lily exploit the many complex capabilities of the aircraft. Indeed, it may be necessary to ...... Reading, MA: Addison-Wesley Publishing Co. Kotulak, J., and Rash, ...
USAARL Report No. 97-18
Aircraft Multifunction Display and Control Systems: A New Quantitative Human Factors Design Method for Organizing Functions and Display Contents
BY Gregory Francis Purdue University
and Matthew J. Reardon Aircrew Health and Performance Division
April 1997
Approved
for public
releese,
distribution
unlimited.
U.S. Army Aeromedical Research Laboratory Fort Rucker, Alabama 36362-0577
Notice
Oualified reauesters Qualified requestersmay obtain copies from the Defense Technical Information Center (DTIC), Cameron Station, Alexandria, Virginia 223 14. Orderswill be expedited if placed through the librarian or other person designatedto requestdocumentsfrom DTIC. Change of address Organizationsreceiving reports from the U.S. Army Aeromedical ResearchLaboratory on automatic mailing lists should confirm correct addresswhen correspondingabout laboratory reports. Disposition Destroy this document when it is no longer needed. Do not return it to the originator. Disclaimer The views, opinions, and/or findings contained in this report are those of the author(s) and should not be construedas an official Department of the Army position, policy, or decision, unless so designated by other official documentation. Citation of trade names in this report does not constitute an offkial Department of the Army endorsementor approval of the use of suchcommercial items. Human use Human subjectsparticipated in these studiesafter giving their free and informed voluntary consent. Investigators adhered to AR 70-25 and USAMRDC Reg 70-25 on Use of Volunteers in Research. Reviewed:
WlWk JEFFREY C. RABIN LTC, MS Director, Aircrew Health and Performance Division Releasedfor publication:
eview Corm-i-&tee
Col&el, MC, k&S Commanding
.
Unclassified SECURITY CLASSIFICATION OF THIS PAGE
REPORT DOCUMENTATION PAGE
I
OMB No. 070+0188
lb. RESTRlCrmE M4RKlNGS
Unclassified 2a. SECURITY CLASSIFICATION AUlHORllY
3. DlSTRlBUTlON /AVAlLABlLllY OF REPORT
Approved unlimited
I 2b.DECLASSlFlCATlON
I DOWNGRADING SCHEDULE
for
public
release,
distribution
. 4. PERFORMING ORGANlZATlON REPORT NUMBER(S)
USAARL Report
No.
5. MONITORING ORGANKATlON REPORT NUMBER(S)
97-18
. 6a. NAME OF PERFORMING ORGANKATlON Army Aeromedical U.S. Research Laboratory
8b. OFFICE SYMBOL
7a. NAME OF MONITORING ORGANKATION
U.S. Army Medical Command-
I 6cADDRESS@ly,State, endZ/PCode) P.O. Box 620577 Fort Rucker, AL 36362-0577
Research
and Materiel
7b. ADDRESS (City, Sfate, and ZIP Code)
I
. 6a. NAME OF FUNDING I SPONSORlNG ORGANlZATlON
6b. OFFICE SYMBOL (~applicable)
.
Fort Detrick Frederick, MD
21702-5012
9. PROCUREMENT INSTRUMENT IDENTlFlCATlON NUMBER
I
10. SOURCt OF FUNDING NUMBtRS
6~. ADDRESS (City, State, andZiP code)
PROGRAM ELEMENT NO.
ACCESSION NO.
3M162787A879 11. TlTLE (h/udeSewlify-)
.
0) ,Aircraft multifunction design model for organizing
display and control systems: A new quantitative functions and display contents
human factors
12. PERSONAL AUTHORIS)
.
Gregory
Francis
and Matthew
13a. TYPE OF REPORT
Reardon
13b. TIME COVERED
14. DATE OF REPORT (Year, A&WI, Day)
15. PAGE COUNT
Final 16. SUPPLEMEHTAL NOTATION
17.
COSATl CODES FIELD
GROUP
01
02
24
07
16. SUBJECT TERMS (Co&me on mwtse
necessary and Ueniify by thck numbetj
SUB-GROUP
19. ABSTRACT(Continueonnsmuseifneoessary
Cockpit design, workload
hierar:hy
, multifunction
displays,
andithtifybybincicnumbetj
The objectives of this study were to review the current state of aircraft multifunction display and control system (MF'DCS) design methods and develop a quantitative method of Reports in the issues. designing MFDCSs that incorporate important human factors literature indicate that MF'DCS design can influence flight performance. However, current MFDCSs in design methods rely primarily on the designer's intuition and experience. aircraft cockpits use computer-generated graphics and symbology that have integrated and largely replaced the myrtad discrete electromechanical flight instruments found in older While much is known about the physical and visual properties of ME'DCSs, less is aircraft. MFDCSs may result known about which human factors are important for their design and use. in greater workload if the distribution of virtual instruments, graphical and text data, and control functions in an n-dimensional structure of display pages places excessive cognitive and psychomotor demands on pilots during either routine or emergency situations. involving the derivation of a weighted sum of A quantitative method was developed, each of which incorporates the effects of an arbitrary number of separate cost functions, of The method models, using a high level human factors and MF'DCS design guidelines. I
. 20.DlSTRlBUllON
q
I
/ AVAlLABILl-lY OF ABSTRACT UNCLABSIFIEDAINLWTED 0 SAME AS RPT.
ZZa.NAME OF RESPONSlBLE INDMDUAL Science Support Center Chief, IID Form 1473, JUN 88
21. ABSTRACT SECURllY ClASSIFICATlON [3
DTICUSERS
Unclassified
.
22b. TELEPHONE (/n&de Ama code)
1 (334) pmkus-mobsokLa
255-6907
1 22~. OFFICE SYMBOL
1 MCMR-UAX-SI SECURITY CLASSlFlCATlON OF THIS PAGE
Unclassified
19.
Abstract
(Continued)
a pilot's search for specific information or functions among abstraction, alternative hierarchies of MF'DCSdisplay pages. An annealing algorithm was proposed as an effective numerical method for finding the display page hierarchy that minimizes the composite cost function. Further research is needed to determine whether the set of constituent cost functions is sufficient or needs to be expanded. Studies also are needed to determine specific values for cost function coefficients and to validate the overall The quantitative method delineated in this report for designing model. optimal hierarchies of ME'DCScontent pages and functions may become useful for engineers as a design tool'during development of MFDCSs that will maximize pilot performance and minimize errors and excessive in-flight workload.
Table of contents Page
.
Introduction .................................................................. Reducing information workload in the cockpit ......................................
.7
.
Integration ................................................................. HUDS ..................................................................
...8
Pilot’s associate ............................................................
.8
Alternative MFDCS interfaces ................................................
.9
Expandeduse of visual and auditory senses ......................................
.9
MFDCS content and interface design .............................................
10
MFDCS design issues ........................................................
10
Quantitative MFDCS design methods ............................................
14
Current stateof MFDCS design ................................................
19
A new quantitative method for optimizing MFDCS content hierarchies ................. Cost as expectedaccesstime ................................................. Hill-climbing ............................................................
.20 .20
..2 2
Cost for related functions ....................................................
.23
Cost for expectedaccesstime and relatedness....................................
.26
Simulatedannealing.........................................................2 Costfunctions ........................................................... Frequentlyusedfimctions ................................................ Timecriticalfunctions
...................................................
Ideallocations .........................................................
7 ..3 1 ..3 1 ..3 1 ..3 1
Repeated selectionof buttons ..............................................
.32
Minimize number of levels ................................................
.33
Minimize overall accesstime ...............................................
.33
Related functions on close pages ............................................
.33
Consistentlocation of related items ..........................................
.33
.. . 111
Table of contents(continued) Page Relatedfunctionson the samepage .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34
Errors ................................................................
..3 4
Dedicateddisplays......................................................
..3 4
Discussion................................................................
..3 5
Conclusions...............................................................
..3 6
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...37
List of figures
1. AschematicoftheaftcockpitintheAH-lWSuperCobra,Venom..
.3
..................
2. Simulatedpagesfor the proposedVenom Super-Cockpit............................ 3. The designprocessfor the developmentof the V- 22 Ospreycockpitand MFDCS
.4
4. A hierarchialstructurewith threelevelsandthreepossibleoptionsat eachchoicepoint 5. The developmentof a hierarchythroughhi&climbing
12
.......
.............................
...
16 .24
6. The developmentof a hierarchythat minimizesdistancebetweenrelatedfunctions......
.25
7. Hierarchiesfor minimizationof expectedaccesstime andrelatedness................
.26
8. An intuitive descriptionof hill-climbing and simulatedannealing ...................
.28
9. Frequencyof final hierarchycostsfor hill-climbing and simulatedannealing...........
.30
iv
Introduction Everyone,at leastoccasionally,hasprobably experiencedfrustrationwith poorly designed interactive,computer-based,information display interfaces. Modem automatedtelephony systems,for example, have had notoriouslyproblematicuserinterfacedesigns. Users often have hadto listento lengthy, complicatedinstructionsand navigatetheir way throughnumerouslevels of optionmenusto eventually reachor input the information they desired. Likewise, the typically poor interfacedesignsfor many bank automaticteller machines(ATMs) have preventedindividuals from learningto use them (Rogerset al, 1996). This hasfrustratednew ATM usersand, undoubtedly,has been costly to providersof thesetypes of servicessincethey probably lost a portionof thesecustomersto competitorsoffering alternative,easierto use systems. In aviation contexts,the quality of an interfacedesignfor electronicdisplay and control systemscan obviously have greaterimpact than mere inconvenienceand frustration. Military and civilian aircraftdesignedin the 1960’s and 1970’s had so many separategauges, dials,lights, switches,buttons,circuit breakers,controlwheels, and leversin compactaircraft cockpitsthat crewmembersnecessarilyhad to spenda significantamount of time heads-down scanninginstrumentpanelsto find the information and functionsrequiredto maintain safe flight. At that time, display, monitoring, and control functionswere still largely dependenton the useof looselyinterconnectedanalogsystems. With suchtechnology,maintaining continuous,complete, andaccurateawarenessof aircraft statusimposeda heavy psychomotorworkload. It required explicit mental effort to continuouslyintegratethe dynamic information from the many scattered dials,gauges,and advisory or caution lights. Furthermore,an early or subtleemergencysituation probablytook longer to clearly identify, and more stepsto correct,than is usually the casein currentgenerationaircraft. Becauseof high pilot workloadsassociatedwith early generation cockpits,most transportaircraftrequireda flight engineerin addition to the pilot and copilot. The developmentof increasinglycapablemicrocomputers,softwaretools for implementing realtime digital data acquisitionsystems,and advancesin the designand manufactureof small video displaysprovided the technologyfor the evolution of computerizedmultifunction display and controlunits for both military and civilian aircraft. Technologicaladvancesgraduallypermitted replacingthe multitude of separateelectromechanicalstatus,warning, and control deviceswith integratedmultifunction display control systems(MFDCSs). From their inception, MFDCSs were often similar in appearanceand usageto ATMs in that crewmemberspushedbuttonsto move througha hierarchy of displaypagescontaininginstructions,information, or lists of user-activated functions(e.g., data entry). MFDCSs gained increasingacceptanceamong aviatorsand were generallycreditedwith reducingcockpit instrument“clutter” as well asreducingthe time crewmembersspentsearchingfor, and mentally integratingaircraft statusinformation. The reductionin pilot workload due to the introductionof increasinglycapableMFDCSs in the cockpit was a primary factor in eliminating the need for flight engineersin most currentgeneration transportaircraft.
1
The initial impressions of MFDCSs werethattheyreducedpilot workloadduringroutineflight. However,with time,any reductionsin workloadweregraduallyoffsetby the ability of these computer-based cockpitsystemsto encapsulate an increasingnumberof additionalfeatures, functions,andcapabilitiesnot feasiblewith the analogsystemstheyreplaced.This progressive increasein functionalityhasbecomeparticularlyapparentin military aircraft. For example, military combatandelectronicwarfareaircrafthavebeenusingcomputer-based displaysystems sincethe 1970’s and,althoughtoday’s versionsof thesesystemshavemuchgreatercomputational speeds andmemorycapacity,the numberof functionsavailableto usersseemsto haveexpanded proportionately.Most of the expandingarrayof functionsrequiresubstantial crewmember involvement(e.g.,monitoringa largeamountof additional,previouslyunavailable,information; selectingfrom an expandedarrayof optionsandsystemconfigurations; multisensor-based decision making;andtroubleshooting complexsoftware-dominated systems).Therefore,crewmember workloadswith currentstate-of-the-art aircraftMFDCSs in somecircumstances may actuallybe greaterthanthatexperienced in olderaircraftwith lesssophisticated systems. The useof MFDCSs in U.S. Army helicoptersisjust beginningto becomeprevalent.Currently, for example,only the OH-58D scouthelicoptersandversionsof the UH-60 utility transportfor specialoperations havemorethanoneMFDCS in the cockpitinstrument panel. OtherArmy helicoptersareprimarilyequippedwith the moretraditionalarraysof discreteelectromechanical gauges,dialsandswitches. However,helicopterupgrades andentirelynew helicopterdesignsfor the U.S. Army, suchasthe Comanchescout/attack andthe TiltRotortransporthelicopters,will includemultiple,highly integratedcockpitMFDCSs andretainonly a few criticalbackupanalog gaugesto maintainbasicflight capabilityin caseof completeelectronicsystemsfailure. Figure1 is a schematicof the aft (copilot/gunner) cockpitlayoutof the AH-1 W SuperCobra attackhelicopterasproposedfor the BritishArmy (Holley andBusbridge,1995). This is a modern versionof the AH-l Cobragunship,which originallywasdesignedfor, andeffectivelyusedin the Viet Nam War. SuperCobra prototypesincorporatean advancedtechnologymissionequipment packagecalledthe SuperCockpit whichincludestwo largecolorMFDCSswith 26 push-buttons integratedinto the surrounding bevels.Eight of the push-buttons arehard-keyswitcheswhich activatecriticalor frequentlyusedhigh-levelfunctionsor displaymodes. The other 18 pushbuttonsaresoft-keys,meaningthattheir functionsandlabelsmay changeacrossdifferentMFDCS displaypages. Figure2 depictstwo MFDCS displaypagesfor the SuperCockpit (Holley andBusbridge,1995). The left displayshowsreal-timestatusinformationfrom the aircti enginesandotheraircraft systems(SYS). The push-buttons on the right sideof the panelareassociated with softwaregenerated displaylabelsindicatingjumpsto additionaldisplaypagescontainingrelated information.Pressinga soft-keycausesthe MFDCS to displaya newpagecontainingthe informationor functionsindicatedby the key’s label. MFDCSs typicallycontaina wide rangeof singleandmultistepfunctions.The type of objects andinformationdisplayedon the MFDCS, the dataacquisitionchannelsthat are represented by the 2
0
HMD
1
/
@-
I
STOWAGE 1_(16’x5.75’x-6’)~
I
Figure 1. A schematicof the aft cockpit layout in the AH- 1W SuperCobra,Venom. Two MFDCSs display the bulk the information.
3
Figure2. Simulatedpagesfor the proposedVenomSuperCockpit.The systemspage(top left) showsinformationon enginesandincludeslegendsalongthe rightto indicatethatpressingthe associated buttonwill causethe displayto presenttherequested information.Targeting informationis shownin thetop right figure. The hierarchicalstructure corresponding to someof the MFDCS is presented at the bottom. 4
displayedobjects,the setof activedatabase links,aswell asthe functionsthatsoft-keyscanactivate arecommonlygroupedtogetherlogicallyon oneor moreinterconnected displaypagesforminga specificMFDCS mode. Flight crewscancyclethroughthe numerous MFDCS functionalmodes with oneor moreof the surrounding push-buttons. Typical MFDCS modesincludethosefor attitudereferenceandnavigation,communications, movingmapdisplay,systemscontroland status,targetingandweaponsselectionandstatus,aswell assituationalawareness displaysbased on multisensor datafusion.SomeMFDCS modesmay havedisplaypagescontainingclustersof relatedvirtualinstruments suchasattitude,altitude,andairspeedindicators,fuel gauges,moving maps,etc.,with or withoutsymbologyoverlaysfor navigationor weaponsselectionandtargeting. Displayscreens aredesigned to present,for any selectedmode,only a subsetof the total informationfromthemonitoredaircraftsystems.Pilotsdynamicallyselectdisplaymodesbasedon the informationandfunctionalitydesiredto accomplishconstantlychangingflight management or combattaskssuchassituationalawareness, navigation,communications, systems monitoring, battlefieldandthreatmonitoring,andtargeting. An MPDCS canbe conceptualized asa relativelysmalltwo-dimensional windowfor viewinga singlepageof informationselectedfroma muchlargernumberof pagesof staticanddynamicdata arrangedin a multidimensional hierarchy.The informationaccessible via an MFDCS anditshardor soft-keyoptionselectionbuttonshasa virtualstructure thatcanbe represented descriptively, graphically,symbolically,or asmathematical models.For crewmembers to efficientlyusecomplex andextensiveMFDCS dataandfunctionhierarchies, they mustacquirean accuratementalimage andconceptual understanding of how all the dataandfunctionsencapsulated in the available displaymodesaregroupedandinterrelated andhow this structure canbe efficientlyandrapidly traversed usingtheavailablededicatedandsoftwaredefinedbuttons.If the displaypagehierarchy andnavigablepathsbetweenfunctionallyrelatedclustersof displaypagesarenotwell understood, MFDCS usersarelikely to becomelostin the MFDCS’s informationspaceor becomeconfused with regardto the locationof immediatelyneededinformationor functions. Obviously,becominglostin the informationspaceof a poorlydesignedMFDCS wouldonly add to a pilot’s senseof dangerandconfusionduringin-flight emergencies involvingspatial disorientation, serioussystemfailures,or suddenunusualattitudes.Duringcriticalin-flight situations wherecomposure, clarityof thought,andefficientuseof time areessential,getting“lost” in the pagespaceof an MFDCS is likely to precipitatepanicandpreventidentificationand resolutionof the problem.In suchsituations, MFDCS usersmightbeginenteringessentially gunnersin randomMFDCS pagenavigationselections.Similarly,duringcombatoperations, Army attackor scouthelicopters, despitedangerandfear,mustbe ableto rapidlyandaccurately traversethe information(MFDCS mode)subspace relevantto their specialized tasks.Gunnersin highthreatscenarios mustbe ableto cycleveryrapidlythroughvariousMFDCS modesto accomplish suchtasksastargetdetection,recognition,hand-off,ranging,prioriti&ion, weapon selection,targetdesignation (e.g.lasing),weaponsfiring, andeffectassessment. Becoming confusedat anypointduringthesecomplexprocesses, with respectto howto transferbetween modeson the MFDCSsutilizedto performthetasks,couldresultin targetescapeor, of more 5
immediateconsequence, give the adversarysufficienttime to detect,closein, andfire first,with potentiallylethaleffect. ModernMFDCSs aretruly impressiveandseemfunctionallyandestheticallywell designedas depictedin advertisements andduringdemonstrations in circumstances of little or no stress.But, while the modemcockpitrelieson MFDCSs,little hasbeenpublishedregardinghow unusual, critical,or dangerous circumstances affectuser-MFDCSperformance andmissioneffectiveness. Furthermore, therehasnot yet beena systematic evaluationof MFDCSs to enumerateanddefinea taxonomyof the cognitiveandpsychomotor humanfactorissuesthat shouldbe considered during their design.In thisreport,we offer whatwe believeto be a new quantitativemethodfor designing MFDCS displaypagehierarchiesthatoptimizesthe distributionof contentandfunctionsusinga setof weightedprioritiesrepresenting humanfactorsanddesignguidelinesthoughtto be important influencersof user-MFDCSinteractions. MFDCSstradethe workloadassociated with visuallysearchingcockpitinstruments for a cognitiveworkloadassociated with a cognitivesearchthroughmentalimagesof a multidimensionaldatabase of pagesof informationandfunctions.Physicallysearchingfor a display pagecontainingnecessary functionscanbe time consumingandoftenhasthe additionaldrawback of requiringthe coordinateduseof buttons,cursorcontrols,anddataentrykeypads.These activitiescandistractcrewmembers andtemporarilyreducetheir situationalawareness.The SuperCobra andthe AH-64D LongbowApachecockpit includenumerousMFDCS modeselect buttonsandmenuscrolltoggleslocatednot only alongthe bordersof theMFDCS, but alsoon the flight controls(HannenandCloud, 1995). Studiesindicatethattime spentaccessing information from a MFDCS influencesperformance.Sirevaag,et al. (1993) hadfive U.S. Army helicopter pilotsfly simulatednap-of-the-earth (NOE) reconnaissance missionsandreportinformationat specificwaypoints.Reportingthis informationrequiredpagingthroughan MFDCS. Althoughthe pilotsalsohada head-updisplay(HUD) on their helmetthatprovidedaircraftsituationalawareness information(speed,altitude,etc.),flight performancewasadverselyaffectedasthe communication loadincreased.In particular,underhigh communication loads,pilotsspent,on average,8 more seconds perminuteabovethe specifiedNOE altitude. That studyillustratedthatthe time spent accessing informationfrom MFDCSs canadverselyaffectflight performance. Suchfindingsare consistent with concernsaboutthe workloadrequiredin continuously balancingflight andaircraftsystems-management duties.The capabilitiesof an increasingnumber of aircraftrequirecarefulattentionto, andskilleduseof, manyMFDCSs. For example,Dohme (1995) observedthat OH-58D AeroscoutandAH-64 Attackhelicoptersbothusethe airbornetarget hand-offsystem(ATHS), accessed throughan MFDCS unit. The database for the ATHS functions aloneconsists of approximately180differentpagesof menus,inputfields,andinformation(ATHS is alsooneof the optionsin figure2, top right). Dohmeestimatedthatabout300 pagesof informationsupported the entiresetof functionsin the MFDCS. He suggested that learningall the MFDCS modesanddevelopingthe ability to quicklyandefficientlyaccessthe relevantinformation for all potentialtaskswasa formidablechallengefor trainees.
6
Thereareconcerns aboutexcessive aircrewworkloadadverselyaffectingflight performance duringcomplicatedor stressful missions.Duringhighworkloadmissionsegments, crewmembers maybeginto selectivelyignoreelementsof informationwhichmay actuallybe quiteimportant. The nextsectiondiscusses methodsof improvingoverallinformationacquisitionin the cockpit,and thenfocuseson how to incorporate cognitiveandpsychomotor humanfactorissues,aswell as sections proposea newmethodfor designguidelines,intothe MFDCS designprocess.Subsequent includinghumanfactorissuesin determiningan optimaldistributionof MFDCS contentand functions,discusshow to applythe quantitativemethods,andrecommend directionsfor further research.
Reducinginformationworkloadin the cocknit As military aircraftcomplexityandfunctionalcapabilitiesincreased, concernarosethat crewmembers couldbecomemoreeasilyoverwhelmed with informationandtaskoverload.In response to thisconcern,theredevelopeda stronginterestin simplifyingcockpitsystem-user interfaces andassisting pilotsin copingwith theproliferationof flight andmissionrelated functions.A generalgoalfor new aircraftdesignswasto makeit aseasyaspossiblefor crewmembersto access, understand, andefficientlytakeactionon cockpitandsystems-related data. Thissectionreviewsvariousproposed methodsfor improvinginformationtransferto crewmembersto improveflight andmissionperformance andcapabilities. Integration The introductionof computer-driven displayandcontrolsystems into aircraftcockpitsallowed MFDCS designers to createnewanddynamicmethodsof combiningandpresentinginformation fromsystemsandsensors.A singlecockpitdisplaybecamecapableof simultaneously integrating manydifferentsources of information,thusreducingthe workloadrequiredto scana multitudeof separate instruments.Work in thisarealed to novelmethodsof integratingandportrayingflight information(reviewedby StokesandWickens,1988). In supportof theseefforts,a widevarietyof newsymbologywasdeveloped, but oftenit wasonly applicableto specificaircraft(e.g.,Newman, 1995,Appendix). Integratinginformationfrom multiplesources into an MFDCS cangreatlyreducethetime neededfor crewmembers to accessinformation.Additionalimprovement couldbe gainedby refiningthe criteriafor selectingwhichdisplayobjectsandsoft-keyfunctionsshouldbe collated togetherinto functionallyrelatedgroupsof displaypages.The properstrategyin designingthe contents, menus,andbranchingschemefor MFDCS pageshasthepotentialfor reducingthetotal numberof displaypagesor modes.Combiningrelatedinformationandfunctionalityinto relativelyfew coherentdisplaymodescangivecrewmembers a betterunderstanding of the entire informationstructure andallow fasterandmoreefficientuseof MJTDCScapabilities.As a result, automated flight systems informationintegrationcanachievelargesavingsin creweffort. 7
On the otherhand,integratingunrelatedinformationsources andsoft-keyfunctionsinto single displaypagescanhinder,ratherthanhelp,crewmember’sunderstanding of systemsstatus(Stokes andWickens,1988). Likewise,includingan excessivenumberof menuoptionsor soft-key functionsin a singledisplaypageor MPDCS modecanproducedisplayclutterandcomplicatea crewmember’ssearchfor a particularfunction.Decidingwhichfunctionsandinformationobjects to integratetogetherinto a singleor relatedgroupof displaypagesrequiresa thorough understanding of the interrelationship betweenaircraftsystemsandsubsystems, aswell asthe informationandfunctionsrequiredfor performingcockpitprocedures andmissiontasks.However, dataandfunctionintegrationbasedon thesefactorsaloneusuallywill not solveall MFDCS-user interfaceproblems.The MFDCS contentdatabase mustalsobe designedto incorporatedisplay pagesin a way thatmaximizesthe user’sability to efficientlysearchandlocatethe desiredMPDCS functions,options,andpagesor modes. HUDS
HUDs project(via applicationof advancedvideotechnologies) flight informationdirectlyinto the crewperson’line s of sight,therebyreducingthe needfor head-downscanningof cockpitpanel displaysor instruments.HUD systems allow pilotsto continuously trackrelevantflight performance parameters via computergenerated symbologyanddatasuperimposed on the direct line-of-sightimagery. Numerousstudieshavedemonstrated improvedfhght performancewith HUDs (seeNewman 1995 for a comprehensive review). Currently,however,HUDs cannot displayasmuchor aswide a rangeof differenttypesof dataanddisplayobjectsasMFDCSs. This is partlybecauseexcessiveinformationor displayobjectsprojectedon a HUD canleadto severe visualclutter,therebydeterioratinga pilot’s externalview. Therefore,HUDs do not supersede the needfor MFDCSs. Increasinglysophisticated MPDCSswill continueto be the primaryflight and systemsmonitoringandmanagement interfacefor civilian andmilitary pilotsfor many decades into the future. HUD andMFDCSs,however,will undoubtedlybecomeincreasinglyintegratedand complementary. Pilot’s associate A pilot’s associate is an advancedconceptfor assistingpilotswith a software-based systemthat usesdatafusiontechniquesandautomaticallyanalyzescomplexmultisensordata,recommends actions,andimplementspilot’s commands to performcertaintasks.Partof the pilot-associate interfacewill consistof an advancedhighly integratedMPDCS utilizinga largeflatpanelscreenas partof theuserinterface. It will incorporate artificial intelligencemethodsto adaptivelyintegrate multisensor informationanddynamicallyadviseandalert crewsaboutpotentialproblems, solutions,threats,andopportunities (McBryanandHall, 1995). It will alsobe capableof autonomous decisionmakingfor constrained andpredefinedcircumstances. The pilot’s associate will automaticallytrackandanticipatenecessary changesin flight modesandadaptivelyorganize anddisplaythe appropriatetask-oriented informationandfunctions.The developmentof sucha systemhasthe potentialto greatlyreducethe needfor pilotsto searchfor andintegrateinformation andfunctionsscatteredamongthe manydisplaypagesor modesin an MFDCS. 8
While potentially valuable,a pilot’s associatefor advancedmilitary rotary-wing aircraft is still an emergingtechnology. Moreover, similar but lesscomplex types of automationin commercial aircrafthave occasionallyled to seriousproblemswith “mode awareness,”whereby crews have experienceddifficulty determining what the automationwas doing (Sarterand Woods, 1995). Currently,mode identificationoften requirespaging up and down throughdifferent layers of the MFDCS modesto enablethe userto identify the most current settingsfor systemand control variablesas well asto reorientwith respectto location in the MFDCS mode or page hierarchy. Alternative MFDCS interfaces Researchsuggeststhat using an MFDCS function selectinterfaceotherthan push-buttonscan reducethe difficulty of navigatingthrough MFDCSs’ information and function space. Speech recognitiondevicesand pilot electroencephalograghic (EEG) signalsarepotential meansof handsoff interfacingwith an MFDCS. Such methodseventually could replaceor complementthe use of hard and soft-keysfor controllingMFDCS displays,selectingmodes,and activatingvarious functions. These alternativeinput interfaceswould have the advantageof freeing the pilots’ hands for othertasks. However, they will not necessarilylead to improved performancesearchingan MFDCS database.Reising and Curry (1987) found no difference in flight performancefor a speechrecognitioninterfacecomparedto a well-designedpush-buttoninterface. Whatever the interface,limitations in the designof the MFDCS still will likely impact a flight crew’s ability to lily exploit the many complex capabilitiesof the aircraft. Indeed, it may be necessaryto entirely restructurethe MFDCS databaseto obtain optimal performancewith a new interfacemethod. How to do this rationally is not clear and requiresadditionalMFDCS human factorsresearch. Expanded use of visual and auditory senses Another alternativeto the MFDCS interfaceis presentingflight and aircraftsystemsinformation to crewmembersthroughperipheralratherthan foveal vision. Stokesand Wickens (1988) provide a review of studiesthat evaluatedauditory and peripheralvisual displays. Information deliveredvia a peripheraldisplay is designedto be noticeablein the pilots’ peripheralvision. Although potentially useful, the benefitsof suchdisplayshave not yet beenverified in aircraft. Additional researchis neededto define how they can be effectively adaptedto enhancepilot performance, informationprocessing,situationalawareness,and decisionmaking. Simple auditory signalsare commonly incorporatedinto cockpit warning systems. However, more complex warning and advisoryauditory systems,to include three-dimensionalauditory “displays”to assistcrewswith situationalawarenessand threat localization,are being researched. Major drawbacksfor extensiveuse of auditory systemsare their potentialfor interfering with crew communication,the time neededfor listeningto and interpretinglong messages,and their transient nature,which may requirepilots to rapidly refocusattention from othertasksto mentally register the auditorymessage. These are somereasonswhy auditory pilot information systemsare unlikely to completely supersedevisually orientedMFDCS panels.
9
MFDCS contentandinterfacedesign MFDCS systemsaretypicallycomposed of hardwareand softwarecomponents.The hardware components includeaviationcapablecomputerboards,cockpitdisplaypanels,surrounding bevels includereal-timeoperating with push-buttons, andalphanumeric keypads.Softwarecomponents systems,routinesfor generatingdynamicsymbology,mapdatabases, aircraf3systemstiormation, aswell asdatabases for graphicdisplayobjects,soft-keyfunctionmappings,objectinteraction rules,performancelimits, procedures, andvariouschecklists.A governingevent-oriented software programkeepsthe systemcontinuously activeandresponsive to pilot inputsandchangesin aircraft status.The designof this software,andthe databases thatit candynamicallyandadaptivelydraw objectsandinformationfrom, is the focusof ourconcern. MFDCS softwareandassociated databases canbe conceptualized asa multi-dimensional spaceof interconnected pagesof information,menus,andfunctions.The high-leveldesignproblemis how to organizean optimal structure andpatternof interconnections for the informationandfunctionalityassigned to an MFDCS. Becauseof the complexityof thesesystems,it is usuallynecessary to defineoptimal&y with respectto constraints anddesiredperformance criteriaor goals. One of the essentialdesign goalsfor an MFDCS is thatusersbe ableto efficientlysearchthroughits informationspaceto find necessary dataandcontrolfunctionsin urgentsituations. Carefuldesignanddistributionof displayobjects,data,andfunctionsacrossMFDCS pagesand information. For example, modescanmimmizethe time andeffortrequiredto locatenecessary ReisingandCurry(1987) useda realisticF-l 5 simulatorgamewhichprojectedthe out-the-window view on a displayin a cockpitmockupandrequirednonpilottestsubjectsto accessflight, navigational,andsystemsinformationthrougha simulatedMFDCS. They comparedflight performancefor two hierarchicaldesignsof the MFDCS displaypages.They foundsubstantial improvementin flight performance whenthey organizedthe contentsof the pagesaccordingto the differentphasesof the flights,comparedto a fixed organizationthatclusteredthe information accordingto datasourcecharacteristics. Theirresultsindicatedthatdifferenttypesof MFDCS page hierarchiescouldsignificantlyinfluencesimulatedflight performance. Assigningfimctionsto pagesandswitchesis a difficult taskbecause the human-computer interactionsinvolvedin accessing informationfrom an MFDCS arecomplicatedandnot entirely understood.Unfortunately,the frequencyandpatternof MFDCS modeor pageswitchingand functionselectionsduringactualflight havenot beenwell documented.Also, the largenumberof possiblecombinations of pages,functions,andsoft-switches quicklyleadsto combinatorial explosionwhenattemptingto considerall possiblelayouts. The nextsectiondescribes current approaches to MFDCS design. MFDCS designissues Studiesof human-computer interactionhaveinvestigatedmanyimportantcharacteristics of displaysandhumaninformationprocessing.The displaysmustoperatewithin constraints imposed by the humanvisualsystem(e.g.,contrast,resolution,brightness, etc.)andthe propertiesof the 10
interface(e.g., size of knobs,buttonsizes,and resistance)must mesh with pilot abilities and anthropomorphiccharacteristics.Studiesof biophysicalinterfacevariableshave lead to military standardsfor MFDCS design(e.g., military standard(MIL-STD) -1472D). In particular,a great deal is known aboutthe responsesof the aviator visual systemunder variousconditionsin helicoptercockpits(e.g., Frezell, Hohnann, and Oliver, 1973; Frezell et al., 1975; Holly and Rogers, 1982; Behar, Bachman,and Egenmaier, 1988; Kotulak and Rash, 1992; Rabin, 1995,1996; Rabin and Wiley, 1996) and aboutthe electro-opticaland physical propertiesof electronicdisplay devices(e.g., Rash, Monroe, and Verona, 1981; Cote, Krueger, and Simmons, 1982; Rash and Becher, 1982; Rash and Verona, 1989; Kotulak, Morse, and McLean, 1994; Rabin, 1994,1996). This knowledge is clearly importantbecauseit helps ensurethat the variouscomponentsin modem aircraftcockpit instrumentpanelshave propertiesconsistentwith crewmembers’biophysical capabilities. On the other hand, few researchresultsare availablethat define and quantitatethe cognitive dimensionsand problemsrelatingto pilots acquiringand maintaining a clear mental picture of the distributionof information, displayobjects,menus, data entry fields, and functionsacrosshundreds of MFDCS display pages;the n-dimensionalinterrelationshipsbetweenpages;or the most efficient set of actionsto take to navigateto different display pagesor functions. This must become betterunderstoodso that MFDCS designcriteria can be developed in a truly rational manner. Surprisingly,developmentof presentand past generationsof MFDCSs have generallybeen ad hoc, relying on the experienceandjudgment of MFDCS designexperts. Most MFDCS designers organizethe information contentinto a hierarchicalstructureand then deviate from that structure when intuition, experience,or testingsuggeststhat it will be beneficial. The designof an MFDCS is difficult becauseeven a small contentdatabasecan generatean immense number of different hierarchicalstructures.Searchingthroughall the possibilitiesto find the besthierarchycan be very difficult, resourceintensive,and time consuming. Discussionswith membersof currentMFDCS design staffs(e.g., at Honeywell, Sikorsky, Army ResearchInstitute, U.S. Army Aeromedical ResearchLaboratory) indicatethat MFDCS designhas primarily relied on quasi-systematic,nonquantitativetechniqueslearnedthrough experienceand validatedwith trial-and-error. For example, Graf and Holley (1988) describedthe stepstaken to designthe MFDCS in the cockpitof the V-22 Osprey. Figure 3 is a schematicof the development process.The designersstartedwith a mission analysisto determine crewmemberdutiesfor the aircraftand specifiedrangeof missions. The designersdeterminedhow much time crewmembers had to carry out varioustasksduring missions. With this information, the designerscreateda cockpit design(including MFDCSs). They analyzedthe cockpit in two ways. First, they used a computerizedworkload and performanceanalysistool to predict whether the currentdesignwas acceptablefor the aircraft’s missionprofiles. Second,crewmemberstestedthe cockpit design in simulatedflights. These man-in-the-loopsimulator flights provided data for determining problem areasin the designand allowed crewmembersto make commentson positive and negative aspects of the new cockpit design.The designersmodified the cockpit systemsaccordinglyand iteratedthe processuntil cockpit instrumentationcapabilitiesmatchedmission and usability requirements. As 11
canbe imagined,thiswasa lengthyprocess.Repeatedlymeasuringpilot-MFDCS interactionin simulatorsis bothtime consumingandexpensive. Moreover,ad hocchangesto the MFDCS (or otherpartsof the cockpit),that help solveoneproblem,may inadvertentlyintroducenew ones.
p.-.---.-. II
MSSION MlALYSlS _~_._._~_._I_~_._ I
INTERACTIVE ANALYSIS ____._l___.___.-____~-.-.-~-_ !-m
I
. I ;
. ;
i .
;
i .
COCKPIT DES,GN
k-+ ; .
i . I
c
I ;
-,
i. A
CRITIW TASK 4tWLEiI.S
+
cOMPurwRED WORKLOAD PERFORM4hlCE AMLVSIS WAS)
-_--_-----------
OIJTPLIT FEEDS
i l
I I ;
-DESIGNRQMTS -c+i+xmx -SAT -M4WRIM 41s‘ Lsq LSAFI -ETc
i ,;_ I ; h@N-INTHELOOP SIMULATION
i I I
Figure3. The designprocessfor the developmentof the V-22 Ospreycockpitand MFDCS. Designersidentify constraints imposedby missionanalysisandthen iterativelybuilda cockpitthat satisfiesthoseconstraints (modifiedfrom Graf and Holley, 1988). The box with the thick edgeindicateswherethe proposed quantitativemethodwill influencethe designprocess.
Publisheddescriptions of MFDCS designtechniques emphasize thatthe layoutof functionsand pagesshouldfollow generalguidelines,butthey do not explainpracticalmethodsfor satisfyingthe guidelines(Calhoun 1978;Lind, 1981;SpigerandFarrell, 1982;MIL-STD-1472D; Williges, Williges, andFainter,1988;Holley andBusbridge,1995). Someof theseguidelinesare: 1. Frequentlyusedfunctionsshouldbe the mostaccessible, 2. Time criticalfunctionsshouldbethe mostaccessible. 3. Frequentlyusedandtime criticalfunctionsshouldbe activatedby the buttons thatfeel “ideally located”(e.g.,top of a columnof buttons). 4. Programrepeatedselectionof the samebutton. For example,locatethemost commonlyselectedfunctionof a menuon the samebuttonthat calledup that menu. Failing that,programcommonfunctionsto adjacentbuttons. 12
5. The number of levels in the hierarchyshouldbe as small as possible. 6. The overall time to reachfunctionsshouldbe minimized. 7. Functionsthat are usedtogethershouldbe groupedon the same or adjacent pages. 8. Related functionson separatepagesshouldbe in a consistentlocation. 9. Related functionsshouldbe listednext to eachother when on a singlepage. 10. Considerthe types of errorscrewmembersmight make and place functions accordinglyto minimize the effect of thoseerrors. 11. In some cases,frequentlyusedand time critical functionsshouldbe removed from the hierarchicalstructureand be given dedicateddisplays. Many of thesegeneralMFDCS designguidelinesarethe sameas thosefor structuringthe layout of physicalcontrols(Sandersand McCormick, 1987), while others(4,5,8,9, and 11) appearto be uniqueto the designof softwaregeneratedfunction selectionswitchesfor computer-drivendisplay units. Some of theseguidelineshave been investigatedexperimentally. For example, Snowberry, Parkinson,and Sisson(1983) showedthat searchspeedand accuracyincreasedas the numberof levelsin a hierarchyof user-activatedfunctionsdecreased(5). Likewise, Teitlebaum and Granda (1983) demonstratedthat placing relatedfunctionsin inconsistentpositionsresultedin a 73 percent increasein searchtime (8). A literaturesearchfound no reportsdocumentingthe degreeof effectivenessof the remaining guidelines,althoughthey seemreasonableand have face validity. MFDCS designersselectthe guidelinesthey considerto be most important. For example,in the developmentof the MFDCSs for the SuperCobraattackhelicopter, Holly and Busbridge(1995) focusedon guidelines 1,2,5,7, and 8. The designersgroupedrelated functions into one of eight subsystems(which were assignedto the buttonsalong the bottom of the MFDCS as in figure 2). Thesewere further organizedinto two major subgroups.Related information on the samedisplay pagewas functionally grouped,and the sameinformation on different pageswas presentedin the sameposition acrossthe pages. The designersalsoemphasizeda minimum-depth approachand ensuredthat all critical information was no more than two levels from the top of the MFDCS page hierarchy. The most critical information neededto fly and fight was no further than one level from the top of the hierarchy. However, applicationof thesegeneralMFDCS designcriteria is problematicbecausethey often conflict with eachother. For example, shoulda frequentlyused function be placed by itself near the top of the hierarchyof the MFDCS pages(1) or shouldit be placed in a submenuon a secondarypage with its related, but infrequentlyused,functions(7)? Likewise, shouldcriteria3,4 or 7 dominate selectionof a soft-key for a specificfunction? Currently, there doesnot appearto be a quantitativemethod of deducingthe optimal trade-offsso designerstry out different optionsuntil the entire system“feels” good. This is a time consumingtask becausemovement of a single function can requirea cascadeof relatedchangesthroughoutthe MFDCS. With an ad hoc, intuitive, or trial-and-errorapproachto the designof MFDCS datacontentand functionality, operationaltestsmust be usedto judge the performanceof an MFDCS. However, it 13
oftenrequiresa greatdealof effortbothto build new MFDCS layoutsandto measuretheir performance experimentally.Designers,therefore,may nothavesufficienttime or resources to generateandvalidatemany alternativedesigns.Indeed,in the designof the SuperCobra’s MFDCSs,Holley andBusbridge(1995) conclude,“A rapidprototypingcapabilityfor controldisplayformatsis suchan importanttool thatthe designof a ‘glasscockpit’shouldnotbe undertaken withoutone.”Thosedesigners hadaccessto simulators andgraphicsworkstations, and socouldquicklytry differentMFDCS configurations.However,thereis no indicationthatthey hada quantitativeoptimizationmethodfor assigningfunctionsto MFDCS pagesandbuttons.In figure3, the boldbox suggests wherea quantitativemethodfor buildinga MFDCS hierarchywould contributeto the overallcockpitdesignprocess.A quantitativemethodof designmightalsohelp clarify therelativeimportanceandinterrelationships of the designguidelineslistedabove. QuantitativeMFDCS designmethods Navigatingthrougha hierarchyof displaypagescontainingfunctionsmappedto hardor soft keysis a commonlyrequiredtaskfor many familiar applications (e.g.,automatedtellers,computer programmenus,telephoneansweringsystems).This sectionsummarizes somepreviously developedgenericformulasfor analysisanddesignof hierarchicaldatastructures.Because,to date,thesemethodshavenot beenfully developedor validated,they aregenerallyunsuitablefor notation complexpracticalapplicationslike the designof MFDCSs. This sectionalsointroduces for usein subsequent sections. Most MFDCSsincorporatehierarchicalstructures thatdefineorganizationof contentand navigationalpathsbetweendisplaypagesor modes.Navigationthroughthe hierarchyis accomplished via the useof navigationalobjectssuchasmenus,lists,andsoftor hard-keys.In a simplebranchinghierarchy,eachscreencontainsinformation,displayobjects(e.g.,virtual instruments, gauges,andwarninglights,symbology,andtext) andsoftkeysfor variousfunctions. Activatingsoft-keyson the displayor hard-keybuttonson the MFDCS bezelareusedto navigate throughthe hierarchyto the desireddisplaypageshavingthe desiredinformationand/orfurther selections.The top of the hierarchyis the onepagethatis not a selectionfrom any otherpage. Fromthetop page,the usernavigatesthrougha sequence of screens thatis uniquefor eachtarget page. Eachpagein the hierarchyis at a level whichindicateshow manyscreensthe usermustgo throughto reachthepage. Figure2 (hottom)showspartof the hierarchicalstructure in the SuperCockpit MFDCS. The top of the hierarchyis a dummypage,asit containsno informationexceptchoicesto jump to other pages.Many of the buttonsat thistop-levelpage(andotherpages)arenot used,but areincludedin the hierarchicalstructure to represent buttonlocations.The SYS pagepresentssomeinformation (not indicatedin thehierarchy)andoptionsto jump to otherpages,whichareindicatedby links to pagesat the nextlevel. Thesepageswill presentsomeinformationand(may) provideoptionsfor additionalpagesat the nextlevel. Thus,reachingthe MAINT pagefrom the top pagerequirestwo buttonpushes,oneto accessthe SYS pageandanotherto accessthe MAINT page. 14
Hierarchicaldata structuresalso are usedin computerscienceapplicationsfor databasesorting. By arrangingthe contentsof an ordereddatabaseinto hierarchicaltrees,a computercan more quickly searchthe database.A number of algorithmsexist to optimize the layout of a database (e.g., Knuth, 1973; Lorin, 1975). Unfortunately,thesealgorithmswere developedexclusively to satisfyrequirementsfor efficiently searchingthroughstructureddatabases.These early algorithms, however, did not include mechanismsfor optimizing databasestructureswith respectto the numerousand complex details of human-computerinteractions. The databaselayout algorithmsfor efficient automatedsearchingfor simple informationdo not appearto be generalizableto the more difficult problem of human searches.Nevertheless,the notationfor describinga hierarchy is useful in both situations. Considerthe hierarchy in figure 4a. It consistsof n = 3 page levels, (0,1,2) with m = 3 menu options(representedgraphically as the lines emanatingfrom nodes)possiblefrom eachpage (representedasthe circular nodes). Each page,or node, in the hierarchyis indexed as (i, k) which indicatesthe level, 0 I J’< n , of eachpage andposition, 0 I k < m' , in that level (n.b., k=O for the first page or node at each level). The numbersin the hierarchy schematizedin figure 4a suggest this coding scheme. Note that the total numberof pagesin this type of hierarchyis: 2
m’
.
j=O
It will be helpful to discusshow this notationcorrespondsto movement in the hierarchy. The “parent”menu, if it exists,of page (i, k) is at position (j - l,Lk / ml) , where 1x1 is the largest integerlessthan x (i.e., round x downward). Likewise, the “children” of page (j, k) , if they exist, are found at positions (j + 1,rCm)to (i + 1,km + m - 1) . Figures4b and 4c demonstratehow the notationcorrespondsto the positionsin the hierarchicalstructure. This notationonly describesthe positionsof pagesin the hierarchy, it doesnot requirethat a page actually containsa function or jump selection. Somepagesin this hierarchymay containinformation, virtual instruments,or other display objects,in addition to mechanisms(e.g., menu sectionor soft-keys)to jump to other pagesas constrainedby the interconnections. Other pagesalso contain specificfunctionsthat can be activated. These functionsallow the userto interactwith aircraft systemsto perform necessary tasks. Supposethere are v functions in a database.Let i=O, 1,...,v-1 index the functionsand let v-1 q(i) = (i, k) indicate the position of the functionin the hierarchy. Define Q = m(i) asthe setof i=o pageindicescontainingfunctions. With this notation in hand, we can describea simple model of the human-computerinteraction and showhow to minimize expectedfunction accesstime within a restrictedclassof hierarchies. If 15
Level 0
Level 1
Level 2
Level
Level
Level
Level 0
Level 1
Figure4. (a) A hierarchicalstructure with threelevelsandthreepossibleoptionsat each choicepoint. The numbersindicatea codingschemethatidentifiesthe positionof each optionat eachlevel. Eachpositioncanbe identifiedasa coordinatepair (j, k) , wherethe firstnumberindicatesthe level andthe secondnumberindicatesthe positionwithin the level. (b) The notationidentifyingthepositionof theparentto page(2,5). (c) The notation identifyingthepositionsof the childrenof page(1,2).
16
eachfunction, i, is assignedto a uniquepage in a hierarchyand has a probability of being needed, pi ,
and T+) is the time neededto reachpageq(i), then the expected(average)time that it will take
to navigateto a desiredpage containingany randomly selectedseriesof the functionsis:
Accurately estimating q(‘) requiresdetailedknowledgeabout the interactionbetween computer and human systems. Lee and MacGregor (1985) proposedthe following model. Let c indicatethe time neededfor a userto read, mentally interpret,and categorizeone option on a display page. Let s indicatethe time neededto strike a key to selectan option oncethe userknows which option to select. Let Y indicatethe time neededby the computerto producethe next display. Let mjk indicatethe number of optionsat pageposition (j, k) that the usermust categorizebefore making a choice. Then, assumingthat c, s, and Yare constantacrosspages,the time neededto reachpage
(j,k) is:
where the summationis acrossall the levels that the usermust navigate,and the sum identifies how many hierarchylocationsthe usermust categorizebetweenthe top page and page (j, k) . Lee and MacGregor (1985) consideredthe situationwhere the useraccesseseach function equally often, pi = 1 / v ; eachpage hasthe samenumber of options,m; and the user must go througha constantnumber of pages,n; to reacha function. Then, assumingthat searchingthrough m optionsrequires(on average)categorizing (m + 1) / 2 optionsbefore fmding the desireditem, the expectedaccesstime boils down to
E(T) =
~+r+~(~+‘)
. 2
1
Given this analysis,one can determinewhetherit is betterto have a broad design (with many optionsper page) or a deep design (with many levels in the hierarchy). With all the functionsat the bottom level of the hierarchy,it is easyto seethat one needsonly n=-
lnv lnm
levels in the hierarchy. Substitutingthe right sideof this equationfor n above and settingthe derivativeof E(T) with respectto m equalto zero produces: 17
aEtT) =
1’
s+r+c(m+1)/2
hv
am
1 c =O -+ lnm2
m(lnm)* [
A bit of algebrashowsthatthis means:
m(lnm-1)=1+2(S+r).
c
Lee andMacGregor(1985) showedthat if a designermeasures thetermsc, s, andr, thenthe expectedresponse time canbe minimizedby selectingthe numberof navigationor functionoptions suchasNewton’smethodcanbe perMFDCS page,m, that satisfythe aboveequation.Techniques usedto estimatethe valueof m. For reasonable valuesof c, s, andr, Lee andMacGregorfoundthat m rarelygoesaboveeight. PaapandRoske-Hofstrand (1986) considered a variationon the Lee andMacGregoranalysisby hypothesizing thatthe mannerin whichthe navigationalor functionoptionsare groupedon a displaycouldaffectthe time requiredto selectan optionon a menupage. When navigationor functionoptionson a displaypagearegrouped,the effectivenumberof categorizations for each menupagedecreases. This canreducethe overallselectiondecisiontime or converselyallow a largernumberof optionswhile maintainingthe samedecisiontime. For instance,with c = 0.25, s = 0.5, andY= 0.5 (seconds), Lee andMacGregor’sanalysis,thatdoesnot incorporategrouping, suggests settingm = 8. On the otherhand, PaapandRoske-Hofstrand’analysis s that incorporates the improvedefficienciesdueto groupingoptionsgivesm = 38. Unfortunately,theseanalyticdesignresultsareoftenof tangentialrelevanceto manypractical situations because of currentlimitationsin the designmodels.For example,physicalfactorssuch asthe sizeof soft-keysandbezelbuttonsaswell asdisplaysizeandresolutiontypically limits the maximumnumberof optionselectionsperpage. Additionally,functionsearchstrategies at each pagewill likely vary betweenusersbaseduponorganizationof the contentandpreviousexperience (Vandierendonck, et al.,1988). The line of analysisdiscussed abovealsorestrictsitselfto very specifictypesof hierarchies:onesthatuseall availablekey positionson eachpage(compareto figure2) andwhereall the functionsareon the lowestlevel. Thus,evenoptirnalityfrom Lee and MacGregor’sapproachmay not leadto the bestinformationdisplayoverall. Fisher,et al. (1990) proposed an expandedschemefor optimizingthe searchfor specificfunctionsin an information displaysystemwith a largerclassof hierarchies.Unfortunately,their schemeis still too limited in scopefor mostapplications. Expectedfunctionaccesstime is not the only factorthatcanbe minimized. Roske-Hofstrand and Paap(1986) described a methodof buildinga hierarchicalstructure consistentwith a user’s “cognitivemap”of the contentdatabase.Subjectsratedthe similarityof all pairsof the 64 pagesin 18
a database.They convertedthesesimilarity ratingsinto distancesbetweenpages.These values were then used in an algorithm to solve for a hierarchicalstructureof pageshaving minimum accesstime paths. The resultingstructureimproved performancerelative to an already existing hierarchy. Roske-Hofstrandand Paap (1986) demonstratedthe importanceof consideringa user’s mental model of the relationshipsbetweenfunctions,but it is difficult to designhierarchieswith this techniquebecausea generally acceptableand validatedmeasureof a user’s mental model hasnot beendeveloped. While a requirementto satisfya similarity relationshipbetween functions seems to be a useful constraintfor designinga hierarchyof displays,other measuresof how functions complementeach other (e.g., measureof sequentialuse) could also be formulated into valid design constraintsthat would act to offset or exploit relatedcognitive or userinterfacelimitations or _ advantages.Even if designerscould find a consistentlyaccuratemeasureof cognitive distance betweenpage contentsin a databaseof information displays,it is not clear how one would build an appropriatedisplay page hierarchyto minimize that distance. Seidler and Wickens (1992) showed that cognitive distanceinteractedwith other aspectsof a hierarchicalstructurebesidesapparent differencesand similarities. Thus, designof a hierarchyof display contentmust take multiple constraintsinto account. The method usedby Roske-Hofstrandand Paap (1986) is too limited in scopeto deal with suchadditionalcomplexity. Current stateof MFDCS design The literatureon human-factorsaspectsof MFDCS use and design suggestsseveralconclusions. Accessinginformation from MFDCSs with large databasesof display page content and userselectablefunctions can contributesignificantlyto crew workload. The designof MFDCS display pagecontentsand hierarchiesby industryleadersin avionicsseemsto be most frequently performedby applying general“common sense”guidelinesthat experienceddesignersimplement in an ad hoc fashion. A quantitativemethod of balancingthe previously listed guidelines for MFDCS designcould help designersdevelop MFDCSs that have higher probabilitiesof having high function searchefficiency and would have the potential of reducingMFDCS-associated workloads. Current quantitativedesignmethodsfor information display systemsseem to be inadequate. An investigationinto the designof MFDCS hierarchiesof display pagesor modes and embedded functionsshouldhave at leasttwo principal foci. First, a quantitativemethod of designing a hierarchyof MFDCS display pagesmust be elaboratedthat incorporatesasmany human factor and userinterfaceconstraintsand capabilitiesas possible. Without a quantitativedesign tool, designers of MFDCS page contentsand accesshierarchieswill continueto rely on intuition, luck, inefficient trial-and-errorexperience,and reportsfrom the field regardingoperationalproblems with MFDCSs. Substantialamountsof time and resourcesmay be expendedgeneratingwhat quantitativemethodsmight show to be suboptimalhierarchicalstructuresthat could be problematic for pilots in certainhigh-stresscircumstances(e.g., in-flight emergencies). Moreover, without a quantitativeMFDCS designmethod, resultsfrom relatedhuman factor studieswill have little 19
influencebecause thereis no way to ensurethatthe hierarchyreflectstherelativeimportanceof a quantitative factorsfoundto be relevantto effectiveuseof MFDCSs. The nextsectiondescribes methodthatis capableof generatinga hierarchyof displaypagesanduserfunctionsthatwill be optimalwith respectto designerspecifiedcriteria. The secondelementneededto advancemodel-based methodsfor organizingMFDCS page structures is additionalexperimentalstudyto identifyandquantitatetherelevantcomponents of pilot-MFDCS interactions.FutureMFDCS researchalsoshouldinvestigate rigorouslythe previouslylistedMFDCS designguidelinesto determinethe extentto whichthey adequately describeandproperlyweightcognitivefactorsandimportantaspects of the userinterface.Such studieswill be requiredto identifyrealisticvalues(andvariances)for theparameters in the optimizationequations.The humanfactor-related parameters alsomaybe parameter&dby user characteristics (e.g.,agerange,gender,experiencelevels,education,or useof performance enhancingmedications).Likewise,valuesquantitatingthe characteristics andperformanceof the physicalcomponents of the MFDCS couldbe stratifiedby specificmanufacturers anddisplay systems.
A new quantitativemethodfor ontimizingMFDCS contenthierarchies This sectiondescribeswhat we believeto be a new methodof optimizingthe hierarchyof contentpagesanduserfunctionsfor MFDCSs. First, in orderto quantifythe numeroushumanfactorconstraints that couldbe imposedduring the designof the displaypagestructurefor an MFDCS, define an overallcostfor a given hierarchyas a weightedlinear combinationof an arbitrarynumberof costfunctionsdevelopedto satisfyrelatedcriteria:
i=l
Eachconstraint,i, imposesa cost( Ci ) andweights(hi) eachcostaccordingto its significanceas obtainedby the designerfrom humanfactorexpertsfamiliar with the capabilitiesandlimitations of the aircraftfor whichthe MFDCS will be installed. The following sectiondescribeshow to efficiently calculatea costfor expectedaccesstime. Subsequent sectionsdemonstrate how to selecta hierarchythat minimizesthe costfunction. Costas expectedaccesstime Defining a costfunctionfor optimizingan MFDCS pagehierarchydesignrequiresknowing which of the many physicaland software-related propertiesof an MFDCS canhave significant effectson performanceof requiredin-flight duties. Also, one needsto considerthat some 20
MFDCS pagescan show either functions or option menus, but not both. Other MFDCSs (as in figure 2) can simultaneouslydisplay both functions and option menus. For the following discussion,it is assumedthat the MFDCS is similar to those in figure 2 and portrays functions, which allow data or control inputs by the user, and menus simultaneously(selecting a menu option typically causesa jump to anotherdisplay page). As noted above, a designermay want to minimize the expectedaccesstime acrossall pages,so C, might be: V-l
Cl =
c &Pi i=o
where, as before, the time to reach page (j,k) is:
For a nonhomogeneousprobability distribution, calculating Tik requiresmore effort. To simplify matters, assumethat userssearchthe options on a menu page one at a time, and that the pagesare searchedin the order of their indices. Thus, at page (i - 1,Lk / categorizewhichever pagesbetween (i - I + 1, 1k / m’ ] m) and
(j -
nz’1 ) , a usermust
I + 1, 1k / m’-’ 1) contain a
desiredmenu option. The last page is the option that the user must selectto reach page (i, k ) . (While this is not likely a valid model of how userssearchan MFDCS menu page, the following analysisdoes not depend on the user’s searchmethod, only that the designer can identify the method.) It is easyto check for a function at any of thesepositionsby determining if the page in questionis in the set of function position indices Q. However, if a page is not in Q, its contents may still need to be scannedand interpretedbecauseit could contain a menu choice whose descendantsare function pages. Such a page would have a label that must be categorized. There is a recursivealgorithm that considersthesepossibilities. Define the following function:
I 1
Hjk =
h+m-I
if (.L k) E Q or c Htj+l)h > 0 h=km
otherwise,
which returnsa value of one if page (j, k) is either a function or is a menu selection that eventually reachesa function page. The summation simply checksto see if the children of page (i, k) are function pagesor have children that are function pages. Calculation of the H term
21
worksits way downto the bottomof the hierarchyandthenfiltersbackup to the top in a recursivefashion. 1 ]) The numberof optionsthat mustbe mentallycategorizedat menuposition (*--l,[klm’
is
then: m(j_/)[k/m’J
=
Lc ’
H(j-,+l)h
’
h=Lklm’J m
Althoughthe notationis ratherawkwardto look at, it is a relatively simplematterto write a computerprogramto carryout thesecalculations. With theseformulas,it is possibleto calculatethe expectedaccesstime for any layout of functionson a given hierarchicalstructure.In theory,onecouldconsidereverypossiblelayout of functionsand selectthe onewith the lowestcost. In practice,suchan approachwill rarely work becausethe numberof possiblelayoutstypically will be astronomical.In the following sectionswe discussseveralnumericaltechniquesfor solvingcostminimizationproblems. The hill-climbing techniqueis discussed first andsubsequently simulatedannealingwhich works betterfor costfunctionshavingnumerouslocal minima. Hill-climbing When differentiableequations,from which analyticaloptimizationresultscanbe directly obtained,cannotbe formulated,computerscientistsoften applya numericaltechniquecalledhillclimbingto find a globalmaximumfor largecomplexsystems.After selectingan initial MFDCS pagehierarchy,a designercancalculateits cost C(0) usingthe equationsabove. If the designermodifiesthe hierarchyandcalculatesa new cost C(1)sothat C(1) c C(O), thenthe new hierarchyhasa smallercostandshouldreplacethe olderhierarchy. Iteratingthis processwill eventuallyleadto a hierarchy(or setof hierarchies)for whichthe costcannotbe reducedany further. This approachis calledhill-climbing becauseit is analogousto climbinga hill by moving in whateverdirectionis up relativeto your currentposition. An examplewill demonstrate the procedure.Supposeyou want to distributev=5 functionson the hierarchyframeworkin figure4 to minimize C, . Supposethe probabilityof accessingeach functionis: i+l Pi==>
so that functionswith higherindicesare accessed mostoften. To apply the hill-climbing method, calculatethe costof an initial randomlayoutof the functions.Pick a functioni at randomand randomlypick a pagein the hierarchystructure.Move functioni to that page(and if a different 22
functionis alreadyat that page,havethe functionsswappositions).Recalculatethe costand or staysthe same,revertthe system acceptthe changeif the costdecreases.If the costincreases backto its layoutbeforethe move. Continuethis processuntil the systemstopschanging. Figure5 showsthe effectof the hill-climbingprocedure.Figure5a showsan initial random layoutof functions.The layoutis not optimalandhasa costof C, = 0.587. Figure5b showsthe effectof the first movethat decreased the cost. Function3 movedup a level. This reduces searchtime for that functionwithoutaffectingany otherfunction’s searchtime, therebyreducing costto C, = 0.513. Figures5c-f showthe effectsof subsequent movesleadingto decreases in cost. Figure5f showsthe final hierarchyresultingfrom this procedure.The programstopped after onethousandconsecutive movesfailed to decrease the cost. The fmal hierarchyplacesthe mostprobablefunctionat the top, the nextmostprobablefunctionsat level 1, andthe least probablefunctionat level 2. This is an optimallayoutfor this situation.Figure5g showsthe final hierarchywith non-neededpagesremoved. Costfor relatedfunctions The designof an MFDCS may needto considerfactorsotherthanexpectedaccesstime. For example,guidelinesevenfrom page 13 suggests thatthe designershouldplacerelatedfunctions on the samepageor on adjacentpages(i.e., if not on the samepage,onebutton-press away). The relatedness of two functionsi andi, R, , can be estimatedthroughpilot surveysor by MFDCS designexperts. Define the page-distance, yj, betweentwo functions,i andj, asthe maximumnumberof levelsup onemustgo from eitherfunctionto find a menupagethat is parenttobothfunctions. Page-distance canbe calculatedin the following way. Let q(i) = (1,k) and q(j) = (1+ r,h) with Y 2 0 so that functionj is at the sameor lower level asfunctioni. Thenthe pagedistanceis: flj = r + minu E[O,Z]suchthatl-$-]=I--&]}.
As u stepsup from 0 to I, the calculationon the right stepsup from a child to parentpageand is checksto seeif the pathwaysof the two functions’pageshaveconverged.The page-distance the smallestnumberof levelsup for whichthe two pathwaysconverge.For example,when yY = 1 eitherthe two functionscanbe reachedfrom the sameparentpageor one functioncanbe reachedby a selectionfrom the otherpage. Minimization of the following costterm will put relatedfunctionsascloseaspossible: v-l v-l C, = ~,~,RuH$ . i=Oj=O
23
Figure 5. The development of a hierarchy through hill-climbing. (a) The initial layout of functions producesa high cost. (b) Function 3 has moved up a level. This reducesthe number of stepsneeded to reach the function. (c) Function 0 has moved to the left. This frees a menu label at level 1, and reducescategorization time on the way to functions 1 and 2. (d) Function 2 moves up a level. This reducesthe number of stepsneeded to reach the function. (e) Function 1 moves up a level. This reducesthe number of steps needed to reach the function. (f) Functions 1 and 2 swap positions. This places the more probable function in a position to be categorized first. Further changesdo not reduce cost. (g) The fmal hierarchy with non-neededpages removed. [The following parameters were use: c=O.1,1-o. 1, s=O.2.]
24
To demonstrate how thiscosttermworks,let relatedness betweenfunctionsobeythe following formula: if Ji-jl
