scieee Open visual document viewer

An optimal design problem in wave propagation

Bellido Guerrero, José Carlos; Donoso Bellón, Alberto

Abstract

In this paper we consider an optimal design problem in wave propagation proposed in [1] SIGMUND O. and JENSEN J.S. Systematic Design of Phononic Band-Gap Materials and Structures by Topology Optimization, The Royal Society of London, Philosophical Transactions, Series A, Vol. 361, pp. 1001–1019, 2003 in the one-dimensional situation: Given two materials at our disposal with different Young’s modulus and different density, the problem consists of finding the best distributions of the two initial materials in a rod in order to minimize the vibration energy in the structure under periodic loading of driving frequency Ω. We comment on relaxation and optimality conditions, and perform numerical simulations of the optimal configurations. We also prove the existence of classical solutions in certain cases.

Full text

XX Cong eso de Ecuaciones Di e enciales y Aplicaciones X Cong eso de Ma em´ a ica Aplicada Se illa, 24-28 sep iemb e 2007 (pp. 1–8) An op imal design p oblem in wa e p opaga ion J.C. Bellido1, A. Donoso1 1Dp o. Ma em´a icas (ETSII), Uni e sidad de Cas illa la Mancha, Edi icio Poli ´ecnico, A da. Camilo Jos´e Cela s/n, 13071, Ciudad Real. E-mails: [email p o ec ed], [email p o ec ed]. Palab as cla e: Op imal design, wa e p opaga ion, classical solu ions Resumen In his pape we conside an op imal design p oblem in wa e p opaga ion p oposed in [1] in he one-dimensional si ua ion: Gi en wo ma e ials a ou disposal wi h di - e en Young’s modulus and di e en densi y, he p oblem consis s o inding he bes dis ibu ions o he wo ini ial ma e ials in a od in o de o minimize he ib a ion ene gy in he s uc u e unde pe iodic loading o d i ing equency Ω. We commen on elaxa ion and op imali y condi ions, and pe o m nume ical simula ions o he op imal con igu a ions. We also p o e he exis ence o classical solu ions in ce ain cases. 1. In oduc ion In [1] a s uc u al model was p oposed o he design o ini e s uc u es made up o wo gi en ma e ials in he con ex o wa e p opaga ion. The model is an op imal design p oblem in which he dis ibu ion o wo gi en ma e ials is op imized so as o minimize a cos unc ional ela ed o he ib a ion o p opaga ion o wa es along he medium. In p ac ice, his model may be use ul o he sys ema ic design o wa e il e s, damping o wa es, o wa e guides. The au ho s de elop a nume ical me hod o he op imiza ion o hose s uc u es based on opology op imiza ion (see [2]), bu he e is no ma hema ical analysis o he model. In he nume ical examples, hey also obse e he su p ising ac ha no mic os uc u e appea s be ween he wo ma e ials when one ies o design a il e o o minimize he ib a ion ene gy. Ou aim he e is o analyze ma hema ically he model p oposed in [1] in he one-dimensional si ua ion o longi udinal p opaga ion. We conside he one-dimensional wa e equa ion ρw = (Ewx)x, 1 J.C. Bellido, A. Donoso whe e is he ime a iable, x he space a iable, wis he displacemen , and Eand ρ a e he Young’s modulus and ma e ial densi y espec i ely. No ice we do no conside any damping e m in ou equa ion. We a e in e es ed in ime-ha monic solu ions unde pe iodic loading o he p e ious equa ion, i.e. in solu ions o he o m w(x, ) = u(x) exp(iΩ ), whe e Ω is he d i ing equency and u he ampli ude unc ion. Then he ampli ude, u, e i ies he equa ion E(x)u00+ Ω2ρ(x)u= 0. Now, le us assume ha we ha e wo ma e ials a ou disposal, and consequen ly he elas ic coe icien , E(x), and he densi y, ρ(x), ake wo alues on, E1, E2(0 < E1, E2) and ρ1, ρ2(0 < ρ1, ρ2) espec i ely. We make he echnical assump ion E1≤E2, ρ1≤ρ2, and ha a leas one o he inequali ies is s ic (o he wise we would jus ha e one ma e- ial). I we ha e a od o leng h l, ha we ep esen by he in e al [0, l], and we dis ibu e he wo ma e ials in he od, he elas ic coe icien unc ion and he densi y unc ion a e espec i ely E(x) = E1χ(x) + E2(1 −χ(x)), ρ(x) = ρ1χ(x) + ρ2(1 −χ(x)), whe e χis he cha ac e is ic unc ion o he (measu able) subse o he od whe e we place he i s ma e ial. Now i is na u al o conside he ollowing op imal design p oblem, (P) MinAad J(χ), whe e he admissible se o designs is Aad ={χ: cha ac e is ic unc ion o a measu able subse o [0, l]}, he cos unc ional is he ib a ion ene gy J(χ) = Zl 0Ω2(ρ1χ(x) + ρ2(1 −χ(x)))|u(x)|2+ (E1χ(x) + E2(1 −χ(x)))|u0(x)|2dx, and uis compu ed om χ h ough he s a e equa ion [(E1χ(x) + E2(1 −χ(x)))u0]0+ Ω2(ρ1χ(x) + ρ2(1 −χ(x)))u= 0, (E1χ(0) + E2(1 −χ(0)))u0(0) = γ, (E1χ(l) + E2(1 −χ(l)))u0(l) = 0, γis a ixed gi en nonze o numbe . He e we conside he case o bounda y condi ions co esponding o one ex e nal load on he le ex eme o he od, and homogeneous Neumann bounda y condi ions on he o he ex eme. Roughly speaking, he op imal design p oblem consis s o de e mining he bes dis ibu ion (i.e., he bes χ) o he wo gi en ma e ials in he od in o de o minimize he ib a ion ene gy in he s uc u e, so ha we minimize he ib a ion ampli ude along he od. No ice ha ou op imal design p oblem indeed is an op imal design p oblem o an ODE, as we ha e simpli ied he wa e equa ion by conside ing ime ha monic solu ions. 2 Op imal design in wa e p opaga ion We do no conside any olume cons ain in his p oblem, and his is in acco dance wi h he physical na u e o he p oblem since wa es p opaga e be e h ough homogeneous ma e ials han h ough mix u es o wo ma e ials. Howe e he p oblems makes pe ec sense i we impose a es ic ion on he amoun o he ma e ials ha we may use, as o ins ance he cons ain Zl 0 χ(x)dx ≤δ, o Zl 0 χ(x)dx =δ, wi h δ∈(0, l) a ixed cons an . The solu ions o he s a e equa ions a e unde s ood as weak solu ions. Tha equa ion is ob iously nonellip ic; howe e by he F edholm’s al e na i e, we can claim he exis ence o a unique solu ion o any bounda y da a i and only i Ω2is no an eigen alue o he p oblem −(E1χ(x) + E2(1 −χ(x)))u00=λ(ρ1χ(x) + ρ2(1 −χ(x)))u, u ∈H1(0, l),(1a) (E1χ(0) + E2(1 −χ(0)))u0(0) = 0,(1b) (E1χ(l) + E2(1 −χ(l)))u0(l) = 0.(1c) Recall ha , as he ope a o in ol ed is compac and sel -adjoin , we know ha he e exis s an inc easing (maybe no s ic ly) sequence o eigen alues. Fo he sake o simplici y, and in o de o o mula e he op imal design p oblem, we will assume ha Ω2is a away o any eigen alue o he p oblem (1a-1c) o any cha ac e is ic unc ion χ. This hypo hesis makes physical sense (see [1]). I is well-known ha in p inciple we canno hope he p e ious op imal design p oblem o admi op imal solu ions, and in gene al we need o s udy elaxa ion o i (see [4] and he e e ences he ein). The plan o he pape is he ollowing. Sec ion 2 is de o ed o s a e a elaxa ion esul o ou op imal design p oblem and o gi e necessa y condi ions o op imali y o he elaxed o mula ion o he p oblem. In Sec ion 3 we ob ain quali a i e p ope ies o he op imal solu ions o he elaxed p oblem, being able o p o e, in ce ain cases, ha he e a e op imal solu ions in he o m o cha ac e is ic unc ions, wha implies he exis ence o minimize s o he o iginal op imal design p oblem (P). Finally, in Sec ion 4 we pe o m nume ical simula ions by using a g adien me hod based on he sensi i i ies compu ed in Sec ion 2 and analyze nume ically examples o p ac ical in e es . 2. Relaxa ion Relaxa ion o op imal con ol o op imal design p oblems o ODE’s is a classical subjec . In his sense ou op imal design p oblem i s in o his amewo k and elaxa ion o i is s aigh o wa d, and so o p o e he elaxa ion esul s a ed below is s anda d (basic e e ences on he subjec a e [4, 5, 6]). Theo em 2.1 The op imal design p oblem (˜ P) MinA? ad ˜ J(E(θ), ρ(θ)), 3 J.C. Bellido, A. Donoso whe e he admissibili y se is A? ad ={θ∈L∞(0, l) : 0 ≤θ≤1}, E(θ) and ρ(θ) a e gi en by E=E(θ) = E1E2/(E1+θ(E2−E1)), ρ =ρ(θ) = ρ1θ+ρ2(1 −θ), he cos unc ional is ˜ J(E, ρ) = Zl 0Ω2ρ(x)|u(x)|2+E(x)|u0(x)|2dx, and uis compu ed om he pai (E, ρ) h ough he s a e equa ion Eu00+ Ω2ρu = 0,(2a) E(0)u0(0) = γ, E(l)u0(l) = 0 (2b) is a elaxa ion o he op imal design p oblem (P). We will need o he nume ical simula ions he g adien o he cos unc ional ˜ Jin he elaxed p oblem (˜ P). I s compu a ion is di ec by using he classical adjoin me hod. Theo em 2.2 The unc ional ˜ Jin p oblem (˜ P) is Gˆa eaux di e en iable on he se A? ad, and he Gˆa eaux de i a i e a ¯ θin he di ec ion o he admissible a ia ion θ(admissible a ia ions means ha θ∈L∞(0, l) such ha he e exis s h0>0 such ha o e e y h≤h0,¯ θ+hθ ∈ A? ad) is gi en by ˜ J0(¯ θ;θ) = Zl 0 {[E0(¯ θ(x))(¯u0(x)¯p0(x) + |¯u0(x)|2) + Ω2(ρ2−ρ1)(¯u(x)¯p(x)− |¯u(x)|2)]θ(x) −(E0(¯ θ(x))¯u(x)¯p(x)θ(x))0}dx, whe e ¯uis he s a e associa ed o ¯ θ(i.e. he solu ion o (2a-2b) o he pai (E(¯ θ)ρ(¯ θ))), and ¯pis he adjoin s a e, being in his case he unique solu ion o [E(¯ θ)¯p0]0+ Ω2ρ(¯ θ)¯p= 2(Ω2ρ(¯ θ)¯u−[E(¯ θ)¯u0]0),(3a) E(¯ θ(0))¯p0(0) = −2γ, E(¯ θ(l))¯p0(l) = 0.(3b) Using he Taylo expansion o he cos unc ional we ob ain, as a di ec consequence o he p e ious esul , he ollowing maximum p inciple. Co olla y 2.1 I ¯ θ(wi h ¯ θan elemen o he in e io o A? ad) is an op imal solu ion o p oblem (˜ P) hen ˜ J0(¯ θ;θ−¯ θ)≥0 o any θsuch ha θ−¯ θis an admissible a ia ion. 4 Op imal design in wa e p opaga ion 3. Exis ence o Solu ions o P oblem (P) In his sec ion we explo e he consequences o he op imali y condi ions gi en in he o m o a maximum p inciple in Co olla y 2.1 on he op imal solu ions. Ou objec i e is o check whe he i is possible o p o e exis ence esul s o he o iginal op imal design p oblem (P). In he case ha E1=E2(and we assume hen ha E1=E2= 1), he design a iable does no ac on he p inciple pa o he s a e equa ion and he p oblem is ma hema ically much simple , we a e able o p o e ha indeed any op imal solu ion o p oblem (˜ P) is a cha ac e is ic unc ion, so ha he o iginal op imal design p oblem admi s minimize s. We use he op imali y condi ions in his pa icula case o show he esul , which is gi en in Theo em 3.1. The gene al si ua ion is mo e di icul and does no seem possible o use he same a gumen , howe e we gi e a condi ion implying he exis ence o a leas one op imal solu ion in he o m o a cha ac e is ic unc ion. We di ide his sec ion in o wo subsec ions co esponding o he wo cases. 3.1. Case E1=E2 We assume ha E1=E2= 1 and ou p oblem now, ha we deno e as (˜ P1), is he ollowing MinA? ad ˜ J(ρ(θ)), whe e he admissibili y se is A? ad ={θ∈L∞(0, l) : 0 ≤θ≤1}, ρ(θ) is gi en by ρ(θ) = ρ1θ+ρ2(1 −θ), he cos unc ional is ˜ J(θ) = Zl 0Ω2ρ(θ(x))|u(x)|2+|u0(x)|2dx, and uis compu ed om θ h ough he s a e equa ion u00+ Ω2ρ(θ(x))u= 0, u0(0) = γ, u0(l) = 0 Following Theo em 2.1, his p oblem is in he elaxed o m and he e o e i admi s a leas one op imal solu ion. The ollowing esul es ablishes ha any op imal solu ion is ac ually a classical one (classical solu ion means a solu ion in he o m o a cha ac e is ic unc ion). Theo em 3.1 Any op imal solu ion, ¯ θ, o p oblem (˜ P1) akes on only he alues {0,1}, i.e. any op imal solu ion o (˜ P1) is a cha ac e is ic unc ion. 5 J.C. Bellido, A. Donoso 3.2. Gene al case The exis ence o classical solu ions o he gene al si ua ion o (˜ P) is no clea in gene al as a as he au ho s a e able o claim, al hough some hing may be said abou his. We do no s a e a o mal esul in his sec ion, bu show using new ideas ha he p oblem admi s op imal solu ions in ce ain cases. All he echnical de ails a e included in [3], bu we gi e nume ical e idence o his in he nex sec ion. 4. Nume ical App oach and Examples A e he analysis ca ied ou in he p e ious sec ions, his sec ion is de o ed o show se e al nume ical simula ions o ou p oblem. The app oach ha we ollow in he p e- sen wo k is o app oxima e di ec ly he op imal solu ions o he elaxed op imal design p oblem (˜ P) by a g adien me hod. O cou se his implies app oxima ion o he s a es and adjoin s a es, i.e. solu ions o he homogenized s a e equa ion (see Theo em 2.1) and he co esponding adjoin equa ion (see Theo em 2.2). We belie e ha his app oach will ce ainly gi e good app oxima ions o he op imal designs o p oblem (˜ P). The nume ical algo i hm used in his wo k is a a iable s ep-size descen me hod gi en by θk+1 =θk−˜ J0(θk)sk, whe e skis a small posi i e s ep such ha ˜ J(θk+1)<˜ J(θk), ˜ J0(θk) is he g adien o he cos unc ional espec o he con ol, compu ed in Sec ion 2, and he s ep skis looked o as he s ep ha minimizes ˜ J(θk+1(sk)) (we use a descen me hod o compu ing sk). On he o he hand, since he con ol mus e i y ha 0 ≤θk+1 ≤1, hen he upda ing o mula can be easily compac ed by θk+1 = m´ax(0,m´ın(1, θk−˜ J0(θk)sk)), being he same ha we ob ained by using a p ojec ed g adien me hod. Conce ning he s opping c i e ia, i has been checked ha he cos unc ional sligh ly changes i s alue and he e o e, he L∞−no m o he di e ence ˜ J(θk+1)−˜ J(θk) educes o ze o, as we app oxima e o he minimum. All nume ical examples belonging o his wo k ha e been ob ained unde he s opping c i e ia, ol = 10−5, equi ing o con e gence abou 15 i e a ions o he algo i hm, on a e age. I has also been es ed ha e y simila esul s a e ob ained when we un he same nume ical examples wi h di e en mesh sizes. Fo he sake o b ie ness, we illus a e ou app oach h ough wo nume ical examples. In bo h o hem, he design domain is an elas ic od o uni leng h subjec ed o ime ha monic longi udinal exci a ion o ampli ude γin he le end. Ma e ial p ope ies a e E2= 200 GPa, ρ2= 7,800 kg/m3, and E1= 70 GPa, ρ1= 2700 kg/m3co esponding o s eel and aluminum, espec i ely. In p e ious sec ions i has been possible o p o e ha op imized solu ions a e classical when he Young’s modulus Ekeeps cons an along he od and o show ha he e a e cases wi h E a ying i s alue in which s ill he e a e classical solu ions. In gene al, nume ical simula ions indica e ha op imized layou s a e classical, and he e o e, he e is no mic os uc u e among he wo phases we a e mixing. In he i s example we ha e used he equency alue Ω = 50 kHz. E en hough we le he con ol θ a y con inuously be ween 0 and 1 du ing he op imiza ion p ocess, we can 6 Op imal design in wa e p opaga ion see in Figu e 1(a) ha he op imized solu ion ob ained is classical a he end o he p ocess. No ice ha his con igu a ion make he wa e damp i sel in an op imized way, as i a els along he od (see Figu e 1(b)). Simila conclusions as be o e can be ob ained in he second example o Ω = 200 kHz (see Figu e 1(c) and (d)). We can epea he same nume ical examples bu keeping cons an he Young’s modulus, o ins ance, E2=E1= 70 GPa. The nume ical esul s o Ω = 50 kHz and 200 kHz (Figu e 2) co obo a e he exis ence o classical solu ions such as i was analy ically p o ed. 0 0.25 0.5 0.75 11 0 0.5 11 Axial posi ion Design a iable, θ (a) Op imized design 0 0.25 0.5 0.75 11 −5 0 5 10x 10 −6 Axial posi ion Ampli ude, u (b) Wa e ampli ude 0 0.25 0.5 0.75 11 0 0.5 11 Axial posi ion Design a iable, θ (c) Op imized design 0 0.25 0.5 0.75 11 −1 0 1 2x 10 −6 Axial posi ion Ampli ude, u (d) Wa e ampli ude Figu a 1: I is impo an o no ice ha ( o his pa icula case E2=E1) we ini ially ied wi h a nume ical app oach based on op imali y condi ions. Al hough in some si ua ions such algo i hm ga e us he same esul s as he ones ob ained by he g adien me hod, he ac is ha i did no enjoy good con e gence p ope ies in gene al, bu when i did we go simila esul s. In iew o he nume ical esul s we can s a e ha , e en hough he design p oblem ea ed is his pape is a he di e en o he band-gap one, as we commen ed in he In oduc ion, he op imized ini e s uc u es (almos pe iodic in some cases) ob ained ( o minimum ib a ion ene gy) o a pa icula alue o Ω, no a oid (like in he band-gap si ua ion) bu make a he di icul he wa e can p opaga e i sel along he od only a ha co esponding alue o Ω. To conclude, i is wo h emphasizing ha i s , all his analysis makes sense o high equency alues ( ypically g ea e han 10 kHz), o he wise he op imal designs a e “bo- ing” because hey would co espond o ake only he ma e ial wi h highe a io (Ei/ρi), o i= 1,2, and second, he same layou s we p esen he e a e ob ained whene e he p oduc Ω lkeeps cons an . 7 J.C. Bellido, A. Donoso 0 0.25 0.5 0.75 11 0 0.5 11 Axial posi ion Design a iable, θ (a) Op imized design 0 0.25 0.5 0.75 11 −1 0 1x 10 −5 Axial posi ion Ampli ude, u (b) Wa e ampli ude 0 0.25 0.5 0.75 11 0 0.5 11 Axial posi ion Design a iable, θ (c) Op imized design 0 0.25 0.5 0.75 11 −2 0 2 4x 10 −6 Axial posi ion Ampli ude, u (d) Wa e ampli ude Figu a 2: Acknowledgemen The au ho s acknowledge many s imula ing discussions on he subjec o his pape wi h Ma in Bendsøe, Jacob Jensen and Ole Sigmund. Financial suppo was p o ided by MEC (Spain) G an MTM2004-07114, and by Jun a de Comunidades de Cas illa-La Mancha (Spain) G an PAI-05-027. Re e encias [1] SIGMUND O. and JENSEN J.S. Sys ema ic Design o Phononic Band-Gap Ma e ials and S uc u es by Topology Op imiza ion, The Royal Socie y o London, Philosophical T ansac ions, Se ies A, Vol. 361, pp. 1001–1019, 2003. [2] BENDSØE M.P. and SIGMUND O. Topology Op imiza ion: Theo y, Me hods and Applica ions, Sp inge -Ve lag, Be lin, Ge many, 2003. [3] BELLIDO J.C. and DONOSO A. An op imal design p oblem in wa e p opaga ion, o appea in Jou nal o Op imiza ion Theo y and Applica ions. [4] LURIE K.A. Applied Op imal Con ol Theo y o Dis ibu ed Sys ems, Plenum P ess, New Yo k, New Yo k, 1993. [5] WARGA J. Op imal Con ol o Di e en ial and Func ional Equa ions, Academic P ess, New Yo k, New Yo k, 1972. [6] YOUNG L.C. Lec u es on he Calculus o Va ia ions and Op imal Con ol Theo y, W. B. Saunde s Coo po a ion, Philadelphia, Pennsyl ania, 1969. 8