Finite Element Simulation of a Full-Scale Crash Test of aComposite Helicopter

Edwin L. Fasanella, Karen E. Jackson, and Karen H. LyleUS Army Research Laboratory, Vehicle Technology Directorate

e.l.fasanella@larc.nasa.gov , k.e.jackson@larc.nasa.gov , k.h.lyle@larc.nasa.gov Hampton, VA

Abstract

A finite element model of the SikorskyAdvanced Composite Airframe Program (ACAP)helicopter was developed using the nonlinear,explicit transient dynamic code, MSC.Dytran.Analytical predictions were correlated with ex-perimental data obtained from a full-scale crashtest of the Sikorsky ACAP helicopter flight testarticle that was conducted at the Impact DynamicsResearch Facility of NASA Langley ResearchCenter. The helicopter was impacted at 38-ft/svertical and 32.5-ft/s forward velocity with an atti-tude of 6.25° pitch (nose up) and 3.5°-left roll.The objective of the crash simulation was toevaluate the capabilities of a commercially avail-able transient dynamic code in predicting the re-sponse of a composite airframe subjected to im-pact loading. The model was developed from anexisting MSC.Nastran modal-vibration model ofthe helicopter. Considerable modifications weremade in converting the original modal-vibrationmodel to a model for crash simulation. Followingconversion of the model, a two-stage modelingapproach was used to generate analytical predic-tions. Due to the relatively long pulse duration, arigid structural model containing a fairly complexlanding gear model was executed from initial con-tact through landing gear stroke. Prior to fuselagecontact, the nodal displacements and velocitieswere output to a file. Then, a flexible structuralmodel was executed with the nodal displace-ments and velocities used as initial conditions.This paper describes the development of the fi-nite element crash model, the two-stage modelingapproach, and the correlation of the analyticalpredictions with the experimental data from thefull-scale crash test of the ACAP helicopter.

________________________Presented at the American Helicopter Society 56th An-nual Forum, Virginia Beach, Virginia, May 2-4, 2000.This paper is a work of the U. S. Government and istherefore in the public domain.

Introduction

An important aspect of crashworthinessresearch is the demonstration and validation ofanalytical/computational tools for accurate simula-tion of airframe structural response to crash im-pacts. The “validation of numerical simulations”was identified as one of five key technology short-falls during the Workshop on ComputationalMethods for Crashworthiness [1], which was heldat NASA Langley Research Center in 1992.Crash simulation codes can be used during theairframe design phase to certify seats and aircraftto dynamic crash loads, to predict seat and occu-pant response to impact with the probability ofinjury, and to evaluate numerous crash scenariosnot economically feasible with full-scale crashtesting.

The US Army has been active in support-ing the development and utilization of crash mod-eling and simulation codes for many decades.More than 25 years ago, the US Army sponsoredthe initial development of a kinematic crash analy-sis code, KRASH [2], by the Lockheed-CaliforniaCompany. Kinematic codes employ a semi-empirical modeling approach using lumped-masses, beams, and nonlinear springs to repre-sent the airframe structure. These codes relyheavily on test data for definition of spring proper-ties to characterize the crushing behavior of thesubfloor and other structural components. Goodcorrelation between the analytical and experimen-tal data is usually obtained for global parameters,such as engine or landing gear response. How-ever, these codes would be unable to predict lo-calized responses, e.g. the stress level in an air-frame component at a particular time during acrash event.

Currently, a new generation of crashanalysis codes have been developed that will ac-curately simulate the nonlinear, transient dynamicresponse of airframe structures. These finite ele-ment codes, such as LS-DYNA [3], MSC.Dytran[4], and PAM-CRASH [5], use an explicit solver

that eliminates the need to repetitively decom-pose large global stiffness matrices as is requiredfor implicit codes. Explicit codes require an ex-tremely small time step, typically less than a micro-second, whose duration is controlled by thesmallest element in the model. Thus, impactsimulations having a pulse duration on the orderof 30-40 milliseconds can require several CPUhours to solve on an engineering workstation.Consequently, efficient beam, shell, and solidelements are needed to achieve quick run timesfor very large models.

The new codes are capable of modelingnonlinear geometric behavior including largestructural deformations. In addition, these codesare very effective in modeling materials, such asmetals, that deform plastically and that have wellknown failure mechanisms. However, the use oflight weight, high strength composite materials foraircraft construction brings with it difficulties inmodeling material response and failure behavior.Structural composite materials can exhibit a widerange of material responses from linear elastic tocompletely nonlinear anisotropic behavior, de-pending on the individual fiber and matrix proper-ties and laminate stacking sequence. Also, lami-nated composite materials exhibit a wide variety offailure modes including matrix cracking, fiber fail-ure, and delamination that can occur singly or incombination. These failure modes can changedepending on the type and rate of loading. Ingeneral, the initial failure event in a single ply of acomposite laminate does not produce cata-strophic failure. Consequently, the capability tomodel the progressive failure of composite mate-rials, from initial damage to ultimate failure, isneeded. With the increased application of com-posite materials in the construction of advancedaircraft and rotorcraft, it is important to build confi-dence in the computational capabilities of thesecodes through analytical/experimental validation.

In 1996, the US Army initiated a four-phase research program to evaluate the capabili-ties of commercial crash simulation codes formodeling the impact response of a compositehelicopter. As part of this program, a finite ele-ment crash model of the Sikorsky AdvancedComposite Airframe Program (ACAP) helicopter[6, 7] was developed. In this effort, an existingMSC.Nastran [8] modal-vibration model of thehelicopter was converted to an MSC.Dytran modelfor the crash simulation. In 1999, a full-scale crashtest of an ACAP helicopter was performed at theImpact Dynamics Research Facility (IDRF) [9] ofNASA Langley Research Center to generate ex-

perimental data for correlation with the simulation.The present paper will describe: (1) the develop-ment of the helicopter crash model, (2) modifica-tions made to the crash model to better representthe test article, (3) a summary of the experimentalresults from the full-scale crash test of the ACAPhelicopter, and (4) the validation of the crashsimulation through analytical/experimental correla-tion.

Finite Element Model Development

Description of MSC.Dytran Finite Element Code

The commercial code, MSC.Dytran, wasused to perform the crash simulation of the ACAPhelicopter. MSC.Dytran is a three-dimensionalfinite element code for simulating highly nonlineartransient response of solids, structures, and flu-ids. The MSC.Patran [10] pre- and post-processing code was used with the MSC.Dytran“Preference” to build the finite element crashmodel and to post-process the results.

Description of the Sikorsky Modal-Vibration Model

An MSC.Nastran model of the ACAP heli-copter that was originally developed for correlationwith modal vibration data [11] was obtained fromSikorsky Aircraft. The model, shown in Figure 1,had approximately 5000 nodes, 9,500 elements,219 material models including many differentcomposite materials, and over 700 different prop-erty cards. The elements included 5,453 shellelements; 1,956 beam elements; 1,770 rod ele-ments; and 372 concentrated masses. Becausethis model was originally used for modal analysis,extensive modifications were required to convertit for a crash analysis.

Figure 1. Sikorsky Modal-vibration model of theACAP helicopter.

Conversion of the modal-vibration model to acrash model

The initial work in converting the modal-vibration model to an MSC.Dytran input deck con-centrated on combining elements, deleting un-necessary elements, and rediscretizing themodel. The original linear elastic material propertycards had to be modified to include yield stress,strain hardening, density, and maximum strain tofailure, as appropriate. The tail cone was greatlysimplified, and the stabilizer and rudder were bothremoved. A transition mesh was developed toconnect the coarse mesh of the tail cone to thefiner mesh of the fuselage cabin. Many of theoriginal triangular elements were combined andconverted to quadrilateral elements. Triangularelements are not recommended for models thatundergo large deformations as they are typicallytoo stiff.

Most of the original concentrated massesin the modal-vibration model were removed.Other concentrated masses were modified oradded to represent actual lumped masses on thehelicopter including the rotorcraft transmission,engines, anthropomorphic dummies, seats, fuel,instrumentation boxes, batteries, and cameras,etc.

The elements in the original modal-vibration model representing the main and noselanding gear were removed. An external usersubroutine was developed to calculate the mainlanding gear forces as a function of velocity andstroking distance including both the oleo-pneumatic and the crushable honeycomb stages.The original crashworthy nose gear of the ACAPhelicopter had been removed and replaced with anon-crashworthy standard nose gear. Modifica-tions were required to make the existing nosegear more crashworthy. The hydraulic fluid wasdrained, and a thin-walled aluminum tube with ahoneycomb core was inserted inside the gear toprovide some energy absorption. The modifiednose gear was modeled as a spring having a con-stant spring force of 8,000-lb. to represent thecrush strength of the honeycomb-filled aluminumtube. A more detailed description of the landinggear modeling approach is provided in the nextsection.

Since the original modal-vibration modelwas too detailed in some regions and too coarsein other regions, especially near the impact point,some rediscretization of the mesh was made.One of the main energy absorbing devices for the

structure was the crushable lower portion of thekeel beams and subfloor bulkheads. There werefour keel beams beneath the floor, two inner keelbeams and two outer keel beams. The keelbeams and transverse bulkheads beneath thefloor were constructed of two horizontal C-channels, one above the other, with a beaded (orwaffle) web geometry. The upper channel wasconstructed of graphite and the lower 4-inch highbeaded web was constructed of Kevlar. Theouter keel beams were very thick and did notcrush in the test. The lower portion of the innerkeel and bulkhead beams was fabricated of a thin-ner beaded Kevlar construction designed tocrush and absorb energy. In the original model,the crushable Kevlar web was modeled using 4-inch-high shell elements with longitudinal beamelements to represent the flanges of the C-sections. The Kevlar shell elements representingthe keel beam and the bulkheads were rediscre-tized by dividing each original shell element verti-cally into four shell elements. This representation,as shown in Figure 2, allowed crushing to occur[12].

Figure 2. Finite element model of the subfloor ofthe ACAP helicopter.

Following the extensive modifications,the ACAP helicopter crash model executed with atime step of approximately 1.9 microseconds.The model was run many times to track down theelements controlling the time step. These ele-ments were then combined with other elementsto increase the time step to an appropriate dura-tion to allow a reasonable run time for the model.

Description of the crash model

The final crash model of the ACAP heli-copter is shown in Figure 3. The model consistsof 4,128 grid points and 7,346 elements, which

Figure 3. Final crash model of the ACAP helicopter.

include 3,118 beam and rod elements 3,283quadrilateral shell elements, 695 triangular shellelements, and 250 solid elements that representthe impact surface. The number of different mate-rial property cards was reduced from 219 in theMSC.Nastran modal-vibration model to 34 in theMSC.Dytran crash model. In addition, the numberof different property cards for shell elements wassignificantly reduced by combining many of thePCOMP cards. The PCOMP cards are used tospecify the material properties and orientations foreach ply in a laminated composite shell element.Also, the total number of concentrated masses inthe model was reduced from 372 in the modal-vibration model to 98 in the crash model. The 98concentrated masses represent actual lumpedmasses used in the experiment, as indicated inTable1. In the final crash model, the structuralelements weighed 2,838-lb. and the concen-trated masses weighed 5,160-lb., for a totalweight of 7,998-lb.

A flat plate consisting of 250 solid ele-ments with fixed bottom nodes was added to themodel to represent the impact surface. A mastersurface to slave node contact was defined be-tween the flat impact surface and the nodes in thestructural model. The initial conditions includingpitch, roll, yaw, and translational and rotational ve-locities were determined from measurementsmade from photographs taken from the high-speed video and from motion picture analysis ofthe high-speed (400 frames/s) film. The coordi-nate system used in the model was x-axis positive

from nose to tail, z-axis positive up, and y-axis po-sition to the right. The nominal impact conditionsare -38.5 ft/s vertical velocity, -32 ft/s longitudinalvelocity, 9.6 degrees/second angular pitch veloc-ity (rotating backward due to swing), 6.25° pitch(nose up), and 3.5° left roll. For this simulation, itwas more expedient to adjust the position of theimpact surface, rather than the structural model toaccount for the roll and pitch attitude. The posi-tions of the center-of gravity for the helicoptermodel were determined to be x= 203.7-in., y= 0-in., and z= 87-in. Note that these dimensions aregiven in reference to the manufacturer’s body sta-tion (BS), butt-line, and waterline coordinate sys-tem.

Modeling Approach

To perform the simulation, a two stagemodeling approach was employed in which a rigidstructural model of the helicopter was executedduring deformation of the landing gear. Approxi-mately 0.05 seconds before fuselage contact, thex, y, and z- locations of all grid points and the cor-responding nodal velocities in the rigid modelwere output to a file. These initial conditions werethen input as the starting point of the flexiblemodel simulation. The rigid-to-flexible approachwas used to significantly decrease the CPU timerequired to complete the simulation, and becausethe rigid model made the introduction of the pitchangular velocity much easier to input. The devel-opment of the landing gear model, rigid structuralmodel, and flexible structural model are discussedin the following subsections.

Table 1. Summary of the major concentratedmasses in the finite element model.

Concentrated Mass Weight, lb.

Pilot dummy and seat 279.3

Copilot dummy and seat 315.5

Right and left fuel tank(805 lb. each)

1,610.0

Nose gear 52.0

Lifting hard point in the nose 98.0

Right and left main gear(122 lb. each)

244.0

Right and left engine(315 lb. each)

630.0

Main gearbox 832.0

Tail mass 69.0

Main rotor hub mass 352.6

Right and left troops & seats(180 lb. each)

360.0

Landing Gear Model

A landing gear model, which accuratelysimulates the energy absorbed by the gear with-out adding substantial complexity to the model,was developed. For a crash model, the landinggear response can be represented as a spring, asshown in Figure 4(a), where the force is computedin a user-written external subroutine. In general,the “spring” force equation can be dependent onthe relative displacement and the relative velocityof the connected nodes. This flexibility allows auser-defined spring to simulate the effect of thevelocity dependant oleo-pneumatic stage and thecrushable honeycomb stage, which depends onlyon displacement. An assumption is made that thegear is fully deployed, as would be expected in acrash scenario. Alignment of the spring relative tothe aircraft can be a challenging modeling prob-lem. A number of rigid joints, such as sliding, rota-

tional, ball, and universal joints, are currently avail-able as standard capabilities in commercial codes.However, these joints may become unstablewhen large forces such as those experienced bythe landing gear during a severe impact are ap-plied. For this reason, these built-in joints werereplaced in the analysis with a component con-taining several nodes and beam elements overwhich the forces are distributed. Nodal alignmentwas accomplished by creating four contact sur-faces on the outer surface of two perpendicularplates. The beam nodes representing the gearhub and piston were then constrained to remainwithin the intersecting shaded region, as shown inFigure 4(b).

For simplicity of model development, theinitial landing gear models were attached to a sim-ple triangular rigid plate element, which approxi-mated the fuselage. The aircraft mass, center ofgravity, and moments of inertia were explicitlyspecified for the triangular element. Workstationsimulations using this simple model are completedin a few minutes. The predicted nodal accelera-tions, velocities, and displacements at the gearattachment nodes were compared with the corre-sponding experimental data. These comparisonsallowed modifications to be quickly evaluated.Once adequate experimental and analytical cor-relation was achieved, the simple fuselage repre-sentation was replaced by a rigid ACAP fuselagemodel with accurate geometry and mass momentsof inertia about the center-of-gravity.

Rigid Structural Model

The ACAP structural model was maderigid by setting material properties for all elementsto MATRIG, except for those elements formingthe landing gear. As described in the previoussection, landing gear forces were computed usingan external subroutine. This procedure allowedquick runs to be made, of approximately one-hourduration, to refine the correlation of the sequenceof events, and to ensure that the landing gearsubroutine accurately predicted the gear forces.For this model, it was important to ensure that thelocation of the center-of-gravity and moments ofinertia matched the experimental values closely.The rigid model was executed for a duration of 0.2seconds after left gear contact. The archive filesgenerated by the rigid model were post-processed in MSC.Patran with user-

Perpendicular surfaces

Tire

Spring

Fuselage attachment point

Beam elements

Perpendicular surfaces

Landing gear beamsAlignment area

(b) Top view (not to scale)

(a) Isometric view (not to scale)

Figure 4. Diagram of simplified landing gear approach.

written commands to create a file containing nodalvelocities and a file containing nodal positions at0.045 seconds. The nodal position file, whichhad the new GRID locations for each node, wasinput into the flexible model using the INCLUDEstatement. The velocity file was read into theflexible model by an external user-writtenFORTRAN subroutine.

Flexible Structural Model

The model was transitioned from rigid toflexible at 0.045 seconds after left gear impact,which is just after the right landing gear tire makescontact with the impact surface. The flexible struc-tural model was the same as the rigid model ex-cept for the following modifications. The materialcards were non-rigid, the initial GRID coordinateswere the nodal positions of the rigid model at0.045 seconds, and the initial velocity of each gridpoint was read in by the user subroutine from therigid model at 0.045 seconds. Also, during thislatter stage of the analysis, each landing gearforce was modeled using the appropriate constanthoneycomb crushing force, 8,000 pounds for thenose gear and approximately 26,000 pounds foreach of the main gear. The transition occurredwithout incident except for a slight discontinuitythat occurred when the flexible model was initi-ated due to the sudden onset of the landing gearhoneycomb force. Several techniques to mini-

mize this discontinuity (acceleration spike) at thetransition point from rigid to flexible were investi-gated, such as ramping the honeycomb forceover multiple time steps. However, an optimumsolution to this problem was not found.

Experimental Program

A full-scale crash test of the SikorskyACAP helicopter was performed in 1999 to gen-erate experimental data for correlation with thecrash simulation. The development of two ACAPhelicopters was sponsored in the early 1980’s bythe US Army Applied Technology Laboratory.Separate contracts were awarded to Bell Helicop-ter Textron and Sikorsky Aircraft to design, fabri-cate, and test a new helicopter manufactured pri-marily from composite materials as part of the Ad-vanced Composite Airframe Program. The heli-copters were intended to demonstrate the poten-tial of advanced composite materials to saveweight and cost in airframe structures; whileachieving systems compatibility and meeting Armyrequirements for vulnerability, reliability, maintain-ability, and survivability. The Sikorsky ACAP heli-copter was designed as a composite replacementfor the metallic S-76 helicopter. The resultingACAP airframe consisted of 82% composite mate-rials, and the total weight and cost savingsachieved based on the final design were 23% and24%, respectively. A systems approach was used

in designing the helicopter for maximum crashprotection, including energy absorbing landinggear, crushable subfloor structure, and load-limiting seats. However, the primary energy ab-sorbing elements were the landing gear. Thelanding gear were designed to remove 80% of theenergy of a crash performed at the impact condi-tions of 38-ft/s, 10° pitch, and 10° roll as specifiedin Reference [7]. The main landing gear con-tained an aluminum honeycomb tube that woulddissipate kinetic energy during a crash eventthrough stable crushing of up to 18 inches.

In 1987, the Sikorsky ACAP static testarticle was crash tested at the IDRF at NASA Lan-gley Research Center to demonstrate and verifythe crashworthy features of the airframe design[7]. A drop test was performed at 39 ft/s verticalvelocity with a 10-degree pitch and a 10-degreeroll impact attitude. There was no induced pitch orroll rate. Some of the data obtained from this testcould have been used for correlation with theMSC.Dytran crash simulation. However, a moresevere impact condition with both vertical andlongitudinal velocity components was desired forthis correlation. Given that the flight test articlewas available and in good condition, a secondcrash test was performed for the specific purposeof generating experimental data for comparisonwith the MSC.Dytran simulation.

In June of 1999, the second full-scalecrash test of the Sikorsky ACAP helicopter wasconducted at the IDRF. Prior to the drop test,some modifications were made to the airframe tocorrect problems that had occurred in the 1987test [13]. Onboard hardware in the 1999 test in-cluded two different energy-absorbing crew seatsfor the pilot and copilot positions, two energy-absorbing troop seats, three-50th percentile Hy-brid II anthropomorphic dummies, a modified 50th

percentile dummy with a self-contained data ac-quisition system in the co-pilot seat, and two MA-16 inertia reels with triaxial locking mechanisms forpilot and copilot restraint systems. The fuel tankswere filled with water and instrumented with pres-sure transducers to record the hydrodynamicpressure during impact. Finally, an airbag elec-tronic switching unit was mounted to the floor inthe rear of the cabin to indicate the time at ignitionby turning on a lamp. Approximately 90 channelsof data were collected at 10,000 samples per sec-ond using a digital data acquisition system, in-cluding 62 accelerometers, 11 displacementtransducers (string potentio-meters), 5 loads cells(one lumbar and 4 seat restraint), 4 pressuretransducers, 1 radar, and 1 continuity strain gage.A schematic drawing of the helicopter indicatingthe locations of selected accelerometers used inthe analytical/ experimental correlation is shown inFigure 5.

Right gear

Right gear

Right engine

Right & left engine

Left engine

Pilot and copilot

Pilot floor

Copilot floor

Right troop

Left troop

Right and lefttroop floor

BS 255

BS 255

BS 182 floor

BS 182 floor

BS 188 top

BS 188 top

Figure 5. Selected accelerometer locations on the ACAP helicopter.

A double-harness support was used toraise the helicopter to a height of approximately40-ft. at the IDRF. The helicopter was dropped ina pendulum fashion, to achieve the impact condi-tions of 38 ft/s vertical velocity and 32.5 ft/s longi-tudinal velocity, with a +6.25°(nose up) pitch and a3.5° roll (left) onto a rigid impact surface.

All acceleration data from the ACAP crashtest were analyzed and checked for polarity errors,zero-offsets, and noise. Floor acceleration data inthe vertical direction were integrated to obtain thevertical velocity change. The integration also pro-vided a quality check of the data. Those channelsin which the integrated velocity change variedgreatly from the nominal vertical impact velocitywere not used for the correlation with the analysis.

In general, high quality experimental datawas obtained from the crash test. However, thistest was the last one performed using the umbili-cal based data acquisition system at the IDRF. Inaddition to the problems incurred with using a verylong umbilical cable of several hundred feet, morenoise than usual was noted in this test. In severetests such as the ACAP helicopter crash test, it isnot unusual for wires to be cut and for anomalouselectrical transients to be generated in the data.Several acceleration traces were noted to have anumber of high-amplitude transients, and to haveexhibited a zero offset after the test. The chan-nels that showed large zero offset differentialswere not used for comparison with the analysis.

Analytical and Experimental Correlation

Correlation with time sequence of events

An important correlation between theanalytical and experimental data is a comparison ofthe predicted and measured time sequence ofevents. The comparison, shown in Table 2, indi-cates that the simulation correctly predicted thetiming of right gear contact, nose gear contact, tailcone failure, and the peak accelerations of fourdifferent locations on the helicopter. In mostcases, the simulation predicted the event timingwithin seven milliseconds. It should be noted thatevents occurring earlier than 0.045 seconds werepredicted using the rigid model, while events oc-curring after 0.045 seconds were predicted usingthe flexible model.

Table 2. Comparison of the predicted and experi-mental time sequence of events.

Event Simula-tion(sec-onds)

Measuredtest result(seconds)

Left main gearcontact

0.0 0.0

Right main gearcontact

0.016 0.012

Nose gear contact 0.068 0.069Tail breaks 0.075 0.074Time of peak accel . Left troop seat floor 0.110 0.105 Rt. troop seat floor 0.114 0.107 Copilot floor 0.115 0.110 Right engine 0.117 0.116

Comparisons of the predicted helicoptermotion with high-speed film pictures of the actualcrash scenario are shown in Figure 6 for timeequal to 0.0, 0.05, 0.1, and 0.15 seconds afterleft gear contact. The predicted motion of thehelicopter agrees well with the observed crashscenario. In addition, the simulation accuratelypredicts the location and time of tail cone failure.

Correlation between model and structuraldeformation

Several discrete structural failures werenoted to occur during the crash test, includingfailures of the ceiling beams in the fuselage cabin,support beams for the overhead engines andtransmission, and the separation of the tail fromthe fuselage. In general, the crash simulationpredicted the location of these failures. For ex-ample, the undeformed model of the bulkhead atBS 188 is shown in Figure 7 (a). This bulkhead islocated behind the rear of the cabin and beneaththe overhead rotor transmission. The model ofthe bulkhead consists of a ring of shell elementsrepresenting the outer skin, with one row of hori-zontal shell elements representing the floor. Thetwo support beams that form the “V” in the centerof the bulkhead were modeled with beam ele-ments and are illustrated with dark lines. The fail-ure of one of the support beams, as predicted bythe model at time 0.125 seconds, is shown in Fig-ure 7(b). The buckling of the lower portion of thesubfloor is also apparent in the figure. An exam-ple of a support beam failure from a post-test pho-tograph is shown in Figure 7 (c).

Time = 0.0 s

Time = 0.05 s

Time = 0.10 s

Time = 0.15 s

Figure 6. Comparison of predicted and actualcrash scenario.

One of the major goals of this modelingeffort was an accurate prediction of the subfloorcrushing response. Unfortunately, the displace-ment transducers that were used to measure the

dynamic subfloor deformation failed to performduring the test. Consequently, only the post-testfloor deformation was available for comparisonswith predictions. Since no dynamic crush wasmeasured, only a qualitative comparison can begiven. The predicted crush of the lower Kevlarportion of the subfloor is shown in Figures 8 (a)and (b). For clarity, the upper non-crushinggraphite portion of the subfloor is not shown. Theinitiation of crushing at BS 188 at time 0.105 sec-onds, which is shortly after contact of the bottomof the helicopter with the impact surface, is shownin Figure 8 (a). At time 0.125 seconds, all of theinner keel beams exhibit buckling and crushing,as indicated in Figure 8 (b). An examination of thepost-test ACAP fuselage showed that the Kevlarsubfloor had permanent crush of about 0.25-inchat BS 188. The crush increased forward to amaximum of 1.5 inches at BS 143. The perma-nent crush at BS 122 ranged from 0.5 to 1.0inches. A photograph of the subfloor from BS188 through BS 143 is shown in Figure 8 (c) thathighlights the actual post-test deformation.

Rigid Body Model Results

A comparison of predicted and measuredvertical velocities for three locations in the helicop-ter including the top of the right gear, the rightengine location, and the copilot seat floor locationis shown in Figure 9. The predicted responseswere obtained from the rigid structural model, andthe experimental responses were obtained fromintegration of the accelerometer data. The rigidmodel required approximately 1.8 CPU hours tosimulate 0.2 seconds “real time” on a Sun UltraEnterprise 450 computer. The rigid fuselagemodel impacts the contact surface at approxi-mately 0.1 seconds. Thus, the rigid results areonly applicable for the first 0.1 seconds of simula-tion. Since the ACAP impacted with a pitch an-gular velocity, the vertical velocity at a given loca-tion is the sum of the center-of-gravity (cg) velocityplus or minus the rotational component. Sincethe helicopter had a negative pitch rate at impact,the vertical velocity at the copilot cabin floor is ac-tually less than the engine and gear vertical veloc-ity, which is near the cg-location. As can be seenin Figure 9, the model follows the measured mo-tion of the ACAP helicopter quite well.

The gear forces, predicted by the rigidbody simulation for the main and nose gear, areshown in Figure 10. The left gear tire impacts attime zero, the right gear at time 0.16 seconds, andthe nose gear at time 0.68 seconds. These

(a). Model of the bulkhead and the subfloor located at BS 188 prior to impact.

(b). Model showing subfloor crushing and support beam failure at 0.125 seconds.

(c). Overhead mass support structure showing post-test damage.

Figure 7. Model of bulkhead at BS 188 and post-test photo showing failure of support beam.

BeamFailure

Support beams

Floor

LowerSubfloor Region

Support beamFailure

SubfloorCrushing

(a). Model of the lower portion of the subfloor near the fuselage impact point at 0.105 s.

(b). Model of the lower portion of the subfloor region at 0.125 seconds after impact.

(c). Photograph of the ACAP helicopter subfloor between BS 122 and 188.

Figure 8. Deformed plots of the fuselage Kevlar subfloor beams and a post-test photograph showing actualdamage and deformation.

BS 143

BS 188

BS 143BS 188

BS 188BS 143

No crushing

BS 122

CrushedKevlar

BS 122

-500

-400

-300

-200

-100

0

100

200

0 0.05 0.1 0.15

Exp. dataAnalysis

Velocity, in/s

Time, s

(a). Right gear.

-500

-400

-300

-200

-100

0

100

200

300

0 0.05 0.1 0.15

Exp. dataAnalysis

Velocity, in/s

Time, s

(b). Right engine.

-500

-400

-300

-200

-100

0

100

200

0 0.05 0.1 0.15

Exp. (copilotseat floor)

Analysis

Velocity, in/s

Time, s

(c). Copilot seat floor location.

Figure 9. Comparison of predicted rigid modelvelocities with experimental data.

times correlate well with the measured valuesshown in Table 2, which are 0.00, 0.12, and 0.69,respectively.

Flexible Model Results

The flexible model was initiated at 0.045seconds using the grid point position and velocitydata from the rigid model. To run the flexiblemodel for 0.1 seconds real time (from 0.045 to0.145 seconds) required approximately 12 CPUhours on a Sun Ultra Enterprise 450. The flexiblemodel is computationally intense compared withthe rigid model. To conserve resources, a flexiblesimulation such as this one with a time step of ap-proximately one microsecond would normally bestarted a fraction of a millisecond before fuselageimpact. However, to include the tail failure in themodel, which initiates at about 0.025 secondsbefore fuselage impact with the surface, the flexi-ble model had to be started approximately 0.05seconds before fuselage impact.

-120000

-100000

-80000

-60000

-40000

-20000

0

20000

0 0.05 0.1 0.15

Left gearRight gearNose gear

Axial force, lb.

Time, s

Figure 10. Predicted right and left main gear andnose gear axial force time histories.

The flexible model included the effects ofgravity, however friction of the impact surface wasnot included. Friction is difficult to measure ex-perimentally, and would only be significant afterfuselage contact. Since friction was not modeled,no results are shown for the longitudinal or lateraldirections. The correlation between the flexiblemodel z-velocity and the measured vertical veloc-ity for the three locations that were tracked by therigid model (the top of the right main gear, the

-500

-400

-300

-200

-100

0

100

200

0 0.05 0.1 0.15

Exp. (right main gear)Analysis

Velocity, in/s

Time, s(a). Right main gear.

-500

-400

-300

-200

-100

0

100

200

300

0 0.05 0.1 0.15

Exp. (right engine)

Analysis

Velocity, in/s

Time, s(b). Right engine.

-500

-400

-300

-200

-100

0

100

200

300

0 0.05 0.1 0.15

Exp. (copilot seat floor)Analysis

Velocity, in/s

Time, s (c). Copilot seat floor.

Figure 11. Comparison of predicted flexiblemodel velocities with experimental data.

right engine, and the copilot seat floor location)are shown in Figure 11. As would be expected,the flexible model shows vibratory oscillations,even more than measured experimentally. Themagnitude of the oscillations was enhanced dueto the impulse of the landing gear loads being ap-plied at full load at start-up of the flexible simula-tion. The transition from the rigid to the flexiblemodel is not trivial and can cause considerableproblems if not properly administered.

The comparisons of the experimental andpredicted vertical velocity responses of the rightengine, right main landing gear, and copilot floorfor both the rigid and flexible models, shown inFigures 10 and 11, respectively, were performedto validate the two-stage modeling approach. Thiscomparison demonstrated that equivalent resultswere obtained for both the rigid and flexible mod-els for the period before fuselage impact, up to0.1-seconds after left gear contact. The fact thatthe analytical results obtained from both the rigidand flexible models demonstrated the same highlevel of agreement with the experimental data in-dicated that the two-stage modeling approach wassuccessful.

Comparison of Predicted Acceleration TimeHistories with Data

To remove the high-frequency ringingfrom the low-frequency crash pulse, both the ex-perimental data and the predicted accelerationswere filtered using a zero-phase digital 60-Hz low-pass, 3rd order Butterworth filter generated by themathematical code Matlab. To produce zero-phase distortion in time, the algorithm used bothforward and backward filtering.

A major goal of this modeling effort was anaccurate prediction of the floor-level accelerationssince this information is needed for the properdesign of crashworthy seats. The predicted andmeasured vertical acceleration responses for thepilot and copilot seat floor locations are shown inFigures 12 (a) and (b), respectively. The experi-mental data is shown from time zero, correspond-ing to the time of left gear contact. The predictedvalues were obtained from the flexible model,which initiated at 0.045 seconds. The initiationand duration of the two acceleration pulses com-pare quite favorably. The predicted accelerationpeak for the pilot was 115-g, while the experimen-tal value is 90-g. For the copilot, the predictedpeak acceleration also exceeded the experimen-tal, with about 115-g predicted and with 83-gmeasured peak accelerations. The over

-100

-50

0

50

100

150

0 0.05 0.1 0.15

Exp. (pilot seat floor)Analysis

Acceleration, g

Time, s(a). Pilot seat floor location.

-100

-50

0

50

100

150

0 0.05 0.1 0.15

Exp. (copilot seat floor)Analysis

Acceleration, g

Time, s(b). Copilot seat floor location.

Figure 12. Predicted and measured vertical ac-celerations for the pilot and copilot floor locations.

prediction of the peak acceleration for the pilotand copilot suggests that the model is more elas-tic than the actual structure. For simplicity and ro-bustness of the model, the Kevlar subfloor mate-rial properties were represented as elastic-plasticwithout failure. To obtain more accurate results,the Kevlar subfloor model may require more re-fined meshing and a more accurate materialmodel. In addition, since energy is proportional tovelocity squared, the rotational slap-down velocityof the nose portion of the helicopter must be

-40

-20

0

20

40

60

80

0 0.05 0.1 0.15

Exp. (right troopseat floor)Analysis

Acceleration, g

Time, s(a). Right troop seat floor location.

-40

-20

0

20

40

60

80

100

0 0.05 0.1 0.15

Exp. (left troopseat floor)Analysis

Acceleration, g

Time, s(b). Left troop seat floor location.

Figure 13. Predicted and measured vertical accel-erations at the right and left seat floor locations.

modeled accurately to predict the actual accelera-tions more closely.

In contrast to the predicted pilot and co-pilot accelerations, which exceeded the meas-ured values, the predicted peak acceleration forthe right troop seat, shown in Figure 13 (a), wasslightly low at 64-g as compared with the experi-mental value of about 75-g. As shown in Figure13 (b), the predicted peak acceleration of the left

troop seat floor, 51-g, was about 40% lower thanthe measured acceleration, which was 85-g. Theexplanation for this discrepancy was not readilyapparent.

Predicted and experimental accelerationresponses for the floor at BS 182 are plotted inFigure 14. The predicted acceleration pulse isquite similar in shape to the experimental pulsewith a peak acceleration of 54-g compared withthe measured value of 61-g. The predicted peaklags the experimental peak in time by about 0.01seconds. The acceleration time histories for theright and left engine masses are given in Figure15 (a) and (b), respectively. The predicted peakacceleration for the right engine matched the ex-perimental value of 41-g, although the responsewas slightly out of phase in time. The predictedpeak for the left engine is matched with the ex-periment in time, but is slightly lower in magnitude.

-20

0

20

40

60

80

0 0.05 0.1 0.15

Exp. (floor BS182)Analysis

Acceleration, g

Time, sFigure 14. Predicted and measured acceleration

responses of the floor BS182 location.

The predicted and measured vertical ac-celeration responses of the top of the right mainlanding gear, the bulkhead at BS 255 which is justin front of the main gear, and the top of the bulk-head at BS 188, are shown in Figures 16, 17, and18, respectively. The comparisons are favorable,except for the bulkhead acceleration at BS 188.The predicted acceleration response at BS 188 isof shorter duration and higher magnitude than themeasured acceleration. These results indicatethe model was too stiff in this location. A large

number of vertical rod elements were used tomodel the fuselage wall and bulkhead at BS 188in the original modal-vibration model. Similar rodelements were used in the subfloor as well. Forthe crash model, the rod elements in the subfloorwere either removed or replaced with beam ele-ments, since rod elements are too stiff to repre-sent the actual structural crash behavior. It is im-portant to note that the rod elements at BS 188were not removed or modified, which resulted inthe high acceleration shown in Figure 18.

-10

0

10

20

30

40

50

0 0.05 0.1 0.15

Exp. (right engine)Analysis

Acceleration, g

Time, s

(a). Right engine.

-20

-10

0

10

20

30

40

50

0 0.05 0.1 0.15

Exp. (left engine)Analysis

Acceleration, g

Time, s

(b). Left engine.

Figure 15. Predicted and measured accelerationresponses for the right and left engines.

-10

0

10

20

30

40

0 0.05 0.1 0.15

Exp. (top of right main gear)Analysis

Acceleration, g

Time, s

Figure 16. Predicted and measured accelerationresponses of the top of the right main gear.

-10

0

10

20

30

40

50

0 0.05 0.1 0.15

Exp. (bulkhead BS255)Analysis

Acceleration, g

Time, s

Figure 17. Predicted and measured accelerationresponses of the bulkhead BS255 location, in

front of the main landing gear.

-10

0

10

20

30

40

50

0 0.05 0.1 0.15

Exp. (top of bulkhead)

Analysis

Acceleration, g

Time, sFigure 18. Predicted and measured acceleration

responses for the top of bulkhead BS 188.

Concluding Remarks

A finite element model of the SikorskyAdvanced Composite Airframe Program (ACAP)helicopter was developed using the nonlinear,explicit transient dynamic code, MSC.Dytran.Analytical predictions were correlated with ex-perimental data obtained from a full-scale crashtest of the Sikorsky ACAP helicopter flight testarticle, that was conducted at the Impact DynamicsResearch Facility of NASA Langley ResearchCenter in 1999. The helicopter was impacted at38-ft/s vertical and 32.5-ft/s forward velocity withan attitude of 6.25° pitch (nose up) and 3.5°-leftroll at impact.

The objective of the crash simulation wasto evaluate the capabilities of a commercially avail-able transient dynamic code in predicting the re-sponse of a composite airframe subjected to im-pact loading through extensive analyti-cal/experimental correlation. The crash model ofthe ACAP helicopter was developed from an ex-isting MSC.Nastran modal-vibration model of thehelicopter. Considerable modifications were re-quired to convert the original modal-vibration inputdeck to an MSC.Dytran model for crash simulation.

Following the model conversion andmodification, a two-stage rigid-to-flexible modelingapproach was used to generate analytical predic-tions. Due to the relatively long pulse duration, arigid fuselage model was executed in which anexternal subroutine was used to represent theresponse of the landing gear, including both theoleo-pneumatic and crushable honeycomb

stages. Approximately 0.05 seconds before fu-selage contact, the nodal displacements and ve-locities from the rigid fuselage model were output.Then, a flexible structural model was executedwith the nodal displacements and velocities usedas initial conditions. This two-stage modeling ap-proach worked well except for a slight discontinu-ity that occurred when the flexible model was initi-ated due to the sudden onset of the landing gearforces. Comparisons of the experimental andpredicted vertical velocity responses of the rightengine, right main landing gear, and copilot floorwere performed for both the rigid and flexible fu-selage models to validate the two-stage modelingapproach. This comparison demonstrated thatequivalent results were obtained for both the rigidand flexible models for the period before fuselageimpact; i.e., up to 0.1-seconds after left gear im-pact.

The initial correlation between the finiteelement simulation and the experimental resultswas a comparison of the time sequence of events.In general, the simulation accurately predicted thetime of right main gear contact, nose gear contact,tail cone failure, and the timing of the peak accel-erations of the left and right troop seat floor, copi-lot floor, and the right engine locations.

A comparison of the predicted and ex-perimental acceleration responses for the pilotand co-pilot floor, the right and left troop seatfloor, the right and left engine, right main gear,and several other locations on the fuselage weremade. For this comparison, both the experimentaland predicted acceleration data were filtered us-ing a three-pole Butterworth 60Hz low-pass filter.In general, the simulation accurately predicted theoverall shape, magnitude, and pulse duration foreach of the experimental acceleration responsesmentioned previously. The predicted peak accel-eration at the upper bulkhead location was sub-stantially higher than the experimental data at thatlocation. The discrepancy was attributed to thelarge number of rod elements in the bulkhead thatwere unmodified from the original Sikorsky modal-vibration model. Typically, rod elements are ex-tremely stiff and do not accurately reflect the re-sponse of the actual bulkhead beams.

In addition, the crash simulation correctlypredicted the subfloor crushing response andstructural failures of the engine support beams.Overall, the high level of agreement obtained be-tween model and test should build confidence inthe future use of nonlinear, explicit transient dy-

namic finite element codes as a crashworthy de-sign and evaluation tool for aircraft.

Acknowledgements

The authors would like to recognize sev-eral organizations and companies for their supportof this research program. The ACAP helicopterwas provided by the US Army Aviation AppliedTechnology Directorate, located at Ft. Eustis, Vir-ginia. Financial support for the full-scale crash testand simulation activity was provided by the Avia-tion Safety Program of NASA Langley ResearchCenter. The MSC.Nastran modal-vibration modelof the ACAP helicopter was provided by SikorskyAircraft Company. The authors would like to ac-knowledge Charles Clarke and Joseph Shen ofSikorsky Aircraft for their many helpful discus-sions. Finally, the authors wish to acknowledgethe excellent technical support provided by theMSC.Software Company.

References

1. "Computational Methods for Crashworthiness,"NASA Conference Publication 3223, October1993.

2. Gamon, M.; Wittlin, G.; and LaBarge, B.,"KRASH 85 User's Guide- Input/Output Format,”Final Report DOT/FAA/CT-85/10, May 1985.

3. Anon., “LS-DYNA3D User’s Manual,” LivermoreSoftware Technology Company, Livermore, CA,1997.

4. MSC.DYTRAN User’s Manual Version 4.0 , TheMacNeal-Schwendler Corporation, Los Angeles,CA, November 1997.

5. PAM-CRASH, Engineering Systems Interna-tional SA, F-94588 Rungis, France

6. Anon., “ACAP Interim Technical Report, Vol-ume III - Military Characteristics,” Report No.TR17304, Sikorsky Aircraft, March 1982.

7. Perschbacher, J. P.; Clarke, C.; Furnes, K.; andCarnell, B., “Advanced Composite Airframe Pro-gram (ACAP) Militarization Test and Evaluation(MT&E) Volume V - Airframe Drop Test,”USAATCOM TR 88-D-22E, March 1996.

8. Lahey, R. S.; Miller, M. P.; and Reymond, M.,“MSC/NASTRAN Reference Manual, Version 68,”The MacNeal-Schwendler Corporation, 1994.

9. Vaughan, V. L. and Alfaro-Bou, E., “Impact Dy-namics Research Facility for Full-Scale AircraftCrash Testing,” NASA TN D-8179, April 1976.

10. Anon., “MSC.PATRAN, DYTRAN PreferenceGuide,” Publication No. 903077, Version 6, TheMacNeal-Schwendler Corporation, 1996.

11. Stebbins, R. F. and Twomey, W. J., “Plan,Formulate, and Discuss a NASTRAN Finite Ele-ment Model of the Sikorsky ACAP Helicopter Air-frame.” NASA CR 182059, June 1990.

12. Lyle, K. H.; Jackson, K. E.; and Fasanella, E.L., “Development of an ACAP Helicopter ImpactModel,” Proceedings of the AHS National Techni-cal Specialists’ Meeting on Rotorcraft Crashwor-thiness, Phoenix, AZ, September 14-16, 1998.

13. Boitnott, R. L.; Jackson, K. E.; Fasanella, E.L.; and Kellas, S., “Full-Scale Crash Test of theSikorsky Advanced Composite Airframe ProgramHelicopter,” Proceedings of the American Heli-copter Society Forum 56, Virginia Beach, VA, May2-4, 2000.