betaFeedback

WorksPhysics and Astronomy

Optimal sizing of 1D vibrating columns accounting for axial compression and self-weight

Abstract

We investigate the effect of axial compression on the optimal design of columns, for the maximization of the fundamental vibration frequency. The compression may be due to a force at the columns' tip or to a load distributed along its axis, which may act either independently or simultaneously. We discuss the influence of these contributions on the optimality conditions, and show how the optimal beam design, and the corresponding frequency gain drastically change with the level of compression. We also discuss the indirect effect of frequency optimization on the critical load factors for the tip (λPλ_{P}) and distributed (λQλ_{Q}) loads. Finally, we provide some quantitative results for the optimal design problem parametrized by the triple (λPλ_{P}, λQλ_{Q}, Ω2Ω^{2}) of buckling and dynamic eigenvalues.

Introduction

Sizing of beams is a classical topic within optimal structural design, and has been one of the driving applications in the early developments of the field[5, 51, 22]. Despite the simplicity of beam models, closed-form solutions to their optimal design are achievable only in a few cases, involving compliance, volume, or displacement objective or constraints, and linear elastic beams[21, 30, 49, 44], plastic design[51, 15, 50], or design against buckling of uniformly compressed, statically determined columns[34, 47]. In more advanced instances, such as optimization of vibration frequencies, or buckling loads for general boundary conditions and loads, the beams’ optimal size can only be computed by numerical methods.

Nevertheless, important insights about the qualitative behavior of the solution can still be obtained through local asymptotic analysis[26], and bounds on the optimal frequencies or buckling loads can be given by isoperimetric inequalities[43, 55]. For instance, Niordson[39] studied the optimal design of a simply supported beam for maximum fundamental frequency of vibration, showing an optimal improvement of 6.6%. The analysis was extended to the cantilever beam by Karihaloo and Niordson[33], giving a much higher frequency improvement: 425-678%, depending on the relationship between stiffness and inertia, and then to the optimization of higher frequencies by Olhoff[40]. These studies highlighted some delicate matters, which have been the focus of many later works, both dealing with vibration and buckling optimization.

First, at points where the beam’s internal actions vanish, the optimal beam size will also go to zero, causing a singularity in one of the generalized displacement or strain components[41]. In particular, at locations where only the bending moment vanishes a hinge-like point appears (Type I singularity[21]), whereas at locations where both the bending moment and shear force vanish a complete separation of the beam (Type II singularity[40]) can occur. This may compromise the well-position of the Sturm-Liouville operator underlying the optimal design problem[9, 6, 7], and several works have focused on finding the correct setting, and admissible class of optimal size distributions to avoid this[37, 14, 19].

Another delicate point concerns the very existence of non-trivial optimal solutions, depending on the relationship between stiffness and mass, the introduction of design-independent inertia terms, and the particular choice of boundary conditions[46]. These matters have been the focus of many mathematical studies, such as those by Brach[11, 12], Holmåker[27], and Vepa[57].

The literature on optimal beam design against buckling is equally vast[13]. Limiting the discussion to recent times, Keller[34] started it off by giving a closed-form solution for the optimal design of a simply supported column with maximum buckling load, under a uniform tip compression. Then, Tadjbakhsh and Keller[55] extended to other boundary conditions, including the clamped beam, and this latter case was then reassessed by Olhoff and Rasmussen[42], accounting for the bimodal character of the optimal solution. The design of a heavy column, where the compression varies along the beam’s axis according to self-weight, was addressed by Keller and Niordson[35], using a combination of asymptotic and numerical methods. In this case, the same technical caveats encountered for frequency optimization are raised, as discussed by Cox and McCarthy[16, 38], or Barnes[10, 8], who introduced minimum requirements on the optimal beam size to ensure well position of the problem. From a more practical stance, the regularization of the problem by introduction of a minimum size had already been proposed by Adali [1].

Buckling design under the combined effect of self-weight and a tip, design-independent compression, was studied by Atanackovic, [2], also considering the bimodal optimization of the clamped column [3, 4]. Other recent works on the topic are from Sadiku [52, 53], Seyranian and Privalova[54], and Egorov [17, 18].

In the author’s knowledge, no research work has yet systematically addressed the interplay between buckling and frequency optimization, while considering self-weight. The work by Rammerstorffer[47] is the only one we are aware of discussing the influence of a pre-stress on the beam frequency optimization problem. However, the analysis was limited to uniform pre-stress, simply supported configuration, and a linear control, since the design variable was the Young’s modulus.

As the optimal size of frequency optimized beams shows large variations, and the tendency to vanish at some locations, a question about the beams’ stability naturally arises. The designs in Figure 1 suggest that the buckling strength of beams optimized purely for vibrations may be severely compromised, to the extent that even self-weight may trigger instability.

Sketch of a column loaded by the tip force $P$ and distributed axial load $\tilde{q}(s)$, and optimal size distributions for four boundary conditions

Figure 1. (a) Sketch of a column loaded by the tip force PP and distributed axial load q~(s)\tilde{q}(s), considering the cantilever configuration. (b) Optimal size distribution, maximizing the fundamental frequency of vibration Ω1\Omega_1, for the four cases of boundary conditions: simply supported (SS), cantilever (CF), clamped-pinned (CP), and fully clamped (CC). For each case, rΩr_\Omega is the maximum relative frequency gain, given by [39, 40]

Here we aim at providing a broad analysis of the influence of compression on the optimization outcomes. We address the classical problem of maximizing the fundamental frequency of vibration, introducing both constant (size-independent) and variable (size-dependent) compression sources. First, we discuss how these enter the optimally conditions of the problem, also considering the particular case of self-weight. Then, we provide numerical results considering the two compression sources acting either separately or simultaneously. We will show that both substantially change the optimized design, and, generally, also the frequency improvement which can be obtained. Depending on the beam’s boundary conditions, and for a compression which is close to the critical one, the natural frequency increases between 7 to 14.2 times, when the applied compression is uniform, and up to 15 times when it is size-dependent. These improvements are much higher than those achieved when optimizing an unloaded beam[39, 33], making frequency optimization even more appealing for higher compressed beams[47].

We also discuss the indirect effect that the optimal design process has on the buckling capacity of the beam. Again, this heavily depends on the amount of compression considered in the optimization, and it is not obvious that the optimized design is stronger than the uniform one as soon as we account for compression in the optimization process.

Finally, we point out that here we consider the technical Bernoulli beam model, following the majority of the research works in the field. Relatively few works have dealt with the Timoshenko beam[29, 32, 36], or with geometrically nonlinear models[20, 45], or again with non-conservative loads[23, 48].

Governing equations and statement of the optimization problem

We consider the initially slender, straight, non-uniform column depicted in Figure 1(a), whose axis occupies the interval [0,L]⊂R[0,L]\subset\mathbb{R}. The column is subjected to the tip force of magnitude PP, and to the distributed axial load with intensity q~(s)\tilde{q}(s), s∈(0,L)s\in(0,L), both inducing compression.

Assuming the moderately nonlinear kinematic Bernoulli theory, the Lagrangian density associated with the beam’s small amplitude transversal vibrations is

ℓ[s,v~(s,t)]=12[ρAv~t2+(P+∫sLq~(ζ) dζ)v~s2−EJv~ss2](1)\ell[s,\tilde{v}(s,t)]=\frac{1}{2}\left[\rho A\tilde{v}_{t}^{2}+\left(P+\int_{s}^{L}\tilde{q}(\zeta)\,\mathrm{d}\zeta\right)\tilde{v}_{s}^{2}-EJ\tilde{v}_{ss}^{2}\right] \tag*{(1)}

where ss and tt are the space and time coordinates, v~(s,t)\tilde{v}(s,t) is the lateral deflection, A=A(s)A=A(s), ρ=ρ(s)\rho=\rho(s), J=J(s)J=J(s) and E=E(s)E=E(s) are the beam area, density, modulus of inertia and Young’s modulus, respectively. In Equation (1), and later on, subscript lowercase letters will denote differentiation with respect to the corresponding variable.

Taking one characteristic dimension of the cross section as variable over the beam length, according to the function y~(s)∈Y:={y∈C0(0,L)∣y~min⁡≤y~(s)≤y~max⁡}\tilde{y}(s)\in\mathcal{Y}:=\{y\in C^{0}(0,L)\mid\tilde{y}_{\min}\le\tilde{y}(s)\le\tilde{y}_{\max}\} [37, 10], the area becomes A(s)=y0y~(s)A(s)=y_{0}\tilde{y}(s), where y0y_{0} is a constant length [28, 25]. The modulus of inertia can be expressed as

J(s,p)=c1A(s)p=αy02(2−p)[y0y~(s)]p=αy04−py~p(s)(2)J(s,p)=c_{1}A(s)^{p}=\alpha y_{0}^{2(2-p)}\left[y_{0}\tilde{y}(s)\right]^{p}=\alpha y_{0}^{4-p}\tilde{y}^{p}(s) \tag*{(2)}

where c1=αy02(2−p)c_{1}=\alpha y_{0}^{2(2-p)}, α∈R\alpha\in\mathbb{R}, and p∈N+p\in\mathbb{N}_{+} (usually, p={1,2,3}p=\{1,2,3\}). Choosing the y0y_{0} and pp, (2) can be particularized to the case of some commonly used cross sections shapes [44].

Introducing the non-dimensional variables x=s/Lx=s/L, τ=t/T\tau=t/T, v(x,τ)=v~(s,t)/Lv(x,\tau)=\tilde{v}(s,t)/L, y(x)=y0Ly~(s)/Vy(x)=y_{0}L\tilde{y}(s)/V, and q(x)=Lq~(s)/Qq(x)=L\tilde{q}(s)/Q, where VV is the beam’s volume and Q=∫0Lq~(s) dsQ=\int_{0}^{L}\tilde{q}(s)\,\mathrm{d}s the magnitude of the distributed load, the substitution of (2) in (1) gives

ℓ[x,v(x,τ)]=12[ϖyvτ2+(λP+λQ∫x1q(ζ) dζ)vx2−ypvxx2](3)\ell[x,v(x,\tau)]=\frac{1}{2}\left[\varpi yv_{\tau}^{2}+\left(\lambda_{P}+\lambda_{Q}\int_{x}^{1}q(\zeta)\,\mathrm{d}\zeta\right)v_{x}^{2}-y^{p}v_{xx}^{2}\right] \tag*{(3)}

where we have introduced the non-dimensional parameters

λP=PVβ,λQ=QVβ,ϖ=ρLT2β(4)\lambda_{P}=\frac{P}{V\beta},\qquad\lambda_{Q}=\frac{Q}{V\beta},\qquad\varpi=\frac{\rho L}{T^{2}\beta} \tag*{(4)}

and β=EαVp−1y04−2p/Lp+2\beta=E\alpha V^{p-1}y_{0}^{4-2p}/L^{p+2}.

For λP<λP(Cr)\lambda_{P}<\lambda_{P}^{(\mathrm{Cr})} and λQ<λQ(Cr)\lambda_{Q}<\lambda_{Q}^{(\mathrm{Cr})}, where λP(Cr)\lambda_{P}^{(\mathrm{Cr})} and λQ(Cr)\lambda_{Q}^{(\mathrm{Cr})} are the critical multipliers for the tip force and distributed load, respectively, the beam will undergo small amplitude harmonic oscillations. Therefore, we assume v(x,τ)=v(x)e−iωτv(x,\tau)=v(x)e^{-i\omega\tau} and vττ=−ω2v(x)v_{\tau\tau}=-\omega^{2}v(x) (ω>0\omega>0), and imposing stationarity of (3) with respect to δv∈V:=H02(0,1)\delta v\in\mathcal{V}:=H_{0}^{2}(0,1) gives the elasto-dynamic linear field equation governing the beam’s motion

(ypvxx)xx+(λP+λQ∫x1q(ζ) dζ)vxx−λQq(x)vx−Ω2yv=0x∈(0,1)(5)(y^{p}v_{xx})_{xx}+\left(\lambda_{P}+\lambda_{Q}\int_{x}^{1}q(\zeta)\,\mathrm{d}\zeta\right)v_{xx}-\lambda_{Q}q(x)v_{x}-\Omega^{2}yv=0\qquad x\in(0,1) \tag*{(5)}

where Ω2:=ϖω2\Omega^{2}:=\varpi\omega^{2} is the dynamic eigenvalue, proportional to the square of the angular frequency ω\omega. The boundary terms arising from the integration by parts of (3) can be collected in the linear operator

B[v,δv]:=[(ypvxx)x+(λP+λQ∫x1q(ζ) dζ)vx]δv−ypvxxδvx∣x={0,1}(6)B[v,\delta v]:=\left[\left(y^{p}v_{xx}\right)_{x}+\left(\lambda_{P}+\lambda_{Q}\int_{x}^{1}q(\zeta)\,\mathrm{d}\zeta\right)v_{x}\right]\delta v-y^{p}v_{xx}\delta v_{x}\bigg|_{x=\{0,1\}} \tag*{(6)}

corresponding to one of the following pairs of boundary conditions at x={0,1}x=\{0,1\}

eitherv=0or(ypvxx)x+(λP+λQ∫x1q(ζ) dζ)vx=0eithervx=0orypvxx=0(7)\begin{aligned} \text{either}\quad&v=0\quad\text{or}\quad(y^{p}v_{xx})_{x}+\left(\lambda_{P}+\lambda_{Q}\int_{x}^{1}q(\zeta)\,\mathrm{d}\zeta\right)v_{x}=0\\ \text{either}\quad&v_{x}=0\quad\text{or}\quad y^{p}v_{xx}=0 \tag*{(7)} \end{aligned}

When the distributed axial load is proportional to self-weight, we have q~(s)=ρgA(s)=ρgy0y~(s)\tilde{q}(s)=\rho gA(s)=\rho gy_{0}\tilde{y}(s) and (5) and (6) hold with q(ζ)=y(ζ)q(\zeta)=y(\zeta) and λQ=ρgβ−1\lambda_{Q}=\rho g\beta^{-1}.

Optimization problem and solution strategy

We aim at finding the optimal size distribution y∗y_* ∈Y\in\mathcal{Y} maximizing the beam’s fundamental frequency Ω1=Ω1(λP,λQ)\Omega_1 = \Omega_1(\lambda_P,\lambda_Q), for a prescribed total volume fraction VV and compression level, given by the pair (λP,λQ)(\lambda_P,\lambda_Q). This can be stated as

max⁡y∈YΩ12(λP,λQ):={min⁡v∈V−0⟨ypvxx,vxx⟩[0,1]−⟨(λP+λQI[x,y])vx,vx⟩[0,1]⟨yv,v⟩[0,1]}\max_{y\in\mathcal{Y}}\Omega_1^2(\lambda_P,\lambda_Q) := \left\{\min_{v\in\mathcal{V}^{-0}}\frac{\langle y^p v_{xx},v_{xx}\rangle_{[0,1]}-\langle(\lambda_P+\lambda_Q I[x,y])v_x,v_x\rangle_{[0,1]}}{\langle yv,v\rangle_{[0,1]}}\right\}
s.t.⟨y,1⟩[0,1]−1≤0(8)\mathrm{s.t.}\quad\langle y,1\rangle_{[0,1]}-1\leq0 \tag*{(8)}

where we have introduced the shorthand notation ⟨u,v⟩[a,b]=∫abuv dx\langle u,v\rangle_{[a,b]}=\int_a^b uv\,\mathrm{d}x, and set I[x,y]=⟨q(ξ,y(ξ)),1⟩[x,1]I[x,y]=\langle q(\xi,y(\xi)),1\rangle_{[x,1]}. In this latter term we account for a general dependence on both xx and yy, allowing for the case of self-weight.

The objective of (8) is the squared frequency, reduced by both the tip and axially distributed compression, and expressed through the Rayleigh quotient[59]. This can be related to the one of the unloaded beam Ωˉ12\bar{\Omega}_1^2 by

Ω12(λP,λQ)=Ωˉ12−⟨(λP+λQI[x,y])vx,vx⟩[0,1](9)\Omega_1^2(\lambda_P,\lambda_Q)=\bar{\Omega}_1^2-\langle(\lambda_P+\lambda_Q I[x,y])v_x,v_x\rangle_{[0,1]} \tag*{(9)}

where, without loss of generality, we assumed ⟨yv,v⟩[0,1]=1\langle yv,v\rangle_{[0,1]}=1. Equation (9) expresses the well-known inverse proportionality between the squared natural frequency and axial compression[58], and we remark that Ω12=0\Omega_1^2=0 when either λP=λP(Cr)\lambda_P=\lambda_P^{(\mathrm{Cr})} or λQ=λQ(Cr)\lambda_Q=\lambda_Q^{(\mathrm{Cr})}.

Taking the variation of the Lagrangian associated with (8), with respect to δy∈Y\delta y\in\mathcal{Y}, we obtain the adjoint equation

pyp−1vxx2−λQIy[x,y]vx2−Ω12v2−μ=0(10)py^{p-1}v_{xx}^2-\lambda_Q I_y[x,y]v_x^2-\Omega_1^2v^2-\mu=0 \tag*{(10)}

where Iy[x,y]=δI[x,δy]I_y[x,y]=\delta I[x,\delta y] and μ∈R+\mu\in\mathbb{R}_+ is the Lagrange multiplier associated with the volume constraint. Assuming that such constraint is active at the optimum[5], we have

μ∗=(p−1)Ωˉ12+⟨[λP+λQ(I[x,y]−yIy[x,y])]vx,vx⟩[0,1](11)\mu_*=(p-1)\bar{\Omega}_1^2+\langle[\lambda_P+\lambda_Q(I[x,y]-yI_y[x,y])]v_x,v_x\rangle_{[0,1]} \tag*{(11)}

The optimal beam size distribution is then found by solving (10) with respect to yy

y∗(x)=[μ+λQIy[x,y]vx2+Ω12v2pvxx]1−p(12)y_*(x)=\left[\frac{\mu+\lambda_Q I_y[x,y]v_x^2+\Omega_1^2v^2}{pv_{xx}}\right]^{1-p} \tag*{(12)}

and once this is known, the optimal displacement v∗v_* is computed from (5), paired with the appropriate set of boundary conditions from (7), and the eigenvalue Ω1∗2\Omega_{1*}^2 is given by the condition ⟨y,1⟩[0,1]=1\langle y,1\rangle_{[0,1]}=1.

The set of nonlinear field equations governing the state, adjoint and optimal size is summarized here

0=(ypvxx)xx+γ1vxx−γ0vx−Ωˉ12yv0=pyp−1vxx2−γ2vx2−(Ωˉ12−⟨γ1vx,vx⟩[0,1])v2−(p−1)Ωˉ12−⟨(γ1−yγ2)vx,vx⟩[0,1]y=[γ2vx2+(Ωˉ12−⟨γ1vx,vx⟩[0,1])v2+(p−1)Ωˉ12+⟨(γ1−yγ2)vx,vx⟩[0,1]pvxx]1−p(13)\begin{aligned} 0 &= (y^pv_{xx})_{xx}+\gamma_1v_{xx}-\gamma_0v_x-\bar{\Omega}_1^2yv \\ 0 &= py^{p-1}v_{xx}^2-\gamma_2v_x^2-\left(\bar{\Omega}_1^2-\langle\gamma_1v_x,v_x\rangle_{[0,1]}\right)v^2-(p-1)\bar{\Omega}_1^2-\langle(\gamma_1-y\gamma_2)v_x,v_x\rangle_{[0,1]} \\ y &= \left[\frac{\gamma_2v_x^2+\left(\bar{\Omega}_1^2-\langle\gamma_1v_x,v_x\rangle_{[0,1]}\right)v^2+(p-1)\bar{\Omega}_1^2+\langle(\gamma_1-y\gamma_2)v_x,v_x\rangle_{[0,1]}}{pv_{xx}}\right]^{1-p} \tag*{(13)} \end{aligned}

where we have set γ0:=λQq(x)\gamma_0:=\lambda_Qq(x), γ1:=λP+λQI[x,y]\gamma_1:=\lambda_P+\lambda_QI[x,y], and γ2:=λQIy[x,y]\gamma_2:=\lambda_QI_y[x,y].

Equation (13) is a system of non-linear, coupled, integro-differential equations expressing the optimality conditions for the optimization problem of (8). We notice that the design variable yy enters non-linearly in both the state and adjoint equations, whereas the state variable vv appears only linearly in the state equation, as the coefficients γi\gamma_i do not depend on it. Also, we remark that in (13) we have written everything in terms of the eigenvalue of the unloaded beam Ωˉ12\bar{\Omega}_1^2, and explicitly substituted for the Lagrange multiplier expression (11), to fully show the influence of each term, in the most general case. However, we will see in the following example that (13) can be considerably simplified, when tackling specific cases.

Numerical results

We consider a slender beam with aspect ratio L/y0=12L/y_0=12, where y0=0.5y_0=0.5 is the constant cross section dimension, and also the upper bound yˉmax⁡\bar{y}_{\max} for the size distribution. The maximum allowed volume is V=0.25y0LV=0.25y_0L, and the Young modulus and density are E=2.5⋅107 MPaE=2.5\cdot10^7\,\mathrm{MPa} and ρ=2.4 kg/m3\rho=2.4\,\mathrm{kg}/\mathrm{m}^3. We assume sections[44], setting p=2p=2 and α=1/12\alpha=1/12 in (2). Four cases of boundary conditions, also called beam configurations in the following, are considered: cantilever (CF), simply supported (SS), doubly clamped (CC) and clamped-pinned (CP).

The solution strategy for (8) is based on iterating between the state and adjoint equations (13), until both conditions ∥y(k)−y(k−1)∥∞≤10−3\lVert y^{(k)} - y^{(k-1)} \rVert_\infty\leq10^{-3} and ∣Ω(k)−Ω(k−1)∣≤10−4\lvert\Omega^{(k)} - \Omega^{(k-1)} \rvert\leq10^{-4} are fulfilled. This is carried out repeatedly, starting by setting the lower bound y~min⁡=10−3\tilde{y}_{\min}=10^{-3}, then reducing it until y~min⁡=10−10\tilde{y}_{\min}=10^{-10}. In each of these steps, we use the final size distribution from the previous one as initial guess. We will take that a given design shows a Type I, or Type II singularity[40] if its final size distribution y∗y_* hits y~min⁡\tilde{y}_{\min} at an isolated point, or at multiple adjacent points, respectively. The state variable vhv_h is discretized by C1C^1 elements[31], ensuring nodal continuity of both vv and vxv_x, whereas the design variable yhy_h is only continuous at nodes.

In the following, the tip compression is scaled by λP/λPCr(0)∈[0,1)\lambda_P/\lambda_P^{\mathrm{Cr}}(0) \in[0,1), where λPCr(0)=Lc−2π2EJ0\lambda_P^{\mathrm{Cr}}(0)=L_c^{-2}\pi^2EJ_0 is the critical load of the uniform beam, and LcL_c is the characteristic length depending on the boundary conditions[56]. Likewise, the distributed axial load is scaled by λQ/λQCr(0)∈[0,1)\lambda_Q/\lambda_Q^{\mathrm{Cr}}(0) \in[0,1), where λQCr(0)\lambda_Q^{\mathrm{Cr}}(0) is the critical load corresponding to a constant load distribution. To ease the notation, we will denote these multipliers by the symbols λP\lambda_P and λQ\lambda_Q. As we focus on the fundamental vibration frequency, we drop the subscript “1”, and we denote the frequencies of the uniform and optimized size distributions by Ω0\Omega_0 and Ω∗\Omega_*, respectively. Also, we will call “reduced frequencies” those reduced by the axial compression through (9).

Influence of the tip force alone (P 0, Q = 0)

In this case γ0=γ2=0\gamma_0=\gamma_2=0, γ1=λP\gamma_1=\lambda_P, and the state and adjoint equations in (13) simplify to

0=(ypvxx)xx+λPvxx−Ωˉ2yv0=pyp−1vxx2−(Ωˉ2−λP⟨vx,vx⟩[0,1])v2−(p−1)Ωˉ2−λP⟨vx,vx⟩[0,1](14)\begin{aligned} 0 &= (y^p v_{xx})_{xx}+\lambda_P v_{xx}-\bar{\Omega}^{2}yv \\ 0 &= p y^{p-1}v_{xx}^{2}-\left(\bar{\Omega}^{2}-\lambda_P\langle v_x,v_x\rangle_{[0,1]}\right)v^{2}-(p-1)\bar{\Omega}^{2}-\lambda_P\langle v_x,v_x\rangle_{[0,1]} \tag*{(14)} \end{aligned}

and the design-independent compression influences the solution of (8) through the state equation and the Lagrange multiplier μ\mu.

The compression, which is constant during the optimization, is applied in the range λP∈[0,0.995]\lambda_P \in[0,0.995], with ΔλP=1/200\Delta\lambda_P=1/200. We refer to the smallest level of non-zero compression (λP=0.005\lambda_P=0.005) as λP≈0\lambda_P\approx0, and to the last multiplier as λP≈1\lambda_P\approx1. The BLF of the beam, say λPCr\lambda_P^{\mathrm{Cr}}, will be indirectly modified during the optimization, and we must check that

λP≤λPcr\lambda_P \leq\lambda_P^{\mathrm{cr}} at each step (see Subsection 3.4); otherwise the beam will undergo divergence, voiding the assumption of harmonic motion.

Figure 2 shows the influence of the compression level on the frequency of the uniform beam (Ω0\Omega_0, blue dashed line), and of the optimized one (Ω∗\Omega_*, blue continuous line). While the first follows the well-known (nearly) inverse proportionality between Ω02\Omega_0^2 and λP\lambda_P [58], the frequency attained in the optimization decreases less than proportionally with the compression level. The relative frequency gain corresponding to each compression level is given by the ratio rΩ:=Ω∗/Ω0r_\Omega:= \Omega_*/\Omega_0 (see red line in Figure 2). For λP=0\lambda_P = 0 (optimization of the unloaded beam), rΩr_\Omega attains values very close to the theoretical bounds recalled in Figure 1(b). We clearly have rΩ→∞r_\Omega\to\infty as λP→1\lambda_P \to1; however, we recall that we bound λP\lambda_P to be close but slightly below one. The largest absolute rΩr_\Omega is reached by the CF configuration, which for λP≈1\lambda_P \approx1 shows a frequency gain of over 14 times, whereas the SS configuration is the one showing the largest relative improvement between rΩ(0)r_\Omega(0) (unloaded beam) and rΩ(≈1)r_\Omega(\approx1) (highly compressed beam).

Relationship between normalized tip compression and frequency for SS, CF, CS, and CC beams

Figure 2. Relationship between the normalized tip compression λP\lambda_P and the frequency of the uniform (Ω0\Omega_0, blue dashed curve), and optimized beams (Ω∗\Omega_*, blue continuous curve). The frequency gains obtained by solving (8) for each λP\lambda_P, say rΩr_\Omega, are plotted as a red continuous curve. For the CF, CS, and CC beams rΩr_\Omega is non-monotonic and attains the minimum value at λP=l1\lambda_P = l_1 (white dot), whereas λP=l2\lambda_P = l_2 (black dot) corresponds to the condition rΩ(l2)=rΩ(0)r_\Omega(l_2) = r_\Omega(0)

For all the configurations but the SS, the relationship between λP\lambda_P and rΩr_\Omega is non-monotonic. For the CF, CS and CC beams, as soon as λP>0\lambda_P > 0 the frequency attained by the optimized beam (Ω∗\Omega_*) decreases much more rapidly than the frequency of the uniform one, and rΩr_\Omega drops. After reaching a minimum at λP=l1\lambda_P = l_1 (white dot in Figure 2), the frequency gain rΩr_\Omega equals the one for the unloaded beam at λP=l2\lambda_P = l_2 (black dot in Figure 2), then increases further as λP→1\lambda_P \to1.

This non-monotonic behaviour can be explained by looking at how the optimized size distribution y∗(x)y_*(x) changes with the increase of the compression (Figure 3). For the SS beam, the optimized size shows minor changes and only gets thicker near the beam center, and more flat near the pinned ends, as λP\lambda_P increases. For all the other beam configurations, the optimized size shows an abrupt change as soon as λP>0\lambda_P > 0, and immediately adapts to sustain the design-independent tip load. The size distribution is prevented from vanishing according to a Type II singularity, whereas we still have points resembling a Type I singularity, where the bending moment vanishes (see Appendix A). For the CS and CC configurations, these points appear within the beam domain as soon as λP>0\lambda_P > 0 and, for λP≈1\lambda_P \approx1, they are at x≈1/3x \approx1/3 and x≈1/6x \approx1/6, respectively[21]. However, we stress that these points do not strictly correspond to singularities, since we always have y∗>ymin⁡y_* > y_{\min}, as soon as λP>0\lambda_P > 0.

Optimized size distributions for SS, CF, CS, and CC beam configurations

Figure 3. Optimized size distribution y∗(x)y_*(x), relative to the initial one y0y_0, for each beam configuration. On the 3D views, we highlight the designs corresponding to the tip compression values: λP=0\lambda_P = 0 (blue), λP≈1\lambda_P \approx1 (black), and λP=l2\lambda_P = l_2 (red). This latter corresponds to rΩ(l2)=rΩ(0)r_\Omega(l_2) = r_\Omega(0), and the frequency gain equals that of the optimal unloaded beam. The same designs are highlighted on the 2D views, where we also mark the one corresponding to the slightest applied compression λP≈0\lambda_P \approx0 (blue dashed curve)

The optimized design shown as a red dashed line in Figure 3 corresponds to the compression level λP=l2\lambda_P = l_2, where we match the frequency gain of the unloaded beam. We also plot, as green dashed line, the one corresponding to λP=l3\lambda_P = l_3); this is the first designs meeting the same buckling strength of the uniform beam, as will be further discussed in Subsection 3.4.

The above discussion shows that for the CF, CS, and CC beams the magnitude of λP\lambda_P has a huge effect on the frequency optimization. Even the slightest design-independent compression prevents the optimal beam size from vanishing according to a Type II singularity, and this drastically lowers the frequency gain. Then, as the compression level increases, the reduction in the initial frequency is so severe that the frequency gain becomes even higher than the one reached when optimizing the unloaded beam. For the SS beam, the frequency gain is always higher for the compressed beam, compared to the unloaded one.

Influence of the self-weight (λP≥0\lambda_P \geq0, q(x)=y(x)q(x)=y(x), λQ=1\lambda_Q=1)

We repeat the analysis of the previous section, now introducing the self-weight contribution. This amounts to the full system of equations (13), where q(x)=y(x)q(x)=y(x), I[x,y]=⟨y,1⟩[x,1]I[x,y]=\langle y,1\rangle_{[x,1]}, and Iy[x,y]=(1−x)I_y[x,y]=(1-x). Thus, we can simplify γ0=λQy\gamma_0=\lambda_Q y, γ1=λP+λQ⟨y,1⟩[x,1]\gamma_1=\lambda_P+\lambda_Q\langle y,1\rangle_{[x,1]}, and γ2=λQ(1−x)\gamma_2=\lambda_Q(1-x).

We remark that, due to the self-weight, buckling may also happen with λP=0\lambda_P=0. This will be discussed in the next section, and we now fix λQ=1\lambda_Q=1. In the following, we will refer to the beam loaded only by the tip force as the “light beam”, whereas the “heavy beam” also accounts for the self-weight.

We refer to Figure 4 where, against the left axis, we plot ΔΩ(SW):=Ω/Ω(SW)−1\Delta_{\Omega}^{(\mathrm{SW})}:=\Omega/\Omega^{(\mathrm{SW})}-1, which is the relative difference between the reduced frequency for the light beam (Ω\Omega, same as in the previous section), and that of the heavy beam (Ω(SW)\Omega^{(\mathrm{SW})}). For the uniform design, this quantity is clearly larger than one, as the self-weight introduces an additional compression, further reducing the vibration frequency. Also, it is monotonically increasing with λP\lambda_P for all the beam configurations, and this comes from the coupling of λQ\lambda_Q in the vxv_x and vxxv_{xx} terms of the state equation.

Three panels showing the relationship between normalized tip compression $\lambda_P$ and relative frequency reduction. (a) CF ($r_{\Omega}^{(\mathrm{SW})}(\approx 1)=19.513$, $l_4=0.385$, $l_5=0.455$); (b) CS ($r_{\Omega}^{(\mathrm{SW})}(\approx 1)=8.4650$, $l_4=0.375$, $l_5=0.355$); (c) SS ($r_{\Omega}^{(\mathrm{SW})}(\approx 1)=8.489$, $l_4=0$)

Figure 4. Relationship between the normalized tip compression λP\lambda_P and the relative reduction in the frequency due to the self-weight, for the uniform (blue dashed curve), and optimized (blue continuous curve) beams. The red curve shows the relative reduction of the frequency gains ΔrΩ(SW)\Delta r_{\Omega}^{(\mathrm{SW})}. All curves are scaled according to the log-abs transform. The white dot marks the point λP=l4\lambda_P=l_4, where we have the minimum of ΔΩ(SW)\Delta_{\Omega}^{(\mathrm{SW})}, and the black dot marks the point λP=l5\lambda_P=l_5, where ΔrΩ(SW)=0\Delta r_{\Omega}^{(\mathrm{SW})}=0; thus the self-weight has no influence on the frequency gain. The horizontal dotted lines bound the region where ∣ΔrΩ(SW)∣≤10−3|\Delta r_{\Omega}^{(\mathrm{SW})}|\leq10^{-3}. For the CC configuration the trend is basically the same as for the CS, and we refer to Table 1 for numerical values

Let us focus the discussion on the CF configuration first. For the uniform beam (blue dashed curve), the self-weight alone slightly reduces the frequency (ΔΩ(SW)≈0.125%\Delta_{\Omega}^{(\mathrm{SW})}\approx0.125\% for λP=0\lambda_P=0). Then, its effect increases with increasing λP\lambda_P, and for λP≈1\lambda_P\approx1 we have ΔΩ(SW)≈38%\Delta_{\Omega}^{(\mathrm{SW})}\approx38\%. The trend is the opposite when looking at the relative difference in the frequency attained by the optimized beam (blue continuous curve). Now the largest difference happens for λP=0\lambda_P=0 (ΔΩ(SW)≈28%\Delta_{\Omega}^{(\mathrm{SW})}\approx28\%), as the self-weight is the first source of compression perturbing the optimal design from the one corresponding to the unloaded beam. Then, as λP\lambda_P increases and the size-independent compression dominates the self-weight, ΔΩ(SW)\Delta_{\Omega}^{(\mathrm{SW})} quickly drops to a few ‰. However, for the optimized frequency, the ratio ΔΩ(SW)\Delta_{\Omega}^{(\mathrm{SW})} is non-monotonous and after reaching the minimum at λP=l4=0.385\lambda_P=l_4=0.385, the influence of self-weight (slightly) increases again, giving ΔΩ(SW)≈0.415%\Delta_{\Omega}^{(\mathrm{SW})}\approx0.415\% for λP≈1\lambda_P\approx1.

This trend is also reflected by the relative difference between the frequency gains for the light and heavy beams, ΔrΩ(SW):=rΩ/rΩ(SW)−1\Delta r_{\Omega}^{(\mathrm{SW})}:=r_{\Omega}/r_{\Omega}^{(\mathrm{SW})}-1 (see red curves in Figure 4). For low magnitudes of the tip compression we have ΔrΩ(SW)>0\Delta r_{\Omega}^{(\mathrm{SW})}>0, and the optimization gives a higher frequency gain for the light beam, compared to the heavy one. For λP=l5≈0.455\lambda_P=l_5\approx0.455 we have ΔrΩ(SW)=0\Delta r_{\Omega}^{(\mathrm{SW})}=0, and the frequency gains for the heavy and light beams are exactly the same. Then, for higher values of the tip compression we have ΔrΩ(SW)<0\Delta r_{\Omega}^{(\mathrm{SW})}<0 and, for λP≈1\lambda_P\approx1 the heavy beam benefits from the optimization about 27.5% more, compared to the light one. However, we observe that apart form the two regions where λP\lambda_P is close to zero or one, the self-weight has a very little influence on the optimization outcomes. For instance, when 0.175≤λP≤0.6550.175\leq\lambda_P\leq0.655 we have ∣ΔrΩ(SW)∣≤10−3|\Delta r_{\Omega}^{(\mathrm{SW})}|\leq10^{-3}, and therefore the self-weight has a negligible impact on the frequency gain (see horizontal lines in Figure 4).

The CS and CC configurations share the same qualitative behaviour of the CF, tough we have different values for the ratios ΔΩ(SW)\Delta_{\Omega}^{(\mathrm{SW})}, ΔrΩ(SW)\Delta r_{\Omega}^{(\mathrm{SW})}, and location of points l4l_4 and l5l_5 (see also Table 1). Also, both configurations show a much wider range where ∣ΔrΩ(SW)∣≤10−3|\Delta r_{\Omega}^{(\mathrm{SW})}|\leq10^{-3}, and therefore the self-weight has negligible influence on the optimization outcomes (λP∈[0.01,0.88]\lambda_P\in[0.01,0.88] and λP∈[0.01,0.86]\lambda_P\in[0.01,0.86] for the CS and CC beams, respectively). This can be clearly seen by the trend of the red curves in Figure 4, for the CF and CS beams. For the SS configuration, both ΔΩ(SW)\Delta_{\Omega}^{(\mathrm{SW})} and ΔrΩ(SW)\Delta r_{\Omega}^{(\mathrm{SW})} are monotonically increasing and decreasing, respectively. Therefore, as λP\lambda_P increases, we gain systematically more when optimizing the heavy beam, compared to the light one.

λP=λQ=0\lambda_P=\lambda_Q=0λP≥0,λQ=0\lambda_P\geq0,\lambda_Q=0λP≥0,q(x)=y(x),λQ=1\lambda_P\geq0,q(x)=y(x),\lambda_Q=1λP=0,q(x)=y(x),λQ≥0\lambda_P=0,q(x)=y(x),\lambda_Q\geq0
rΩr_\OmegarΩ(λP≈1)r_\Omega(\lambda_P\approx1)(l1,rΩ(l1))(l_1,r_\Omega(l_1))l2l_2l3l_3rΩ(SW)(λP≈1)r_\Omega^{(\mathrm{SW})}(\lambda_P\approx1)l4l_4l5l_5rΩ(λQ≈1)r_\Omega(\lambda_Q\approx1)l6l_6l7l_7
SS1.0667.605(0,1.066)(0,1.066)0-8.489-07.6890-
CF6.81014.196(0.56,2.729)(0.56,2.729)0.9750.719.5130.3850.45515.0780.970.730
CS1.5568.178(0.3,1.132)(0.3,1.132)0.7550.3358.4650.3750.3557.1900.7950.96
CC4.3168.680(0.525,1.597)(0.525,1.597)0.9750.898.9710.4350.425.7750.8950.94

Table 1. Summary of the frequency gains and notable points obtained in the numerical examples. The top labels follow the same naming of Subsection 3.1-Subsection 3.4. rΩ(⋅)r_\Omega(\cdot) is the frequency gain, for a given level of compression. The notable compression values are l1l_1: λP\lambda_P value that gives the minimum frequency gain, (l2l_2 and l6l_6): λP\lambda_P and λQ\lambda_Q value that gives the same frequency gain as for the unloaded beam, l4l_4 minimum frequency gain when considering the self-weight contribution, l5l_5, zero influence of the self-weight on the frequency gain, (l3l_3 and l7l_7) same buckling capacity as the initial beam

The effect of self-weight on the optimized size distributions is shown in Figure 5. Here, the SS beam is not shown, as for this case we have only minor design changes. The plots in the top row compare the relative size distribution of the light (dashed line) and of the heavy beams (continuous line), for three notable tip compression values: λP=0\lambda_P = 0, λP≈1\lambda_P \approx1, and λP=l5\lambda_P = l_5. For all the beam configurations, we can distinguish between y∗(SW)y_*^{(\mathrm{SW})} and y∗y_* only for λP=0\lambda_P = 0, as the self-weight has the effect of reducing the size at the clamped ends, while increasing it at the points where the unloaded beam would develop a Type II singularity. For the compression levels λP=l5\lambda_P = l_5 and λP≈1\lambda_P \approx1, y∗(SW)y_*^{(\mathrm{SW})} and y∗y_* are almost indistinguishable; however, we can refer to the plots of the bottom row, showing a logarithmic plot of their difference, and recognize a similar trend. For λP=0\lambda_P = 0, the CF beam’s sizes show the largest differences due to self-weight. At x=0x = 0 the size is reduced of about 30%, whereas the maximum increase of about 13.5% occurs at x≈0.5x \approx0.5. For the CS beam, the size’s maximum decrease and increase are lower (4.3% and 5.5%), and both occur within the domain, at about x≈0.75x \approx0.75. For the CC beam, the size is reduced of about 24% at the clamped ends, and the maximum increase (10%) occurs, interestingly, at the symmetrical locations x≈0.25x \approx0.25 and x≈0.75x \approx0.75, and not at the mid-point. In all cases, for the higher compression levels the size relative differences quickly drops to very small values. However, it is interesting to note that for the CS configuration we still have size differences up to ±2%\pm2\%, corresponding to λP=l5\lambda_P = l_5, without affecting the frequency gain.

Comparison between optimized size distributions for CF, CS, and CC beams

Figure 5. Comparison between the optimized size distribution for the heavy beam (continuous line) and for the light beam (dashed line), for three values of the tip design-independent compression λP\lambda_P. We remark that λP=l5\lambda_P = l_5 correspond to the tip compression at which the self-weight has no influence on the frequency gain. The plots in the bottom row show the log-scaled relative difference between the two size distributions

Summarizing, the self-weight contribution sensibly impacts the frequency optimization both when considering zero or high values of the design-independent compression. In particular, frequency optimization becomes highly beneficial for heavy and highly compressed columns. Otherwise, the influence of self-weight on the optimization outcome is very small.

Influence of the distributed axial load alone (P=0P = 0, q(x)=y(x)q(x) = y(x), Q0Q 0)

We now remove the size-independent compression, while keeping the distributed axial load proportional to self-weight; thus the coefficients in (13) become γ0=λQγ1\gamma_0 = \lambda_Q \gamma_1, γ1=λQq[x,1]\gamma_1 = \lambda_Q q_{[x,1]}, and γ2=λQ(1−x)\gamma_2 = \lambda_Q(1-x).

We increase the axial load multiplier up to λQ=0.995\lambda_Q = 0.995, with ΔλQ=1/200\Delta\lambda_Q = 1/200. Again, we have an inverse relationship between Ω02\Omega_0^2 and λQ\lambda_Q and, for the uniform design, the multiplier causing buckling under self-weight (and thus Ω0=0\Omega_0 = 0) is λQcr={942.97,397.52,2664.6,3781.8}\lambda_Q^{\mathrm{cr}} = \{942.97, 397.52, 2664.6, 3781.8\}, for the SS, CF, CS and CC configurations, respectively [56].

We point out that scaling λQ\lambda_Q, when considering self-weight, can still be given a meaning as a “tallest column” design problem [35, 16]. Indeed, from (4) we have λQ∝L4\lambda_Q \propto L^4 and ϖ/λQ∝L\varpi/\lambda_Q \propto L. Therefore, we may alternatively interpret this as the optimal design of a tallest column with the same frequency and buckling strength of the initial one. However, the main difference with the classical tallest column problem is that now λQ\lambda_Q is set as a parameter.

Figure 6 (a,b) show the trend of the optimized frequency Ω∗=Ω∗(λQ)\Omega_* = \Omega_*(\lambda_Q) and frequency gain rΩ=rΩ(λQ)r_{\Omega} = r_{\Omega}(\lambda_Q), for the four beam configurations. For comparison, we also display, as dashed lines, the same curves corresponding to the case of tip compression, identical to those in Figure 2. For all the beam configurations, we highlight the points λQ=l6\lambda_Q = l_6, where rΩ(l6)=rΩ(0)r_{\Omega}(l_6) = r_{\Omega}(0) (see also Table 1). As for the case of tip compression, we have l6=0l_6 = 0 for the SS beam, and the frequency gain is monotonically increasing with the load parameter.

Optimized frequency and frequency gain for four beam configurations

Figure 6. Optimized frequency (a), and the frequency gain (b) for each level of the axial load multiplier λQ\lambda_Q and for the four beam configurations. The dashed curves correspond to the case of tip compression λP\lambda_P, and are the same as in Figure 2. The black dots in (b) mark the points λQ=l6\lambda_Q = l_6, where rΩ(l6)=rΩ(0)r_{\Omega}(l_6) = r_{\Omega}(0), where the frequency gain matches that of the unloaded beam. (c) shows the ratio between the frequency gain factors for the cases of distributed axial load and tip compression

For the SS and CF beams, the optimized frequency Ω∗\Omega_* corresponding to a given λQ\lambda_Q is always larger than that obtained for the same value of λP\lambda_P; also, we have rΩ(λQ)>rΩ(λP)r_{\Omega}(\lambda_Q) > r_{\Omega}(\lambda_P). Therefore, for these two configurations frequency optimization leads to larger gains when the beam is loaded by its own self-weight, compared to the tip force compression. The opposite is true for the CS and CC beams, as we now have Ω∗(λQ)<Ω∗(λP)\Omega_*(\lambda_Q) < \Omega_*(\lambda_P) and rΩ(λQ)<rΩ(λP)r_{\Omega}(\lambda_Q) < r_{\Omega}(\lambda_P), for all values of the load multiplier.

These trends are clearly visible from Figure 6(c) showing the ratio rΩ(λQ)/rΩ(λP)r_{\Omega}(\lambda_Q)/r_{\Omega}(\lambda_P), which is always above one for the SS and CF beams, and always below one for the CS and CC. For the SS configuration the ratio only shows a mild variation, whereas, for the CF beam we have a sudden increase, as the slightest amount of tip compression reduces Ω∗\Omega_* much more than the slightest amount of axial load. Then we reach the maximum ratio rΩ(λQ)/rΩ(λP)≈1.15r_{\Omega}(\lambda_Q)/r_{\Omega}(\lambda_P) \approx1.15 for λQ=0.495\lambda_Q = 0.495, and a slight decrease after. For the CS and CC configurations the decrease of this ratio is monotonic, and very strong for the CC. For the case λQ≈1\lambda_Q \approx1 we have frequency gains 1.1% and 6.2% larger, for the SS and CF configurations, and 22% and 36% smaller, for the CS and CC configurations, compared to the case λP≈1\lambda_P \approx1.

Figure 7 displays the comparison between the optimized size distribution corresponding to the axially distributed and tip load. The designs are compared for nearly critical compression values (λQ≈1\lambda_Q \approx1 and λP≈1\lambda_P \approx1), where the two show the largest differences. We see that the optimal size distributions change from those in Figure 3, to a design closer to the tallest column one [16], clearly showing a magnification of the self-weight effect. We also show the designs corresponding to the two notable compression level λQ=l6\lambda_Q=l_6, where the frequency gain matches the one for the unloaded beam, and λQ=l7\lambda_Q=l_7, where the critical load matches that of the initial design (see discussion in the next section).

Comparison between optimal size distributions for nearly critical compression

Figure 7. Comparison between the optimal size distribution for nearly critical compression, considering the tip force alone (λP≈1\lambda_P \approx1), or the distributed axial load proportional to self-weight alone (λQ≈1\lambda_Q \approx1). The two designs shown as a dashed and dotted lines correspond to the compression level λQ=l6\lambda_Q = l_6, where rΩ(l6)=rΩ(0)r_{\Omega}(l_6) = r_{\Omega}(0), and to λQ=l7\lambda_Q = l_7, respectively. This latter design has the same buckling strength of the uniform one (see Subsection 3.4)

We point out that, in none of these cases the design develops a Type II singularity and even for λQ≈0\lambda_Q \approx0 we have y∗=ymin⁡y_* = y_{\min} only at isolated points. This, which may look counter-intuitive, is due to the fact that the multiplier λQ\lambda_Q is here a constant parameter, and is not optimized for as in the classical tallest column formulation[35, 16, 38].

Indirect effect on the buckling load factor

Finally we investigate the indirect effect of maximizing Ω1\Omega_1 at a given compression level, on the beams’ critical load. This latter, which is not directly optimized, may either increase or decrease during the optimization. Figure 8 shows the ratios rλP:=λP(∗)Cr/λP(0)Crr_{\lambda_P}:=\lambda^{\mathrm{Cr}}_{P(*)}/\lambda^{\mathrm{Cr}}_{P(0)} (viz. rλQ:=λQ(0)Cr/λQ(∗)Crr_{\lambda_Q}:=\lambda^{\mathrm{Cr}}_{Q(0)}/\lambda^{\mathrm{Cr}}_{Q(*)}), between the critical load of the optimized beam and that of the uniform one, depending on the magnitude of the tip compression λP\lambda_P, or axial distributed load λQ\lambda_Q. The critical load multiplier of the compressed, uniform beam clearly follows the relationship λP(0)Cr=1/λP\lambda^{\mathrm{Cr}}_{P(0)}=1/\lambda_P (viz., λQ(0)Cr=1/λQ\lambda^{\mathrm{Cr}}_{Q(0)}=1/\lambda_Q), plotted as a black dashed line in Figure 8. Since both rλPr_{\lambda_P} and rλQr_{\lambda_Q} are always above this line, our assumption of harmonic motion is valid.

Ratio between the critical load of the optimized and uniform beams for tip force and distributed load

Figure 8. (a) Ratio between the critical load of the optimized and uniform beams, depending on the compression level considered in the optimization Equation (12). The ratios rλPr_{\lambda_P} and rλQr_{\lambda_Q} are always larger than the current critical load (λCr)−1(\lambda^{\mathrm{Cr}})^{-1} (dashed line at 45∘45^\circ). The black dots mark the points where rλP=1r_{\lambda_P}=1, and the critical load of the optimized beam equals that of the initial one. The dashed lines in plot (b) report the curves in (a), for comparison

For the SS beam, rλP>1r_{\lambda_P}>1 and rλQ>1r_{\lambda_Q}>1 for all compression levels, so for this configuration the maximization of the vibration frequency always gives, as a side effect, the increase of the critical load. Also, the range of variation of the two ratios is very narrow for the SS beam, especially when considering the tip compression, for which 1.328≤rλP≤1.3331.328 \leq r_{\lambda_P} \leq1.333. This is expected, as in this case the optimized beam size is only minimally modified (see Figure 3(a)). On the other hand, we observe a larger range of variation when considering the axially distributed load, for which 1.23≤rλQ≤1.4061.23 \leq r_{\lambda_Q} \leq1.406, and this is consistent with the more marked modification of the beam’s profile in this case (see Figure 7(a)).

For the CF, CS, and CC configurations the buckling capacity of the optimized design heavily depends on the compression level considered in the optimization. Up to the values λP=l3\lambda_P=l_3 and λQ=l6\lambda_Q=l_6, respectively, we have rλP<1r_{\lambda_P}<1 and rλQ<1r_{\lambda_Q}<1; therefore the optimized beam will show a lower buckling capacity than the uniform one. Then, for higher compression levels, the optimization also improves the beam’s critical load. The designs corresponding to points l3l_3 and l6l_6, featuring the same critical load as the uniform beam, are marked by green dotted lines in Figure 3 and Figure 7, respectively.

The first configuration to meet the condition rλP=1r_{\lambda_P}=1, at l3=0.335l_3=0.335, is the CS, whereas the last is the CC, at l3=0.89l_3=0.89. For the CF configuration, the value of rλPr_{\lambda_P} is almost always between those of the CS and CC beams, and we have l3=0.7l_3=0.7. We also observe that rλPr_{\lambda_P} has an approximately linear trend for the CF and CC configurations, whereas it shows a marked non-linear trend for the CS case. Moreover, for λP≈1\lambda_P \approx1 the CS beam attains a critical load which is slightly below (less than 1%) that of the SS beam, whereas the other two beam configurations attain lower values.

When considering the axially distributed load, we see some differences. First, for all beam configurations the same critical load of the uniform design is now reached for higher values of compression (thus l6>l3l_6>l_3). The trend of rλQr_{\lambda_Q} for the CF and CC beams shows little deviations compared to the case of tip compression. On the contrary, the trend for the CS configuration is very different, and this configuration is now the last to reach rλQ=1r_{\lambda_Q}=1, for a compression level close to the critical one (l6=0.96l_6=0.96).

Conclusions

We have investigated the effect of axial compression on the optimal sizing of beams for maximum fundamental vibration frequency, considering both a tip force generating a constant, control-independent compression, and a distributed axial load proportional to self-weight, giving a control-dependent compression. Some notable numerical results discussed throughout Section 3 are summarized in Table 1.

Both sources of compression have an impact on the frequency gains attainable by the optimization and, for all beam configurations but the SS, also on the optimal size distribution. As soon as compression is introduced, the optimal design deviates from that obtained for the unloaded beam (pure vibrations), and is prevented from developing singularities anywhere in the beam domain. This determines a substantial reduction of the optimal frequency gain, compared to the case of an unloaded beam. However, when looking at high compression levels, frequency optimization becomes much more beneficial, compared to the unloaded case. When the compression is just below the critical one, the frequency gains relative to those for the unloaded beam range between 2 to 7.1 for the tip force, and between 11.34 to 7.2 times for the distributed load proportional to self-weight.

When paired to the design-independent compression, the self-weight has a very little contribution for most of the “intermediate” values of the first. However, it does have an influence when the compression is close to the critical value, where again, its account leads to higher frequency gains, especially for the cantilever and simply supported configuration.

Acknowledgements

The author gratefully acknowledges the support of the Villum Foundation, through the Villum Investigator project “InnoTop”.

Remarks on the local behaviour of the dynamic solution

Here we summarize the classical local asymptotic analysis for the optimal design of a purely vibrating beam, recalling the definition of the two types of singularities occurring in this case[21, 33, 40].

For the optimal design of an uncompressed beam (λQ=λP=0\lambda_Q=\lambda_P=0), for maximum frequency of vibration, the system governing the optimality (13) reduces to

(ypvxx)xx−Ωˉ2yv=0pyp−1vx2−Ωˉ2v2−Ωˉ2(p−1)=0B(y,v)=0[Mvx]xk=0,k=1,…nI[Tv]xk=0,k=1,…nII(15)\begin{aligned} (y^p v_{xx})_{xx}-\bar{\Omega}^2yv&=0\\ py^{p-1}v_x^2-\bar{\Omega}^2v^2-\bar{\Omega}^2(p-1)&=0\\ B(y,v)&=0\\ [Mv_x]_{x_k}&=0,\qquad k=1,\ldots n_{\mathrm{I}}\\ [Tv]_{x_k}&=0,\qquad k=1,\ldots n_{\mathrm{II}} \tag*{(15)} \end{aligned}

where we have also introduced the boundary operator B(y,v)B(y,v), and we allowed for a finite number of jumps in the optimal energy distribution associated with the bending moment M:=ypvxxM:=y^pv_{xx} and shear force T:=(ypvxx)xT:=(y^pv_{xx})_x. Continuity of these internal actions must hold everywhere within the beam domain, as we do not consider point loads[24], but depending on the vanishing of either or both of them, we can identify two types of points where the optimal size y∗y_* →0\to0, and some of the response components becomes singular.

The first set of points is defined as follows:

Figure A.9. Optimal response functions, size distribution, and internal actions near a Type I (black curves, plotted against the left axis), and Type II (red curves, plotted against the right axis) point, according to (A.4) and (A.6)

Definition (Type I singularity). Let xk(I)∈[0,1]x_k^{(\mathrm{I})} \in[0,1] be a point where M∗(xk(I))=0M_*(x_k^{(\mathrm{I})}) = 0, and T∗(xk(I))≠0T_*(x_k^{(\mathrm{I})}) \ne0. Then, y∗(ξ)=O(ξ1/3)y_*(\xi) = O(\xi^{1/3}), where ξ=∣x−xk(I)∣\xi= |x-x_k^{(\mathrm{I})}|. Also, v∗v_* is continuous, its derivative v∗,xv_{*,x} has a finite jump, and the curvature v∗,xxv_{*,xx} is singular at xk(I)x_k^{(\mathrm{I})}.

At such points, the continuity of the bending moment is automatically fulfilled, M∗(xk(I))=0M_*(x_k^{(\mathrm{I})}) = 0, and therefore v∗,x(xk(I))v_{*,x}(x_k^{(\mathrm{I})}) can have a finite jump, provided that y∗(xk(I))=0y_*(x_k^{(\mathrm{I})}) = 0, such that the energy product [ ⁣[y∗2v∗,xx] ⁣]xk(I)[\![y_*^2v_{*,xx}]\!]_{x_k^{(\mathrm{I})}} remains finite. Mechanically, this point mimics the introduction of a hinge in the beam’s domain.

The second set of points leading to a response singularity can be introduced as follows:

Definition (Type II singularity). Let xk(II)∈[0,1]x_k^{(\mathrm{II})} \in[0,1] be a point where M∗(xk(II))=T∗(xk(II))=0M_*(x_k^{(\mathrm{II})}) = T_*(x_k^{(\mathrm{II})}) = 0. Then, y∗(ξ)=O(ξ2)y_*(\xi) = O(\xi^2), where ξ=∣x−xk(II)∣\xi= |x-x_k^{(\mathrm{II})}|. Also, v∗v_* has a jump, and both v∗,xv_{*,x} and v∗,xxv_{*,xx} are singular at xk(II)x_k^{(\mathrm{II})}.

At such points both internal actions are vanishing, and this allows for a jump on v∗v_*. Then, the singularity of v∗,x(xk(II))v_{*,x}(x_k^{(\mathrm{II})}) and v∗,xx(xk(II))v_{*,xx}(x_k^{(\mathrm{II})}) requires y∗((xk(II)))=0y_*((x_k^{(\mathrm{II})})) = 0, in order to have finite energy products [ ⁣[y∗2v∗,xx] ⁣]xk(II)[\![y_*^2v_{*,xx}]\!]_{x_k^{(\mathrm{II})}} and [ ⁣[(y∗2v∗,xx)xv] ⁣]xk(II)[\![(y_*^2v_{*,xx})_xv]\!]_{x_k^{(\mathrm{II})}}. Mechanically, this second singularity represents a complete separation of the beam.

The behaviour of the response functions near these two types of points can be deduced by an asymptotic analysis on the following non-linear differential equation

[(Ωˉ2(v∗,x2+p−1)pv∗,xx2)p(1−p)v∗,xx]xx−(Ωˉ2(v∗,x2+p−1)pv∗,xx2)1−pv=0(16)\left[\left(\frac{\bar{\Omega}^2(v_{*,x}^2+p-1)}{pv_{*,xx}^2}\right)^{p(1-p)}v_{*,xx}\right]_{xx} -\left(\frac{\bar{\Omega}^2(v_{*,x}^2+p-1)}{pv_{*,xx}^2}\right)^{1-p}v=0 \tag*{(16)}

obtained by eliminating the expression for y∗y_* from the state equation in (A.1). Then we represent the deflection v∗(ξ)v_*(\xi) according to a regular, or a singular expansion [26].

In the first case we assume v(I)(ξ)=a0+a1ξrv^{(\mathrm{I})}(\xi) = a_0+a_1\xi^r, where r∈Qr \in\mathbb{Q} and (a0,a1)∈R(a_0,a_1) \in\mathbb{R}, and substituting within (A.2), we obtain the residual

R[v(I)]=Ωˉ2ξ4−3rt0[(a04+3(1+2a02))t1+o(ξ)]=0(17)\mathcal{R}[v^{(\mathrm{I})}] = \frac{\bar{\Omega}^2\xi^{4-3r}}{t_0}\left[(a_0^4+3(1+2a_0^2))t_1+o(\xi)\right]=0 \tag*{(17)}

where t0=4a2r3(r−1)3t_0=4a_2r^3(r-1)^3 and t1=3(r−2)(3r−5)t_1=3(r-2)(3r-5). Non-trivial solutions to (A.3) require t1=0t_1=0, while t0≠0t_0\ne0. This correspond to the pair of solutions r1=5/3r_1=5/3 and r2=2r_2=2, which plugged in (A.3) give

lim⁡ξ→0+R{r1,r2}={0,(a0+a03)Ωˉ28a02}\lim_{\xi\to0^+}\mathcal{R}\{r_1,r_2\}=\left\{0,\frac{(a_0+a_0^3)\bar{\Omega}^2}{8a_0^2}\right\}

and therefore only the solution r=5/3r=5/3 is acceptable. The behaviour of the functions near a Type I points is (see Figure A.9)

v∗(ξ)=a0+a1ξ+a2ξ5/3,v∗,x(ξ)=a1+a2ξ2/3,v∗,xx(ξ)=a3ξ−1/3y∗(ξ)=a4ξ1/3+O(ξ),M∗(ξ)=ξ+O(ξ2),T∗(ξ)=a5+O(ξ)(18)\begin{aligned} v_*(\xi)&=a_0+a_1\xi+a_2\xi^{5/3}, & v_{*,x}(\xi)&=a_1+a_2\xi^{2/3}, & v_{*,xx}(\xi)&=a_3\xi^{-1/3}\\ y_*(\xi)&=a_4\xi^{1/3}+O(\xi), & M_*(\xi)&=\xi+O(\xi^2), & T_*(\xi)&=a_5+O(\xi) \tag*{(18)} \end{aligned}

where the expression of the aia_i coefficients can be found in [40].

If we consider the singular expansion v(II)(ξ)=b0ξ−sv^{(\mathrm{II})}(\xi)=b_0\xi^{-s}, s∈N+s\in\mathbb{N}^{+}, b0∈Rb_0\in\mathbb{R} the residual of (A.2) becomes

R[v(II)]=Ωˉ2ξ4−sq0[−b14q1+o(ξ2s)]=0(19)\mathcal{R}[v^{(\mathrm{II})}]=\frac{\bar{\Omega}^{2}\xi^{4-s}}{q_0}\left[-b_1^{4}q_1+o(\xi^{2s})\right]=0 \tag*{(19)}

where q0=4b13s3(s+1)3q_0=4b_1^3s^3(s+1)^3 and q1=(s+15)(s−2)q_1=(s+15)(s-2). Now the only acceptable solution is s=2s=2, and the functions near a Type II point behave as follows (see Figure A.9)

v∗(ξ)=b0ξ−2,v∗,x(ξ)=−b1ξ−3,v∗,xx(ξ)=b2ξ−4y∗(ξ)=b3ξ2,M∗(ξ)=b4ξ4,T∗(ξ)=b5ξ3(20)\begin{aligned} v_*(\xi)&=b_0\xi^{-2},\quad v_{*,x}(\xi)=-b_1\xi^{-3},\quad v_{*,xx}(\xi)=b_2\xi^{-4}\\ y_*(\xi)&=b_3\xi^2,\quad M_*(\xi)=b_4\xi^4,\quad T_*(\xi)=b_5\xi^3 \tag*{(20)} \end{aligned}

where the expression of the bib_i coefficients can be found in [40].

References

  1. [1]Adali, S.: Optimal shape and non-homogeneity of a non-uniformly compressed column. International Journal of Solids and Structures 15(12), 935–949 (1979). DOI https://doi.org/10.1016/0020-7683(79)90023-4
  2. [2]Atanackovic, T., Glavardanov, V.: Optimal shape of a heavy compressed column. Structural and Multidisciplinary Optimization 28(6), 388–396 (2004). DOI 10.1007/s00158-004-0457-1. URL http://link.springer.com/10.1007/s00158-004-0457-1DOI
  3. [3]Atanackovic, T.M.: Optimal shape of column with own weight: bi and single modal optimization. Meccanica 41, 173–196 (2006)DOI
  4. [4]Atanackovic, T.M., Seyranian, A.P.: Application of Pontryagin’s principle to bimodal optimization problems. Structural and Multidisciplinary Optimization 37(1), 1–12 (2008). DOI 10.1007/s00158-007-0211-6. URL http://link.springer.com/10.1007/s00158-007-0211-6DOI
  5. [5]Banichuk, N.V.: Problems and Methods of Optimal Structural Design. Springer US, Boston, MA (1983). DOI 10.1007/978-1-4613-3676-1. URL http://link.springer.com/10.1007/978-1-4613-3676-1DOI
  6. [6]Barnes, D.C.: Extremal Problems for Eigenvalue Functionals. SIAM Journal on Mathematical Analysis 16(6) (1985). DOI 10.1137/0516092DOI
  7. [7]Barnes, D.C.: Extremal Problems for Eigenvalue Functionals. II. SIAM Journal on Mathematical Analysis 19(5), 1151–1161 (1988). DOI 10.1137/0519080. URL http://epubs.siam.org/doi/10.1137/0519080DOI
  8. [8]Barnes, D.C.: The shape of the strongest column is arbitrarily close to the shape of the weakest column. Quarterly of Applied Mathematics 46(4) (1988). DOI 10.1090/qam/973378DOI
  9. [9]Barnes, E.R.: Some min-max problems arising in optimal design studies. In: Control Theory of Systems Governed by Partial Differential Equations, pp. 177–208. Elsevier (1977). DOI 10.1016/B978-0-12-068640-7.50010-6. URL https://linkinghub.elsevier.com/retrieve/pii/B9780120686407500106DOI
  10. [10]Barnes, E.R.: The shape of the strongest column and some related extremal eigenvalue problems. Quarterly of Applied Mathematics 34(4) (1977). DOI 10.1090/qam/493674DOI
  11. [11]Brach, R.: On the extremal fundamental frequencies of vibrating beams. International Journal of Solids and Structures 4(7), 667–674 (1968). DOI 10.1016/0020-7683(68)90068-1. URL https://linkinghub.elsevier.com/retrieve/pii/0020768368900681DOI
  12. [12]Brach, R.M.: On optimal design of vibrating structures. Journal of Optimization Theory and Applications 11(6), 662–667 (1973). DOI 10.1007/BF00935565. URL http://link.springer.com/10.1007/BF00935565DOI
  13. [13]Bulson, P.: Optimal structural design under stability constraints. Engineering Structures 12(1), 69–70 (1990). DOI 10.1016/0141-0296(90)90047-V. URL https://linkinghub.elsevier.com/retrieve/pii/014102969090047VDOI
  14. [14]Carmichael, D.: Singular optimal control problems in the design of vibrating structures. Journal of Sound and Vibration 53(2) (1977). DOI 10.1016/0022-460X(77)90468-0DOI
  15. [15]Charrett, D., Rozvany, G.: Extensions of the Prager-Shield theory of optimal plastic design. International Journal of Non-Linear Mechanics 7(1), 51–64 (1972). DOI 10.1016/0020-7462(72)90021-2. URL https://linkinghub.elsevier.com/retrieve/pii/0020746272900212DOI
  16. [16]Cox, S.J., McCarthy, C.: The Shape of the Tallest Column. SIAM Journal on Mathematical Analysis 29(3), 547–554 (1998). DOI 10.1137/S0036141097314537. URL http://epubs.siam.org/doi/10.1137/S0036141097314537DOI
  17. [17]Egorov, Y.V.: On the Lagrange problem about the optimal form for circular hollow columns. Comptes Rendus - Mecanique 331(10) (2003). DOI 10.1016/j.crme.2003.07.001DOI
  18. [18]Egorov, Y.V.: On the tallest column. Comptes Rendus. Mécanique 338(5), 266–270 (2010). DOI 10.1016/j.crme.2010.05.001DOI
  19. [19]Gajewski, A., Piekarski, Z.: Optimal structural design of a vibrating beam with periodically varying cross-section. Structural Optimization 7(1-2), 112–116 (1994). DOI 10.1007/BF01742515. URL http://link.springer.com/10.1007/BF01742515DOI
  20. [20]Gierliński, J., Mróz, Z.: Optimal design of elastic plates and beams taking large deflections and shear forces into account. Acta Mechanica 39(1-2) (1981). DOI 10.1007/BF01173194DOI
  21. [21]Gjelsvik, A.: Minimum weight design of continuous beams. International Journal of Solids and Structures 7(10) (1971). DOI 10.1016/0020-7683(71)90054-0DOI
  22. [22]Haftka, R.T., Gürdal, Z.: Elements of Structural Optimization, Solid Mechanics And Its Applications, vol. 11. Springer Netherlands, Dordrecht (1992). DOI 10.1007/978-94-011-2550-5. URL http://link.springer.com/10.1007/978-94-011-2550-5DOI
  23. [23]Hanaoka, M., Washizu, K.: Optimum design of Beck’s column. Computers & Structures 11(6), 473–480 (1980). DOI 10.1016/0045-7949(80)90054-1. URL https://linkinghub.elsevier.com/retrieve/pii/0045794980900541
  24. [24]Hartmann, F.: The Mathematical Foundation of Structural Mechanics. Springer Berlin Heidelberg, Berlin, Heidelberg (1985). DOI 10.1007/978-3-642-82401-2. URL http://link.springer.com/10.1007/978-3-642-82401-2DOI
  25. [25]Haug, E.J., Rousselet, B.: Design Sensitivity Analysis in Structural Mechanics. I. Static Response Variations. Journal of Structural Mechanics 8(1), 17–41 (1980). DOI 10.1080/03601218008907351. URL http://www.tandfonline.com/doi/abs/10.1080/03601218008907351DOI
  26. [26]Hinch, E.J.: Perturbation Methods. Cambridge Texts in Applied Mathematics. Cambridge University Press (1991)
  27. [27]Holmåker, K.: Some Mathematical Aspects on a Problem of the Optimal Design of a Vibrating Beam. SIAM Journal on Mathematical Analysis 18(5), 1367–1377 (1987). DOI 10.1137/0518098. URL http://epubs.siam.org/doi/10.1137/0518098DOI
  28. [28]Huang, N.C.: Optimal design of elastic structures for maximum stiffness. International Journal of Solids and Structures 4(7), 689–700 (1968). DOI https://doi.org/10.1016/0020-7683(68)90070-X. URL http://www.sciencedirect.com/science/article/pii/002076836890070X
  29. [29]Huang, N.C.: Effect of shear deformation on optimal design of elastic beams. International Journal of Solids and Structures 7(4) (1971). DOI 10.1016/0020-7683(71)90106-5DOI
  30. [30]Huang, N.C.: Optimal Design of Elastic Beams for Minimum-Maximum Deflection. Journal of Applied Mechanics 38(4), 1078–1081 (1971). DOI 10.1115/1.3408925. URL https://asmedigitalcollection.asme.org/appliedmechanics/article/38/4/1078/423506/Optimal-Design-of-Elastic-Beams-fDOI
  31. [31]Hughes, T.J.R.: The Finite Element Method: Linear Static and Dynamic Finite Element Analysis. Prentice–Hall (1987)
  32. [32]Kamat, M.P.: Effect of shear deformations and rotary inertia on optimum beam frequencies. International Journal for Numerical Methods in Engineering 9(1), 51–62 (1975). DOI 10.1002/nme.1620090105. URL https://onlinelibrary.wiley.com/doi/10.1002/nme.1620090105DOI
  33. [33]Karihaloo, B.L., Niordson, F.I.: Optimum design of vibrating cantilevers. Journal of Optimization Theory and Applications 11(6), 638–654 (1973). DOI 10.1007/BF00935563. URL http://link.springer.com/10.1007/BF00935563DOI
  34. [34]Keller, J.B.: The shape of the strongest column. Archive for Rational Mechanics and Analysis 5(1), 275–285 (1960). DOI 10.1007/BF00252909. URL http://link.springer.com/10.1007/BF00252909DOI
  35. [35]Keller, J.B., Niordson, F.I.: The tallest column. Jour. Math. Mech. 16(5), 433–446 (1966)DOI
  36. [36]Keng-tung, C., Hua, D.: On dynamic optimization of Timoshenko beam. Applied Mathematics and Mechanics 4(1), 69–77 (1983). DOI 10.1007/BF01896714. URL http://link.springer.com/10.1007/BF01896714
  37. [37]Krein, M.G.: On certain problems on the maximum and minimum of characteristic values and on the Lyapunov zones of stability (1955)DOI
  38. [38]Maeve McCarthy, C.: The tallest column — optimality revisited. Journal of Computational and Applied Mathematics 101(1-2), 27–37 (1999). DOI 10.1016/S0377-0427(98)00188-5. URL https://linkinghub.elsevier.com/retrieve/pii/S0377042798001885
  39. [39]Niordson, F.I.: On the optimal design of a vibrating beam. Quarterly of Applied Mathematics 23(1), 47–53 (1965). DOI 10.1090/qam/175392. URL https://www.ams.org/qam/1965-23-01/S0033-569X-1965-0175392-8/DOI
  40. [40]Olhoff, N.: Optimization of Vibrating Beams with Respect to Higher Order Natural Frequencies. Journal of Structural Mechanics 4(1), 87–122 (1976). DOI 10.1080/03601217608907283. URL http://www.tandfonline.com/doi/abs/10.1080/03601217608907283DOI
  41. [41]Olhoff, N., Niordson, F.I.: Some problems concerning singularities of optimal beams and columns. Zeitschrift fuer angewandte Mathematik und Mechanik 59 (1979)
  42. [42]Olhoff, N., Rasmussen, S.H.: On single and bimodal optimum buckling loads of clamped columns. International Journal of Solids and Structures 13(7), 605–614 (1977). DOI 10.1016/0020-7683(77)90043-9. URL https://linkinghub.elsevier.com/retrieve/pii/0020768377900439DOI
  43. [43]Payne, L.E.: Isoperimetric Inequalities for Eigenvalues and Their Applications. In: Autovalori e autosoluzioni, pp. 107–167. Springer Berlin Heidelberg, Berlin, Heidelberg (2011). DOI 10.1007/978-3-642-10994-2{\_}-3. URL http://link.springer.com/10.1007/978-3-642-10994-2_3DOI
  44. [44]Pedersen, P., Pedersen, N.L.: Analytical optimal designs for long and short statically determinate beam structures. Structural and Multidisciplinary Optimization 39(4), 343–357 (2009). DOI 10.1007/s00158-008-0339-z. URL http://link.springer.com/10.1007/s00158-008-0339-zDOI
  45. [45]Plaut, R.H., Virgin, L.N.: Optimal desing of cantilevered elastica for minimum tip deflection under self-weight. SMO 43, 657–664 (2011)DOI
  46. [46]Prager, W., Taylor, J.E.: Problems of Optimal Structural Design. Journal of Applied Mechanics 35(1), 102–106 (1968). DOI 10.1115/1.3601120. URL https://asmedigitalcollection.asme.org/appliedmechanics/article/35/1/102/387171/Problems-of-Optimal-Structural-DesignDOI
  47. [47]Rammerstorfer, F.: On the optimal distribution of the Young’s modulus of a vibrating, prestressed beam. Journal of Sound and Vibration 37(1), 140–145 (1974). DOI 10.1016/S0022-460X(74)80064-7. URL https://linkinghub.elsevier.com/retrieve/pii/S0022460X74800647DOI
  48. [48]Ringertz, U.T.: On the design of Beck’s column. Structural Optimization 8(2-3), 120–124 (1994). DOI 10.1007/BF01743307. URL http://link.springer.com/10.1007/BF01743307DOI
  49. [49]Rozvany, G., Yep, K., Ong, T., Karihaloo, B.: Optimal design of elastic beams under multiple design constraints. International Journal of Solids and Structures 24(4), 331–349 (1988). DOI 10.1016/0020-7683(88)90065-0. URL https://linkinghub.elsevier.com/retrieve/pii/0020768388900650DOI
  50. [50]Rozvany, G.I.N.: Optimal Plastic Design of Beams with Unspecified Actions or Reactions. pp. 115–159 (1989). DOI 10.1007/978-94-009-1161-1{\_}-4. URL http://link.springer.com/10.1007/978-94-009-1161-1_4DOI
  51. [51]Rozvany, G.I.N.: Structural Design via Optimality Criteria, vol. 8. Springer Netherlands, Dordrecht (1989). DOI 10.1007/978-94-009-1161-1DOI
  52. [52]Sadiku, S.: Buckling load optimization for heavy elastic columns: a perturbation approach. Structural and Multidisciplinary Optimization 35(5), 447–452 (2008). DOI 10.1007/s00158-007-0144-0. URL http://link.springer.com/10.1007/s00158-007-0144-0DOI
  53. [53]Sadiku, S.: On the Application of the Perturbation Method in the Buckling Optimization of a Heavy Elastic Column of Constant Aspect Ratio. Journal of Optimization Theory and Applications 146(1), 181–188 (2010). DOI 10.1007/s10957-010-9663-8. URL http://link.springer.com/10.1007/s10957-010-9663-8DOI
  54. [54]Seyranian, A., Privalova, O.: The Lagrange problem on an optimal column: old and new results. Structural and Multidisciplinary Optimization 25(5-6), 393–410 (2003). DOI 10.1007/s00158-003-0333-4. URL http://link.springer.com/10.1007/s00158-003-0333-4DOI
  55. [55]Tadjbakhsh, I., Keller, J.B.: Strongest Columns and Isoperimetric Inequalities for Eigenvalues. Journal of Applied Mechanics 29(1), 159–164 (1962). DOI 10.1115/1.3636448. URL https://asmedigitalcollection.asme.org/appliedmechanics/article/29/1/159/424846/Strongest-Columns-and-IsoperimetricDOI
  56. [56]Timoshenko, S.P., Goodier, J.N.: Theory of Elasticity, third edn. McGraw Hill International Editions (1970)
  57. [57]Vepa, K.: On the existence of solutions to optimization problems with eigenvalue constraints. Quarterly of Applied Mathematics 31(3) (1973). DOI 10.1090/qam/428893DOI
  58. [58]Virgin, L.N.: Vibration of axially loaded structures. Cambridge University Press (2007)DOI
  59. [59]Washizu, K.: Variational Methods in Elasticity and Plasticity, second edn. 01. Pergamon Press (1975)

Paper details

vlnc:W_2q54wm8k
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents