simulation fidelity

te

e manoeu-ging cuesdeveloped.ed. Motions

k gains aretrials withion

Aerospace Science and Technology 9 (2005) 400–408

www.elsevier.com/locate/aesc

Measuring simulation fidelity through an adaptive pilot model

Gareth Padfield∗, Mark White

Flight Simulation Laboratory, Department of Engineering, The University of Liverpool, Brownlow Hill, Liverpool L69 3GH, UK

Received 21 June 2004; accepted 31 March 2005

Available online 23 May 2005

Abstract

The paper presents a new approach to the quantification of simulation fidelity based on an analysis of pilot guidance strategy. Thvre guidance portrait is conceived as the solution to a low-order equivalent system and to properly allow for pilot adaptation to chanand task demands, the model parameters are allowed to vary. Thus the concept of the Adaptive Pilot Model (APM) is proposed andThe theoretical foundation to the concept is developed using the familiar spatial variables in flight control, such as distance and speis then transformed into temporal variables and drawing on the theory ofτ (t)-coupling from visual flow theory (τ (t) is the instantaneoutime to contact) the APM model is transformed into a much simpler algebraic relationship when the pilot maintains constantτ during adeceleration. If we make assumptions about the separation of guidance and stabilisation control strategy, pilot guidance feedbacthen closely related to the frequency and damping of the APM structure. Results are presented from the analysis of simulationpilots flying an acceleration-deceleration manoeuvre that show strong correlation with theτ (t)-based guidance strategy. The interpretatof the theory in terms of simulation fidelity criteria is discussed. 2005 Elsevier SAS. All rights reserved.

Keywords: Flight simulation; Pilot modelling; Simulator validation

n-eir

AR-d-, thebutinteingub-the

ism.-nenesspts

841.

kd inues

craftityal.s forndanand

gsInlop acesdi-

[3]rere-tion

1. Introduction

The level of fidelity of Flight Simulators, or, more geerally Synthetic Training Devices (STD), determines thfitness for purpose and is quantified in documents like JSTD-1H [3] in terms of performance criteria for the indiviual components, e.g. the motion/visual/sound systemsmathematical model. Component fidelity is importantthe standards also require piloted assessment of thegrated system with typical mission sorties flown coverthe training aspects for which the system will be used. Sjective opinion here is important too because it reflectsvalue that an experienced pilot places on the level of realQuantifying overall simulation fidelity is more difficult however, but is equally important because, arguably, compoor sub-system fidelity can only be properly related to fitnfor purpose if connected by measure to the whole. Attem

* Corresponding author. Tel.: (44) 151 794 4800; fax: (44) 151 794 6
E-mail address: [email protected] (G. Padfield).
1270-9638/$ – see front matter 2005 Elsevier SAS. All rights reserved.doi:10.1016/j.ast.2005.03.004

-

t

to quantify overall simulation fidelity within the frameworof handling qualities engineering have been presentea number of forms in recent years. Hess and colleag[7,8,19] have developed an approach based on pilot-airmodelling and introduced the handling qualities sensitivfunction as the basis of a quality metric. McCallum etpropose the use of the ADS-33 [2] performance standardderiving metrics [12]. Within the JSHIP project, Advani aWilkinson [1], and Roscoe and Thompson [18] presentapproach using comparative measures of performancecontrol activity, correlated with handling qualities ratingiven for the same tasks flown in simulation and flight.all these approaches, the philosophy has been to deverational and systematic approach to identifying differenbetween tasks performed in simulation and flight, hencerecting attention to simulation deficiencies. While Ref.is directed at the training community, fidelity criteria aequally applicable to the use of simulation in design,search and development. In these areas, flight simula
can be a primary source of data from which knowledge is

G. Padfield, M. White / Aerospace Science and Technology 9 (2005) 400–408 401

Nomenclature

g gravitational constantKR,KX pilot gain relating pitch attitude command

to range errorKR,KX pilot gain relating pitch attitude command

to range ratek τ coupling parameterR rangeRc range command(= X0)

s Laplace transform variableT manoeuvre timeX distance to goX rate of change of distance to go (velocity)X accelerationXu surge damping derivative

YAθ transfer function relating pitch attitude to rangeYPR transfer function relating range error

to pitch attitude commandYPθ transfer function relating attitude command

to pitch attitudeθ aircraft pitch attitudeθc aircraft pitch attitude commandτθ pitch response time constant (inverse

of pitch bandwidthωθ )ζR, ζX closed loop dampingωR,ωX closed loop frequencyτ(t) time to contactτg τ guideτ rate of change ofτ with time

com

ap-onthevail-troliva-het is

ndara-liedustionsaftt thethe

Thusce

vre.mo-

tialultsnoa-

tionsPMwithRe

ion.el

n arel in

h.ms

ofme

e-100). Intask,g tot theionid-

true,tioned-re-

hintaskft,

caneds tothe

ma-lledowsof

cted

derived, decisions are made and significant resourcesmitted.

This paper presents the initial developments in anproach for quantifying overall simulation fidelity basedan analysis of pilot visual guidance strategy, identifyingcontrol loops utilised, levels of abruptness and the cues aable to support anticipation. The premise is that if the constrategy adopted to perform the same flying task is ‘equlent’ in flight and simulation, then the fidelity is good and ttraining device fit for purpose. The meaning of equivalendeveloped in terms of what we describe as theAdaptive Pi-lot Model (APM) concept, whereby the combined pilot aaircraft is modelled and comparisons made of model pmeters identified from the same curve fitting process appto data from flight and simulation tests. As with previostudies, the research is thus concerned with approximafor describing the behaviour of the combined pilot-aircrsystem. However, in the present work, it is assumed thapilot adapts control strategy during the manoeuvre, withadaptation reflected in the changing model parameters.the changing pilot gains relating to velocity and distancontrol, for example, are tracked through the manoeuThe concept is then extended under the premise thattion control by the pilot follows temporal rather than spaguidance principles, as described in Ref. [15]. The respresented in Ref. [15] indicate that pilots strictly haveneed for velocity or distance information, per se, when mnoeuvring close to a surface. Instead, they use informaabout time to close on surfaces,τ(t), to make judgementabout relative motion and control requirements. The Astructure and temporal guidance approach is illustratedreference to an acceleration-deceleration manoeuvre.sults are shown for several test cases from flight simulat

The theoretical foundations of the Adaptive Pilot Modconcept as applied to the manoeuvres under investigatiodeveloped, followed by a re-interpretation of flight contro
terms ofτ(t) and its derivative. Results are presented from
-

-

flight simulation tests, illustrating the utility of the approacThe topic of simulation fidelity is then discussed in terof open and closed loop criteria, and future directionsthe present research activity are outlined, followed by soConcluding remarks.

2. The adaptive pilot model concept

2.1. Theoretical formulation

A pilot’s task can be divided into three functions, with dscending orders of timescale magnitude; navigation (O(sec)), guidance (O(10 sec)) and stabilisation (O(1 sec)this paper we are essentially interested in the guidancethe manoeuvring around and over obstacles and comina stop in particular areas. We make the assumption thanavigation function is too long term, and the stabilisatfunction too short term to cause interference with the guance strategy. These assumptions will not always beof course. The overlap of control demands for stabilisaand guidance is known to be a source of pilot-inducoscillations [14] and the spare capacity for guidance canduce significantly when the pilot loses his or her way. Witthe framework of the stated assumptions, the guidanceinvolves control of the velocity and position of the aircrarelative to the Earth, in the inertial frame.

The concept of the adaptive pilot model for guidancebe traced back to the work of Heffley [5,6], who examinstopping manoeuvres using low-order equivalent systemrepresent the coupled aircraft-pilot system. Consideringhover-to-hover re-positioning, acceleration-decelerationnoeuvre, aircraft motion can be displayed on a so-caphase-plane portrait of velocity against range. Fig. 1 shexamples of different cases to highlight the generalitythis concept. Results are taken from flight tests condu
on the Bo105 and Bell 412 helicopters, together with simu-

402 G. Padfield, M. White / Aerospace Science and Technology 9 (2005) 400–408

lye atti-

ross

. 1ass

ion.

,nted

.

onsedelrityu-

and;

t-ed

andng

and

dis-

at-dy-and

t tolem

a-

to-

thethe

he

ntu-the. In

Fig. 1. Phase-plane portraits for accel-decel manoeuvres.

Fig. 2. Kinematics of the acceleration-deceleration manoeuvre.

Fig. 3. Closed-loop control of aircraft range.

lation cases with Lynx, Bo105 and UH-60. The relativesimple and similar portraits for these manoeuvres hidcomplex pilot control strategy and associated aircraft atude response, and widely varying aircraft dynamics acthe low speed range.

Heffley recognised that the form of the portraits in Figresemble the free response of a second order (spring-mdamper) system released from an initial displaced conditIf the distance travelled by the aircraft is the rangeR, thedistance to stopX and the total rangeRc, as shown in Fig. 2then the closed-loop pilot-aircraft system can be presein the transfer function form given in Fig. 3.

The pilot initiates the manoeuvre under the commandRc

and concludes when the error(Rc − R) is reduced to zeroIn Fig. 3, θc is the commanded pitch attitude andθ the ac-tual pitch attitude. The linear transfer function formulatiis used for convenience in this description. It is recognithat the non-linear behaviour of the APM, i.e. pilot modparameters varying with the motion, means that the lineaassumption breaks down. A non-linear time-domain form
lation is used at this stage. The pilot transfer functionYPR
-

is assumed to take the form of a lead, with proportionaldifferential gains (KR andKR) on range error and velocity

YPR = −KR − KRs. (1)

The transfer functionYPθ represents the pilot-aircraft, shorterm pitch dynamics (stabilisation function) and is assumto take the form of a first order lag with time constantτθ

(bandwidthωθ), written as;

YPθ = 1

1+ τθ s. (2)

The aircraft transfer function between range responsepitch attitude is approximated in first order form, includithe drag derivativeXu;

YAθ = −g

s(s − Xu). (3)

The open loop transfer function between range errorrange is then given by;

Y = YPRYPθYAθ = − g

s(s − Xu)

1

(1+ τθ s)(KR + KRs). (4)

The dynamics of the free response of the system to aplaced initial range are given by the equation 1+ Y = 0, or;

s3

ωθ

+(

1− Xu

ωθ

)s2 + (gKR − Xu)s + gKR = 0. (5)

Applying the further approximation that the closed looptitude dynamics are much faster than the translationalnamics (i.e. stabilisation much faster than guidance),thatωθ � −Xu, the system reduces to 2nd order form,

s2 + 2ζRωRs + ω2R = 0 (6)

where the pilot gains are related to the natural frequencyωR

and damping ratioζR by the expressions,

KR ≈ 2ζRωR

g, KR ≈ ω2

R

g. (7)

In the continuing analysis and discussion it is convenientransform the system into an equivalent initial value probin the time domain, in terms of the distance to go in the mnoeuvre,X (see Fig. 2), rather than rangeR; thus we write,

d2X

dt2+ 2ζXωX

dX

dt+ ω2

XX = 0, X0(0) = −Rc. (8)

Initially the aircraft is at rest in the hover; the commandRc

is transformed into an initial condition, causing the pilotcommand a pitch down attitude throughKR and so the acceleration phase of the manoeuvre begins. At this stagecontrol inputs are almost open loop so we might expectgain to be relatively low. As the velocity builds up so tmotion is damped through the gainKR , an effect that wemight expect to strengthen for the deceleration phase. Iitively, we might also expect the pilot gain to increase asstopping point is approached and loop closure tightenedRef. [6], Heffley estimates constant values ofKR of 4◦/kt
andKR of 1 deg/ft for a UH-1H performing a quick-stop


de-ant,itchthinandr tohes atid-ctivrlyticise

the, re-s.

is-ualara-eptintsoc-isesan-e,

t isair-b toflowni-oeslos-ing’of

icalnch,de-

inand

ofm-rnceuvreegyrat-

-

t-

t theaten-

uting

eu-

forbe-

from 40 kts; peak nose up attitude during the aggressiveceleration was 40◦. These gains correspond to a constnatural frequency and damping ratio of 0.8 rad/sec and 0.7respectively. Heffley goes on to argue the need for a pattitude bandwidth(ωθ ) of about 2.5 rad/sec to ensure thapilots utilising this level of aggressiveness can do so witLevel 1 handling qualities (i.e. a separation of guidancestabilisation frequencies). This work was conducted priothe publication of ADS-33, Ref. [2] that eventually set tpitch bandwidth requirement for hover/low-speed task2 rad/sec. By inferring stabilisation requirements from guance requirements, in a sense we are able to set a protemargin against adverse aircraft-pilot couplings. It is cleaimportant that a flight simulator gives the pilot a realissensation, providing realistic cues, in this regard, otherwaircraft designs or training outputs could be flawed. Innext section we continue the theoretical developmentscasting the guidance cues from spatial to temporal form

3. τ -coupling guidance strategy

The re-formulation of the motion model in terms of dtance to go in Eq. (8) facilitates an examination of visguidance strategy through the direct visual perception pmeters in the optical flow. Gibson [4] introduced the concof optical flow as the way in which patterns change or pomove on the surfaces over and around which motion iscurring. The perception system that picks up and organthese ‘cues’ has evolved to be robust and efficient in theimal world as a key function in the survival game. Likewisan important requirement for pilots to maintain safe flighthat they are able to predict the future trajectory of theircraft far enough ahead that they can stop, turn or climavoid a hazard; the pilot needs to be able to see opticalwell into the future. In Ref. [9], Lee suggested that an amal’s ability to determine the time to close on an object dnot depend on knowledge of the size of the object, the cing speed or distance. Lee hypothesised that the ‘loomof the object, or the ratio of its size to the rate of growthits image on the retina, is actually the fundamental optvariable used in nature. As for a bird approaching a brafor the pilot in the accel-decel manoeuvre the looming isfined in terms of the instantaneous time to contactτ(t), as;

τ(t) = X

X. (9)

The time to contact information can readily be scaledterms of eye-heights, and using a combination of surfaceobjectτ(t)’s, afford animals (and pilots) with knowledgethe height of the surrounding terrain with respect to theselves. In Ref. [15],τ(t) theory was applied to helicoptemanoeuvring to gain a better understanding of guidastrategies. Initially the deceleration phase of the manoewas examined in isolation to model the guidance stratduring stopping and, in particular, to establish if the st
egy aligned with evidence in nature that birds come to a stop
e

Fig. 4. Velocity and range for Lynx flying an accel-decel (Ref. [15]).

Fig. 5. Correlation ofτ(t) with time for the stopping phase in Fig. 4.

while maintaining a constant rate of change ofτ(t) [10]. Therate of change ofτ(t) with time can be obtained by differentiating Eq. (9),

τ = 1− XX

X2. (10)

With X < 0 (see Fig. 2), thenτ > 1 corresponds to acceleraing flight, τ = 1 corresponds to constant velocity andτ < 1corresponds to a deceleration. It can be shown [15] thaaircraft will come to a hard stop (deceleration maximum lin the manoeuvre) ifτ > 0.5 or a soft stop (deceleratiomaximum early in manoeuvre) ifτ < 0.5. A constant deceleration throughout the manoeuvre implies thatτ = 0.5. Theextent to which pilots hold a constantτ during the stoppingphase of the accel-decel can be established by compthe correlation betweenτ(t) and time. Fig. 4, taken fromRef. [15], shows results for Lynx across the whole manovre; a peak pitch attitude of about 15◦ occurred when thevelocity had reduced to about 20 kts in the deceleration.

The corresponding correlation fit is shown in Fig. 5the final 11 seconds of the manoeuvre. For consistency
tween runs, the initial and final 10% of the data were re-


endverat

ting

ro-enible

s

osegythe

in-ustent

eon-filearchtegym aers,on

ni-

ourringdinailot

ntal

tanth abees-he

as--od, the

ole

t-a-

by

moved from the fit (i.e. the manoeuvre was taken towhen the velocity reached 10% of the peak value). Othis range, the correlation coefficient is remarkably highR2 = 0.983 and the fit coefficient orτ = 0.505. The pilot istrackingτ very closely.

In terms of the adaptive pilot model, the constanτguidance strategy has a particular significance. MultiplyEq. (8) byX/X2, we can write,

XX

X2= 1− τ = −2ζX(ωXτ) − (ωXτ)2. (11)

With τ constant, Eq. (11) implies that the product

ωXτ = const. (12)

or that ωX, and hence the pilot gain, are inversely pportional to the time to stop. For the limiting case whτ = 0.5 (constant deceleration), real solutions are posswhenζX � 0.707. WhenζX � 0.707, we have,

ωXτ = 0.707 (13)

so that whenτ is 4 seconds (see Fig. 5),ωX = 0.18 rad/secand whenτ is 1 second,ωX = 0.707 rad/sec. These valuecorrespond to pilot feedback gains of,

τ = 4 seconds, KX = 0.06◦/ft, KX = 0.8◦/kt;τ = 1 second, KX = 0.9◦/ft, KX = 3.0◦/kt.

The values close to the stopping point are similar to thderived by Heffley [6], but there a constant pilot stratewas assumed throughout the manoeuvre. Very close tostopping point, the value of pilot gain cannot increasedefinitely of course, and a different guidance strategy mswitch in. This region is outside the scope of the presanalysis.

Further evidence that pilots adoptτ -based guidancstrategies can be found in Ref. [13]. Flight tests were cducted at NASA to derive the optimum deceleration profor helicopters approaching a landing pad. The resewas conducted to establish the preferred guidance straadopted by pilots for use in director-based displays. Frowide range of tests conducted using 3 different helicoptthe deceleration profile was found to fit the curve basedthe function,

X = kX2

Xn. (14)

Re-arranging terms and substituting forτ , this relationshipcan be written in the form,

τ = 1− kX1−n (15)

wherek andn are parameters that vary as a function of itial range and airspeed. Withn = 1, a constantτ strategyis adopted, but the data from Ref. [11] predictedn to varybetween 1.2 and 1.7. Clearly pilots do not always favthe τ constant strategy and there is a suggestion that, dudecelerating, descending approaches, the need for coortion between horizontal and vertical motion leads the p
to adopt a differentτ -based, strategy.
-

In the development of generalτ -theory, Lee [11] hasrecognised this in the concept ofτ -coupling. Quoting fromRef. [15], . . . “General tau theory posits that the closure ofany type of gap, using any form of sensory input, is guidedby sensing and constantly adjusting the tau of the gap. Thetheory shows, for example, that information solely about τx

is sufficient to enable the gap X to be closed in a controlledmanner, as when making a gentle landing”. In the case of theaccel-decel manoeuvre the pilot effectively initiates a memodel of the manoeuvre, described as an intrinsicτ -guide,and locks onto this throughout the manoeuvre. The consτ strategy can be shown to result from the coupling witconstant velocityτ -guide. The whole accel-decel cannotflown like this however, but it can be shown that the necsary guiding motion is one with constant acceleration. Tconstant acceleration guide has the form Ref. [15],

τg = 1

2(t − T 2

t) (16)

and

τg = 1

2

(1+

(T

t

)2). (17)

Taking the same manoeuvre shown in Fig. 4 andsuming the relationship,τ = kτg , the fit with the constantacceleration guide is shown in Fig. 6. The fit is now goover the whole manoeuvre and, during the final stagesconstant velocity and acceleration guides converge.

The quadratic relationship in Eq. (11) holds for the whmanoeuvre, with the general solution given by,

ωXτ = −ζ ±√

ζ 2 − (1− τ). (18)

In the very initial stages of the manoeuvre, whent � T ,Eq. (18) can be written in the approximate form,

ωX ≈ −√

τ

τ≈ 1

T

√2

k. (19)

According to the APM model, the initial closed loop naural frequency is therefore inversely proportional to the mnoeuvre time (one might intuitively expect this) scaled

Fig. 6. Correlation ofτ(t) with τ g for the whole accel-decel in Fig. 4.


c-

heler-eainleraer ispa-ary

iedreanduaren.as-ingbou

0 ftcon

ghte

de-tem-

s.ith)in

ion

the coupling coefficientk. For the case shown in Fig. 6, acording to Eq. (19), the initial value ofωX is then about0.13 rad/sec, or very similar to the value predicted by tconstantτ guidance model at the beginning of the deceation phase(0.12= 0.707/τ). The results suggest that thpilot may adopt a strategy that keeps the positional gconstant (frequency remains constant) during the accetion phase and then stiffens to a maximum as the hovapproached. We now examine the applicability of the stial and temporal APM model approaches in a preliminanalysis of simulation test data.

4. Preliminary results from flight simulation tests

Fig. 7 shows the first results of the APM model applto Lynx piloted simulation data shown in Fig. 4. The figushows the velocity profile and the estimated dampingsfrequencies using 2-second data windows and a least sqfit process. The longitudinal cyclic history is also showA fairly constant frequency is accompanied by an increing damping during the acceleration phase. At the beginnof the deceleration phase, the frequency has settled to a0.2 rad/sec (cf. with 0.13 rad/sec from Eq. (9)), with thedamping staying constant at about 0.75 until the final 5of the manoeuvre. The results are therefore reasonablysistent with the simple theoretical predictions.

Piloted tests have also been flown on the Liverpool flisimulator, shown in Fig. 8. This facility is described in somdetail in Ref. [17] and includes 6 axes of motion, 5 outsiworld visual channels and an electric control loader sys– all programmable. The FLIGHTLAB modelling and simulation package is used to build, analyse and run model

As part of the APM research, tests have been flown wBo105 and the FLIGHTLAB generic rotorcraft (UH-60 likemodels. The APM results for the UH-60 are presentedFigs. 9 and 10, the latter showing the correlation of motτx(t) with guideτg(t) over the whole accel-decel.

Fig. 7. Adaptive Pilot Model applied to Lynx flying accel-decel manoeuvre.

-

s

t

-

Fig. 8. The Liverpool flight simulator.

Fig. 9. Adaptive Pilot Model applied to UH-60 flying accel-decel manoeu-vre.

Fig. 10. Correlation ofτ x and constant accelerationτ g for UH-60.


eu-

st,citytantcyare

stop

e is

11s asaseve-ainssllsthe

eas-

ex-ing

anyten-

eennceordis

s as

s-atedancepo-y ofen-Inredentula-elmpleownthero-PMpted)y.

llingtion

m’andch.dels isionearsce

[3]shedbe-ctiv-nper

ionet-

B

Fig. 11. Adaptive Pilot Model applied to Bo-15 flying accel-decel manovre.

Fig. 12. Correlation ofτ x and constant accelerationτ g for Bo-105.

The accel-decel is of longer duration than the Lynx tewith a distance of 1100 ft covered and a maximum veloof about 60 ft/sec reached and held approximately consfor a significant portion of the manoeuvre. Similar frequenand damping variations, compared with the Lynx results,predicted across the manoeuvre. Pilot gains close to theping point rise to approximately 2◦/kt and 0.3◦/ft. The timeto stop correlation with the constant acceleration guidagain strong with a coupling coefficient of about 0.46.

The equivalent Bo105 results are shown in Figs.and 12. The frequency and hence positional gain risethe manoeuvre is completed but here the damping increabove critical, corresponding to the pilot increasing thelocity feedback as the stopping point is reached. Pilot grise to about 3◦/kt and 0.5◦/ft. A curious feature occurabout 200 ft from the stopping point when the pilot puback rapidly and during this phase of the decelerationdamping and frequency reduce momentarily, before incr
ing again during the final stages.
-

s

The time to stop correlation is again high but a closeamination of Fig. 12 reveals that the fit close to the stopppoint is poor.

The results shown in Figs. 9–12 are preliminary andobservations made or conclusions drawn are reported astative at the time of writing. Several other test runs have banalysed that suggest variations in pilot gains and guidastrategy throughout the manoeuvre that do not fully accwith the simple APM, and a more thorough investigationunderway to shed light on the pilot adaptation proceswell as the parameter estimation process adopted.

5. Simulation fidelity – a discussion

The value of the APM approach in simulation fidelity asessment will be measured by the sensitivity of the estimclosed-loop system parameters to changes in pilot guidstrategy, brought about by changes in simulation comnent characteristics, e.g. model accuracy or the qualitthe simulation visual and vestibular motion cues. This ssitivity must also correlate with pilot opinion of course.the continuing investigations these aspects will be exploboth in the context of changes to the simulation environmand in the context of direct comparisons between simtion and flight. It will be important to calibrate the modfor changes in task demands on the one hand, for exathe level of pilot aggressiveness used, and also for knchanges in simulation component fidelity, for example indetails of the rotor modelling or the visual cues. To bebust, the method should feature systematic changes in Amodel parameters (and hence the guidance strategy adoderived from systematic changes in the simulation fidelit

The interpretation of motion control in terms ofτ(t) inthis paper has also enabled a simpler, more direct modescheme – mirroring the direct process of visual perceppresent in the natural world. Modelling throughτ(t) cou-pling appears to offer the potential for achieving ‘optimuharmony between the different motion cueing systemsthis aspect will also be explored in the continuing resear

The question of how accurate a mathematical moneeds to be to satisfy different simulation requirementto some extent still an open question. In a series of ‘ActGroups’ GARTEUR has addressed this topic over the ywith a particular emphasis on modelling for performanand handling qualities prediction (e.g. Ref. [16]). Ref.is now a published standard but there has been no publianalysis, to the authors’ knowledge, of the relationshiptween the performance measures and fidelity. Such an aity forms part of the work of the current GARTEUR ActioGroup HC-AG12, and the research reported in this paforms an element of that work.

Our research continues with a focus on flight-simulatcomparisons and the development of useful fidelity mrics from the APM model structure. The FLIGHTLA
Bo105 developed at Liverpool is being compared with flight


test

lineit-e to

hecesolim-

timeidesicalandn-

force

romor-

henell

gy.les

op,tccel-nal

en-uvreov-me-edla-on

s tol-ndlitythe

theare

ngata.deellnkss

est

-raft

ecifi-S

u-

in,

ver,9-

li-Heli-

ator

in1)

Fig. 13. Response to longitudinal cyclic in hover; comparison of flightand FLIGHTLAB simulation – Bo105.

test data provided by the DLR Braunschweig. A baseFLIGHTLAB model is close to release at the time of wring and a typical comparison, showing the pitch responslongitudinal cyclic doublet in hover, is shown in Fig. 13.

The pitch response match falls outside the±10% errorband required in Ref. [3] but this is not untypical of tblade-element modelling standard used. The usual proin the continuing refinement of fidelity is to include contrgearing and other non-physical parametric changes toprove the match. The refinement process is often veryconsuming and there are no formalised best practice guavailable. On a more positive note, the better the physmodel, the less non-physical corrections will be requiredthe pursuit of this level of modelling fidelity has to be a cotinuing priority in the simulation community.

6. Concluding remarks

This paper has outlined the theoretical foundationsthe adaptive pilot model concept for flight motion guidanand presented the first results from application to data fpiloted simulation trials. It has been shown that a secondder system approximation of the APM can be used wthe timescales for flight guidance and stabilisation are w
separated. The system frequency and damping ratio are the
s

directly related to pilot gains in the feedback control strateTransforming the system from spatial to temporal variabresults in a first order differential equation in the time to stτ(t). It then follows that when the pilot follows a constanτguidance strategy during the deceleration phase of an adecel manoeuvre, the natural frequency (and pilot positiogain) varies inversely withτ(t), a strategy which clearlybreaks down very close to the stopping position. A more geral guidance approach to the whole accel-decel manoeis described where the pilot locks onto a motion guide ming with constant acceleration. Initial results derived fropiloted simulation data confirm the principles of motion dscribed by the APM structure. The APM is being developin the current research at Liverpool for application to simution fidelity assessment. Criteria will be developed basedthe sensitivity of the pilot gains used in closed-loop tasksimulation fidelity. The continuing work will exam the utiity of the APM in detecting fidelity changes from visual avestibular motion cues and also simulation model fidebased on a Bo105 helicopter, with test data provided byDLR Braunschweig.

Acknowledgements

The research reported in this paper is funded byUK EPSRC, through grant GR/R02603/01. The authorsgrateful to the DLR Braunschweig, particularly Dr Wolfgavon Grunhagen, for making available Bo105 flight test dThe Lynx simulation data published in Ref. [15], was maavailable by QinetiQ, thanks to Malcolm Charlton. The B412 data was made available by the NRC Ottawa, thato Bill Gubbels. The test pilot for the Liverpool simulationwith the UH-60 and Bo105 was Andy Berryman, ex-RN tpilot.

References

[1] S.K. Advani, C.H. Wilkinson, Dynamic interface modelling and simulation – a unique challenge, The Challenge of Realistic RotorcSimulation, RAeS, London, November 2001.

[2] Anon., Aeronautical Design Standard-33E-PRF, Performance Spcation, Handling Qualities Requirements for Military Rotorcraft, UArmy AMCOM, Redstone, Alabama, March 21, 2000.

[3] Anon., JAR-STD 1H – Helicopter flight simulators, Joint Aviation Athorities, JAA, April 2001.

[4] J.J. Gibson, The Perception of the Visual World, Houghton MifflBoston, 1950.

[5] R.K. Heffley, A model for manual decelerating approaches to hoin: Proc. 15th Annual Conference on Manual Control, AFDL-TR-73134, November 1979, pp. 545–554.

[6] R.K. Heffley, A pilot-in-the-loop analysis of several kinds of hecopter acceleration-deceleration manoeuvres, NASA CP 2216,copter Handling Qualities, 1982.

[7] R.A. Hess, T. Malsbury, Closed loop assessment of flight simulfidelity, J. Guidance Control Dynam. 14 (1) (1991) 191–197.

[8] R.A. Hess, W. Siwakosit, Assessment of flight simulator fidelitymulti-axis tasks including visual cue quality, J. Aircraft 38 (4) (200

n 607–614.


s.

on-80

. 10

im-tion,

ual

ley,naly-ean

ow

s),

d-rafte-m,

htoc.

llingHA

er-

fi-00)

[9] D.N. Lee, The optic flow-field: the foundation of vision, Phil. TranRoy. Soc. London B 290 (1980) 169–179.

[10] D.N. Lee, M.N.O. Davies, P.R. Green, F.R. van der Weel, Visual ctrol of velocity of approach by pigeons when landing, J. Exp. Biol. 1(1993) 85–104.

[11] D.N. Lee, Guiding movement by coupling taus, Ecological Psychol(1998) 221–250.

[12] A.T. McCallum, M.T. Charlton, Structured approach to helicopter sulator acceptance, The Challenge of Realistic Rotorcraft SimulaRAeS, London, November 2001.

[13] G.C. Moen, D.J. DiCarlo, K.R. Yenni, A parametric analysis of visapproaches for helicopters, NASA TN D-8275, December 1976.

[14] G.D. Padfield, J.P. Jones, M.T. Charlton, S.E. Howell, R. BradWhere does the workload go when pilot attack manoeuvres? An asis from flying qualities theory and experiment, in: 20th EuropRotorcraft Forum, Amsterdam, October 1994.

[15] G.D. Padfield, D. Lee, R. Bradley, How do helicopter pilots kn

when to stop, turn or pull up? (Developing guidelines for vision aidJ. Amer. Helicopter Soc. 48 (2) (2003).

[16] G.D. Padfield, A.T. McCallum, H. Haverdings, A.-M. Dequin, D. Hadon, K. Kampa, P.-M. Basset, W. von Grünhagen, Predicting rotorcflying qualities through simulation modelling; a review of key rsults from GARTEUR AG06, in: 22nd European Rotorcraft ForuBrighton, England, September 1996.

[17] G.D. Padfield, M.D. White, Flight simulation in academia; Helifligin its first year of operation, Aeronautical J. Roy. Aeronautical S(September 2003).

[18] M.F. Roscoe, J.H. Thompson, JSHIP’s dynamic interface modeand simulation systems: a simulation of the UH-60A helicopter/Lshipboard environment task, in: Proc. 59th Annual Forum of the Amican Helicopter Society, Phoenix, AZ, May 2003.

[19] J.A. Schroeder, W.W.Y. Chung, R.A. Hess, Evaluation of motiondelity criterion with visual scene changes, J. Aircraft 37 (4) (20580–587.

simulation fidelity

Documents