ContentslistsavailableatScienceDirect
Computer
Methods
and
Programs
in
Biomedicine
journalhomepage:www.elsevier.com/locate/cmpb
Beam-inside-beam
contact:
Mechanical
simulations
of
slender
medical
instruments
inside
the
human
body
Marco
Magliulo
a,
Jakub
Lengiewicz
a,b,
Andreas
Zilian
a,
Lars
A
.A
.
Beex
a,∗a Institute of Computational Engineering, Faculty of Science, Technology and Communication, University of Luxembourg, Maison du Nombre, 6, Avenue de la
Fonte, 4364 Esch-sur-Alzette, Luxembourg
b Institute of Fundamental Technological Research of the Polish Academy of Sciences (IPPT PAN), ul. Pawinskiego 5B, 02–106 Warsaw, Poland
a
r
t
i
c
l
e
i
n
f
o
Article history: Received 15 January 2020 Revised 6 April 2020 Accepted 29 April 2020 Keywords: Surgical simulation Contact mechanics Beam-inside-beam Artery Cochleaa
b
s
t
r
a
c
t
BackgroundandObjective: Thiscontributionpresentsarapidcomputationalframeworktomechanically simulatethe insertion ofa slendermedical instrumentin atubular structure such as an artery, the cochleaoranotherslenderinstrument.
Methods: Beamsareemployedtorapidlysimulatethemechanicalbehaviourofthemedicalinstrument and the tubular structure.However, the framework’snovelty isits capabilityto handlethe mechani-calcontactbetweenaninnerbeam(representingthemedicalinstrument)embeddedinahollowouter beam(representing the tubular structure). This“beam-inside-beam” contact framework, whichforces two beamsto remain embedded,is the firstofits kindsince existingcontact frameworks for beams are“beam-to-beam” approaches,i.e.theyrepelbeamsfromeachother.Furthermore,weproposecontact kinematicssuchthatnotonlyinstrumentsandtubeswithcircularcross-sectionscanbeconsidered,but alsothosewithellipticalcross-sections.Thisprovidesflexibilityfortheoptimizationofpatient-specific instruments.
Results: Theresultsdemonstratethattheframework’srobustnessissubstantial,becauseonlyafew incre-mentspersimulationandafewiterationsperincrementarerequired,eventhoughlargedeformations, largerotationsandlargecurvaturechangesofboththeinstrumentandtubularstructureoccur.The sta-bilityoftheframeworkremainshighevenifthemodulusoftheinnertubeisthousandtimeslargerthan thatoftheoutertube.Ameshconvergencestudyfurthermoreexposesthatarelativelysmallnumberof elementsisrequiredtoaccuratelyapproachthereferencesolution.
Conclusions: Theframework’shighsimulationspeedoriginatesfromtheexploitationoftherigidityof thebeams’cross-sectionstoquantifytheexclusionbetweentheinnerandthehollowouterbeam.This rigiditylimitstheaccuracyoftheframeworkatthesametime,butthisisunavoidablesincesimulation accuracyandsimulationspeedaretwocompetinginterests.Hence,theframeworkisparticularly attrac-tiveifsimulationspeedispreferredoveraccuracy.
© 2020TheAuthors.PublishedbyElsevierB.V. ThisisanopenaccessarticleundertheCCBY-NC-NDlicense. (http://creativecommons.org/licenses/by-nc-nd/4.0/)
1. Introduction
Mechanicalsimulationsofsurgicalinterventionscanbeusedto train surgeons, reveal patient-specificcomplicationsthat may oc-cur during surgery and plan interventions patient-specifically. In the future,mechanical simulations of surgical interventionsmay even be used to optimize medical instruments for each patient (e.g. shape and stiffness) and be exploited to autonomously per-forminterventions.
∗ Corresponding author.
E-mail address: Lars.Beex@uni.lu (L.A .A . Beex).
Numerous frameworks to numerically simulate surgical inter-ventions can be found in the literature. For instance, one set of frameworkssimulatescuttingthroughsofttissuesinreal-time[4– 6,34,42]toprovidehaptic feedbackto thetraineeperforming the “intervention”. Another set of approaches aims to simulate the insertion of needles [2,3,8,10,36]. These frameworks mayalso be usedtoprovidehapticfeedbackand/or tohelp toaccurately steer theneedletothetargetofinterestduringsurgery.
However,theframeworkpresentedinthiscontributionfocuses onmechanicalsimulationsthat involvetheinsertion(or removal) of a slender medical instrument inside a tubular structure such as an artery, the cochlea or another slender instrument such as https://doi.org/10.1016/j.cmpb.2020.105527
proposed framework and the frameworksof [1,9,15,27,39] is that our framework represents both the slender medical instrument andthe tubular structure as beams, whereas the frameworks of [1,9,15,27,39]only representthe slendermedicalinstrument with beamswhilst conventional 3D finiteelements are usedto repre-sentthetubularstructure.
A wide variety of schemes to handle contact between beams canbedistinguishedintheliterature.Allexisting“beam-to-beam” contact frameworks are formulated to repel penetrating beams. Severalof these beam-to-beam contact frameworks are only ap-plicable if the beams’ cross-sections are circular, shear deforma-tionsareignoredandthecontactarearemainssmall,since unilat-eralcontactconditionsareenforcedattheclosestpairofcentroid points[33,41,43].Thus,iftwobeamscollide,acontactforceis ap-pliedattheclosestpairofcentroidpointstorepelthetwobeams. Ifthebeams’cross-sectionsareelliptical,the considerationofthe centroid-linesalone isinsufficient to determinethe contact loca-tions.Instead,contactforcesmustbeappliedattheclosestpairof surfacepointswherethetangentplanesofthecontactingsurfaces areparallel.This wasdemonstrated byGay Netoetal.[12,13]for frictionlessandfrictionalcases,respectively.
Furthermore, in case of non-localized contact (e.g. for paral-lel beamsin contact),theassumption of point-wisecontactdoes not hold. Meier et al. [30] have therefore proposed a contact frameworktohandlenon-localizedbeam-to-beamcontact,butthe cross-sectionalshapeisrestrictedto(rigidly)circularandshear de-formationisnotaccountedfor.Theserestrictionsenable quantify-ing the penetration solely usingthe centroid-lines of the beams, whichyieldsrapidsimulations.
Magliuloetal.[28,29]presentedothermaster-slaveframeworks forbeam-to-beamcontactapplicabletobothshear-deformableand shear-undeformablebeams,withboth circularaswellaselliptical cross-sections.Bothschemes considerthebeam’s surfaces explic-itly,whichhasresultedinwider applicabilitythanthe schemeof Meieretal.[30],albeitatthe expenseofthesimulation speed.A two-halfpassalgorithmwasfurthermoreformulatedin[29]to re-movethebiasofmaster-slaveapproachesforbeam-to-beam con-tact,butwithlimitedbenefitsfortheresults.
The beam conglomerates of interest to this contribution dif-ferfromthe aforementionedworks [28,29],since the focusis on “beam-inside-beam” contact instead of“beam-to-beam” contact.In otherwords,thebeamsmustremainembeddedforthebeam con-glomeratesofinterestinourcontribution,whereasexisting beam-to-beamcontact frameworks repel penetrating beams. The mea-sureof penetration in our beam-inside-beam framework, on the other hand, shows similarities with the measure of penetration forthe“beam-to-beam” contact frameworkof[28].Penetrationis measuredbetweensectionsdistributedalong theinnerbeamand theinteriorsurfaceoftheouterbeam.Incaseofcontact,unilateral constraintsareregularizedwiththepenaltymethod,whichbrings compliancetotheotherwiserigidcross-sections.
The outline of the remainder of this contribution is as fol-lows. In Section 2, the contact framework is presented in the space-continuous setting along with the associated contact vir-tual work. Also in this section, the spatial discretization applied
The contactkinematics employed inthiscontributionare pre-sented in this section [12,13,28,29]. First, the beam’s surface parametrizationisexplained.Then,theproceduretoquantify pen-etrationisdetailed.Theformulationofthecontactvirtualworkis presentedforapenaltyapproach.
2.1.1. Parametrizationofthesurface
The geometrically exact beam (GEB) Simo-Reissner theory [7,16,17,31,35,37,38] is used in this contribution. This entails that the beamsare shear-deformable andthat rigidcross-sections are considered,whichcannotwarp.
The surface of beam B is parameterized with two convective parametersh=
h1,h2T.h1∈[0,L],denotesthearc-lengthparam-eterofthebeam’scentroidlinex0c:
(
0,L)
→R3whileh2∈[0,2π
]isa circumferentialparameterofthe perimeterofcross-section C attachedtox0c
(
h1)
(see[12]andFig.1).Ldenotestheinitiallengthof the beam. The location ofa surfacepoint in the undeformed configurationintheglobalcoordinatesystem, x0,canbeobtained
from:
x0
(
h)
=x0c(
h1)
+v0(
h)
, (1)wherev0 denotes a vector containedin C.Here, we assume that
v0alwaysconnectsx0c toasurfacepoint.IncaseC iselliptical,v0
canbeexpressedas:
v0
(
h)
=acos(
h2)
e01(
h1)
+bsin(
h2)
e02(
h1)
, (2)where a and b denote the dimensions of the elliptical cross-sectionsinitsprincipaldirections.e01ande02 areorthonormal
ba-sisvectorsoftheplanecontainingC.e03denotesthenormalvector
toC.Thetriad{e01,e02,e03}formsanorthonormalbasis.
Duetothehypothesisofrigidsections,thelocationofthesame materialpoint inthedeformedconfigurationcan similarlybe ob-tainedfrom:
x
(
h)
=ϕ(
x0(
h))
=xc(
h1)
+v(
h)
, (3)where:
xc=x0c+u, (4)
denotesthelocationofthecentroidpointinthedeformed config-uration.u:
(
0,L)
→R3denotesthecentroid-line’sdisplacement.ϕ
Fig. 1. Beam kinematics for the current configuration. The beam’s centroid-line is presented with a dashed line. Local basis vector e 3 is not aligned with the vector tangent to the centroid-line ∂xc
∂h1 due to shear deformation. A typical vector v lying
denotes thedeformation mappingrelatingthecurrentlocationof a point toits location intheundeformed configurationsuch that
x
(
h)
=ϕ
(
X(
h))
.visobtainedfrom:v
(
h)
=(
h1)
· v0
(
h)
=acos(
h2)
e1+bsin(
h2)
e2, (5)where
: (0, L)→SO(3), with SO(3) the rotation group, is a ro-tation tensor that rotates e0i to ei for i∈{1,2, 3}.Because shear
deformationcanbepresent,e3isnotnecessarilyalignedwiththe
tangenttothecentroid-line(seeFig.1).Insuchcases,i.e.:
e3×
∂
xc
∂
h1 =0, (6)where × denotesthecrossproduct.
For further use, we define two (covariant) tangent vectors to the surface ofB at x
(
h)
, denoted byτ
1=∂∂hx 1 andτ
2=∂∂hx 2 (seeFig.1).Ingeneral,
τ
1andτ
2arenotorthogonaltoeachother.2.1.2. Contactkinematics
Wefocushereonasystemconsistingoftwobodies:BIdenotes
the thininner beamandBJ denotes thehollow outer beam. We
assumeherethatbothBIandBJaremodeledasaGEBwithplain
andhollowcross-sections,respectively. Wedenoteby
∂
BJ thein-teriorsurfaceofBJ.
ToquantifythepenetrationofBIwith
∂
BJ,andtoquantifythecontactareaoverwhichthispenetrationoccurs,we: 1. Seedsectionsalongthecentroid-lineofBI(seeFig.3),
2. For each seeded section, we solve a projection (local) prob-lemtodetermineifitpenetrates
∂
BJandifso,byhowmuch.Thisprojection problemyields twosurfacepoints:one onthe perimeteroftheseededsectionandoneon
∂
BJ.Thesepointsare usedto establisha measure ofpenetration, whichinturn determines theamplitudeofthecontactforces(if penetration ispresent).
AsBIandBJ havea differentrole,theproposed frameworkis
a master-slaveapproach.We call BI theslave andBJ themaster
[40]. Next,we discusshow to compute ifa givensection of BI,
denoted by C withperimeter
∂
C,penetrates∂
BJ and ifso, howtheamountofpenetrationiscomputed.
Itmustbenotedthattheproposedcontactalgorithmcanonly be usedifone contactarea occursforeachcross-sections(leftin Fig. 2). The contactframework can thus not handlescenarios as presentedontherightinFig.2.Ifthecross-sectionsofBI is
per-fectlyalignedwiththecross-sectionsofBJ,onlyonecontactarea
occursif:
(
ainner)
2binner
<
(
bouter)
2aouter
ainner≥ binner aouter≥ bouter, (7)
wheresubscriptinnerreferstotheinnerbeamandsubscriptouter
referstotheinnercross-sectionsoftheouterbeam.
WenowintroducenI,anoutward pointingunitvectornormal
to
∂
BI.Itisdefinedasfollows: nIhI=τ
I1(
h I)
×τ
I 2(
h I)
τ
I 1(
h I)
×τ
I 2(
h I)
. (8)nJontheotherhandistheinwardpointingunitvectornormalto
∂
BJ.Itisdefinedasfollows: nJhJ=τ
J1(
h J)
×τ
J 2(
h J)
τ
J 1(
h J)
×τ
J 2(
h J)
. (9)Localproblemandmeasureofpenetrationbetween
∂
C and∂
BJ.WewillnowinvestigateifC,thecross-sectionuniquelydefinedby convectivecoordinatehC,penetrates
∂
BJ andifso,byhowmuch.Theso-calledgapvectorgconnectsapointontheperimeterofC, xI∈
∂
C,toasurfacepointon∂
BJ,xJ:g
(
hC,hI2,hJ1,hJ2)
=xJ(
hJ1,hJ2)
− xI(
hC,hI2)
. (10)Alocal(or projection)problemmustnowbesolved,whichyields points¯xIand¯xJ,suchthatanappropriatemeasureofpenetration
isestablished.Fourconvectivecoordinatesareinvolvedinthelocal problem:hC,whichisfixed,aswellashI2,hJ1 andhJ2 thatareto
bedetermined.
Apossibilitytodeterminetheunknownconvectivecoordinates wouldbe tosolvean optimizationproblembyminimizing an ob-jectivefunction.Anotherpossibility[12–14,21,22,33]istosolvefor a set of equations that does not stem from an objective func-tion,generallyunilateralorbilateralorthogonalityconditions. Pre-viously, we have shown that the latter makes the resolution of thelocal problem20–30% faster to solve forbeam-to-beam con-tact[28].Wethereforeconsiderasimilarapproach forthe beam-inside-beamcontactofthiscontribution.Threeoftheequationswe solveforareexpressedas:
f1
(
¯q)
=¯xJ− ¯xI− ¯g¯nI=0, (11)where:
q=
hI2,hI1,hJ2,gT
, (12)denotesthearrayofunknowns.Hereandinthefollowing,an over-headbaroveraquantityindicatesthatitisevaluatedatthe solu-tionoftheprojectionproblem.Thus, ¯qdenotesthearray solution ofEq.11.The independentvariable ¯gquantifiespenetration mea-sured in the normal direction ¯nI
(
hC,¯hI2)
, usually denoted asgN
anddefinedas:
gN= ¯g=
(
¯xJ− ¯xI)
· ¯nI. (13)Asfourvariablesarepresentforonlythreeequations,the sys-temof Eq.11 isunder-determined. An additional equation is re-quired.Tothisend, weintroduce plane P spanned bynI and
τ
I 2withthefollowingnormalvector:
˜
nI=
τ
I2× nI. (14)
Also,wedefinenJpasthenormalizedprojectionofnJonP:
nJp= nJ−
(
nJ· ˜nI)
n˜I nJ−(
nJ· ˜nI)
n˜I. (15)Fig. 3. (a) Two beams in contact. (b) Sections for which penetration has been detected.
Fig. 4. Perpendicular view to the plane P of ∂C. Surface points ¯x I and ¯x J , obtained
after solving Eq. 17 , are presented as red dots. Vectors ¯n I and ¯n Jp are both orthog-
onal to ¯τJ
2 . (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
At thesolution of the local problem, we want nI and nJp to be
orthogonalto
τ
J2(see Fig.4).Thisistrueifthefollowingequation
holds:
f2
(
¯q)
=aC(
¯nI+¯nJp)
· ¯τ
J2=0, (16)whereaC denotes the dimension ofC along its largestsemi-axis whichisusedtoscalef1such thatf1 andf2 havethesameunits.
Thesetofequationstosolveforisnowabbreviatedasfollows: f
(
¯q)
=f1(
¯q)
,f2(
¯q)
T
=0. (17)
The setof equations to solve forin Eq.17 is nonlinear. Tosolve it,we apply Newton’s method for which we linearise residual f
whichrequiresthefollowingJacobian: H
(
q)
=∂
f∂
q =⎡
⎣
∂f 1 ∂q∂f2 ∂q
T
⎤
⎦
. (18)Tocompute the componentsof H,we need to introduce the fol-lowingquantities:
1.ContravariantcomponentsMKi jofMK,themetrictensorofthe surfaceofbodyK,read:
MK11 MK12 MK21 MK22 =
MK11 MK12 MK21 MK22 −1 =
τ
K 1·τ
K1τ
K1·τ
K2τ
K 2·τ
K1τ
K2·τ
K2 −1 . (19)2. Thesecondordersurfacederivatives:
τ
Ki j=
∂
τ
Ki
∂
hKj. (20)3. ThecovariantcomponentsofcurvaturetensorCK: CKi j =
τ
K i j· nK (21) 4. Weingarten’sformula:∂
nK∂
hj =−MK jkCK kiτ
Kj. (22)MakinguseofEq.22,thepartialderivativesoff1 withrespect
toqinEq.18yield:
∂
f1∂
q =∂
f1∂
hI2,∂
f1∂
hJ1,∂
f1∂
hJ2,∂
f1∂
g =−τ
I 2+g(
MIjkCIk2τ
Ij)
,τ
J1,τ
J2,−nI . (23)Thedifferentiationoff2withrespecttoqgives:
∂
f2∂
q =∂
f2∂
hI2,∂
f2∂
hJ1,∂
f2∂
hJ2,∂
f2∂
gT
. (24)Thecorrespondingexpressionsaremorecomplicated,inparticular becausenJpdependsonhI2,hJ1 andhJ2.
WenowrewritenJpas:
nJp=Nn Jp ln Jp, (25) where: Nn Jp=nJ·
(
I− ˜nIn˜I)
=nJ· D, (26) and: ln Jp=Nn Jp. (27)ThepartialderivativeofnJpwithrespecttothekth-surface
where
δ
denotes theKronecker symbol (nottobe confused with thevariationsymbol).Theterm ∂n ˜I∂hlk,where: ˜ nI=N ˜ n J lN n˜J, (31) with: Nn ˜J =
τ
I 2× nI, (32) and: lN n˜J =τ
I 2× nI, (33) reads:∂
n˜I∂
hlk = 1 lN n˜J I−˜nIn˜I·∂
N˜ nJ∂
hlk =F·∂
Nn˜ J∂
hlk. (34) Wecanwrite:∂
Nn ˜ J∂
hlk =δ
lIτ
I 2k× nI+τ
I2× −MIi jCI jkτ
Ii =c, (35) suchthat:∂
n˜I∂
hlk =F· c. (36)Finally,thisyields:
∂
nJp∂
hlk = 1 ln Jp I−nJp nJp ·∂
Nn Jp∂
hI2 =E·δ
lJ−MJlmC mkτ
K· D− nJ·(
F· c)
˜nI+n˜I(
F· c)
=dlk. (37)Using Eq. 37, compact expressions forthe components of ∂f2
∂q in
Eq.24canbeobtained:
∂
f2∂
hI2 =aC∂
nI∂
hI2+∂
nJp∂
hI2 ·τ
J 2 =aC−MIjkCI k2τ
Ij+dI2 ·τ
J 2, (38)∂
f2∂
hJ1 =aC∂
nJp∂
hJ1 ·τ
J 2+nJp·∂
τ
J 2∂
hJ1 =aCdJ1·τ
J 2+nJp·τ
J21 , (39)∂
f2∂
hJ2 =aC∂
nJp∂
hJ2 ·τ
J 2+nJp·∂
τ
J 2∂
hI2 =aCdJ2·τ
J 2+nJp·τ
J22 . (40)CombiningEqs.23,38,39,40,Hreads:
H= ⎡ ⎣ −τ I 2+g(MIjkCIk2τIj) τJ1 τJ2 −nI aC −MIjkCI k2τIj+d I2 ·τJ 2aC d J1·τJ 2+n Jp·τJ21 aCd J2·τJ 2+n Jp·τJ22 0 ⎤ ⎦. (41) First-guessprocedure
As statedabove, Eq.17issolved iteratively.Aninitial guessof the local parameters q must be provided to the solver. We em-ployasimpletwo-stepprocedureto establishan appropriatefirst guess:
1. We(approximately)findthecentroidpointofthemasterbody that is the closest to the centroid point of the slave cross-section,xI
c
(
hC)
.Asimplewayofachievingthisisby samplingcross-sectionpointsalong themasterbeam’scentroid-lineand picktheclosestcentroid-pointfromxI
c
(
hC)
.Theassociatedcon-vectiveparameteroftheclosest sampledcentroidpoint is de-notedbyhJ1, f g.
2.To determine the initial values of circumferential parameters
hI2, f g andhJ2, f g, we locate a pair of perimeterpoints on the
cross-sectionsattachedtoxI
c
(
hC)
andxJc(
hJ1, f g)
.Thisprocedureis depictedinFig. 5.We startby samplingfourpointson the perimeter of both cross-sections. The pair of closest points is then chosen. Next,for each cross-section,we seed a point on the middleofeachsub-curveattachedto thepointpreviously selected ((c) inFig. 5). Again, theclosest pair ofpoints is se-lected.Thisprocedureisrepeatedseveraltimes.Inour simula-tions, itwasrepeateduntiltherelative changeofthedistance betweenthepairofclosestpointsfallsbelow10%.
Notethatinpractice,theapproachtoestablishtheinitialguess isonlyperformedforagivenslave cross-sectionifitisnotactive butclose to the master surface. Ifa slave cross-section is active (meaningthat it alreadypenetrated the master surfacein a pre-vious contactdetection),the solutionofthe localproblemof the previouscontactdetection(¯q)isusedasthefirstguess.
2.1.3. Contactconstraintsandvirtualworkequationincludingcontact
The impenetrability of
∂
C and∂
BJ is enforced via unilateralcontactconditions:
gN≥ 0 TN<0 gNTN=0, (42)
where TN denotes the nominal contact traction, meaning that it
referstothereferencesurfacearea.
Incaseofcontact,acontactvirtual work,
δ
c,isaddedtothe
virtual work equation for thetwo-body systemand thespace of admissiblevariationsV ismodified[40].Inthequasi-staticsetting, thevirtualworkreads:
δ(
pIJ,δ
pIJ)
=δ
BI
(
pI,δ
pI)
+δ
BJ(
pJ,δ
pJ)
+
δ
c(
pIJ,δ
pIJ)
=0,∀
δ
pIJ∈V, (43)where
δ
Bi denotes the internal and external virtual work of
beamBi (excluding contactinteractions). Kinematic variables
as-sociated with Bi are denoted by pi=
ui,θ
iT:(
0,L)
→R3× R3andtheassociatedtest functionsby
δ
pi.ui denotesthedisplace-mentfieldoftheBi’scentroid-lineand
θ
iitsfieldofrotationvec-tors parametrizing SO(3) [16]. pi is only admissible if pi
(
XBi)
=pi D
(
XBi
)
,∀
XBi∈∂
BiD.
∂
BDi denotesthepartoftheboundaryof∂
BiwhereDirichletboundaryconditionsareimposed[40].
pIJ=
pI,pJT gathersthe kinematicvariables ofboth beams.Similarly,testfunctionsaregatheredin
δ
pIJ=δ
pI,δ
pJT.InEq.43,thevirtualworkduetocontact,
δ
c,accountsforall
thesectionspenetrated.Theinfinitesimalvirtualworkproducedat asinglesection,denotedbyd
δ
c,canbewrittenas[28]:
d
δ
c=TNδ
gN∂
xI 0c∂
hI1 dhI1, (44)where
δ
gN denotes the variationof gN that dependson allkine-matic variables in pIJ, and dhI1 denotes the differential of the
slave’scentroid lineparameter associatedto C. dhI1 isrelated to
thedifferential length of the undeformed centroid-lineaccording to:
dLBI=
∂
xI0c∂
hI1 dhI1, (45)Next,wediscusshowtocalculateTNforthepenaltyapproachused
in this contribution andalso how to compute
δ
gN. The nominalFig. 5. Illustration of the procedure to determine a good initial guess for the local problem. (a) Determination of centroid point x J
c(hJ1, f g) that must be as close as possible
to x I
c(hC) ; (b) Determination of the closest pair of points amongst sampled points on ∂C and the perimeters of the cross-section attached to x Jc(hJ1, f g) ; (c) Determination of
the closest pair of points amongst the closest pair of points from (b) (in orange) and points in the middle of the curve connected to theses points (in red). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
notdepend on pIJ,thelinearizationof
δ
yields shorter
expres-sionsthanifthecurrenttractionvectorwasemployed[26].
Penaltymethod
If a penalty formulationisused, contacttraction TN, actingat
thepairofsurfacepoints¯xJand ¯xI,isgivenby:
TN=−
N
−gN, (46)
where
N>0 denotes the penalty stiffness and
• denote the
Macaulay brackets, representing the positive part of its operand [25]. The fact that
N must be selected can be seen as a
weak-nessofthepenaltymethod.Indeed,other constraintenforcement methods likethe Lagrange multipliersmethod do not need such user-definedparameters.Ontheotherhand,inthecontextof con-tact frameworks for beams that are characterized by rigid cross sections,the penalty parametercan beinterpreted assome com-plianceofthebeamsinthetransversaldirections[28,33].Inserting Eq.46intoEq.44,thevirtualworkofthecontactforcereads: d
δ
c=−N
−gN
δ
gN∂
xI 0c∂
hI1 dhI1. (47)The virtual work of all penetrated sections follows fromthe in-tegration ofthe infinitesimal virtual work of a singlepenetrated section,d
δ
c,alongthecentroid-lineofBIasfollows:
δ
c= h1I U h1I L dδ
c, (48) whereh1IL andh1UIarethelowerandupperboundsoftheintegral,
respectively.
Variationofthenormalgap,
δ
gNgN=¯g· ¯nI,measured fora fixed hI1, dependson pIJ butalso
onthelocalvariablesinq,whichinturnimplicitlydependonpIJ.
Eventually,
δ
gNmustbesolelyexpressedintermsofthevariationsofthe primary variables, here
δ
pIJ. The variational operatorap-pliedtogNgives:
δ
gN=δ
¯g· ¯nI+¯g·δ
¯nI. (49)Also,wehave:
¯g·
δ
¯nI=gN¯nI·
δ
¯nI=0, (50)as
δ
¯nI· ¯nI=0.Noting that ¯gdependson pIJand ¯q(
pIJ)
,we get:δ
¯g=∂
g∂
pIJ q=¯qT
δ
pIJ+∂
g∂
q q=¯qT
δ
q (51) =∂
g∂
pIJ q=¯q Tδ
pIJ+∂
g∂
q q=¯q T dq dpIJ q=¯qδ
pIJ. (52)TherelationshipsbetweenqandpIJcanbefoundonthebasisof
thestationarityof f
(
pIJ,¯q(
pIJ))
withrespecttopIJ:df dpIJ=
∂
f∂
pIJ q=¯qδ
pIJ+⎛
⎜
⎜
⎜
⎝
∂
f∂
q H q=¯q⎞
⎟
⎟
⎟
⎠
δ
¯q=0, (53)whichleadsaftersomerearrangementto:
δ
¯q=− H−1 q=¯q∂
f∂
pIJ q=¯qδ
pIJ. (54)Forfurtheruse,wedefineAasfollows: A=−H−1
q=¯q
∂
f∂
pIJ q=¯q, (55) suchthat:δ
¯q=Aδ
pIJ. (56)InsertingthisinEq.52yields:
δ
¯g=∂
g∂
pIJ q=¯qT
+∂
g∂
q q=¯qT
Aδ
pIJ. (57)Insertingthisinto49yields:
δ
gN=δ
¯g· ¯nI=∂
¯g∂
pIJT
δ
pIJ· ¯nI+∂
¯g∂
qT
Aδ
pIJ· ¯nI (58) =(δ
pIJ)
T∂
¯g∂
pIJ· ¯n I+AT ∂¯g∂
q· ¯n I (59) =(δ
pIJ)
Tz. (60)Noteforfurtherusethat ∂∂p¯g IJ¯nIreads:
∂
g∂
pIJ q=¯q· ¯n I=∂
¯xJ∂
q T · ¯nI−∂
¯xI∂
q T · ¯nI (61) =!
0τ
J 1· ¯nIτ
J 2· ¯nI"
−!
τ
I 1· ¯nI 0 0"
(62) =!
0τ
J 1· ¯nIτ
J 2· ¯nI"
−!
0 0 0"
, (63)whereisitobviousfromEq.63that ∂¯xI
∂q T
Allinall,wecanwrite:
δ
c= h1I U h1I L −N
−gNzT
δ
pIJ∂
xI 0c∂
hI1 dhI1. (64) 2.2. Spatialdiscretization2.2.1. Interpolationofthebeams’surface
The FiniteElementMethod(FEM)isthe discretisationmethod used in this work. Each beam is now discretized with a set of beamfiniteelements(BFEs)[16,37,38].Therotationvectorsarethe primary rotational kinematic variables [16] that are interpolated. Fora BFE E,we denote by Xh
c
(
h1)
:0,LE
→R3 the interpolatedposition of its centroid-line, where LE denotes the length of the centroid-line in the undeformed configuration. uh
c
(
h1)
: 0,LE→ R3andθ
h(
h1)
:0,LE→R3 denotetheinterpolateddisplacementfieldofthecentroid-lineandtheinterpolatedfieldofrotation vec-tors,respectively.Rodrigues’formulaisusedtoobtainrotation ten-sor
fromtheinterpolatedrotationvector1,denoted
θ
h.Thedis-placement ofnode K isdenoted by uˆEK andits rotationvector by ˆ
θ
EK.ThekinematicvariablesofelementE aregatheredin:pE=
uˆE1,...,uˆEnu,θ
ˆE 1,...,θ
ˆ E nθ T , (65)where nu and nθ denote the number of nodes used to
interpo-late the displacement andthe rotation vector, respectively. Since thenodalvariablesofallBFEsofBIandBJaredenotedby pˆ,the
associatedvariationsaredenotedby
δ
ˆp.2.2.2. Contactresidualandstiffness
ThediscretizedformofthevirtualworkinEq.43leadstoaset ofnonlinearequations.Newton’smethodisgenerallyusedto iter-ativelydetermine thesolution ˆpsol of thevirtual work statement. ThisrequiresthelinearizationofEq.43aroundanestimateofpˆsol,
ˆ
p,whichyields:
δ(
ˆp+pˆ,δ
pˆ)
δ(
pˆ,δ
pˆ)
+δ(
ˆp,δ
ˆp)
pˆ=δ
pˆT(
rg+Kgˆp)
0,(66)
where rg and K
g denote theglobal residual force and the global
stiffness matrix, respectively. pˆ denotes an increment of the nodalvariables.Theglobalforcesread:
rg
(
ˆp)
=fint(
ˆp)
+rc(
pˆ)
− fext(
ˆp)
, (67)where fint denotes the internal force vector stemming from the
contributionofallBFEs,and fext theexternalforcevector. rc
con-tainsallthecontactcontributionsfromallcontactelements,where acontactelementreferstoaseededsectionattachedatan integra-tion point (see below) along BI andits projection on discretized
surface
∂
BJ.Assuming here that fext does not depend on pIJ, the global
stiffnessobtainedafterthelinearizationofrg,canbedecomposed
asfollows: K
g=Kint+Kc, (68)
whereKint,denotesthestiffnessmatrixoftheBFEs,andKcdenotes thestiffnessmatrixofallcontactelements.
Thecontactvirtualworkisthesumofallcontactcontributions:
δ
c(
pˆ,δ
pˆ)
=#
e∈S
δ
c,e(
pˆe,δ
pˆe)
, (69)1 It is well known that the corresponding finite element formulation does not share the strain-invariance property of the underlying geometrically exact theory [18] .
whereS denotesthe setofactive contactelements(i.e.thosefor which gN<0), and
δ
c,e denotes the contact virtual work
asso-ciated withcontact element e.If no smoothingprocedure of the surfaceisrequired[26],eachcontactelementsinvolvestwo BFEs, one which is part of the discretization BI, and the other which
ispartofthediscretizationofBJ.However, ifasmoothingofthe
beam’s surface is performed, each contact element directly de-pends on several BFEs of BI and of BJ. The set of elements of
BIandBJusedtoconstructcontactelementearedenotedbyM
andN, respectively.The involvednodalvariables are gatheredin array ˆpe=
pˆM,pˆNT. Similarly, theinvolved nodalvariations are denotedbyδ
ˆpe.Thelinearizationof
δ
creads:
δ
c(
ˆp+ˆp,δ
pˆ)
= # e∈Sδ
c,e(
ˆpe+ˆpe,δ
ˆpe)
≈# e∈Sδ
c(
pˆe,δ
ˆpe)
+δ
c(
ˆpe,δ
ˆpe)
ˆpe (70) ≈# e∈Sδ
ˆpTe(
rce+Kcepˆe)
. (71)Next,we discusshow to constructelement contributions rce and
Kcetotheglobalresidual(force)vectorandtheglobalstiffness ma-trix,respectively.
Forcevectorandstiffnessofcontactelemente
We numerically integrate Eq. 39 witha quadrature (Gauss or Lobato-type).Tothisend,weplacenMIP integrationpointsofa sin-glesubdomainalongthecentroid-lineofM[21,33].Thisyields:
δ
c=−N 1 −1
−gN
(η)
δ
gN(η)
J(η)
dη
(72) ≈ −N nM IP # k wk
−gN
(η
k)
δ
gN(η
k)
J(η
k)
, (73) ≈ nM IP # k wk−gNk
δ
gNkJk, (74)where
η
∈[−1,1]denotesthecentroidpointcoordinateinthe pa-rameterspaceandJ= ∂x I0c∂η .Theweightandcoordinatesofthekth
integrationpointaredenotedbywkand
η
k,respectively.Giventhe section along M attachedto xc(
η
)
for whichwe solve the localproblemofEq.17,twosurfacepoints, ¯xMand ¯xN,arecomputed. InEq.73,rcekisexpressedas:
rcek=−
N
−gN
dgN
dpˆe. (75)
If the Automatic Differentiation (AD) formalism is employed [23,24],the dependency ofthe local variables withrespect to pˆe aswell asrelationsofEq.63 canbe directlyincorporated as fol-lows:
δ
gN=δ
¯g· ¯nI (76) =δ
¯xJ· ¯nI−δ
¯xI· ¯nI (77) = ˆ∂
xJ ˆ∂
pˆe ∂q ∂pˆe=A −∂
ˆxI ˆ∂
pˆe ∂q ∂pˆe=0 · ¯nI, (78) whereoperator ∂ˆˆ∂w denotesthe AutomaticDifferentiation(AD) of
Fig. 6. Example 1: (a) Initial configuration; (b) configuration halfway through the loading, and (c) final configuration.
Kcek,stemmingforthecontributionofthekth integrationpoint
placedalonghM1 ,canbeobtainedusingADasfollows: Kcek=
∂
ˆrcek ˆ∂
ˆpe ∂¯q ∂ˆpe=A . (79) rcandKc,whichcontainthecontributionsofall contactelements
insetS,areobtainedfrom: rc=A # e∈S nM IP # k wkrcek=A # e∈S rce, (80) Kc=A # e∈S nM IP # k wkKcek=A # e∈S Kce. (81)
where A $ denotes the finite-element assembly operator. Note that the exception indifferentiation in Eq.81 allows to properly linearisercek suchthattheexception ∂q
∂pˆe =0inEq.78isreplaced
by ∂q
∂pˆe =Aduringlinearization.
3. Results
Inthissection,thebeam-inside-beamcontactschemeisapplied totwo numericalexamples.First,a thinbeamispulledout from anotherbeaminwhichitisinitiallyinserted.Second,athinbeam isinsertedinacurvedbeam.Forthesetwoexamples,asingle in-tegrationpoint isusedto evaluaterce andKce in Eqs.80and81, respectively.
In both examples, large relative displacements ofthe contact-ingsurfacestakeplace.Thisimpliesthatforacontactelement,if projection point ¯xJ lies on thesurface ofBFE M, thisprojection
mightgooff theboundofthesurface
∂
M.Ifthishappens, projec-tionpoint ¯xJ shouldlie onan adjacentelement’ssurface, namelythesurfaceofelementBM+1orBM−1.However, astwo-node
ge-ometricallyexactbeamelementsareemployed,gapsandoverlaps ofthedifferent BFE’ssurfaces arepresentifthe beamisnot ini-tiallystraight.Forthisreason,itmightbedifficultorimpossibleto definethenewlocationof ¯xJ.
To palliate this problem, a dedicated surfacesmoothing tech-niquewas introduced in[29]. The resultingauxiliary surface has
C1-continuitywhichisconvenientforcontacttreatment.This
pro-cedureisusedinthefollowingexamples.
Due to the discretization of the contact kinematics, a sud-denloss of contactnear the inlet andoutlet of the outer beam may result in loss of convergence. The methodology presented inAppendix Ais thereforeadopted inthe following examplesto avoidthesecomplications.
3.1. Example1:Pullout
In the first example, two elastic beams with the same initial centroid-lineformapartofahelix (Fig.6a).Theyhavethesame centroid-line as a parameterized helix. The outer beam is hol-low with an elliptical cross-section defined by a=22.3mm and
b=17.8mm.Its wall thicknessis 0.2mm. Its Young’smodulus is settoE=0.15MPa[20]and75BFEsareemployedtodiscretizeit. Thethinbeamhasanellipticalsectionwithsemi-axesa=3.9mm is set to andb=3mm. It is stiffer than the hollow beamas its Young’s modulus is set to E=1.5 MPa. Its Poisson’s ratio is set to
ν
=0.3 and it is discretized with 100 BFEs. A penalty stiff-nessof1N/misused.Thedisplacementsandrotationsofbothend nodesoftheouterbeamarerestrained.Oneendofthethinbeam is pulledaway fromthe outer beamby 1200mm in300 equally spacedincrements.Contact interactions substantially deform the thin beam, (Fig. 6b). As the (prescribed) end node ofthe thin beam contin-uestomove awayfromtheouter beam, slidingofthecontacting surfaces occurs until the final configuration is reached(Fig 6 c). Astheprescribedbeamdeformsandmovesalongtheouterbeam, the number of penetrated sections changes (Fig. 7). The compo-nentsofthereactionforceandtorqueattheprescribedendofthe thinbeamarereportedinFig.8aandFig.8b,respectively.
3.2. Example2:Insertion
Thesecondexampleinvolvesaninitiallystraightthinbeamthat ispushed in a hollow,largely circularbeam(see Fig. 9). Initially, onlya smallpartofthethinbeamisinsertedinthe hollowone. The kinematic variables of the outer beam’s end node near the thinbeamare restrained.Thez-displacementoftheinnerbeam’s endnodefurthestawayfromtheouterbeamisprescribedtoreach 270mmin300increments,whilsttheotherkinematicvariablesat thisendnodearerestrained.
This inner beamhas a length of54cm, anda Young’s modu-lus of 1.5MPa.The cross sectional shape is given by a=5.4mm andb=4.3mm. The outer hollow beamis more compliant with
E=0.15 MPa. Its wall thickness is 1mm and its cross-sectional semi-axes are a=20mm and b=16mm. 100 and 180 BFEs are employedtodiscretizetheinnerandouterbeam,respectively.The Poisson’sratioofbothbeamsissetto0.33.
Fig. 7. Example 1: top: number of global iterations to reach the convergence criteria fint + r c − f ext< 10 −8 ; bottom: number of penetrated sections as a function of the
displacement of the end node of the thin beam. The peak in the number of iterations corresponds to the sliding of a cross-section of the slave beam in contact out of the hollow beam. The peak in the number of iterations corresponds to the sliding of a cross-section of the slave beam in contact out of the hollow beam. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 8. Example 1: (a) components of the reaction force and (b) reaction torque at the prescribed end of the thin beam.
Fig. 9. Example 2: Initial configuration.
determinethecontactlocationaredifficulttoapplyinthese situ-ations.Thereasonisthattheyrelyonthedeterminationofa min-imumofadistancefunctionthatisalmostconstantifsurfacesare closetoeach other onafiniteregion. Thetopdiagram ofFig.11, furthermore,showsthatonlyafewiterationsarerequiredtoreach convergence. This was the same as for the first example, as re-vealed in the top diagram of Fig. 7. The force-displacement and torque-displacementdiagramsofFig.12clearlyshowthatdifferent
regimesarepresent,whereeachregimeisgovernedbyadifferent numberofcontactareas.
Inthecurrentexample,theYoung’smodulusoftheinnerbeam is tentimes larger than that ofthe outer beam. To demonstrate that the framework is also robust for an entirely different ra-tio of Young’s moduli, the example is repeated withexactly the samegeometrical, material and numericalparameters, except for theYoung’smodulusoftheinnerbeam.Insteadof1.5MPa,weset themodulusto150MPasuchthatitisthousandtimeslargerthan thatoftheouterbeam.
The figures for this additional test case are reported in AppendixB.Theyshowthat, althoughthepredicteddeformations as well as the force- and torque-curves are completely different (cf.Figs.10,11and12),themaximumnumberofiterationsisagain notmorethantwo.
Meshconvergencestudy
Fig. 10. Example 2: (a) Initial configuration; (b) configuration halfway through the loading, and (c) final configuration.
Fig. 11. Example 2: Top: number of global iterations to reach convergence criteria: fint + r c − f ext< 10 −8 ; Bottom: evolution of the number of contact interactions
between sections of the thin beam and the surface of the outer beam.
differentmeshes,thedisplacementfieldoftheinnerbeamis com-paredtoareferencesolution,whichisobtainedwith180elements fortheinnerbeam.Figs.aandbshowthatthedisplacementfields convergetothereferencedisplacementfields.
4. Discussion
The beam-inside-beam contact framework presented in this contributionisthefirstapproachtoensurethataninnerbeam re-mainsembeddedinsideanouter,hollowbeambecauseallexisting contactframeworks forbeams aimto achieve the opposite: they repelbeamsfromeachother.
Theadvantageofusingbeamsinsurgicalsimulationsover con-ventional3Dfiniteelementsisthepotentialtoachievefaster sim-ulations (although our particular implementation can surely not compete with frameworks such as SOFA [11]). The disadvantage of using beams over conventional finite elements is a reduction of the simulation accuracy. Hence, the framework proposed in thiscontribution may be perceived to be beneficial if simulation speed is preferred over simulation accuracy. On the other hand, ourframeworkcanhandleslendermedicalinstrumentswithboth circular and elliptical cross-sections, whereas the frameworks of [1,9,15,27,39]haveonlybeendemonstratedtohandlecircularones. The reason that beams, formulated to rapidly simulate the me-chanicalbehaviorofslenderbodies,arefasterformechanical sim-ulationsinvolvingcontactistwofold.First,beamscomewithfewer degreesoffreedom(i.e.kinematicvariables)thanifthetubes’ sur-faces are representedby conventional finite elements, becausein most beam theories the beams’ cross-sections are rigid (i.e. the cross-sections cannot deform). Hence, the entire beam’s
Fig. 13. Effect of the mesh refinement on the final displacement of the nodes of the thin beam. The displacements in the final configuration are compared to the ones with the finest mesh (180 nodes) (a) Difference of the Y -displacements ; (b) Difference of the Z-displacements.
try can be constructedfrom its centroidlinedescription andthe cross-sections’ orientation. Thisdrasticallyreducesthenumberof degreesoffreedomnecessarytodiscretiseaslenderbody.Second, andseemingly even moreimportantis thefact that the penetra-tion can be quantified for an entire (rigid)cross-section atonce, whichdrasticallyreducesthenumberoflocalproblemsthatneeds tobesolved.
Onepossibleextensionofthebeam-inside-beamcontact frame-workistheincorporationoffluidflowsinsidethehollowbeamto representblood flowincaseofarteries andveins.Frictional slid-ingbetweentheinnerandouterbeamalsoseemslikeanecessary extensionforthefuture.Surroundingtissueswerefurthermore ne-glectedinthepresentedsimulations,whichdidnotfocusona par-ticulartype ofinterventioninorderto highlightthegeneralityof theframework.
The accuracy of the framework is not as high as that of the frameworks presented in [9,19,27,39]. However, simulation speed andsimulationaccuracyaretwocompetinginterestsandifspeed is preferredover accuracy,our newframeworkis apromising al-ternativetoexistingframeworks.
DeclarationofCompetingInterest
Theauthorsdeclarethattheyhavenoknowncompeting finan-cialinterests.
Acknowledgments
MarcoMagliulo,AndreasZilianandLarsBeexwouldliketo ac-knowledge thefinancial supportof theUniversity ofLuxembourg forprojectTEXTOOL.JakubLengiewiczgratefullyacknowledgesthe financial supportofEUHorizon2020Marie SklodowskaCurie In-dividual Fellowship MOrPhEM (H2020-MSCA-IF-2017, project no. 800150). This work was partially prepared in the framework of the DRIVEN project funded by the European Union’s Horizon 2020research andinnovation programmeunder grantagreement No.811099.
AppendixA. Treatmentofcontactattheendsoftheouter beam.
InthenumericalexamplesofSection3,thesectionatthetipof theinnerbeamslidesalongthewalloftheouterbeamuntilit ex-itstheouterbeam(firsttwoimagesinFig.14).Theoccurrenceof
thisis monitored,because whenit occurs, the contactconstraint betweenthe section attachedto the last integrationpoint of the innerbeamandthe outer beammustbe deactivated andthe in-crementrepeated.Then, nocontactinteractionsto embedthe in-ner beam inside the outer beam maybe left. Ifthis is thecase, a suddenrelease ofthe inner beammayoccur which makes the simulationdiverge(bottomleftinFig.14).
Toavoidthis,anadditionalconstraintisaddedtotheoutletof thehollowbeam. Itenforces thesection attheedgeoftheouter beamto be incontact withthe surface of the inner beam (bot-tom right in Fig. 14). The local problem to that must be solved to quantifypenetration is againgiven by Eq.17, exceptthat this timeBIdenotestheouterbeamandBJtheinnerone.Themethod
ofLagrangemultipliersisusedtoenforcethisconstraint. The rea-sonis thatifthe penalty methodisemployed andthesection at the end of the outer beam is detached from the surface of the
Fig. 15. Example 2 with a stiffer inner beam: (a) Initial configuration; (b) configu- ration halfway through the loading, and (c) final configuration.
AppendixB.FiguresforExample2withE=150MPaforthe innerbeam.
(a)
(b)
Fig. 16. Example 2 with a stiffer inner beam: Top: number of global iterations to reach convergence criteria: fint + r c − f ext< 10 −8 ; Bottom: evolution of the number of
Fig. 17. Example 2 with a stiffer inner beam: components of the reaction force (a) and torque (b) at the prescribed end of the thin beam.
References
[1] T. Alderliesten, M.K. Konings, W.J. Niessen, Simulation of minimally invasive vascular interventions for training purposes, Comput. Aid. Surg. 9 (1–2) (2004) 3–15, doi: 10.3109/1092908040 0 0 06408 .
[2] R. Alterovitz, K. Goldberg, A. Okamura, Planning for steerable bevel-tip needle insertion through 2D soft tissue with obstacles, in: Proceedings - IEEE Interna- tional Conference on Robotics and Automation, volume 20 05, 20 05, pp. 1640– 1645, doi: 10.1109/ROBOT.2005.1570348 .
[3] T.R. Coles, N.W. John, D. Gould, D.G. Caldwell, Integrating haptics with aug- mented reality in a femoral palpation and needle insertion training simulation, IEEE Trans. Haptic. 4 (3) (2011) 199–209, doi: 10.1109/TOH.2011.32 .
[4] S. Cotin, H. Delingette, N. Ayache, Real-time elastic deformations of soft tis- sues for surgery simulation, IEEE Trans. Vis. Comput. Graph 5 (1) (1999) 62–73, doi: 10.1109/2945.764872 .
[5] H. Courtecuisse, J. Allard, P. Kerfriden, S.P.A. Bordas, S. Cotin, C. Duriez, Real- time simulation of contact and cutting of heterogeneous soft-tissues, Med. Im- age Anal. 18 (2) (2014) 394–410, doi: 10.1016/j.media.2013.11.001 .
[6] H. Courtecuisse, H. Jung, J. Allard, C. Duriez, D.Y. Lee, S. Cotin, GPU-Based real- time soft tissue deformation with cutting and haptic feedback, Prog. Biophys. Mol. Biol. 103 (2–3) (2010) 159–168, doi: 10.1016/j.pbiomolbio.2010.09.016 . [7] M.A. Crisfield, G. Jeleni ´c, Objectivity of strain measures in the geometrically
exact three-dimensional beam theory and its finite-element implementation, Proc. R. Soc. Lond. Ser. A 455 (1983) (1999) 1125–1147, doi: 10.1098/rspa.1999. 0352 .
[8] S.P. DiMaio, S.E. Salcudean, Needle insertion modeling and simulation, IEEE Trans. Robot. Automat. 19 (5) (2003) 864–875, doi: 10.1109/TRA.2003.817044 . [9] C. Duriez, S. Cotin, J. Lenoir, P. Neumann, et al., New approaches to catheter
navigation for interventional radiology simulation, Comput. Aid. Surg. 11 (6) (20 06) 30 0–308, doi: 10.3109/10929080601090623 .
[10] C. Duriez, C. Guébert, M. Marchal, S. Cotin, L. Grisoni, et al., Interactive simula- tion of flexible needle insertions based on constraint models, in: Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelli- gence and Lecture Notes in Bioinformatics), volume 5762 LNCS, 2009, pp. 291– 299, doi: 10.1007/978- 3- 642- 04271- 3 _ 36 .
[11] F. Faure, C. Duriez, H. Delingette, J. Allard, B. Gilles, S. Marchesseau, H. Tal- bot, H. Courtecuisse, G. Bousquet, I. Peterlik, S. Cotin, SOFA: a multi-model framework for interactive physical simulation, 2012, pp. 283–321, doi: 10.1007/ 8415 _ 2012 _ 125 .
[12] A. Gay Neto, P.M. Pimenta, P. Wriggers, A master-surface to master-surface for- mulation for beam to beam contact. Part I: frictionless interaction, Comput. Method. Appl. Mech. Eng. 303 (2016) 400–429, doi: 10.1016/j.cma.2016.02.005 . [13] A. Gay Neto, P.M. Pimenta, P. Wriggers, A master-surface to master-surface for- mulation for beam to beam contact. Part II: frictional interaction, Comput. Methods Appl. Mech. Eng. 319 (2017) 146–174, doi: 10.1016/j.cma.2017.01.038 . https://linkinghub.elsevier.com/retrieve/pii/S0045782516303218
[14] A. Gay Neto, P. Wriggers, Computing pointwise contact between bodies: a class of formulations based on mastermaster approach, Comput. Mech. (0123456789) (2019), doi: 10.10 07/s0 0466- 019- 01680- 9 .
[15] O. Goury, Y. Nguyen, R. Torres, J. Dequidt, C. Duriez, Numerical simulation of cochlear-implant surgery: towards patient-specific planning, in: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelli- gence and Lecture Notes in Bioinformatics), volume 9900 LNCS, Springer Ver- lag, 2016, pp. 500–507, doi: 10.1007/978- 3- 319- 46720- 7 _ 58 .
[16] A. Ibrahimbegovi ´c, F. Frey, I. Kožar, Computational aspects of vector-like parametrization of three-dimensional finite rotations, Int. J. Numer. Method. Eng. 38 (21) (1995) 3653–3673, doi: 10.1002/nme.1620382107 .
[17] H. Irschik, J. Gerstmayr, A continuum-mechanics interpretation of Reissner’s non-linear shear-deformable beam theory, Math. Comput. Model. Dyn. Syst. 17 (1) (2011) 19–29, doi: 10.1080/13873954.2010.537512 .
[18] G. Jeleni ´c, M.A. Crisfield, Geometrically exact 3D beam theory: implementation of a strain-invariant finite element for statics and dynamics, Comput. Method. Appl. Mech. Eng. 171 (1–2) (1999) 141–171, doi: 10.1016/S0045-7825(98) 00249-7 .
[19] G. Kazmitcheff, Modélisation dynamique de l’oreille moyenne et des interac- tions outils organes pour la conception d’un simulateur appliqué à la chirurgie otologique, 2014 Ph.D. thesis. https://hal.inria.fr/tel-01105262
[20] T. Khamdaeng, J. Luo, J. Vappou, P. Terdtoon, E.E. Konofagou, Arterial stiffness identification of the human carotid artery using the stress-strain relationship in vivo, Ultrasonics 52 (3) (2012) 402–411, doi: 10.1016/j.ultras.2011.09.006 . [21] A. Konyukhov, O. Mrenes, K. Schweizerhof, Consistent development of a beam-
to-beam contact algorithm via the curve-to-solid beam contact: analysis for the nonfrictional case, Int. J. Numer. Method. Eng. 113 (7) (2018) 1108–1144, doi: 10.1002/nme.5701 .
[22] A. Konyukhov, K. Schweizerhof, Computational Contact Mechanics, Lecture Notes in Applied and Computational Mechanics, volume 67, Springer Berlin Heidelberg, Berlin, Heidelberg, 2013, doi: 10.1007/978- 3- 642- 31531-2 . [23] J. Korelc, Multi-language and multi-environment generation of nonlinear
finite element codes, Eng. Comput. 18 (4) (2002) 312–327, doi: 10.1007/ s0 0366020 0 028 .
[24] J. Korelc, P. Wriggers, Automation of Finite Element Methods, Springer Interna- tional Publishing, Cham, 2016, doi: 10.1007/978- 3- 319- 39005- 5 .
[25] T.A. Laursen, J.C. Simo, A continuum-based finite element formulation for the implicit solution of multibody, large deformation-frictional contact prob- lems, Int. J. Numer. Method. Eng. 36 (20) (1993) 3451–3485, doi: 10.1002/nme. 1620362005 .
[26] J. Lengiewicz, J. Korelc, S. Stupkiewicz, Automation of finite element formula- tions for large deformation contact problems, Int. J. Numer. Method. Eng. 85 (10) (2011), doi: 10.10 02/nme.30 09 .
ory(2016) 1–92. 1609.0011910.1007/s11831-017-9232-5.
[33] C. Meier, W.A. Wall, A. Popp, A unified approach for beam-to-beam contact, Comput. Method. Appl. Mech. Eng. 315 (August) (2017) 972–1010, doi: 10.1016/ j.cma.2016.11.028 .
[34] C. Mendoza, C. Laugier, Simulating soft tissue cutting using finite element models, in: Proceedings - IEEE International Conference on Robotics and Au- tomation, volume 1, Institute of Electrical and Electronics Engineers Inc., 2003, pp. 1109–1114, doi: 10.1109/robot.2003.1241741 .
[35] I. Romero, The interpolation of rotations and its application to finite element models of geometrically exact rods, Comput. Mech. 34 (2) (2004) 121–133, doi: 10.10 07/s0 0466-0 04- 0559- z .
[40] P. Wriggers, Computational Contact Mechanics, Springer Berlin Heidelberg, Berlin, Heidelberg, 2006, doi: 10.1007/978- 3- 540- 32609- 0 .
[41] P. Wriggers, G. Zavarise, On contact between three-dimensional beams under- going large deflections, Commun. Numer. Method. Eng. 13 (6) (1997) 429–438, doi: 10.1002/(SICI)1099- 0887(199706)13:6 < 429::AID- CNM70 3.0.CO;2- X . [42] W. Wu, P.A. Heng, An improved scheme of an interactive finite element model
for 3D soft-tissue cutting and deformation, Vis. Comput. 21 (8–10) (2005) 707– 716, doi: 10.10 07/s0 0371-0 05-0310-6 .