Simulation Modeling in Support of a European Airspace Study

JOHN MORRIS, JOHN JAMES, JAMES DeARMON,KELLY CONNOLLY, PAULA MAHONEY

The MITRE Corporation202 Burlington Road Rte. 62

Bedford, MA 01730-1420USA

iemorris(a),mitre.orghttp://www.mitre.org

,.

Abstract: To address some questions of usage of European airspace by United States Air Forceaircraft, two simulation models are employed. The first model offers a wide array of functions torepresent aircraft movement and management of aggregates or "flows" of aircraft. The secondmodel uses a Petri net approach to represent the complexity of a flight planning/replanningoperation, in order to estimate staffing requirements.

Key-words: Simulation model, Petri net, air traffic management, airspace congestion.

1 IntroductionA recent study by The MITRE Corporationused two corporate owned and developedsimulation models, which we describehereunder. The MITRE Corporation is anot-for-profit research and systemsengineering firm, chartered to perform workonly in the public interest, i.e., working forgovernment, either domestic or foreign.Two major sponsors are the United StatesAir Force (USAF) and the Federal AviationAdministration (FAA), who supported thedevelopment of the simulation models.

Our team was tasked with answeringquestions regarding airspace usage by USAFmilitary flights over Europe. The USAFmust make investment decisions regardingavionic capabilities (i.e., electronicequipment on aircraft-radios, altimeters,navigation equipment, etc.), and ourresearch looked into the advantages of fullavionic equipage of the fleet (about a dozentypes of aircraft). With full avionicequipage, civilian Air Traffic Control (ATC)provides the best, expedited service. Theproblem was made more complex via therequirement that differing types of aircraft,departing from different airfields,rendezvous at a given time and place.

The two simulation models to bedescribed in this paper are the CollaborativeRouting and Coordination Tools (CRCT,pronounced "circuit") and MSim. CRCTwas used to model civilian air traffic andairspace, as well as USAF military flights'intersections with airspace sectors. MSimwas used to model arrival times of aircraft,as well as the staffing requirements for theflight planning/replanning function.

2 Background on Airspace StudyThe success of the USAF in the internationalarena hinges on the ability to access civilianairspace in a wide array of sovereignpolicies. Onboard aircraft equipagedetermines the level of ATC services andaccess to airspace--in general, the better theequipage, the better the service and access.The term Communication NavigationSurveillance / Air Traffic Management(CNS/ATM) is used to refer to these avioniccapabilities. It is not cost effective for theUSAF simply to equip every aircraft withthe best-available avionics. Rather, a trade-off analysis emerges: what level ofspending for upgrades coincides with agiven level of performance? Lackingaccess to the best routes and altitudes,

military aircraft are subject to increasedmixing with the civil fleet, and may suffercongestion delay. We modeled thisphenomenon for two future years by usingtwo different simulation models.

Several hypothetical scenarios wereconsidered. For each scenario, cases wheremilitary aircraft were "CNS capable"(enabled to meet CNS requirements to theextent possible) and "CNS not-capable"(lacking on one or more CNS capabilities)were examined. A hypothetical flight sortiewas considered, whereby various aircraft areto rendezvous at a given time and location.Full results have been presented in otherpapers [1, 2], and are not repeated here.Rather, this paper discusses the underlyingsimulation models.

In general terms, CRCT was used tomodel civilian air traffic over Europe, andusing some published equations for sectorloading, the times and locations ofcongestion were computed. Next, thesubject USAF military flights were likewise"flown" in the simulation model, and thetimes of sector-by-sector intersection wererecorded. This data on congestion andmilitary flight paths was next presented tothe MSim simulation for processing. MSimmodeled the dynamic interactions-militaryflights being delayed via reroute aroundcongested sectors, or queuing for tankers toperiodically refuel. Estimates of lateness offlights was used as input into a secondMSim model, to be used to model flightplanning/replanning staffing at an operationscenter. This second MSim model is the onedescribed below.

3 CRCTIn the field of air traffic management, twomajor functions exist: ATC providesseparation services between pairs of aircraft,or between an aircraft and airspace; trafficflow management (TFM), by contrast,manages system resources such as airports,routes, and airspace. It is the job of TFM toensure that the demand for resources doesnot exceed the available capacity.

Generally, in light of a predicted demand-over-capacity situation, aircraft can bemoved in either time (via delay assigrunent)or space (via reroute, in the case of airspacecongestion).

CRCT is a set of functions designed toassist TFM personnel in their quest tobalance air traffic demand with the capacityof airspace resources to accommodate thatdemand. Using a graphical user interface,flow managers may visualize airspace andforecast air traffic, and assess potentialairspace congestion. To amelioratecongestion, a traffic flow manager typicallyworks with "flows" or aggregates, asopposed to individual flights. CRCTsupplies a "what-if' capability, allowing atraffic flow manager to try variousstrategies, or variants on pre-storedinitiatives, in a virtual mode. The trafficflow manager would consider not onlyresource capacities, but also the impact onair carrier preferences. For example, someair carriers prefer a route deviation over aground delay. Once satisfied that acandidate initiative would likely succeed,the traffic flow manager would disseminatethe solution to airspace managers and users,expecting compliance from users, such aspilots and commercial air carriers. AlthoughCRCT was developed for real-time, real-world application, it also supports an offlinemode which will support simulation studies.

Developed by the Center for AdvancedAviation System Development (CAASD) atThe MITRE Corporation, CRCT currentlyexists on a research platform in the CAASDcomputer laboratories, and as a concept-evaluation platform at several federal ATCfacilities. For this project, we accessed theoffline and playbook features of the softwaresystem-all of the real-time, real-worldtraffic management functions in CRCT canbe used for simulation modeling. In theparlance of simulation modeling (see [3]),we have a dynamic. deterministic simulationmodel. It is dynamic in that the situation(state of flights, sector demand, etc.) changeover time. It is deterministic in that thereare no random variables-a set of flight planinputs create the same outputs each time the

simulation model is run. Specifically, weused the following functions in oursimulation modeling.

3.1 Traffic Flow and Demand AnalysisThe CRCT adjunct supporting infrastructureperformed flight plan and trajectoryprocessing for the subject military flights.(A trajectory is the estimated future path of aflight, consisting of geographic position andaltitude per time).

In addition to the subject military flightpaths, multiple complete days of civiliantraffic were modeled. Figure I gives asnapshot-a "freeze" in time--of this busy

"'.':!8. 'J~"':lWl\"1I'~~;.\':"~['j'<~5!'~,'?'''~W~~t;~iF~''~p;'

+

.; ~ A!f- Rar>9<'~ ..-: 600

W-61 00

... 1--00, ,-6100..............

-4100 -3100. I . I . I ,

-2100..............

-~IOO...............

,........0600Z '0600Z ' o7boz

..........--.oeooz

..........--.0900Z

Planning Sets 17 All ..J None

environment. In the lower right of thefigure, the label "FUTURE" is displayed,indicating that the user is in forecast mode.Note the very busy nexus on the continent-these are busy civilian airports in cities suchas Paris and Frankfurt. Note also the flowson the far left, headed west. This is themorning (10 a.m. local time) push ofoutbound flights to North America.

To assess sector congestion and thepotential impact on military flights, an arrayof flight geometry information per sectorwas captured and analyzed. Per publishedequations and thresholds, sector loading wasassessed using: traffic level, number ofaltitude transitions, number and type ofpairwise aircraft proximity events, etc.

~

.

Tools

~

~

>+-

¥.+2:00..............

+6:00..............

+3:00'--'--' +4:00..............

~O:10192 'uboz..........--.~oz

..........--.~z

..........--.~oz

..........--.~400Z

Fig. 1 CAPER Traffic Display for Europe

3.2 Aircraft Reroute DefinitionIt was decided, for our model representation,that military flights would avoid congestedsectors by re-routing around them. It washence necessary to determine the additionalflight time for congestion-avoiding flights.The re-routes would not be representeddynamically in the CRCT modeling-rather,a probability distribution of typical re-routetimes was developed: several analysts

..

played the role of traffic managers. Usingthe CRCT graphical user interface (GUI),the analysts selected 30 sectors at random,and worked-out re-route paths around thesesectors. These 30 delay times were thenfitted to a log-normal distribution, and usedin MSim, as described below. Figure 2shows military routes overlaid on civiliantraffic, at a zoom-in magnification greaterthan that of Figure 2.

~

.

y .lItr~:';f:lC."-'1:.""!'..I_.~

Display Maps

''\.

~~'""""...;.;.;;.;;;,;,,~:.;;;1:.a::..._

Fig. 2 Military Routes Overlaid on Civilian Traffic

3.3 Capture of Dynamic DataCRCT allows capture and recording ofdynamic data. In the simulation mode, thedynamism is a function of events scheduledand executed per the simulation systemclock (In the real-time mode, the dynamicsare obviously a function of real-worldactions and events unfolding). Dynamicdata on sector-specific flight geometrieswere captured and evaluated to determinetimes and locations of congested sectors.

3.4 Data Transfer from CRCT toMSimIn summary, three sets of information aretransferred from the CRCT simulation to theMSim simulation runs:

Times and locations of congestedsectorsParameters for the probabilitydistribution representing congestedsector delaysMilitary flight trajectories withsequences of sector entry and exit times

MSim uses these datasets, in concert withinfonnation about tanker capacities, their enroute locations, to model the dynamical timeordering of events, in light of a complexnetwork of interacting entities. The goal isto assess the probability of rendezvous ofspecific sets of airborne assets, i.e., thenumber of late arrival flights. As a secondapplication of MSim, these latenessestimates are used to help estimate therequirements for flight planning staff, asdescribed below.

4 MSim and the AOC ProcessModelIn the USAF, flight planning/replanningtakes place at an Air and Space OperationsCenter (AOe). An elaborate sequence andsynchrony of infonnation flows anddecisions must precede any execution ofcoordinated flight planning activities.

An AOC process model was developedusing MSim, which is based on Petri netmethodology. Petri nets were developed for

systems in which communication,synchronization, and resource sharing playan important role. See Figure 3 for thesimplest atomic example. Starting from theleft, when the two input places (note InputPlace 1 has two precedent activities) have atoken, then the transition fires, and tokensare moved to output places. The transitioncan be configured to consume some amountof simulation clock time.

MSim is a prototype simulation tool thatwas developed at The MITRE Corporation,initially, to model the perfonnance ofdistributed computer systems, but later usedto analyze the perfonnance of businessprocesses [4]. MSim may be categorized asa dynamic, stochastic simulation model. Itis dynamic in that the situation (rendezvoussuccess rate, requirement for flight planningstaff, etc.) changes over time. It isstochastic in that random variables, e.g.,time to refuel or time to re-plan acoordinated flight package, create "chanceeffects" and a probabilistic outcome.

Given Transition

Fig. 3 A simple two-input, two output Petri Net

MSim also utilizes a high leveldefinition of system "threads" to specify therouting of tokens in the Petri net. A threadis a path through the model of the system fora given system stimulus, (For readersfamiliar with the Unified ModelingLanguage (UML), this corresponds to thetenn "scenario," as used by the ObjectManagement Group [5].) It may be open or

closed and is drawn in the model diagramsby associating a thread of a specific colorwith a set of edges, as illustrated in Figure 4.Threads also have a name and a priority, andmodel diagrams can show multiple threadsat the same time, which is a valuable feature,both for debugging and for appreciation ofcomplex interrelationships.

I~ ~ AaMum8r(",..'"

. ~...~~:~> I.

"'

.

"'

.+1 --

H

i

.~

~'~.

.. I

~-::;~ .

...~:. - II

. . . I~I

-.I. I

III. I,

I:

~ ~, ~_!

.' -. '-

..

....:-:-"'-

,;' ~ ~

.. . ~

.' ,I--~

I .

.- I. --~~

. '- 1--;-;--: --4 ~. J

.. - '00-... . -

.'

I

f . ..'--

--'-~ ~ "

~

Fig. 4 Portion of AOC Petri Net Model

For a given scenario, MSim producesperformance metrics such as resourceutilization, component throughput, andthread response time. These metrics can beused to: (1) determine if the processmodeled meets its operational performancerequirements, (2) to find the performancebottlenecks, and (3) to evaluate theperformance effectiveness of proposedprocess changes. Metrics may be exporteddirectly to Excel for plotting and subsequentuse in other automation products. MSimmodels can participate as a member of asimulation federation [6] through integrationusing High Level Architecture (HLA) suchthat federation time and MSim time aresynchronized.

Figure 4 shows part of the AOC modelthat is implemented in MSim. Petri netTransitions (rectangular box) model theactivities that do work and produce outputs.Places (circles or ellipses) represent the typeof data that the Transition needs for inputand the type of data that the Transitionproduces as output. Tokens represent theinstances of data created and consumed bythe Transitions. Tokens are also associatedwith thread types and their movement is

restricted along Arcs that are associated withthe same types.

The execution behavior of an ordinaryPetri net follows two simple rules: (1) Onceall the input Places to a given Transitionhave a token, then the Transition can fire(occur) and is said to be enabled, and (2)When a Transition fires, it takes a tokenfrom each input place and puts a token ineach output Place. A timed Petri net allowsdiscrete event simulation to be modeled, andin this case, Transitions usually have timingfunctions that introduce time delays inprocessmg.

The timed Petri net methodology isuseful for modeling the performance of real-time distributed systems partly because ofthe ability to analyze concurrency andresource contention in a manner appropriateto real-time systems. The uniquecontribution of MSim is that it distinguishesbetween data and resource tokens and

provides priority preemptive scheduling onthe resources within the tool. Moreover itutilizes system threads to route the tokens.This is a fundamental architectural featurethat should not be hidden in low-level logic.Finally, it provides in-place hierarchy to

represent model hierarchy. This makes thecontext easier to understand when thecurrent focus is down several levels. Themodels that result from using the Petri netparadigm are similar to the physical designthat they represent. This is an alternative tothe models developed from the process-oriented simulation paradigm, where modelstend to be an abstraction based on theexecution flow, and are harder to correlate to

the physical design. This closeness of theMSim model to the real design makes themodel easier to verify. Table 1 identifies theMSim's Petri net features, which are moregeneral than ordinary Petri nets.

In summary, MSim was successfullyused to model military flight arrivals underconditions of a rendezvous requirement, aswell as the staffing requirements for theflight planning/replanning function.

~

-

Table 1 The MSim Petri net features

5 ConclusionSimulation modeling is one of the mostprominent analytical solutions available formodem, complex applications. It has beensaid: "When all else fails, simulate",suggesting that closed-fonn and otheralgorithmic solutions, though desirable, may

be inadequate to represent the vagaries ofthe real world.

Two quite different simulation modelshave been described. CRCT was developedas a real-time tool for air traffic managers,but offers a wide array of offline facilitiesthat can support simulation modeling.

Feature DescriptionHigh level --A Place is marked by a multi-set of structured tokens. The tokens have aPetri net thread type and can carry a data structure as well as a synchronization value

(i.e., control fork and join operations)Timed Petri net --The Transition firing takes a user-defined amount (i.e., distribution) of timeArcs --Associate thread type with the Arc of the net and only allow movement of

tokens along edges having the same type as the tokenTransitions --Every Transition has a code expression that can have a Boolean guard

function that must evaluate to true for the Transition to fire. The codeexpression can change the tokens type and set values in the data structurewithin the token

Specify the --Allows fused Places and Transitions so that a Petri net can be made up of aPetri net set of pages with common features.

--Hierarchy is used as a short hand way to specify a set of Places connected to agiven Transition or a Place connected to a set of Transitions.--Binding is a set of input data tokens-one from each input place. It representsthe necessary data and resources required to do the work in a Transition. It hasan inherent thread type associated with its tokens and an associated priority asdefined by the thread-type. The inherent thread type of a Ready-to-RunTransition controls the type of resource tokens used by the firing of theTransition. Thus, a different resource could be used depending on the type ofdata flowing through the Transition

Resource --A built in, optional resource allocation algorithm determines the highestallocation priority bindings that should be running with the required resources. Thealgorithm algorithm is a variation of a standard combinatorial problem called the

Provisioning Problem [7]. Each binding corresponds to the items beingprovisioned and the resources are the provisions. Their cost is the binding'spriority. This algorithm is run whenever there are Transitions that could fire toselect the Transitions that will be running next

These facilities were used for our Europeanairspace study - civilian traffic for twofuture years, and associated sectorcongestion, and flight path trajectory ofmilitary flights pursuing a rendezvous pointwith minimal delay.

MSim, by contrast, was used torepresent time sequencing and dynamicprocesses such as: queuing for refuelingresources, or demand for flightplanning/replanning staff. Using a Petri nettechnology, MSim supports classicaldiscrete event simulation, plus allowstransparency with respect to "threads" ortransaction sequence paths through multipleheterogeneous processes.

References:[1] Wigfield, E., Connolly, K., Alshtein, A,

DeArmon, J., Flournoy, R., Hershey,W., James, J., Mahoney, P., Mathieu, J.,Maurer, J., Ostwald, P., "Missioneffectiveness and European airspace:U.S. Air Force CNS/ATM Planning forFuture Years," Integrated Commandand Control Research and TechnologySymposium (ICCRTS), September 26-28, 2006.http://www.dodccro.org/eventsl11th ICCRTS/htmllpapers/139.pdf

[2] Wigfield, E., Connolly, K., Alshtein, A,DeArmon, J., Flournoy, R., Hershey,W., James, J., Mahoney, P., Mathieu, J.,Maurer, J., and Ostwald, P. "Missioneffectiveness and European airspace:U.S. Air Force CNS/ATM planning forfuture years," Accepted for publicationin the Journal of Defense Modeling andSimulation, San Diego, CA, 2006.

[3] Law, A and Kelton, W. D., SimulationModeling and Analysis, McGraw-Hill,New York, 2000.

[4] James, J. H., and Schaffner, S. C.,"Visualization of the dynamics ofperformance simulations," Proceedingsof the Summer Computer SimulationConference, LaJolla, CA, 1994.

[5] Object Management Group, Website onUnified Modeling Language, 2007.http://www.uml.org/

[6] Kuhl, Weatherly, and Dahmann,Creating Computer Simulation Systems:An Introduction to the High LevelArchitecture, Prentice, Jersey City, NJ,1999.

[7] Lawler, E., Combinatorial Optimization:Networks and Matroids. University ofCalifornia at Berkeley, Holt, Rinehart,and Winston, 1976.

~ D()lo~

Ow~

OW,D