lo, 111. Forward-looking wind sensors, such as
airborne Doppler weather radars, produce a series
of wind measurement sequences in a volume directly
forward of an aircraft (Fig. 7). Because the aircraft is
moving through the wind field, predicting wind shear
intensity from these measurements is improved by
accounting for motion-induced correlations between
the measurement sequences. The prediction algorithm
consists of multiple parallel Kalman filters that
exchange data to improve the hazard estimate as the
aircraft moves toward an area of wind shear, updating
hazard predictions (Fig. 8).
A statistical analysis of the prediction algorithm's
detection characteristics [ l l ] aids the refinement of the
algorithm and provides data for the Bayesian network
model. Observed statistics for microburst intensity
and structure and a parametric microburst model [22]
were used to generate an ensemble of microbursts
for estimating the algorithm's probability of missed
detection, PMD,and mean-square error as a function
of a critical design parameter. An analytical method
was used to estimate the probability of false warning,
PFW.Fig. 9 presents estimakes of PMDand PFWfor
the final prediction algorithm design as a function
of the design threshold. This result is used to define
120
Fig. 8. Stochastic prediction algorithm measurement process
Design W r h o l d of Algorithm
Fig. 9. Statistical analysis of detection algorithm reliability.
the matrix (of conditional probabilities in the Bayesian
network linking the forward-looking sensor node to the
wind shear node.
SIMULATION OF THE WIND SHEAR SAFETY
ADVISOR
Results of simulating the WSSA in two
wind-shear -encounter scenarios are presented. 'The
scenarios, based on Weather Evaluation Exercises in
the Windshear Training Aid, demonstrate the WSSA's
ability to evaluate a variety of information and give
timely advance warnings in wind shear situations.
WSSA functions currently are implemented on a
Symbolics 3670 LISP machine. Alerts and warnings
generated by the WSSA appear as text displays on a
multiple-window user interface (Fig. 10). This interface
was designed to develop and evaluate the WSSA
programs and is not a cockpit display.
IBBE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 31, NO. 1 JANUARY 1995
DJdancc information and h e r hftraSilon WMOW
h i . MDniiotinp window
In lhls Window, guidance information of interest
lo the flight crew is displayed, and a user can
interact with the WSSA by enterina oommands with
keyboard or mome. For eiample:
Finished wiih the tutorial?
h Lhi. wvM01. t h W S S will print wi I
'i*i"n'
'"*'Y
I*.mru.d.
.
I
~
~
~
~
~
chooses an answer with the mouse. for example:
INO
~
~
~
~
I
I
Sensor hiom"ion W d o r
Siaiui htormailon Window
I. ...........................................
In this window, $ensor inlormatim of interest
to e flight crew 18 displayed. for example:
/In this window. status i n f m e t i o n for t h e flbhl
is displayed. curing takeoff or final approach,
~.........r.........*.........................
WEATHER ADVISORY INFORMATION
* A report ha3 been received frcm data link.
.
I*
*
* A LLWAS MB-ADVISORY was reported
* 0.2 minutes ego
a1 0.4 miles.
Awaiting takeoff from OMAHA.
Takeoff soheduled io begin
In 3.0 MINUTES
.
....................~*.~~~~*.~*.**...........
* Risk of Wind Shear Encounter is LOW.
Risk of Heavy Preoipitation is LOW.
....................*...****..........*
~~~~
~
Fig. 10. WSSA user interface.
(4
(b)
Fig. 11. Simulated microburst encounter during takeoff.
The first simulation, based on the sixth Weather
Evaluation Exercise, begins 2 min before a summer
afternoon departure from Denver's Stapleton Airport
(Fig. ll(a)). ATIS information indicates dry surface
conditions with a dewpoint depression of 38'. Twelve
sec later, virga (evaporating precipitation associated
with dry microbursts) is sighted near the departure
path. The WSSA determines that the risk of wind
shear encounter is medium, and it recommends
precautions. Thirty sec before takeoff, a TDWR Wind
Shear Advisory is received. This increases the risk
level to high, and the WSSA issues a Wind Shear
Advisory Alert, recommending a delay (Fig. 12). If a
15 min delay were accepted, the risk level would be
reduced to medium. Otherwise, if the flight crew elects
not to delay, they encounter microburst wind shear
after liftoff (Fig. ll(b)).
As departure commences, the WSSA recommends
an aborted takeoff to avoid potential wind shear.
The TDWR issues a Microburst Advisory 30 s into
takeoff, but by this time the takeoff has progressed
so far that aborting could cause the aircraft to roll
off the end of the runway; therefore, the WSS,4
recommends continuing the takeoff. After takeoff,
the aircraft encounters performance-increasing wind
shear, followed by performance-decreasing wind shear.
The WSSA declares a Time-Critical Executivc-Level
Alert, directing the flight crew to execute a recovery
procedure. As the aircraft descends to 50 ft, a
Ground-Proximity Warning is declared, and the WSSA
recommends further increases in pitch attitude and
thrust. The wind shear has been transited 12 s later,
and the WSSA displays an advisory recommending the
issuance of a pilot report.
The second simulation, based on the ninth Weather
Evaluation Exercise, takes place during an evcning
approach to Stapleton Airport (Fig. 13(a)). The aircraft
is equipped with an airborne Doppler weather radar
and a lightning detection system. ATIS reports that
rainshowers and thunderstorms are in the area, and
reports a low dewpoint depression, indicating that dry
microburst wind shear is less likely. Airborne radar
indicates an area of moderate radar reflectivity near
the flight path. The WSSA determines that thc risk
S'IRA"ON & STENGEL: REAL-TIME DECISION AIDING: AIRCRAFT GUIDANCE FOK WIND SI1I:AK AVOIL>AUCI
121
~
~
~
.............................................
.......................
..
Ouidanoe Inlormation and User
h t r r a c t l o n Window
Ih. halard 8s w w displayd.
P U W I f f i & h.z.rd la 10 M d1.pl.y.d
to Ih.IlWt ~ r . i ,
SO Ih h z w d Is l l o w diIpl.yed.
PlzI"IN6. *n mvotdmnc. .Ir.t.gy
I. r.qu1r.d
fw
I h I N X t Illghl W S . .
10 t k . . c a m n d . a
.voiaanc. .tr.t.gy
is to delay.
$0
WINDSHEAR ADVISORY ALERT
* RISK OF WIND SHEAR ENCOUNTER DURING
' TAKEOFF AT
*
:
DENVER IS HIGH, DUE TO:
DRY-SURFACE
VIRGA
TDWR. WS-ADVISORY
*
~
~
~
!
YES
NO
~
~
~
~
.
~
~
~
~
~
~
~
~
Will the n e x t flight phase be delayed?
.......................
Senrot Lnformrtlon wu*bor
.............................................
WEATHER ADVISORY INFOAMATION
.
A report has been received from data link.
A TDWR WS-ADVISORY was repwted near the
TAKEOFF path at DENVER
' 0.2 minuies ago.
..............................................
.......................
I
Castu YlformallonW h d a w
I
~~?~~.~~~!~?~*!~.~~?!~~******..
Awailing takeoff from DENVER
Takeoff scheduled lo begin
in 0.7 MIMITES
.............................................
*
Risk of Wind Shirar Encounter Is HIGH.
Risk of Heavy Pl'ecipltallan is LOW.
Fig. 12. WSSA determines risk is high.
Siiuauon. E v c n q Approach 10 Denver Slapleton
Fig. 13. Simulated microburst encounter duiring approach
of wind shear encounter is medium, and displays
a Weather Advisory recommending precautions.
Over the next fifteen seconds, the area of reflectivity
increases in intensity and moves toward the flight
path, and lightning is detected in the vicinity of the
radar contour. This information raises the risk of
heavy precipitation to high, and the WSSA issues a
Heavy Precipitation Warning recommending a missed
approach.
Seconds later, the WSSAs prediction algorithm
detects moderate wind shear from Doppler wind
measurements. A microburst is descending from
the base of the thunderstorm onto the flight path
(Fig. 13(b)). The WSSA determines that the risk of
wind shear encounter is high, and it issues a Wind
Shear Caution, recommending an immediate missed
approach. Over the next 20 s the radar contour
approaches the intended flight path, and severe wind
shear is predicted from wind measurements. The
WSSA declares a Wind Shear Warning, advising an
immediate flight crew response (Fig. 14). Additional
messages are displayed as high radar reflectivity and
lightning move to intersect the flight path. Seventy
122
seconds after the WSSA first advised the flight crew
to execute il missed approach, the microburst is
encountered, and a Time-Critical Executive-Level
Alert is declared. After the wind shear encounter,
the WSSA displays a message recommending that the
flight crew report the incident.
These simulations show that wind shear detection
systems ancl conventional weather indicators can work
together to increase the margin of flight safety, given
an effective strategy for integrating information in
uncertain situations. Computational systems with
the capability to reason with uncertainty could assist
time-critical decision-making related to vehicle control,
mission planning, and failure diagnosis. The feasibility
of the WSSA concept could be established through
operational testing and refinement of the knowledge
base, and through piloted simulation of a real-time
WSSA. Other uses of the WSSA logic remain to
be explored; it could form the basis of a cockpilor simulator-based pilot training aid, and it could
be incorporated in an automated system for haiard
detection and flight guidance.
IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 31. NO. 1 JANUilRY 1995
*
~
~
~
.......................
Quidancc Inlormat~onand User InteractIan Window
Rule Monm4r1g Window
.......................
Ih. hallld In m w SkIWed.
PLANNING A k i n i d I. to D. d1apl.y.d
to tht fliphi c r w .
a0 Itae n m r d I. mr dllplmycd.
PUNNING ThWa I.
lorward-loo* .I.it.
m n d m1ss.d l p p M L h IS posllble.
th r.comll*nd.d
~.id.n~.
'0
FORWARD-LOOK WINDSEAR WARNING
* RISK OF WIND SHEAR IS HIGH, DUE TO:
*
WEATHER-RADAR, EAW-PRECIPITATION
LIGHTNINO-DETECTION, HIQH-FLASH-RATE
FORWARD-LOOK. WARMNG
RAlNSfiOWERS
AVOlOANCE STRATEGY: MISSED APPROACH
.
. . I . .........................................
119
a missed approach being called?
.......................
.I,.L.~~
il
m1n.d m ~ o . c h .
MS
NO
I
I
I
I
Srniw InformationWindow
I
Slalut Inram"ion W W w
.............................................
.
.
'
.............................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
WEATHER ADVISORY INFORMATION
1
Awalllng approach at DENVER.
Final approach scheduled to begin
in 0.3 MINUTES.
Risk of Wind Shear Encounter is HIGH.
Risk of Heavy Preoipitetion Is HlQH
I
I
Fig. 14. Wind shear warning issues.
CONCLUSl ONS
An integrated architecture of symbolic and
numerical processes can aid strategic decisions
in uncertain time-critical environments, using
declaratively represented knowledge to combine a
variety of available data. Probability theory provides
a basis for modeling uncertainty in dynamic situations
where timeliness affects the reliability of information.
Both qualitative and quantitative stochastic models
are incorporated in this logic-development framework.
The computational structure of the Wind Shear Safety
Advisor is composed of explicit assumptions that
can be individually refined and validated through
observational study or computational statistical
analysis. The refined logic could be applied for
real-time decision-aiding once a suitable pilot-machine
interface is designed.
REFE:RENCES
Townsend, J. (committee chairman) (1983)
Low-Altitude Wind Shear and Its Hazard to Aviation.
Washington, DC: National Academy Press, 1983.
[Z] Fujita, T. T. (1985)
The downburst: Microburst and macroburst.
Satellite and mesometeorology research project, University
of Chicago, Chicago, IL, 1985.
[3] Wolfson, M. M. (1990)
Understanding and Predicting Microbursts.
Ph.D. Thesis, M.I.T., Cambridge, MA, Feb. 1990.
(41 Bowles, R. L. (1990)
Windshear detection and avoidance: Airborne systems
survey.
In Proceedings of the 29th Conference on Decision and
Control.
Honolulu, HI, Dec. 1990, pp. 708-736.
[5] Turnbull, D., McCarthy, J., Evans, J., and Zmic', D. (1989)
The FAA terminal Doppler weather radar (TDWR)
program.
In Proceedings of 3rd International Conference on the
Aviation Weather System.
Anaheim, CA: Feb. 1989, pp. 414419.
[l]
(1987)
Windhear Training Aid.
U.S. Department of Transportation, Federal Aviation
Administration, Associate Administrator for Development
and Logistics, Washington, DC, Feb. 1987.
Psiaki, M. L. (1987)
Control of flight through wind shear using deterministic
trajectory optimization.
Ph.D. dissertation, Princeton Ilnivewty, Report 1787-T,
Sept. 1987.
Stengel, R. E, and Stratton, D. A . (1989)
An expert system for wind shear avoidance.
Engineering Applications of Artificial Intelligence, 2, 3
(Sept. 1989), 190-197.
Stratton, D. A., and Stengel, R. E (1990)
Probabilistic reasoning for intelligent wind shear
avoidance.
J. Guidance, Control, and Llynamics, 15, 1, (Jan.-Feb.
1992), 247-254.
Stratton, D. A., and Stengel, R. F. (1990)
Stochastic prediction techniques for wind shear hazard
assessment.
J. Guidance, Control, and Dynamics, L5, 5 . (Sept.-Oct.
1992), 1224-1229.
Stratton, D. A., and Stengel, R. E (1991)
Robust Kalman filter design for predictive wind shear
detection.
IEEE Trans. Aerospace and Electronic Systems, 29, 4,
(1993), 1185-1194.
Cullen, J. A., and Wolfson, M. M. (1990)
Predicting summer microburst hazard from thunderstorm
day statistics.
Presented at the 16th Conference on Severe Local Storms,
American Meteorological Society, Oct. 1990.
Hjelmfelt, M. R. (1988)
Structure and lifecycle of microburst outflows observed in
Colorado.
Journal of Applied Meteorology, 27, 8 (Aug. 1988), 900-927.
Byers, H. R., and Braham, R. R. (1949)
The Thunderstorm.
Dept. of Commerce, Washington, DC, 1949.
Wolfson, M., et al. (1990)
Characteristics of thunderstorm-generated low-altitude
wind shear: A survey based on nationwide Doppler
weather radar testbed measurements.
In Proceedings of the 29rh Conference on Decision and
Control, Vol. 2,Honolulu, HI, 1990, pp. 682688.
STRATTON & STENGEL: REAL-TIME DECISION AIDING: AIRCRAFI GUIDANCE FOR WIND SHFAR AVOIDANCI I
123
[16]
[17]
[18]
[19]
Barab, J. D., Page, R. D., Rosenburg, B. L., Zurinskas, T. E.,
and Smythe, G. R. (1988)
Evaluation of enhancements to the low level windshear
alert system (LLWAS) at Stapleton International Airport.
DOT/FAA/PS-88/14,Springfield, VA, July 1987-Mar. 1988.
Bernella, D. M. (1991)
Terminal Doppler weather radar operational test and
evaluation Orlando 1990.
Lincoln Laboratory project report ATC-179,
D O T / F M R - 9 1 D , Apr. 1991.
Handelman, D. A. (1989)
A rule-based paradigm for intelligent adaptive flight
control.
Ph.D. dissertation, Princeton University, Princeton, NJ,
MAE report 1858-T, Apr. 1989.
Pearl, J. (1988)
Probabilistic Reasoning in Intelligent Systems: Networks of
Plausible Inference.
San Mateo, C A Morgan Kaufmann, 1988.
[20]
[21]
[221
Anderson, B. D. O., and Moore, J. [I. (1979)
Optimal Filtering.
EngUewood Cliffs, NJ: Prentice-€tall, 1979.
Stengel, R. E (1994)
Optimal Control and Estimation,
New York: Dover, 1994.
(Originally published as Stochstic Optimal Control:
Theory and Application. New York Wiley, 1986.)
Oseguera, R. M., and Rowles, R. L. (1988)
A simple, analytic 3dimensional downburst model based
on boundary layer stagnation flow.
NASA TM 100632, Washington, DC, 1988.
D. Alexander Stratton received his bachelor’s degree in Aeronautical Engineering
from Rensselaer Polytechnic Institute, Troy, NY,in May 1985. He received his
Ph.D. in Oct. 1992 from Princeton University, Princeton, NJ.
He is currently employed as an engineer with Parker Hannifin Corporation’s
Gull Electronics Division, a manufacturer of avionics and fuel systems. He designs
and develops integrated aircraft systems for flight inspection of radio navigation
aids. His interests include satellite and inertial navigation systems, aircraft guidance
and control, and flight safety.
Robert Stengel (M’77-SM’83-F‘93)
received the S.B. degree from Massachusetts
Institute of Technology, Cambridge (1960) and M.S.E., M.A., and Ph.D. degrees
from Princeton University, Princeton, NJ (1965, 1966, 1968).
He is currently Associate Dean of Ehgineering and Applied Science and
Professor of Mechanical and Aerospace: Engineering at Princeton University,
where he directs the Topical Program on Robotics and Intelligent Systems and the
Laboratory for Control and Automation. He also has served with The Analytic
Sciences Corporation, Charles Stark Draper Laboratory, USAF, and NASA. He
was a principal designer of the Apollo Lunar Module manual control logic, and he
contributed to Space Shuttle control design.
Dr. Stengel is an Associate Fellow of the American Institute of Aeronautics
and Astronautics, an Associate Editor at Large of the IEEE Pansactions on
Automatic Control, North American Etlitor of the Cambridge University Press
Aerospace Series, and Member of the Program Council for the New Jersey Space
Grant Consortium. He wrote the book, Optimal Control and Estimation (Dover,
1994).
124
ICCC TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 31, NO. 1 JANUARY 1995