Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 47, 3, pp. 685-699, Warsaw 2009 OPTIMAL DESIGN OF SANDWICH PANELS WITH A SOFT CORE Robert Studziński Zbigniew Pozorski Andrzej Garstecki Poznan University of Technology, Institute of Structural Engineering, Poznań, Poland e-mail: andrzej.garstecki@put.poznan.pl Themain issue taken up in the paper is to find optimal designs ofmulti- span sandwich panels with slightly profiled steel facings and polyure- thane foam core (PUR), which would satisfy conflicting demands of the market, i.e. minimal variance in types of panels, maximum range of ap- plication andminimumcost.The aim is to finddimensional andmaterial parametersof panelswhichgenerateminimumcost andmaximumlength of spanunder prescribed loads in ultimate and serviceability limit states. The multi-criterion optimization problem is formulated in such a way, where the length of the span plays two roles, namely a design variable and a component of a vector objective function. An evolutionary algo- rithm is used. Numerous inequality constraints are introduced in two ways: directly and by external penalty functions. Key words: sandwich panels, soft core, Pareto optimization, soft compu- ting, genetic algorithm 1. Introduction In recent years, we have observed an increase in practical applications of san- dwichpanels in civil engineeringandmechanical constructions (Hassinen et al., 1997;Magnucka-Blandzi andMagnucki, 2007). High bending stiffness coupled with small weight and very good thermal and damping properties make san- dwich panels an attractive structure for designers (Craig and Norman, 2004). Easiness of transport and assembly in all weather conditions are additional advantages. These aspects have also generated a growth in investment outlays for computational and experimental research. 686 R. Studziński et al. Sandwich panels most often used in civil engineering consist of two ste- el facings and a soft core. The facings carry normal stresses, while the three principal roles of the core are to carry shear stresses, to protect the com- pressed facing against buckling and to provide expected thermal insulation (Chuda-Kowalska et al., 2007). Note that a greater elasticity modulus of the core, which usually improves the mechanical response of a slab, is accom- panied by greater weight and higher cost of the material, as well as wor- se insulation coefficient. Another conflict is observed between mechanical fe- atures and price of a panel when variable thickness or depth of the profi- ling of facings is considered. The above conflicting relations make the tra- ditional design difficult and give reasons for using methods of the optimal design. Optimal design of sandwich panels has been widely discussed in litera- ture. Bozhevolnaya and Lyckegaard (2006) examined sandwich panels with junctions between different core materials. The application of different co- re materials in the same panel allows designers to improve the capability of the core to withstand varying external shear loads. The optimal cho- ice of the core material was also discussed in Czarnecki et al. (2008), whe- re the basic aim was to find plates of minimal compliance. The compari- son of sandwich panels behaviour with various combinations of materials in three-point bending was described in Craig and Norman (2004). This analysis served as a basis for creating failure maps which make it possi- ble to optimize the geometry and mass of sandwich panels. Tan and Soh (2007) discussed the two-criterion optimization (minimization of the weight and heat transfer) for sandwich panels with a prismatic core using genetic algorithms. The present paper extends the class of problems discussed in literature by considering conflicting market demands for panels of minimum cost and the widest range of application. To fulfill the latter demand, we will look for solutions applicable as one- or multi-span structural elements satisfying design requirements for the maximum length of the span in typical load con- ditions. A two-span plate can be considered as a representative case, because it exhibits simultaneously the field and support phenomena, which are do- minant in design. It is assumed that the panels are subjected to mechani- cal and thermal loading. Thermal action induces high strains and stresses in multi-span panels. Its destructive effects can be reduced by using me- thods of optimal structural design. The interrelation of optimal values of de- sign variables can be used in manufacturing of panels and their engineering applications. Optimal design of sandwich panels with a soft core 687 2. Sandwich panel theory This paper discuses only panels with parallel facings and soft core. The panel under considerationwith the assumed supportand loading conditions is shown in Fig.1. Fig. 1. Two-span sandwich panel loadedmechanically and thermally. q represents uniformly distributed external load. T1 and T2 denote changes of temperature of the upper and lower facings, respectively In this case of support, uniformly distributed loading and thermal actions, a sufficiently adequate theory is the Timoshenko beam theory generalized to sandwich sections. This is valid however, only when the ratio of the span length L to the plate width b is greater than 2. According to the modified Timoshenko andReissner theories (Stamm andWitte, 1970) we assume that: the strains are small; thematerials (steel in the facings and foam in the core) are isotropic, homogeneous and linearly elastic; normal stress in the foam core is negligible (σxC = σyC = 0). From the last assumption it follows that the shear stresses in the core are constant along the transverse axis z (τxzC = τyzC = const). The mechanical model for the structural analysis, i.e. for the evaluationof stresses, strains anddisplacements, dependsonthe typeof facings (flat or profiled). Deeply profiled facingsmade of thin sheets fall into the same category as thick facings. Therefore, panels which have at least one deeply profiled facing must be considered separately from those that have two flat facings. Our attention will be focused on panels with flat or slightly profiled facings.However, for the sake of completeness, we start thediscussionwith the general case, in which at least one facing is deeply profiled or thick. The cross sectional equilibrium condition can be written in the form of two uncoupled differential equations (2.1)1 for vertical deflection w and (2.1)2 for the shear strain γ (Stamm andWitte, 1970) − BF1+BF2 GCAC wVI + B BS wIV = q BS − q′′ GCAC −θ′′ (2.1) − BF1+BF2 GCAC γIV + B BS γ′′ =− q′ GCAC − BF1+BF2 GCAC θ′′′ 688 R. Studziński et al. where w and γ are functions of the position coordinate x, which is measured along the length of the panel (see Fig.1). The denominator GCAC represents the shear stiffness of the core, q is the uniformly distributed transverse load and θ is the initial curvature inducedbya temperaturedifference ∆T between the lower and upper face sheets. θ is positive when associated with stretching of the lower face. The total bending stiffness B is given by B=BF1+BF2+BS +BC (2.2) where BF1,BF2 are the bending stiffness of the upper and lower facings with respect to their own centre lines (2.3)1, BS is the bending stiffness of the facings with respect to the global centre line of the sandwich panel (2.3)2 and BC is the bending stiffness of the corewith respect to its own centre line (2.3)3 BFi = bt3i 12 EFi BS = btie 2 iEFi (i=1,2) (2.3) BC = bd3 12 EC Table 1 demonstrates typical ratios of stiffness parameters encountered in sandwich panels with soft core. Table 1. Stiffness parameters Stiffness ratios Range [ ] BC/BS 0.470-0.630 ∑ BFi/BS (flat facings) 0.008-0.015 ∑ BFi/BS (slightly profiled facings) 0.020-0.150 ∑ BFi/BS (profiled facings) 1.00-15.00 This table demonstrates that BF1 and BF2 in sandwich panels with flat and slightly profiled facings aswell as BC are negligible. Neglecting BF1,BF2 and BC in (2.1) and (2.2), one arrives at (2.4) for sandwich beams with flat or slightly profiled facings wIV = q BS − q′′ GCAC −θ′′ γ′′ =− q′ GCAC (2.4) Integrating twice (2.4) and using equilibrium equations M ′ =Q,Q′ =−q, we obtain constitutive equations (2.5) of the Timoshenko beam M =BS(γ ′−w′′−θ) Q=GCACγ (2.5) Optimal design of sandwich panels with a soft core 689 where M = My and Q = Qz denote the bending moment and shear force, respectively. The normal stresses σF1, σF2 in the metal face sheets and shear stress τ in the core are expressed by (2.6)1 and (2.6)2, respectively (see Fig.2) σFi =(−1) i M eAFi τ = Q AC (i=1,2) (2.6) where AFi (i=1,2) represents cross-sectional areas of the face sheets and AC represents the cross-sectional area of the core. The shear stresses τ across the facings t1 and t2 have the 2-nd order parabolic distribution (Fig.2). Since t1 and t2 are very small (usually 0.5mm), the shear stresses in the facings can be neglected. This is represented by (2.6)2. Fig. 2. Distribution of normal stresses (σ1,σ2) and shear stress (τ) in a sandwich panel with thin flat facings 3. Formulation and solution of the optimization problem Consider a two-span sandwich panel with flat or slightly profiled steel facings and with equal span lengths L. The panel is subjected to temperature action and to a uniformly distributed load, resulting from snow or wind actions. The temperature of the face exposed to sunshine depends on the colour of the face. The design vector x comprises five variables which specify geometric and material parameters of the panel and the length of spans x= [L,t1, t2,d,GC] (3.1) where L, t1, t2, d and GC denote the length of the span, thickness of the external and internal face sheets, the thickness and the shear modulus of the soft core, respectively. 690 R. Studziński et al. Theaimof theoptimization is tofinda structure of themaximumlength L andminimumcost. Hence, the optimal design problem is formulated as a two- criterion optimization with a vector objective function Γ . Our task is to find theoptimal designvector x, satisfying the constraints andprovidingminimum of Γ Γ(x)= [Γ1,Γ2] = [FC,α/L]→ min x∈X0 (3.2) Here FC is the total cost, L is the span length and α is the scaling coefficient. X0 is the allowable domain of design vectors x, specified by constraints gi(x)¬ 0 i=1,2, . . . ,n (3.3) Constraints (3.3) fall into one of two categories. The first one represents beha- vioral constraints, which are implicit functions of x. They follow from requ- irements of the ultimate and serviceability limit states and are listed inTables 2 and 3. Table 2.Constraints following from the ultimate limit state Constraint Comments gi(x)= |τi| fcvd −1¬ 0 τi – shear stress fcvd – design shear strength of core gi(x)= |σMi | fyd −1¬ 0 σMi – normal stress in span or at support fyd – design yield strength of face sheet gi(x)= |σwi | fswi −1¬ 0 σwi – wrinkling stress in span or at support fswi – design wrinkling stress in span or at support with consequential failure gi(x)= |σdi | fccd −1¬ 0 σdi – crushing stress at support fccd – compressive strength of core gi(x)= FLi fdl −1¬ 0 FLi – reaction at support fdl – design fastener capacity The second group takes the form of box conditions on the components of x, they are explicit functions. The total cost FC(x) in (3.2) is expressed by the cost c0 and specific costs of materials c1, c2. Here c0 contains constant manufacturing costs independent of the design variables x. The parameters c1 and c2 specify the cost of steel in face sheets and of PUR in the core. The parameters c0, c1 and c2 were evaluated based onmarket data FC(x)= c0+ c1(t1+ t2)+ c2(GC)d (3.4) Optimal design of sandwich panels with a soft core 691 Table 3.Constraints following from the serviceability limit state Constraint Comments gi(x)= |σMi | fyk −1¬ 0 σMi – normal stress in span or at support fyk – characteristic yield strength of face sheet gi(x)= |σwi | fkswi −1¬ 0 σwi – wrinkling stress in span or at support fkswi – characteristic wrinkling stress in span or at support without consequential failure gi(x)= umaxi ulim −1¬ 0 umaxi – maximum deflection in span ulim – limit deflection in span It is assumed that c0 = 35m −1, c1 = 19200m −2. The cost of material c2 is considered to be a function of the shear modulus GC: c2(GC)= [a(GC −G 0 C)+100]m −2 G0C =2500 kPa (3.5) The function c2(GC) with different values of the parameter a is illustrated in Fig.3. Fig. 3. Unit cost of corematerial c2 as a function of the shear elasticity modulus of the core GC The implicit behavioral constraints are introduced by means of external penalty functions. Penalized objective function (3.2) takes the following form ΓP1 (x)=Γ1(x)+ 24 ∑ i=1 Pi(x)→min x (3.6) 692 R. Studziński et al. where Pi(x) are penalty functions (i=1,2, . . . ,24) Pi(x)= { 0 for gi(x)¬ 0 βigi(x) for gi(x)> 0 (3.7) The implicit constraints are evaluated from the stress and displacement, com- puted using FEM based on the modified Timoshenko theory (Pozorski et al., 2007). In practical applications, the specific costs and box conditions can be va- lidated by manufacturing companies. The explicit constraints for t1, t2, d and GC were assumed: t1, t2 ∈ 〈0.0004m,0.0010m〉, d∈ 〈0.06m,0.20m〉 and GC ∈ 〈2500kPa,4500kPa〉. The above box conditions refer to current the technology of manufacturing and demands of the market. These constraints were explicitly introduced into the optimization algori- thm. The optimization problem was solved using the DPEA (Distributed Pa- rallel Evolutionary Algorithms) program (Burczyński and Kuś, 2004). The DPEA can operate on many subpopulations. The following evolutionary pa- rameters and operators were defined: number of subpopulation 1, number of chromosomes 20, number of genes 4, probability of mutation 0.1, probability of crossover 1. A tournament selection was performed. 4. Optimization results: discussion Within a general framework of the above presented formulation, a number of specific problems have been solved in order to study the influence of rigidity and variable cost of the core material, the temperature level and intensity of the external load. Table 4 describes these three classes of problems. Assumed temperatures on the lower and upper face sheets follow demands of design codes. Both face sheets have the same thermal expansion coefficient α = 0.0000121/◦C. 4.1. Optimal shear rigidity of the core Pareto optimal solutions for different parameters a (Table 4) specifying the cost of the core material as a function of GC are shown in Fig.4. Theplots showhowthe total cost of apanel dependson its allowable length of the span L. A drastic increase of the cost is observed for L> 3.5m. The Optimal design of sandwich panels with a soft core 693 Table 4.The optimization problems q [kN/m] ∆T =T2−T1 [ ◦C] a [1/(kPam2)] Variable a in the cost function of the core material 0.385 −55◦C (summer), +50◦C (winter) 0.000 0.385 −55◦C (summer), +50◦C (winter) 0.050 0.385 −55◦C (summer), +50◦C (winter) 0.100 Variable temperature T1 and T2 depending on the colour of the facing 0.385 −40◦C (summer) 0.050 0.385 −55◦C (summer) 0.050 – −40◦C/−65◦C/−90◦C 0.000 Variable load level q 0.385 −40◦C (summer) 0.050 1.540 −40◦C (summer) 0.050 0.75/1.00/2.00 – 0.000 Fig. 4. Pareto optimal solutions for different costs of the core material plots can be a good basis for a compromise decision as to what type of panel is worth producing. This compromise decision must account for a percentage demand for panels of greater L, however L=3.5mor 4.0m can be suggested. Note that panels with L > 5.9m are unacceptable because they violate the behavioral constraints in the presence of the assumed box conditions. This is clearly shown in the window of Fig.4. The external penalty function is equal to zero only in the admissible region, where all constraints are satisfied. It is unexpected and interesting that all three plots shown in Fig.4 are nearly identical. Hence, they demonstrate that the parameter a in (3.5) and in Fig.3 694 R. Studziński et al. has no influence on the total cost of the panel. The reason can be found in the window of Fig.5. Fig. 5. Optimal slab thickness d for variable span length L It can be seen that in the whole range of variable L the optimal value of the shear coefficient is equal to its lower bound G = G0 = 2500kPa. One exceptional point for L=2.0m in Fig.5 does not disturb the cost function C in Fig.4, because it is connected with the case when a=0. The plot of the optimal depth of the slab d is shown in Fig.5. It is intere- sting that it attains its lower bound d=0.06m for L¬ 3.0m. In the region L> 3.0m small differences in d for variable a result from the interchange of active constraints. Figure 6 shows the optimal thickness of facings. The thickness of the upper facing attains its lower bound t1 = 0.4mm for L ¬ 4.0m. The thickness of the lower facing t2 slightly increases for L=3.0m.This is an effect of wrinkling at themiddle support. For L=4.0m, the wrinkling effect is not so decisive because the plate depth d is twice as great. In the region of L> 4.0m, a rapid increase in thickness of both facings is observed. 4.2. The influence of temperature and load intensity on the optimal design Pareto optimal solutions referring to a uniformly distributed loading com- bined with two temperature levels are shown in Fig.7. The Pareto solutions for the two load levels combined with temperature action (see Table 4) are presented in Fig.8. Optimal design of sandwich panels with a soft core 695 Fig. 6. Optimal thickness of facings for variable span length L: (a) upper facing t1, (b) lower facing t2 The respective penalty functions are shown in the windows. As it was expected, the cost increases for more severe loading and temperature condi- tions, yet the quantitative relations between L and cost C presented in Fig.7 and Fig.8 are valuable, because they enable making compromise decisions with regard to the final optimal design. It is surprising that variation of the temperature conditions does not influence the Pareto solution (Fig.7) in the range L< 3m.This can be explained by the fact that for small L the optimal cross sectional variables attain their lower bounds. To better illustrate the in- fluence of loading and temperature action on the optimal design of a sandwich slab next exampleswere solved. Figure 9 presents thePareto optimal solutions 696 R. Studziński et al. Fig. 7. Pareto optimal solutions for two temperature conditions.Window: respective penalty functions Fig. 8. Pareto optimal solutions for two load levels.Window: respective penalty functions when the sandwich is subjected solely to the load or temperature actions of variable intensity specified in the Figure. In the case ofmechanical load, the plots (continuous lines) demonstrate an exponential increase of cost with respect to the span L. In the case of thermal loads, the wrinkling failure at the internal support is an active constraint. The plots (dashed lines) demonstrate a nearly linear relation between the span L and the cost. Note that for small L, the temperature actionmaymore significantly contribute to the total cost than the loading. Optimal design of sandwich panels with a soft core 697 Fig. 9. Pareto optimal solutions for slabs subjected solely to loading (continuous line) or to temperature action (dashed line) 5. Concluding remarks Thepaperpresents anoptimization problem,where a compromisebetween the cost of the product and its widest range of applications can be reached. The problem was formulated as a two-objective optimization of sandwich panels under prescribed loading and temperature induced distortions, aiming at the minimum cost andmaximum length of the span. The length of the panel span plays two roles, namely the design variable and a component of the vector objective function. Pareto optimal solutions were computed. The associated optimal values of design variables were presented and discussed. The quantitative representation of the Pareto solutions can be used in making compromise decisions by producers of sandwich panelswith soft cores. Inmaking final decisions, the actual cost factorsmust be used in the objective function, because Pareto solutions and associated optimal values of design variables depend on the cost function. Nevertheless, the analysis carried out in the paper allow us to draw general conclusions. Since both the facings and the core contribute to the total cost and to the bearing capacity of the panels, they are mutually interrelated depending on external actions and length of the span. In principle, an increase of the thickness of facings generates higher cost than an increase of thickness of the core. Therefore, in the analysed cases, the optimal thickness of the facingswas often equal to its lower bound. This is in agreement with general policy of producers to make the facings as thin as possible. 698 R. Studziński et al. Several optimization problems (different forms of the cost function, various load and temperature actions) for a two-span panelwere carried out. The ana- lysis showed that for design parameters referred to the current technological limitations, a compromise optimal panel should be designed for the range of span lengths L up to 3.5m or 4.5m. A further increase of the span length leads to a drastic increase of cost generated by greater thickness of the facings and core. To respond to the market demand for panels applicable to cases of larger spans – reaching 6.0m – a separate type of panels should be designed. Manufacturing one type of panels to cover the spans L∈ 〈2.0m,6.0m〉 is not economically viable. The paper also shows how the interaction of the external load and tempe- rature influences the optimal solutions. Former experience of authors allows one to state that in the case when the temperature dominates, a smaller shear stiffness GC of the core is preferable. This is a general principle valid formul- tispan panels. For one-span simply supported panels, cores with higher GC are optimal. This is observed particularly in the case, when deflection under a mechanical load is an active behavioral constraint. Acknowledgement The financial support by Ministry of Science and Higher Education under grant No. N506396835 is greatly acknowledged. References 1. Bozhevolnaya E., Lyckegaard A., 2006, Local effects at core junctions of sandwich structures under different types of loads, Composite Structures, 73, 24-32 2. Burczyński T., Kuś W., 2004, Optimization of structures using distributed andparallel evolutionaryalgorithms,LectureNotes onComputational Sciences, 3019, 572-579 3. Chuda-Kowalska M., Pozorski Z., Garstecki A., 2007, Identification of stiffness parameters of sandwich panels with soft core, Proc. of the I Polish Congress of Mechanics, CD [in Polish] 4. Craig A.S., Norman A.F., 2004, Material selection in sandwich beam con- structions, Scripta Materialia, 50, 1335-1339 5. Czarnecki S., Kursa M., Lewiński T., 2008, Sandwich plates of minimal compliance, Computer Methods in Applied Mechanics and Engineering, 197, 51/52, 4866-4881 Optimal design of sandwich panels with a soft core 699 6. HassinenP.,MartikainenL.,BernerK., 1997,On thedesignandanalysis of continuous sandwich panels,Thin-Walled Structures, 29, 129-139 7. Magnucka-BlandziaE.,MagnuckiK., 2007,Effectivedesignofa sandwich beamwith ametal foam core,Thin-Walled Structures, 45, 432-438 8. Pozorski Z., Chuda-KowalskaM., Studziński R., Garstecki A., 2007, Numerical modeling of sandwich panels with deep profiled facings,Proc. of the 17th International Conference on Computer Methods in Mechanics, 315-316 9. StammK.,WitteH., 1970,Sandwichkonstruktionen: Berechnung, Fertigung, Ausführung,Wien 10. Tan X.H., Soh A.K., 2007,Multi-objective optimization of the sandwich pa- nels with prismatic cores using genetic algorithms, International Journal of Solids and Structures, 44, 5466-5480 Optymalne projektowanie płyt warstwowych z miękkim rdzeniem Streszczenie Wpracypodejmuje sięproblemoptymalizacjiwieloprzęsłowychpłytwarstwowych z rdzeniemz poliuretanu (PUR) i okładzinami stalowymi lekko profilowanymi.Poszu- kuje się rozwiązań, które spełnią sprzeczne wymagania rynku, mianowicie: minimali- zację typoszeregu płyt, maksymalizację zakresu ich zastosowania oraz minimalizację kosztu produkcji. Celem optymalizacji jest znalezienie parametrów geometrycznych imateriałowychpłyt warstwowych, któreminimalizują koszt orazmaksymalizują do- puszczalną rozpiętość dla ustalonych obciążeń i przy spełnieniu stanów granicznych nośności i użytkowalności. W wielokryterialnym sformułowaniu problemu optymali- zacyjnego rozpiętość pełni dwie funkcje. Jest ona równocześnie zmienną projektową i składowąwektora funkcji celu. Jakonarzędzieoptymalizacjiwykorzystanoalgorytmy genetyczne. Ograniczenia nierównościowewprowadzono do procedury optymalizacyj- nej za pomocą zewnętrznej funkcji kary oraz jawnie. Manuscript received February 11, 2009; accepted for print March 18, 2009