Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 46, 4, pp. 813-828, Warsaw 2008 OPTIMIZATION OF ELASTIC ANNULAR PLATES SUBJECT TO THERMAL LOADINGS WITH RESPECT TO VIBRATION Antoni Gajewski Cracow University of Technology, Institute of Physics, Kraków, Poland e-mail: dantek@fizyk.ifpk.pk.edu.pl In the paper, the optimization of annular plates subject to circularly symmetric distribution of thermal loading with respect to vibration fre- quency is considered.What is searched for is such a distribution of plate thickness of a constant volumewhichmaximizes its lowest vibration fre- quency. The studies are confined to a plate clamped at the inner and outer edges and to one type of non-homogeneous temperature distribu- tion. The inequality constraints on the minimal and maximal values of the plate thickness are taken into account. The applied method of so- lution is the Pontryagin maximum principle combined with sensitivity analysis and a gradient procedure. Key words: optimization, annular plate, thermal loading 1. Introduction The optimization of circular and annular plates compressed by uniformly distributed conservative loadings with respect to stability were considered in several papers, e.g. by Frauenthal (1972), Grinev and Filippov (1977), Rzegocińska-Pełech and Waszczyszyn (1984), Mermertas and Belek (1990), Wróblewski (1992). Some non-conservative problems of the optimization of circular plates were presented by Gajewski and Cupiał (1992) and Gajewski (2002a,b). Recently, a formulationof theoptimizationproblemofannularplatesunder thermal loadings with respect to buckling was presented by Gajewski (2001) who considered a non-uniformelastic annular plate subject to a non-uniformly distributed temperature field. Similar problems were analysed by Krużelecki and Smaś (2006) who obtained optimal solutions for different modes of sup- ports and different ratios of inner and outer radii. The methods of moving 814 A. Gajewski asymptotes and simulated annealingwere used. Amore general problem of an annular plate optimization was formulated by Gajewski (2002b) who optimi- zed the radially distributed thickness of a plate h(x) with respect to vibration under a constant volume condition and constant temperature loading. To this end, the Pontryagin maximum principle with sensitivity analysis and an ite- rative procedure were used. The principal aim of this paper is a broader presentation of some new results of numerical calculations obtained for a similar problem. 2. Formulation of the optimization problem The principal aim of the present paper is to optimize the radially distributed thickness h(r) of a thin annular plate with respect to vibration (or buckling) under a constant volume condition. The plate supported in different ways is subject to thermal loading by an increment of temperature, which can be ra- dially distributed in a specific way T(r) (Fig.1).We look for such a thickness distribution so as tomaximize the lowest frequency of vibration under a given constant thermal loading and a given volume of the plate. In particular, if the first frequency of vibration is equal to zero we can maximize the first critical thermal loading. Generally, the equations of the precritical and vibration sta- tes should be analysed. The constant volume condition will be written in a dimensionless form 1∫ β xh(x) dx =1 (2.1) where the dimensionless functions and parameters are defined by (4.3) and (4.5). Fig. 1. An annular plate under thermal loading Optimization of elastic annular plates... 815 3. The optimization methods 3.1. Pontryagin maximum principle To solve the presented optimization problem, we use the Pontryagin ma- ximum principle in its classical form, combined with a sensitivity analysis formulation. To this end, the membrane and vibration state equations with boundary conditions will be written in the forms (cf. Gajewski and Życzkow- ski, 1988): Y ′i = Gi(x,Yi,h,P) i =1,2 µ̃1γYγ(β)= 0 ν̃2γYγ(1)= 0 γ =1,2 (3.1) Z′i = Aiα(x,h,P,ω)Zα i =1, . . . ,4 α =1, . . . ,4 µjαZα(β)= 0 νjαZα(1)=0 j =1,2 (3.2) where Yi(x) and Zi(x) denote membrane and vibration state variables, re- spectively, Aij is amatrix dependent on the loading parameter and frequency of vibration, Gi are certain functions (in our case, the generally nonlinear functions Gi are linear with respect to Yi) and µ̃1γ, ν̃2γ, µjα, νjα are given matrices of constant parameters. In the paper, the summation convention is used.Therefore, theHamiltonian andappropriate adjoint equations connected with the state equations and constant volume condition (2.1) are as follows H = χαGα+ψβAβγZγ +Λxh (3.3) χα =− ∂H ∂Yα ψβ =− ∂H ∂Zβ (3.4) and α =1,2, β =1, . . . ,4, γ =1, . . . ,4. The optimal control function h(x) should be determined from the supre- mum condition of the Hamiltonian M(χ,ψ,Ỹ ,Z̃)= sup h∈h̃ H(h,χ,ψ, Ỹ ,Z̃) (3.5) 3.2. Sensitivity analysis Calculating variations of (3.1) and (3.2), multiplying them by χ and ψ, respectively, and integrating, we may eliminate δY and δZ, and derive the following equation for sensitivity operators (in matrix notation) 816 A. Gajewski 1∫ β ( χ ⊤ ∂G ∂h +ψ⊤ ∂A ∂h Z ) δh dx+ δP 1∫ β ( χ ⊤ ∂G ∂P +ψ⊤ ∂A ∂P Z ) dx+ (3.6) +δω 1∫ β ( ψ ⊤ ∂A ∂ω Z ) dx =0 For example, in a particular case of a prescribed frequency of vibration, we obtain δP = 1∫ β g(x)δh dx (3.7) where g(x)=− χ⊤∂G ∂h +ψ⊤∂A ∂h Z 1∫ β ( χ⊤∂G ∂P +ψ⊤∂A ∂P Z ) dx (3.8) If the volume of the plate is constant and the plate thickness is normalized according to formula (2.1), the Pontryagin maximum principle is equivalent to the sensitivity analysis method. In order to determine the optimal plate thickness distribution, thePontry- agin maximum principle and an iterative procedure have been used. Starting from a uniform plate, the improved shapes have been calculated from the formula (cf. Biess et al., 1980) hn+1 = hn+ ǫ ∂H ∂h ∣∣∣ hn (3.9) where H isHamiltonian (3.3) connectedwith state (3.1), (3.2) andappropriate adjoint equations (3.4), ǫ is a small parameter, n – iteration number. It is seen that ∂H ∂h = g(x)+Λx (3.10) where Λ is a constant to be determined by constant volume condition (2.1), and that the same formulamaybeused for optimization under either constant loading or constant frequency constraints. Optimization of elastic annular plates... 817 4. Equations of state 4.1. Equations of the membrane state The basic equations of the membrane and vibration states of thin annu- lar plates can be evaluated on the basis of the monograph by Ogibalov and Gribanov (1968). In a basic pre-critical state, it evokes in-plane radially distributed internal compressive forces and displacements. They are determined by the following boundary value problemwritten in a dimensionless form u′ =− ν x u+ 1−ν2 ehx n− (1+ν)t nr = n x n′ = eh x u+ ν x n+eht nθ = n ′ (4.1) and α1u(β)+α2n(β)= 0 α3u(1)+α4n(1)= 0 (4.2) We confine our consideration to plates rigidly clamped at the inner and outer edges, i.e. we assume: α2 = α4 =0. The following dimensionless variables, parameters and functions have been introduced: — independent space variable and internal-to-external radii ratio x = r a β = b a (4.3) — temperature distribution t(x)= T(x) T0 (4.4) where, in fact, the dimensional value T(x) denotes the increment of tempera- ture distribution over its value in the stressless state, T0 denotes the reference increment of temperature, — variable thickness and Young’s modulus distributions h(x)= h h0 e[x,T(x)] = E[x,T(x)] E0 (4.5) where some possible dependences of the Young modulus on the independent variable x and temperature T have been assumed, 818 A. Gajewski — temperature loading parameter, reference plate rigidity and slenderness parameter P = α̃ α T0 D0 = E0h 3 0 12(1−ν2) (4.6) α = D0 E0h0a 2 = 1 12(1−ν2) (h0 a )2 where ν, α̃, E0, h0, T0 denote Poisson’s coefficient, coefficient of thermal expansion, referenceYoung’smodulus, referenceplate thickness and increment of temperature, respectively, — displacement and compressive force u =−Pαau(x) N =−P D0 a n(x) (4.7) The coefficients α1, . . . ,α4 are certain constants. 4.2. Equations of the vibration state It can be assumed that as a result of vibration (or buckling), the stress and strain components of the basic membrane state are subject to small va- riations. In contradistinction to the membrane state, the deflection and all internal forces in the vibration state are not circulary symmetric. In general they depend on the radial and angular independent variables. Thewell-known equation of small deflection superimposed on themembrane state of the plate may be written in the form ∂2(rMr) ∂r2 + 2 r ∂2(rMrθ) ∂r∂θ + 1 r ∂2Mθ ∂θ2 − ∂Mθ ∂r +rNr ∂2w ∂r2 + (4.8) +Nθ (1 r ∂2w ∂θ2 + ∂w ∂r ) −ρrh ∂2w ∂t 2 =0 where w = w(r,θ,t) is the deflection of the plate. The increments of internal forces (superposed on zero initial values) are expressed as follows Mr =−D0eh 3 [∂2w ∂r2 +ν (1 r ∂w ∂r + 1 r2 ∂2w ∂θ2 )] Mθ =−D0eh 3 [ ν ∂2w ∂r2 + (1 r ∂w ∂r + 1 r2 ∂2w ∂θ2 )] (4.9) Mrθ =−D0eh 3(1−ν) ∂ ∂r (1 r ∂w ∂θ ) Optimization of elastic annular plates... 819 Now, we introduce new dimensionless variables and parameters: — independent time variable t = t/t0, where the dimensional constant para- meter t0 may be treated as a unit of time t0 = √ ρ0a 4 D0 = √ ρ0a 2 αE0 (4.10) — frequency of vibration: ω = ωt0, — plate deflection and internal forces w̃ = w a Mj = aMj D0 for: j = r,θ,rθ (4.11) —mass density distribution: ρ(x)= ρ/ρ0. Next, we separate the functions of independent variables and we use the substitutions suggested by Grinev and Filippov (1977) w̃(x,θ,t)= w(x)eiωt cos(mθ) Mr(x,θ,t)= M̃r(x)e iωt cos(mθ) Mθ(x,θ,t)= M̃θ(x)e iωt cos(mθ) Mrθ(x,θ,t)= M̃rθ(x)e iωt sin(mθ) ˜̃ Mr(x)= M̃r(x)+ 1 2 Nr(x)w(x) ˜̃ Mθ(x)= M̃θ(x)+ 1 2 Nθ(x)w(x) ˜̃ Mrθ(x)= M̃rθ(x) ϕ = dw dx = w′ (4.12) M = x ˜̃ Mr Q = M ′+2m ˜̃ Mrθ− ˜̃ Mθ N = xNr =−Pn(x) Therefore, equation (4.8) can be transformed into a set of four ordinary differential equations in a non-dimensional form w′ = ϕ ϕ′ = (νm2 x2 − Pn 2xD ) w− ν x ϕ− 1 xD M M ′ = [ (1−ν)(3+ν) m2D x2 − 1 2 PS (u x + t )] w+ −(1−ν)(1+ν +2m2) D x ϕ+ ν x M +Q (4.13) Q′ = [ (1−ν)(2+νm2+m2) m2D x3 −Pm2 S x (u x + t ) − P2n2 4xD −ω2ρxh ] w+ − [ (1−ν)(3+ν) m2D x2 − 1 2 PS (u x + t )] ϕ+ (νm2 x2 − Pn 2xD ) M where: D = eh3, S = eh. 820 A. Gajewski To equations (4.13) appropriate boundary conditions should be added. In this paper, we assume that the plate is rigidly clamped at the inner and outer edges w(β) = 0 ϕ(β) = 0 w(1)= 0 ϕ(1)= 0 (4.14) Of course, a variety of boundary conditions might be considered, as in the paper by Krużelecki and Smaś (2006). Boundary value problems (4.1), (4.2) and (4.13), (4.14) consist of the so- called state equations. 4.3. Adjoint equations of the vibration state Since the type of loading by thermal stress is conservative, the adjoint boundary value problem of the vibration state is strictly the same as state bo- undaryvalue problem (4.13), (4.14). It can be shownbymeans of the following substitutions to Eqs (3.4) ψ1 = kQ̂ ψ2 =−kM̂ ψ3 = kϕ̂ ψ4 =−kŵ (4.15) where k is an arbitrary constant. Therefore, the state variables (ŵ, ϕ̂,M̂,Q̂) can be identified with the adjoint state variables, and we can conclude that boundary value problem (4.13), (4.14) is self-adjoint. 4.4. Adjoint equations of the membrane state Substituting new adjoint membrane state variables χ1 = k1n̂ χ2 =−k1û (4.16) the system of adjoint differential equations of the membrane state can be written in the form û′ =− ν x û+ 1−ν2 xeh n̂+ P 2xeh3 (2Mw+Pnw2) (4.17) n̂′ = eh x û+ ν x n̂+Peh (1 x wϕ− m2 x2 w2 ) Here, the appropriate adjoint boundary conditions are the same as for the state functions û(β)= 0 û(1)=0 (4.18) Optimization of elastic annular plates... 821 5. Numerical calculations and results 5.1. Numerical calculations Numerical calculations can be performed for some types of temperature distributions, suggested by Ogibalov and Gribanov (1968), for example t(x)= 1−x2 1−β2 or t(x)= 1− (x−β 1−β )n (5.1) However,we confinedour calculations to thefirst formulaand for boundary conditions (4.2) and (4.14). In all calculations β was set equal to 0.2 and Poisson’s ratio ν wasassumedtobe0.25.Thereferencedensitywasassumedto be equal to the plate density and, as a result, ρ ≡ 1. All differential equations were solved by the Runge-Kutta-Gill integration method of the fourth order, using the transfer matrix method (cf. Gajewski, 2002). The gradient function was calculated from the formula ∂H ∂h = [ Λx+ρxω2w2+e (u x + t )( P m2 x w2−Pwϕ− û )] + −h2 { 3(1−ν)e [ (2+νm2+m2) m2 x3 w2+(1+ν +2m2) 1 x ϕ2+ (5.2) −2(3+ν) m2 x2 wϕ ]} − 1 h2 [(1−ν2) ex nn̂ ] − 1 h4 3 ex ( M + 1 2 Pnw )2 which can be used both under constant loading and constant frequency con- straints. Moreover, we have imposed additional geometrical constraints on the plate thickness h11 ¬ h(x)¬ h22 (5.3) where: h11 =0.5 and h22 =3.0. 5.2. Results 5.2.1. A plate of constant thickness At first, the dependence of the compressive force parameter (temperature) on the frequency of vibration is presented in Fig.2. In the picture, the first and second vibration frequency for m =0, . . . ,4 in relation to thermal loading are plotted. From the first picture in Fig.2, one can conclude that for P . 200 the starting point to the optimization procedure should be a plate of constant 822 A. Gajewski Fig. 2. Thermal loading versus: (a) the first frequency and (b) second frequency of vibration thickness with m = 0. However, for greater values of P we must start with m =1 or even m =2. During the iteration process, values of circumferential waves m can undergo changes. 5.2.2. The optimization with respect to frequency of vibration under constant loading As the first example of the optimization procedure, we present in Fig.3 plate shapes obtained in consecutive iterations for a very small value of tem- perature loading. In fact, it is an optimization with respect to the vibration frequency only. It is seen in the Fig.4 that the convergence of the iteration process is regular and very quick. The frequency of vibration increases from ωprism ≈ 72 for the plate of constant thickness up to ωopt ≈ 95 for the optimal plate. The optimization has been performed for m =0. Fig. 3. Plate thickness in consecutive iterations (P =1, h11 =0.5, h22 =3.0, ω →max, m =0) Optimization of elastic annular plates... 823 Fig. 4. (a) Characteristic curves for the optimal plate (P =1, h11 =0.5, h22 =3.0, ω →max) and (b) convergence of the optimization process (P =1, ω →max) 5.2.3. The optimization with respect to thermal loading under constant frequency constraint As the second example, we present in Fig.5 plate shapes obtained in con- secutive iterations for a very small value of the vibration frequency. In fact, it is an optimization with respect to thermal loading only (under stability constraints). Fig. 5. Plate thickness in consecutive iterations (P →max, Popt =532, ω =10, ε =0.05) It is seen in Fig.6 that the convergence of the iteration process is not so regular as in the previous example. The buckling thermal loading incre- ases from Pprism ≈ 326 for the plate of constant thickness to Popt ≈ 532 for the optimal plate. The optimization had to be performed for m = 2. Moreover, the optimal shape is practically a three-modal solution, for which: Popt(m =0)≈ 532.5, Popt(m =1)≈ 531.9, Popt(m =2)≈ 534.0. 824 A. Gajewski Fig. 6. (a) Characteristic curves for the optimal plate (Popt =532, ω =10) and (b) convergence of the optimization procedure (P →max, Popt =532, ω =10) 5.2.4. Optimal plate shapes for various thermal loading levels The calculations similar to those presented in the previous subsections enable one to construct the final results. They are shown in Fig.7, where the optimal plate shapesobtained forvarious thermal loading levels are illustrated. Of course, the optimal shape under the buckling constraint (for ω = 0) is quite different from that obtained under the vibration frequency constraint (for P =0). Fig. 7. Optimal plates for various thermal loading levels 5.2.5. The influence of temperature on the Young modulus In all results presented above, theYoungmoduluswas treated as a certain constant independent of temperature. However, it should be noted that the increase of temperature (loading parameter P) causingbuckling of the plate is rather high. Therefore, the Youngmodulus changes its value according to the Optimization of elastic annular plates... 825 increase of temperature. This phenomenon can influence the optimal shapes of the plate. To consider this, we have fitted the experimental data presented in Fig.8 (given by Odquist, 1966), with a polynomial of the third degree E = E0e(T)= E0(1−aT + bT 2 − cT 3 ) (5.4) where E0 =197MPa a =3.92 ·10−4[◦C]−1 b =1.77 ·10−6[◦C]−2 c =3.30 ·10−9[◦C]−3 (5.5) Fig. 8. Dependence of the Youngmodulus on teperature As it is seen inFig.8., function (5.4) describes quitewell the real dependen- ce of the Young modulus on temperature. In numerical calculations, formula (5.4) is written in the dimensionless form e[T(x)] = 1−a[ξPt(x)]+ b[ξPt(x)]2− c[ξPt(x)]3 (5.6) in which the dimensional parameter ξ has been introduced ξ = α α̃ [ξ] = [◦C] ξ ≈ (1÷20)[◦C] (5.7) To illustrate the influence of the relationship between the Youngmodulus and temperature on the optimal plate shape, we performed calculations for several values of the parameter ξ for a constant thermal loading P = 300 and m =1. Moreover, in this case, the membrane state without any stress is assumed at the temperature 0◦C. 826 A. Gajewski Fig. 9. (a) Optimal plate shapes and (b) convergence of the optimization procudure for various values of the coefficient ξ (P =300, h11 =0.5, h22 =3.0, ω →max, m =1) 6. Conclusions In thepaper, a significantdifferencebetween theoptimalplate shapesobtained under buckling or vibration constraints has been demonstrated. It should be noted that for somecases ofboundaryconditions theoptimizationwith respect to the lowest frequency of vibration (or the lowest buckling thermal loading) mayresult in the loweringof ahigher-order eigenvaluebelowthefirstone.Then the result of a unimodal optimal design (i.e. with respect to single eigenvalue) is false, and a multimodal optimal design should be employed. Additionally, it has been shown that the dependence of theYoungmodulus on temperature can have great influence on the optimal shapes. Acknowledgment Partial support of this work under Grant 5T07A-003-24 by the State Committee for Scientific Research is gratefully acknowledged. References 1. Biess G., Erfurth H., Zeidler G., 1980,Optimale Prozesse und Systeme, BSB B.G.Teubner, Leipzig 2. Frauenthal J.C., 1972,Constrained optimal design of circular plates against buckling, J. Struct. Mech., 1, 115-127 3. Gajewski A., 2001, Optimization of elastic annular plates subject to thermal stresses under buckling constraints, ”Optimization in Industry III”, Ref. Num- Optimization of elastic annular plates... 827 ber 06, US United Engineering Foundation, June 17-22, Ciocco-Barga, Italy (Eds.: I.C.Parmee, P.Hajela) 4. Gajewski A., 2002a,Optimization of elastic annular plates subject to thermal stresses under vibration constraints,Proc. Vibration in Physical Systems,XXth Jubilee Symposium, Poznań-Błażejewko,May 21-25, 2002, 150-151,Politechni- ka Poznańska, Poznań 5. Gajewski A., 2002b, Vibration and stability of annular plates in non-linear creep conditions, J. Sound and Vibr., 249, 3, 447-463 6. Gajewski A., Cupiał P., 1992,Optimal structural design of an annular pla- te compressed by non-conservative forces, Int. J. Solids Structures, 29, 10, 1283-1292 7. Gajewski A., Życzkowski M., 1988,Optimal Structural Design under Sta- bility Constraints, Kluwer Academic Publishers, Dordrecht 8. Grinev V.B., Filippov A.P., 1977, Optimal design of circular plates in sta- bility problems, Stroit. Mech. i Raschot Sooruzeniy, 2, 16-20 [in Russian] 9. Krużelecki J., Smaś P., 2006, Optimal annular plates with respect to their stability under thermal loadings, Struct. Multidisc. Optim.,32, 111-120 10. Mermertas V., Belek H.T., 1990, Static and dynamic stability of variable thickness annular plates, J. Sound and Vibr., 141, 435-448 11. Odqvist F.K.G., 1966,Mathematical theory of creep and creep rupture, Cla- rendon Press, Oxford 12. OgibalovP.M.,GribanovW.F., 1968,Thermo-stability of plates and shells, MoscowUniversity Press [in Russian] 13. Rzegocińska-Pełech K., Waszczyszyn Z., 1984, Numerical optimum de- sign of elastic annular plates with respect to buckling, Computers & Struct., 18, 369-378 14. Wróblewski A., 1992, Optimal design of circular plates against creep buc- kling,Eng. Opt., 20, 111-128 Optymalizacja płyt pierścieniowych z uwagi na drgania przy obciążeniach termicznych Streszczenie W pracy badano problemy optymalizacji płyt pierścieniowych poddanych działa- niu równomiernie rozłożonych obciążeń temperaturowych, z uwagi na częstość drgań. 828 A. Gajewski Poszukiwano takiego rozkładu grubości płyty (kołowo symetrycznej) o stałej objęto- ści, który prowadzi do maksymalizacji jej częstości drgań. Ograniczono się do pły- ty utwierdzonej na brzegu zewnętrznym i brzegu wewnętrznym oraz do jednego ty- pu niejednorodności rozkładu temperatury. Przyjęto również nierównościowewarunki geometryczne nałożone naminimalną i maksymalną grubość płyty. Jakometodę roz- wiązania zagadnienia przyjęto zasadę maksimum Pontryagina połączoną z analizą wrażliwości i metodą gradientową. Manuscript received March 19, 2008; accepted for print May 19, 2008