Académique Documents
Professionnel Documents
Culture Documents
Kamesh Subbarao‡
The University of Texas at Arlington, Arlington, TX 76019
and
Jane Thipphavong§
NASA Ames Research Center, Moffett Field, CA 94035-1000
An important step in the design of future air traffic system is that of assessing the impact
of trajectory uncertainty and precision on traffic flow efficiency. Queueing models provide
an efficient alternative to Monte Carlo simulations in this endeavor. The approach models
individual sources of uncertainty such as varying aircraft performance parameters, weather,
navigation sensor and control techniques. Once stochastic models capturing the effects of the
trajectory uncertainties and precision are developed, statistical methods are used to relate
them to the queueing network model parameters. The uncertainties are first expressed in
terms of the errors in position and velocity vectors. These are then transformed into service
rate distributions in the queueing network models. The queueing network model can then be
used to analytically quantify the traffic flow efficiency. Preliminary results of the
uncertainty modeling effort are presented here. Advanced uncertainty models and their
translation to multi-resolution queueing model parameters will be presented in the final
version of the paper.
I. Introduction
T he national air traffic system operates under various uncertainties arising from weather, aircraft performance
variations, navigation sensors and control systems. These lead to trajectory uncertainties consequently affects
the overall air-traffic flow efficiency. Understanding the impact of these uncertainties on air traffic flow efficiency is
crucial for the efficient design of future air traffic systems. In all that follows, the traffic flow efficiency will be
defined as the degree to which traffic flow distribution in the air traffic system is affected due to trajectory
uncertainties, when compared with the traffic flow under nominal conditions.
Monte-Carlo simulations using software packages such as FACET2 and ACES3 can be used to quantify these
effects, however, they are generally time consuming, and do not provide explicit relationships that can be employed
in trade studies. On the other hand, queueing models of the air traffic system can provide explicit relationships
between traffic flow efficiency and trajectory uncertainties, facilitating tradeoff studies in an effective and time-
efficient manner. Queueing models are in fact one of the earliest developments in the now well-established field of
Operations Research. According to Reference 4, much of this theory is attributed to the early works of Erlang5 in
1917, on the problems in telephony. Although most of applications continued to be in telephony and surface
transportation, post WW-II surge in aviation lead to several applications of the queuing theory to air traffic6-8. Since
then, this modeling methodology has been adopted for addressing various aspects of the aviation system by the
airlines, air cargo fleet operators, and air traffic system designers.
*
Research Scientist, 868 San Antonio Road, Member AIAA.
†
Chief Scientist and President, 868 San Antonio Road, Associate Fellow AIAA.
‡
Assistant Professor-Mechanical & Aerospace Engineering, 500 W. First Street, Mail Box: 19018, Member AIAA.
§
Research Engineer, Aviation Systems Division, Mail Stop 210-6, AIAA Member.
1
American Institute of Aeronautics and Astronautics
The operating characteristics of queuing systems are largely determined by statistical properties of the two
queueing model parameters, namely, the probability distribution of inter-arrival times and the service times9,10.
These distributions can take almost any form in real queuing systems. However, in order to formulate a queueing
model as a representation of the real system, it is necessary to specify the assumed form of each of these
distributions. To be useful, the assumed form should have adequate fidelity, so that the model provides reasonable
predictions while at the same time being sufficiently simple so that it remains analytically tractable.
Although several air traffic queuing models have been described in the literature, none of them have directly
considered the effects of trajectory uncertainties due to aviation operations and precision of navigation and control
on the traffic flow efficiency. The present paper seeks to address this issue, and focuses on developing a systematic
procedure for evaluating trajectory uncertainties and its effect to the air-traffic flow efficiency.
Various factors affecting trajectory uncertainties are described in Section II, an overview of the trajectory
uncertainty model is presented in Section III, and preliminary results on modeling the uncertainties due to aviation
operations and precision of navigation and control are presented in Section Error! Reference source not found..
Conclusions are given in Section Error! Reference source not found.. The trajectory uncertainty models discussed
in this paper will augment the queueing models of the national airspace system discussed in a companion paper16.
Flight Control
Control Aircraft
Plans System Aircraft
System
Air Wind
Traffic Traffic
Density Control
Navigation
Navigation
System
System
GPS/INS Other
Nav. Aids
The uncertainties due to precision originate from the navigation system as well as the control system. Precision
of navigation determines the number of aircraft that can safely operate in the airspace, while the precision of control
determines the degree to which aircraft can adhere to their flight plans. Aircraft with lower precision in its
navigation and control will have larger uncertainty in the arrival and service times.
Future air traffic system is expected to employ Automatic Dependent Surveillance – Broadcast (ADS-B) or its
successor, and enhanced Cockpit Displays of Traffic and Weather Information (CDTWI) which include spacing
guidance and current spacing information20. Consequently, air traffic management will depend on the navigation
systems on-board aircraft such as GPS and INS to a higher degree than in the current system. Moreover, the traffic
2
American Institute of Aeronautics and Astronautics
density will amplify or attenuate the effects of navigation precision on air traffic management. Error sources and
their distributions in these navigation systems are often specified in terms of Gaussian distributions21.
Quantifying the effects of controls on the trajectory uncertainties is a little more complex. Aircraft controls
consist of two components, firstly, the control of the aircraft by the Flight Management System or the pilot using on-
board navigation system. Secondly, the air traffic control system may provide inputs for strategic as well as tactical
control inputs. Air traffic controls may arise from ground-based systems as well as from on-board self-separation
systems18. Since the air traffic control system is expected to employ the aircraft navigation system through ADS-B,
the precision of navigation system will have an impact on the trajectory uncertainty due to control.
Once models capturing the effects of the trajectory uncertainties and precision are developed, statistical
methods can be used to relate them to the queuing network model parameters. The queuing network model can then
be used to quantify the traffic flow efficiency through the air traffic system16.
Aerodynamics Atmosphere
Power
Plant Error Models for:
Position
Weight Airspeed Queuing
Heading Transformations Model
Wind Climb Rate To Parameters
Error Models Queuing Model
FMS/Pilot Parameters
Air
Traffic
Control
GPS/INS Other
Nav. Aids
Figure 2. Models for Uncertainties in Aviation Operations and Precision of Navigation and Control
The following subsections will outline the derivation of the error models and the transformation to queuing
model parameters.
3
American Institute of Aeronautics and Astronautics
Y
X
Y
Uncertainty X
Footprints
X Z
Z
Y
r
The navigation errors on the aircraft position-velocity vector r may be modeled as additive Gaussian noise
components with varying mean errors:
r r r r
r = rno min al + nNav + b (t ) (1)
An initial model of the trajectory uncertainties introduced by the precision of control can be formulated as in
Equation (1). Firstly, navigation errors will cause the pilot or the autopilot to make airspeed, heading and altitude
rate corrections. The dynamics of these corrections depend on the aircraft dynamics and the pilot/autopilot control
laws. Since it is unrealistic to include all these effects in the queuing model, simplified models capturing the main
effects must be formulated. Following the existing literature on stochastic systems14,15, linear shaping filters can be
used to capture the essential dynamic effects. The time constants of the shaping filter can be chosen to match the
autopilot bandwidths of commercial aircraft.
Statistics of some of these uncertainties can be extracted from historic air traffic data gathered from NASA,
FAA, airlines and other sources, while the others can be derived from physical considerations.
The relationship between airspeed uncertainty in cruise and ambient wind uncertainty can be derived using
geometrical considerations similar to that in Reference 11. Typically, along-track wind uncertainties NWind can be
modeled as Gaussian noise:
V = VNo min al + NWind (2)
The effects of aircraft performance parameters on the climb and descend trajectories can be modeled using the
well-known fuel-optimal economy climb and maximum range descent trajectory models [12, 13], given by:
4
American Institute of Aeronautics and Astronautics
∂ V (TC lim b − D )
=0
∂h (3)
m& W E = Const .
∂ (TIdle − D )
=0
∂h (4)
W E = Const .
These expressions can be used to derive optimal climb rate and descent rate schedules, and form the basis for
climb-descent paths flown by flight management systems. The drag D in the above expressions is computed using
the fact that transport aircraft climb and descent with near-unity load factors (Lift≈weight). Thus, the perturbations in
aircraft weight translate into drag perturbations, and the fuel burn rate m & . These expressions also permit the
modeling of aircraft engine uncertainties.
Following the well-known literature on stochastic systems14,15, the effects of the aircraft weight, powerplant and
aerodynamic uncertainties on the aircraft climb and descent rates can be modeled as Gaussian probability density
functions. These distributions can be used in conjunction with optimal climb/descent schedules to derive the impact
of each error source. Moreover, the principle of superposition can be employed to simplify the analysis. In this case,
the climb-descent uncertainty models will be of the form:
h& = h&opt + N weight + NThrust + N Aero (5)
The variance of the noise components N can be extracted using aircraft operation data, gathered from airlines
and other commercial aircraft operators.
Trailing AC Leading AC
Separation Distance
Metering
Point
5
American Institute of Aeronautics and Astronautics
0 = AP + PA T + Q − PH T R −1 HP (7)
Based on the above mentioned error dynamic model, the service rate distributions can be calculated, which will
be used as an input parameter for the stochastic analysis of the queueing network model proposed in this study.
Assume that aircraft follow a lined path connecting two metering points (see Figure 4). Throughout the flight,
errors and uncertainties are assumed to exist only in the longitudinal direction along the path. The service rate for
this 1-Dimensional air-traffic model can be defined as
V
µ= (8)
d
where V is the aircraft speed, and d is the separation distance. Each variable is approximated as a sum of a constant
value at the nominal operating condition and a small perturbation.
V0 ∆V ∆d
µ 0 + ∆µ ≈ 1 + − (9)
d0 V0 d 0
By assuming ∆V and ∆d as random variables which have Gaussian distributions, the distribution of the
service rate perturbation, ∆µ , can be expressed as Gaussian.
V V 2 σ 2 σ 2 Cov(V , d )
µ ~ N 0 , 02 V2 + d2 − 2
(10)
d 0 d 0 V 0 d 0 V 0 d 0
The variances, σ V2 and σ d2 , reflect the degree of uncertainty in speed and separation distance, respectively.
They are dependent on the accuracy of navigation information from various sensors and the control uncertainty
coupled with the sensor measurement accuracy. Thus the errors in navigation sensors and control can be converted
to service time distributions.
Example service rate distributions for different INS sensor accuracy settings are shown in Figure 5, with the
assumption that uncertainty characteristics are identical for all aircraft. Higher sensor accuracy leads to lower
uncertainty, which eventually results in narrower service rate distributions.
0.35
Nominal Case (r )
0.3 IRS
2
r *0.5
IRS
0.25 r *2.02
IRS
0.2
0.15
0.1
0.05
0
35 40 45 50 55 60 65
µ : service rate (1/hr)
6
American Institute of Aeronautics and Astronautics
shown to have similar performance parameters as the above, totally accounting for about 99 % of all European air
traffic.
An aircraft’s performance depends on its configuration layout, aerodynamic performance, atmospheric
conditions, aircraft weight and power-plant performance. Thus, a change in any one of these parameters can alter
the performance of the aircraft, which would immediately translate to variability in the queuing parameters: arrival
and service rates and split probabilities.
Currently, the effect of uncertainty in the above mentioned parameters on aircraft climb performance is modeled.
It should be noted that this analysis procedure carries over directly to other segments of the airplane’s operating
mission.
Figure 6 shows the RMS error of all the climb rates measured through the airplane’s operating envelope. This
was the data used to generate the parameters for the BADA model. It can be seen that for a B747 aircraft, the
average RMS error through the climb phase from ground level to 35000 ft. is 200 ft/min. The nominal climb rate for
a B747-400 is 2500 ft/min. If the climb rate varies by 200 ft/min over or below 2500 ft/min, the time taken to climb
to 35000 ft varies from 12.963 to 15.218 minutes. At 2500 ft/min, the time to climb is 14 minutes. This translates
to a 16.1 % difference between the maximum expected climb time and the minimum expected climb time (or
uncertainty).
The recommended vertical separation is 1000 ft. So the climb phase can then be divided into 35 different
segments, each of which becomes a queue of its own.
7
American Institute of Aeronautics and Astronautics
dU
m = T − D − W sin γ (11)
dt
.
T − D ∂U (12)
h = U sin γ = U − mU
W ∂t
∂γ (13)
mU = L −W
∂t
Equation (12) represents the climb rate of the aircraft. If the required vertical separation is H, the service rate for
that part of the queue is given by:
H (14)
µ= .
h
The variability in service rate is then given by performing a Taylor series expansion for µ . This can be done
recursively in a tree structure at any given time as follows:
Service Time, T
Wind
Speed, Ug
8
American Institute of Aeronautics and Astronautics
14 18
16
12
14
10
12
Number of Samples
Number of Samples
8 10
8
6
6
4
4
2 2
0
0 400 500 600 700 800 900 1000 1100 1200 1300 1400
1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8
Time (seconds)
Take-off Weight (kgs) 5
x 10
Figure 8: Monte Carlo Simulations, distributions of aircraft weights (left) and distributions of climb times
(right)
1400
Simulation Results
Linear Fit
1200
1000
Time (seconds)
800
600
400
200
1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8
Weight (kgs) x 10
5
9
American Institute of Aeronautics and Astronautics
T = Tt + Tw + Tto , (17)
Similarly, while landing
T = Tla + Tt (18)
In the above equations
T : Total time spent on the airport surface (gate pushback to wheels off for take off
/ wheels on to gate dock for landing)
Tt : Total taxi time
At this point, it is worth highlighting the factors that determine the magnitude of these times. The time taken to
taxi-out in the case of departures and taxi-in in the case of arrivals is a function of the airport layout, and the surface
traffic. Most of the delay due to traffic movement is eventually assimilated in the wait time at the runway. Thus,
Tt = f ( Airport layout ) (19)
The wait-time at the runway is dependant on the runway’s service rate, and the current level of demand for take-
off and landing.
Tw = f (demand , runway service rate) (20)
Finally, the time taken for take-off or landing depends on airplane parameters (weight, aerodynamic coefficients,
power plant) and atmospheric conditions (Pressure Altitude, Temperature, Head winds). Landing time is also
dependant on the same set of parameters.
Tto = f ( Airplane Paramenters, Atmospheric Conditions ) (21)
Tla = f ( Airplane Parameters, Atmospheric Conditions ) (22)
10
American Institute of Aeronautics and Astronautics
Taxi-out time modeling
In this section, a procedure developed to estimate the taxi-out time is discussed. Figure 10 shows the average
taxi-out times of 16 different airlines (denoted by airline ids 1 through 16) operating at the San Francisco Airport. It
can be seen that the average taxi time is different for different airlines. This depends on the layout of the airport,
and the proximity of the airline gate location to the runway. Another factor affecting this is airline operating
procedure. Any model for estimating taxi-times should account for both these factors.
22
Airline id:14
20
18
Airline id:12
Airline id:2
Airline id:10
Airline id:3
Airline id:1
Airline id:5
Airline id:13
Airline id:8
16
Average Taxi Time
Airline id:11
Airline id:4
14
Airline id:6
Airline id:15
12
Airline id:7
Airline id:9
Airline id:16
10
0 2 4 6 8 10 12 14 16
Airline ID
V. Conclusions
This paper presented a trajectory uncertainty model considering aviation operation and precision of navigation
and control. The results will be used for creating probabilistic distributions of the input parameters for queuing
11
American Institute of Aeronautics and Astronautics
analysis of the NAS described in a companion paper16. Since the uncertainty model discussed in this research is
based on simplifying assumptions, their validity and fidelity must be checked before being used in the queuing
models. The associated validation results will be presented in the final version of this paper.
Acknowledgments
This research is supported under NASA Contract No. NNA07BC55C.
References
1
Swenson, H., Barhydt, R., Landis, M., “Next Generation Air Transportation System (NGATS) Air Traffic Management
(ATM)-Airspace Project (External Release)”, NASA Report, June 2006.
2
Bilimoria, K. D., Sridhar, B., Chatterji, G. B., Sheth, G., and Grabbe, S., “FACET: Future ATM Concepts Evaluation
Tool,” 3rd USA/Europe Air Traffic Management R&D Seminar, Naples, Italy, June 2000.
3
Raytheon ATMSDI Team, “Airspace Concept Evaluation System Build 2 Software User Manual,” NASA Ames Research
Center, Moffett Field, CA, November 2003.
4
Saaty, T. L, Elements of Queueing Theory with Applications, Dover, New York, NY, 1961.
5
Erlang, A. K., “Solution of Some Probability Problems of Significance for Automatic Telephone Exchanges,”
Elektroteknikeren, Vol. 13, pp. 5-13, 1917.
6
Pearcey, T., “Delay in Landing of Air Traffic,” Journal of the Royal Aeronautical Society, Vol. 52, pp. 799-812, 1948.
7
Bell, G. E., “Operations Research into Air Traffic Control,” Journal of the Royal Aeronautical Society, Vol. 53, pp. 965-
978, 1949.
8
Bell, G. E., “Queueing Problems in Civil Aviation,” Operations Research Quarterly, Vol. 3, pp. 9-11, 1952.
9
Cassandras, C. G., and LaFortune, S., Introduction to Discrete Event Systems, Kluwer, Boston, MA, 1999.
10
Hiller, F. S., and Lieberman, G. J., Operations Research, Holden-Day, San Francisco, 1974.
11
Tandale, M. D., Menon, P. K., and Grabbe, S. R., “Estimation of Weather-Induced Arrival Delay Statistics through Monte-
Carlo Simulations”, Paper accepted for presentation at the 2007 AIAA Guidance, Navigation, and Control Conference, Hilton
Head, SC, August 20-23.
12
Menon, P. K., Sweriduk, G. D., and Bowers, A., “A Study of Near-Optimal Endurance-Maximizing Periodic Cruise
Trajectories,” Presented at the 2005 AIAA Guidance, Navigation, and Control Conference, August 15 - 18, San Francisco, CA.
Also to Appear in the Journal of Aircraft, 2007.
13
Kelley, H. J., “Aircraft Performance Optimization by Reduced-Order Approximation,” Control and Dynamic Systems,
edited by C. T. Leondes., Academic Press, New York, NY, 1973, pp. 131-178.
14
Jazwinski, A. H., Stochastic Processes and Filtering Theory, Academic Press, New York, NY, 1970.
15
Gelb, A., Applied Optimal Estimation, MIT Press, Cambridge, MA, 1989.
16
Tandale, M. D., Menon, P. K., Cheng, V. H. L., Rosenberger, J., and Thipphavong, J., “Queueing Network Models of the
National Airspace System”, paper communicated to The 8th AIAA Aviation Technology, Integration, and Operations (ATIO)
Conference, Anchorage, Alaska, 14 - 19 Sep 2008.
17
Aircraft Performance Summary Tables for the Base of Aircraft Data (BADA), Eurocontrol Experimental Centre, EEC Note
No.12/04, September 2004. http://www.eurocontrol.int /eec/public/related_links/ACE_bada_documents.html.
18
Sweet, D. and Bushman, D., “NASA Next-Generation Air Transportation System (NextGen) System Level Concept
Design (NNSLCD): En Route Domain System Design,” Sensis Corporation Draft Document, April 2007.
19
Kayton, M., and Fried, W. R., Avionics Navigation Systems, John Wiley, New York, NY, 1997
20
Frolow, I. and Sinnott, J., “National Airspace System Demand and Capacity Modeling,” Proceedings of the IEEE, pages
1618-1624, 77 (11), November 1989.
21
Wieland, F., “Limits to Growth: Results From The Detailed Policy Assessment Tool,” AIAA/IEEE Digital Avionics
Systems Conference - Proceedings, pages 9.2-1-9.2-8, 2, 1997.
22
http://www.bts.gov/programs/airline_information/airline_ontime_statistics/
23
“Advanced Aircraft Performance Modeling for ATM: Enhancements to the BADA Model”, Angela Nuic, Chantal Poisnot,
Mihai-George Iagaru, Eduardo Gallo, Francisco Navaro, Carlos Querejeta, 24th Digital Avionics System Conference, Washington
D. C., 2005
24
National Climatic Data Center: http://www.ncdc.noaa.gov/oa/ncdc.html
25
FAA OPSNET: http://www.apo.data.faa.gov
26
“Relationship between Weather, Traffic and Delay based on Empirical Methods”, Banavar Sridhar and Sean S. M. Swei,
AIAA 2006-7760.
27
"Climb trajectory prediction enhancement using airline flight-planning information," Richard A. Coppenbarger, AIAA-99-
4147, AIAA Guidance, Navigation, and Control Conference, Portland, Oregon, August 1999.
28
FAA Airport Capacity information for San Francisco, http://www.faa.gov/about/office_org/headquarters_offices
/ato/publications/bench/DOWNLOAD/doc/SFO_2004.doc
12
American Institute of Aeronautics and Astronautics