• Aucun résultat trouvé

Using the witness method to detect rigid subsystems of geometric constraints in CAD

N/A
N/A
Protected

Academic year: 2021

Partager "Using the witness method to detect rigid subsystems of geometric constraints in CAD"

Copied!
24
0
0

Texte intégral

(1)

HAL Id: hal-00691703

https://hal.archives-ouvertes.fr/hal-00691703

Submitted on 26 Apr 2012

HAL

is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or

L’archive ouverte pluridisciplinaire

HAL, est

destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires

Using the witness method to detect rigid subsystems of geometric constraints in CAD

Dominique Michelucci, Simon Thierry, Pascal Schreck, Christoph Fünfzig, Jean-David Génevaux

To cite this version:

Dominique Michelucci, Simon Thierry, Pascal Schreck, Christoph Fünfzig, Jean-David Génevaux.

Using the witness method to detect rigid subsystems of geometric constraints in CAD. Symposium

on Solid and Physical Modeling, 2010, Haïfa, Israel. pp.91-100, �10.1145/1839778.1839791�. �hal-

00691703�

(2)

subsystems of geometri onstraints in CAD

Dominique Mihelui

Pasal Shrek

Simon E.B. Thierry

Christoph Fünfzig

*

Jean-David Génevaux

September1 st

3 rd

,2010

Abstrat

Thispaperdeals withtheresolution ofgeometri onstraintsystems

enounteredinCAD-CAM.Themainresultsarethatthewitnessmethod

an be used to detet that a onstraint system is over-onstrained and

thattheomputationofthemaximalrigidsubsystemsofasystemleads

toapowerfuldeompositionmethod.

In a rst step, we reall the theoretial framework of the witness

methodingeometri onstraint solvingand extendthis methodto gen-

erateawitness. Weshowthenthat itanbe usedto inrementallyde-

tetover-onstrainedness. Wegiveanalgorithm toeientlyidentifyall

maximalrigidpartsofageometrionstraint system. Weintroduethe

algorithmofW-deompositiontoidentifyallrigidsubsystems: itmanages

to deompose systemswhihwerenotdeomposableby lassial ombi-

natorialmethods.

Keywords:GeometriConstraintsSolving,witnessonguration,Ja-

obianmatrix,rigiditytheory,W-deomposition

1 Introdution

GeometrionstraintssolvinginComputer-AidedDesign(CAD)aimsatyield-

ingagurewhihmeetssomeinideneandmetrirequirements(e.g.distanes

betweenpointsoranglesbetweenlines),usuallyspeiedingraphialform.For-

mally,ageometrionstraintsystem(GCS)onsistsinonstraints(prediates),

unknowns (geometri entities) and parameters (metri values). Solutions are

returnedastheoordinatesofthegeometrientities. Theleftofgure1shows

anexampleofatehnialsketh, andtherightshowsapossiblesolution.

Theliteraturedesribesanumberofdierentapproahestosolvegeometri

onstraintsystems:

LE2I,UMRCNRS5158,UniversitédeBourgogne

LSIIT,UMRCNRS7005,UniversitédeStrasbourg

(3)

p1

p1 p2

p2 p3

p3

p4 p4

p5 p5

p6 p6

7

5

9

8 6

7 135

120

115

Figure1: A2Dtehnialsketh(left)andapossiblesolution(right).

algebraimethods onsistin translatingthe GCSinto aset ofequations andworkingontheequationsystem,thusforgettingthegeometrialbak-

ground. Algebraimethodsanbelassiedinnumerialmethods[22℄(it-

erativeomputations onverging to anapproximate solutionfrom initial

valuesgivenby theuser)andsymbolimethods[2, 11℄(diret omputa-

tionsontheequations thesemethods areseldom usedbeauseoftheir

omplexity),

geometrimethodsusethegeometriknowledgetosolvethesystem:graph-

basedmethods[6,9,22,28,29,31℄ompilethisknowledgeintoalgorithms

whih onsider only ombinatorial and onnetivity riteria, rule-based

methods[3,17℄dedueonstrutionsplansbyanexpliituseofgeometri

rules,

hybridmethods[4,8,18℄alternatealgebraiandgeometriphasesofom-

putationstousethepowerofbothapproahes.

For moredetailsongeometri onstraintsolving,see[12℄. Ageneraltrend,

both to redue omplexity and to enhane resolution power, is to deompose

theGCS into solvablesubsystems and to assembletheir solutions[4, 5, 9, 13,

15,22,28,29,31,33℄. Forinstane,onthe2Dexampleofgure1,itiseasyto

separatelysolveeahtriangle (p1p2p6,p2p3p4 andp4p5p6)andthenassemble

them. Foradetailedsurveyofdeompositionmethods,see[16℄.

Notiethat,ontheexampleofgure1,ifone removesoneof thetriangles,

sayp2p3p4, andthen triesto solvetheremainingsystem,one needstoadd in-

formationfromthesolvedsubsystem,otherwisetheremainingsystembeomes

artiulated. Thispieeofinformationisalledtheboundary[24℄. Althoughsev-

eralmethodsexisttondtherelevantinformationinspeiresolutionframe-

works [28℄, no generalalgorithm yet exists to omputethe boundary without

addingtoomuh information.

Indeed, it is important for resolution methods, espeially for graph-based

methods, that the system does not have too few or too many onstraints.

Looselyspeaking, asystemisalled

(4)

under-onstrainedifit hasan innitenumberof solutionsbeausethere arenotenoughonstraintstopindowneverygeometrientity,

over-onstrainedifithasnosolutionbeauseofonstraintontraditions,

well-onstrainedifithasanitepositivenumberofsolutions.

Invarianeofrigidsystemsbydisplaementsisgenerallytakenintoaountby

anhoring a point and a diretion in 2D, a point and two diretions in 3D.

Thepointandthediretionarealledareferene forthedisplaements. Other

transformationgroupsmaybeonsidered [30℄.

Alotofworkhasbeendoneaboutthedetetionofover-onstrainedness[14,

27℄orunder-onstrainedness[19,32,37℄ andmoregenerallyaboutthehara-

terization of rigidity [21, 20,30, 35℄. Yet, methods desribed in the literature

may fail to onsider the onsequenes of mathematial theorems that are not

expliitly taken into aount in the onstrution of the resolution algorithm.

Sine a theorem list annot be exhaustive, it is impossible to develop arule-

basedor graph-based algorithm that detets geometri properties induedby

mathematialtheorems.

Inthisartile,weextendthewitnessmethod[25℄toaddressseveralproblems

ited above: how to determine the onstrainedness level of a GCS without

being triked by mathematial theorems (see for instane gure 6); how to

eientlydetetallmaximalwell-onstrainedsubsystemsofagivenGCS;how

to deompose a well-onstrained system into the set of all its minimal well-

onstrainedsubsystems.

For onisenessreasons,in the rest ofthis paper,weonsider 2D systems,

unlessexpliitlymentionnedotherwise. Yet,allalgoritmsanbeextendedto3D

systemswithnearlynohangesand,mostofthetime,theonlymodiationto

bemadeforthetexttobevalidin 3Distoexhangementionsofthreedegrees

offreedom/parameterswithmentionsofsixdegreesoffreedom/parameters.

Thisartileisorganizedasfollows: setion2reallsthepriniplesofthewit-

nessmethodandgivesawaytogenerateawitness;setion3demonstratesthat

aninremental versionof the Gauss-Jordanelimination has the sameompu-

tationalostthantheoriginalversionbutallowstodetetoveronstrainedness

in allases;setion 4givesalgorithmsto eiently identify themaximalrigid

subsystems of anartiulated system; setion 5dedues from these algorithms

a method to further deompose a rigid system into rigid subsystems; nally,

setion7onludesandgivesperspetivestothiswork.

(5)

2.1 Priniple

ThewitnessmethodomesfromideasofStruturalTopology,orRigidityThe-

ory [10℄where the questionofrigidityis studied throughthenotion offrame-

works. Aframework isatriple(V, E, p)where(V, E)isagraphandp:V →Rd

arealizationofthegraph,whihmapsthevertiesofV topointsofdimension d. Thinking ofgraphedgesasrigidbarsandofvertiesasartiulationpoints, themain goalof ombinatorial rigidityis toanswerIs (V, E, p)rigid?, i.e. it

admitsonlyrigidmotionsasawhole,nodeformations.

Innitesimal exion. InRigidity Theory, aninnitesimal exion is amap

q : V → Rd suh that (p(i)−p(j))·(q(i)−q(j)) = 0, for eah (i, j) ∈E. A

framework is alled innitesimally rigid, if theonlyinnitesimal exionsarise

fromthediretisometriesofRd,i.e. thetranslationsand rotations.

Under mild assumptions onerning inidene relationships, if one frame-

work (V, E, p0) isinnitesimally rigidthen almostall frameworks (V, E, p)are

innitesimally rigid. And theinnitesimal rigidity impliesthe rigidity of the

framework. Note thatthere areounter-examplesfor theonverse, whih on-

tainspeialinidenes.

Inotherwords,aframeworkinrigiditytheoryorrespondstotherealization

ofageometrionstraintsystemwhereallonstraintsarepoint-to-pointdistane

onstraints: suhasystemisgeneriallywell-onstraineduptodiretisometries

ifit is generiallyrigid. This was generalizedby Mihelui et al. [25, 26℄ to

metrionstraintsoverpoints,lines,et.(distanesandangles)andtoinidene

onstraints(olinearitiesin2Dand3D,oplanaritiesin3D).

InCAD whenthe designer drawsasketh, he/she hasasolutionX0 for a

systemF(X, Ae) = 0,withsomeparametervaluesAereadonthesketh. Then

the goalis a solutionfor the system F(X, Aa) = 0, where Aa are the values

givenforthedimensioning.

Witness. Let F(X, A) = 0 be a onstraint system, where X are the un-

knowns and A theparameters. We suppose that F(X, A) is dierentiable. A witness isthenasolutionX0 ofF(X, A) = 0forsomeparametervaluesAe.

Using aTaylorexpansionfora small perturbation aroundthe solutionX0

ofF(X, Ae) = 0,wehave

F(X0+εv, Ae) =F(X0, Ae) +εF(X0, Ae)v+O(ε2)

wherev analso beseenasthe instantveloityof eah objetinvolved in the

systemandεisasmalltimestep. Thus,ifaninnitesimallysmallperturbation isanothersolutionofF(X, Ae),wemust have

F(X0, Ae)v= 0

Thespaeoftheinnitesimalmotionsallowedbytheonstraintsatthewitness

isthengivenbyker(F(X0, Ae)). Notethat

thematrixF(X0, Ae)is knownastheJaobianofsystemF(X, Ae) = 0

takenatpointX0;

(6)

whenallonstraintsarepoint-to-pointdistanes,theJaobianistherigid- itymatrixonsideredin RigidityTheory;

for otheronstraintswith parametersthegeneriityonditionsare more

ompliatedthan intheombinatorialase: aparametervalueAeand a

orrespondingsolutionX0aregeneriiftherootisanimpliitfuntionof

theparametersinsomeopenneighborhoodof(X0, Ae);forinstane,fora

trianglespeiedwiththreelengthparameters,thisonditionforbidsthat

onelengthisthesumoftheothers;moregenerallythisonditionimplies

thatthematrix

∂F(X, A)/∂X ∂F(X, A)/∂A

0 Id

has thesamerankin anopen neighborhood of (X0, Ae)It remains that

the generi parameter values are dense in the set of parameter values

orrespondingtoarealization.

Wegivesomeexamplesfortheformulationofgenerionstraints. Forpoint,

line,planeinidenes,weassumethattheorrespondingonstraintsarespei-

edexpliitlywithoutparameters. Thisistoavoidexpressingpoint-pointini-

denesbyadistaneonstraint(P1,x−P2,x)2+ (P1,y−P2,y)2=d2withdistane

parameterd= 0. Foradistaneonstraint(P1,x−P2,x)2+ (P1,y−P2,y)2=d2,

the parameterd = 0 is not generi, asthe onstraint is singular at the solu-

tionpoint. Foran angle onstraint angle(P1, P2, P3) = θ, i.e. P1P2·P3P2 = lP1P2lP3P2cosθ, the parameter valuesθ =±π, θ = ±π/2, and θ = 0 are not

generi. Similarly, point-line, line-plane inidenes and line-line, plane-plane

parallelism/orthogonalityonstraintsarenotexpressedbyangleonstraintsbe-

auseitwouldintroduenon-generiangles.

Typiality. A witnessistypial ifit isrepresentativeforthe searhedsolu-

tion,i.e. it hasthesameombinatorialproperties (oinidenes,ollinearities,

oplanarities, et.). So a random solution (X0, Ae), {(X, A) : F(X, A) = 0}

with the speiedombinatorial properties is typial with probability 1 for a

setof witnesssolutions. Notethat systemsexistwith witnesssolutions,whih

aredierentinombinatorialproperties,and noontinuousdeformationexists

totransformoneintotheother. Foranexampleofsuhasystemsee gure14

in[16℄.

We an then study the degrees of freedom of the system by studying the

rank of the Jaobian F(X0, Ae) on a typial witness X0, and in the ase of

under-onstrainedness,thestrutureoftheallowedinnitesimalmotionsanbe

deduedfromthestudyofthekernelofF(X0, Ae).

In the rest of this paper, we onsider that rows of the Jaobian matrix

representonstraintsand olumnsrepresentoordinatesofthe unknowns. We

lassiallydenotebymthenumberofrowsandbynthenumberofolumnsof

thematrix.

(7)

Theskethisusuallyawitnessbutduetoimpliedinidenesthismaynotbethe

ase. Inthisase,wesolvetheunder-determinedsystem{(X, A) :F(X, A) = 0}

forawitness(X0, Ae). Inthesubdivision solverpresentedin [7℄, thenonlinear monomialsx2i andxixjfori < jarereplaedbyadditionalvariablesxi,iandxi,j,

whihareenlosedin apolytopeBD(xi, xi,i, xi,j,i<j)≥0 withhalfspaesgiven

bythe non-negativity of relevant Bernstein polynomials (Bernstein polytope).

The quadrati onstraint system beomes a polytope S(xi, xi,i, xi,j,i<j) ≥ 0

after rewriting into the additional variables xi,i and xi,j. Thesubsript D of BD(xi, xi,i, xi,j,i<j)≥0indiatesthatthispolytopedependsonthedomainD.

In this way, bounds for the solutiondomain of quadrati polynomialsan be

expressedastwolinearprograms

minxi and maxxi

S(xi, xi,i, xi,j,i<j)≥0 BD(xi, xi,i, xi,j,i<j)≥0

Domainboundsareomputedbylinearprogramminginordertoreduethe

urrent solutiondomain D. If the feasible set is empty, whih is detetedby

linearprogramming,thentheurrentdomainboxontainsnosolution. Other-

wise,weanperformasequeneof redutions andbisetions ofdomain boxes

untilthedomain boxD = [x1, x1]×. . .×[xn, xn] isδ-small: (xi−xi)< δ for

alli. These δ-smallboxesoverthesolutionsetpieewise.

The subdivision solverrequires adomain box to start thesearh. The in-

tervalsfor generi parameter valuesof onstraintsare easy to nd: angle pa-

rameterscosθ(cosθinsteadofθtoavoidtrigonometrifuntions inthesolver) arein [−1 +ǫ,−ǫ]or [ǫ,1−ǫ] with asmall, arbitrary ǫ; intervalsfor distane

parametersdanbeobtainedfrommagnitudeboundsofthepointoordinates.

Findingaboundonthemagnitudeofanyroot[36℄,wouldbeneessarytoprove

thatthe systemhasnosolution. Fortheproblems here,abound onthepoint

oordinatesisknownbeforehand.

Inordertoenumerateallsolutionsofasystem,weusedmid-bisetionofthe

largestintervalin [7℄, whih minimizestheheightof theexplorationtreewhile

ylingthroughdimensions. Fortheaseofdeterminingasinglesolutionasfast

aspossible,thehoie ofthesmallestinterval(greaterorequalδ)isbeneial

assettingvariablestovaluesallowingsolutionsimprovestheeetivenessofthe

domainredutionstep.

We selet the next domain box (of smallest minimum side length greater

thanδ)for redutionand bisetion atrandom. Inthis way,wendasolution

boxontaininga randomsolution, and we take thebox enter projetedonto

thesolutionsetasawitness.

As examples, weshow two systems of dierent diulty. In gure 2, two

triangles with a ommon point p0 are speied by six side lengths. In the

randomsolution,thesidelengthsare alldierent. Ingure3,four pointsand

velineswith10point-lineinidenesarespeiedbyfourangleparametersand

adistane parameter. Theleft partshowsasolutionwithsymmetriandnie

(8)

p0 p1

p2 p3

p4

Figure2: Thebuttery: 2Dsystemwith5pointsand 6distane parameters

d(p0, p1),d(p1, p2),d(p2, p0), d(p0, p3),d(p3, p4),d(p4, p0).

p

q r

c

p

q r c

Figure3:2Dsystemof4pointsand5lineswith10point-lineinidenes,4angle

parameterangle(qp, cp),angle(cp, rp),angle(rq, cq),angle(cq, pq)and1distane

parameterd(r, c). Symmetri solution(left)and random,typialwitnesssolu-

tion(right).

(9)

thetrianglepoints. Intherightpart,atypialwitnesssolutionisshown,whih

wasfoundat random. Itisusedforfurther analysis.

3 Over-onstrainedness

Wealreadyshowedin setion1that thedetetionofover-onstrainednessis a

ompliatedyet essentialproblemin theeld ofgeometrionstraintssolving.

In this setion, we show that the use of the witness method leads to an

eientandrobustdetetionofredundanyin geometrionstraints.

Wealsoshowtheusefulnessofthewitnessmethodtoenhanerobustnessof

deompositionmethodsbyanaurateomputationoftheboundary.

3.1 Inremental detetion of redundany

Weshowedin[25℄ thatitispossibleto interrogateawitnessinordertodetet

whetherasetofonstraintsisdependentornot.Indeed,itispossibletoompute

the rank of the Jaobian matrix at the witness and to ompare it with the

number of onstraints. However, nding a maximal independent subset of a

dependentset isnotatrivialproblem. Working onthewitness,thenaiveidea

wouldbetotryand removeonstraintsonebyoneand,ateahstep,ompute

therankagainto determineif theonstraintisredundant withtheremaining

set. If therank of S −c equalsthe rankof S, then onstraintc isredundant

andanberemoved. Performedthisway,theremovalofredundantonstraints

is expensive. Yet, onsidering an inremental onstrution of the geometri

onstraint system allows to identify the set of redundant onstraints with no

additionalostsinomparisonto thebasidetetionof redundany.

Indeed, onsider a geometri onstraintsystem S with no redundany be-

tweentheonstraints. ApplyingtheGauss-Jordan eliminationmethod on the

Jaobianmatrix at the witnessleads to amatrix J = (IP) with I am×m

diagonal matrixand P a m×f matrix, f =n−m being the number of a-

tualdegrees of freedom of the system. This method hasa known omplexity

of O(min(n, m)nm). Let us nowonsider a systemS with S ⊂ S. Inorder

to knowifS is over-onstrained,oneonlyneeds to inrementallyadd thege- ometrientities andtheonstraints(bearing in mindthat aonstraintanbe

insertedonlywhen the geometrientities it onernsare all in thesystem) of

S− S to S and applying Gauss-Jordanagain. Sinethe leftmost partof the

matrixis thediagonal, thenumber ofoperations is at most2 min(m, n)f: for

eahrowofI,eahnon-zeroelementofP mustbemultipliedandaddedtothe

newrow. Thenumberof operationsisin fat farsmaller,sinethenumberof

zeroelementsin thenewrowofthematrixishigh.

Proeeding inrementallydoesnot raise thenumber ofoperations: it only

hangestheorderoftheoperations. Indeed,thelassialGauss-Jordanelimina-

tionmethodonsistsinolumn-by-olumnoperations: foreaholumnc,divide

rowc byJc,c,thensubstratJr,ctimesthisnewrowfromrowrforeveryr,so

Références

Documents relatifs

The parameters you can change in set up include communication and printing features, and also such functions as terminal control mode, display characteristics,

More specifically, we explored potential candidate, job and employer-side moderators of hiring discrimination against burnout patients and developed a more

in which the disks are copied is you sequence of the following instructions, undesirable Log onto a disk drive which contains the TurboDOS occur.. COPY.CMD or .COM program in

of file name display on page - default default is ON, so turns off name display Htext - Define Heading text to appear at the top of each page deliin = for example user Files

the Instant-C program, the library source files, a stand-alone version of the Instant-C editor, and screen programs your keyboard for configuring PC have IBM an or compatible, if

If no damage is apparent, open the shipping container and remove documentation, mounting hardware, and cables.. Check materials received in option

IN GLOBAL RE' ER£NCE ILLEGAL NOT ON GLOBAL ?ILE DISK ROUTINE NOT ON DISK DISK I/O ERROR ROUTINE IS UNNAMED ROUTL1E ALREAD'Í j;q LIBRARY REMOVE OR SAVE ROUTINÍ COMMAND ONLY USED

COmmunications Management User's Guide binder for more specific information about remote access connections and setups using the optional AT&amp;T UNIX PC