Professional Documents
Culture Documents
Evolutionary Natural Frequency Optimization of Two-Dimensional Structures With Additional Non-Structural Lumped Masses
Evolutionary Natural Frequency Optimization of Two-Dimensional Structures With Additional Non-Structural Lumped Masses
two-dimensional structures
233
with additional non-structural Received September 1995
lumped masses Revised May 1996
Chongbin Zhao
CSIRO Division of Exploration and Mining, Nedlands,West Australia,
Australia
G.P. Steven
Finite Element Analysis Research Centre, University of Sydney,
New South Wales, Australia and
Y.M. Xie
Department of Civil and Building Engineering,
Victoria University of Technology, MMC, Victoria, Australia
Introduction
In recent years, the structural topology optimization has become a very
important topic for two main reasons. First, the structural topology
optimization usually produces a much better structural configuration than the
conventional structural optimization such as the structural sizing optimization
or structural shape optimization. Second, the advent of both advanced
computer technology and the sophisticated structural analysis method enables
the complicated structural topology optimization problem in continuum
mechanics to be solved using numerical methods.
Thus, several optimization methods such as the ground structure approach
(Rozvany and Zhou, 1993), the homogenization method (Bendsøe, 1989; Bendsøe
and Kikuchi, 1988; Diaz and Kikuchi, 1992; Suzuki and Kikuchi, 1991) and the
bubble method (Eschenauer et al., 1993) have been developed and used to tackle
some structural topology optimization problems. Since these methods are based
on some conventional optimization concepts and techniques, the concepts of an
objective function, design variables, constraints and optimality criteria have
been employed in the process of developing them. In addition, based on the
evolutionary principle, the evolutionary structural optimization method (Xie
and Steven, 1993; 1994) has recently been proposed for solving two-dimensional
structural topology optimization problems. The basic idea behind the
evolutionary structural optimization method is that for a structure subjected to Engineering Computations,
Vol. 14 No. 2, 1997, pp. 233-251.
static loading, the material with small stress levels is not used efficiently © MCB University Press, 0264-4401
EC because it carries only a small amount of load. Thus, this material can be
14,2 removed so as to make the structural material more efficient. When the
inefficient material of a structure is removed gradually, the topology of the
structure is slowly evolving and the stress distribution within the resulting
structure becomes more and more uniform. For the purpose of keeping the
solution of high accuracy, the design domain which is a continuum should be
234 discretized into a very fine mesh of finite elements and the number of elements
removed at each iteration should be small. This will lead to a smooth transition
between any two adjacent generations of the structure. Realizing the simplicity
and easy implementation of the evolutionary structural optimization method,
Hinton and Sienz (1995) applied this method to the fully stressed topology
design of structures and produced very interesting results for the topology of
several two-dimensional structures under static loading conditions.
The main purpose of this paper is to extend the evolutionary structural
optimization method to the solution for the natural frequency optimization of
two-dimensional structures with additional non-structural lumped masses. The
statement of a structural natural frequency optimization problem is described
as follows. For a given design domain of the problem and a prescribed weight of
a target structure, the optimal topology of the target structure, which
corresponds to the maximum natural frequency for a concerned structural
vibration mode, needs to be found within this design domain. Generally, the
weight of the design domain is greater than that of the target structure. That is
to say, for the topology design of a target structure, there are infinite choices in
the continuum sense but a large number of choices in the finite element sense,
depending on the finite element mesh used. Once the finite element method is
employed in the process of the structural topology optimization, the basic
problem posed is how to choose the optimal topology of the target structure
among a large number of possible structural configurations, which are located
within the design domain and satisfy the prescribed structural weight
requirement. Since the finite element method is an approximate one, the finer
the finite element mesh used, the more accurate the optimal topology obtained.
In this regard, the term optimal topology only holds true in the finite element
sense. In order to find the optimal topology of a target structure using the
evolutionary structural optimization method, it is necessary to evaluate the
contribution of each element to the concerned natural frequency of the finite
element discretized design domain. If the finite element mesh used to model the
design domain is fine enough and the number of elements removed at each
iteration is very small, it is possible to find the optimal topology of the target
structure by simply removing the inefficient material/elements step by step
until the prescribed structural weight requirement is met. Unlike a static
optimization problem, the efficiency of material used in a free vibration system
is not represented by its stress level but represented by its contribution to the
concerned natural frequency of the system.
Keeping the above in mind, the expression for the contribution factor of an
element to the concerned natural frequency of a system as well as some new
criteria for the structural natural frequency optimization have been derived and Natural
outlined in this paper. Using the contribution factor of an element, the efficiency frequency
of any particular part of material in the design domain can be determined. This optimization
leads to a general method for the evolutionary natural frequency optimization
of a structure. Finally, two examples have been presented to demonstrate the
feasibility of the extended evolutionary structural optimization method when it
is used to deal with the natural frequency optimization problem of a structure 235
with additional non-structural lumped masses.
(1)
236
where u and v are the in-plane displacements of the structure; uA and vA are the
corresponding amplitudes of the displacements; ω is the circular frequency at
which the structure vibrates.
Under the condition of harmonic motion, the total strain energy and total
kinetic energy of a finite element discretized system can be expressed as:
(2)
where {∆A} is the displacement amplitude vector of the system; Ep and Ek are
the total strain energy and total kinetic energy of the system; [K] and [M] are the
global stiffness matrix and mass matrix of the system respectively.
From equation (2), the total energy of the system can be obtained as
(3)
(5b)
In any iteration, if an old system represents the system before removing an
element at the beginning of the iteration and a new system represents the
system after removing the element at the end of the iteration, the natural
frequency difference between the old and the new systems indicates the
contribution of this element to the natural frequency of the old system for a
particular vibration mode. This is the basic idea used for deriving the Natural
contribution factor of an element to the concerned natural frequency of a system. frequency
For the old system, the following equation exists: optimization
(6)
where
237
(7a)
(7b)
where [K]o and [M]o are the global stiffness matrix and mass matrix of the old
system respectively; ω0 is the concerned natural frequency of the old system;
{∆A}o is the displacement amplitude vector of the old system.
Similarly, for the new system, there exists the following equation:
(8)
where
(9a)
(9b)
where [K]n, and [M]n, are the global stiffness matrix and mass matrix of the new
system; ω n , and {∆ A } n are the concerned natural frequency and the
displacement amplitude vector of the new system.
It is noted that the following relationships exist between the old and the new
systems.
(10)
where [Ke] and [Me] are the enlarged stiffness matrix and mass matrix of an
element. These enlarged matrices of the element have the same order as the
global matrices of the system. δω expresses the circular frequency difference
between the new and old systems and is called as the contribution factor of an
element to the circular frequency of the system. { δ ∆ A } represents the
displacement amplitude difference between the new and old systems.
It is obvious that the following equation can be obtained from equations (6)
and (8).
(11)
EC Substituting the relevant equations into equation (11 ) yields the following
14,2 equation:
(12)
238 The governing equation of motion for the new system under the undamped free
vibration condition can be expressed as
(13)
Substituting equation (13) into equation (12) yields the following equation:
(14)
where [me] and [ke] are the mass matrix and stiffness matrix of an element; {∆A
e}
is the displacement amplitude vector of the element.
From equation (14), the following expression can be derived.
(15)
where α can be expressed as:
(16)
If the first-order variations are only considered in equation (14), the contribution
factor of an element to the concerned natural frequency of an undamped free
vibration system can be approximately expressed as:
(17)
If the ith natural frequency and its corresponding vibration mode of the old
system are expressed as ωi and {di}, equations (15) and (17) can be further
simplified using the following equation in structural dynamics:
(18)
where Mi is the modal mass in correspondence with the ith natural frequency of
the system.
Considering equations (15), (16) and (18) simultaneously yields the following
equation:
(19)
where
(20)
where δω i is the exact contribution factor of an element to the ith natural Natural
frequency of the old system; {die} is the modal shape vector of the element in the frequency
old system; {δdi} is the ith modal shape vector difference between the old and optimization
the new systems. The other symbols in equation (20) are of the same meaning
as mentioned before.
Similarly, considering equations (17) and (18) simultaneously yields the
following equation: 239
(21)
where ωold and ωnew are the concerned natural frequency of the old and the new
systems respectively; β1is the maximum variation tolerance of the concerned
natural frequency between the old and the new systems.
The smooth change criterion is used to check the variation of the concerned
natural frequency of the system in a single iteration. If the check indicates that
the smooth change criterion is violated, it means that the number of the elements
removed in the iteration is too large or the elements are wrongly removed from
the system. In the former case, the elements removed in the iteration are returned
to the system and the maximum number of elements removed in a single
iteration need to be introduced. In the latter case, the elements removed need to
be returned to the system as a part of the non-design domain.
(25)
where δω– i* is the average contribution factor of an element to the ith natural
–
frequency
–e
due to all the repeated natural frequencies of the system; ω– i, Mi and
{d i } are the average values of the related quantities due to all the repeated
natural frequencies of the system. They are expressed as:
(26)
Demonstrative examples
Using the extended evolutionary structural optimization method which
contains five basic criteria as mentioned in the last section, the natural
242 frequency optimization of two structures has been carried out under the plane
stress condition in this section. The concerned natural frequency is the
fundamental natural frequency of the structure in the process of the structural
natural frequency optimization. The design domain of the first structure is a
rectangular one with an additional non-structural lumped mass in its centre,
while the design domain of the second structure is a square with a small square
cut-out near its left-hand side and two additional non-structural lumped masses
near its right-hand side. Some detailed results from the natural frequency
optimization of these two structures are outlined below.
Figure 1.
Design domain of a 2D
structure with a lumped
mass
Natural
frequency
optimization
243
Figure 2.
Optimization of a 2D
structure with a lumped
mass
244
Figure 3.
Optimization of a 2D
structure with a lumped
mass
245
Figure 4.
Optimization of a 2D
structure with a lumped
mass
246
Figure 5.
Design domain of a 2D
structure with two
lumped masses
hand side, is discretized into 800 four-node square elements. There are two
additional non-structural lumped masses in the system. One is located near the
top right-hand corner and another is located near the bottom right-hand corner
of the system. These two additional non-structural lumped masses are
connected by a bar which is considered as a part of the non-design domain. In
Figure 5, the elements with dark colour represent the initial design domain,
while the elements with light colour represent the initial non-design domain
before the structural optimization. The left-hand boundary of the design
domain is clamped in the calculation. The following parameters are used for the
whole domain containing the design one and the non-design one: the length,
width and thickness of the large square without the small square cut-out are 3m,
3m and 0.01m, while the length, width and thickness of the small square cut-out
are 1m, 1m and 0.01m respectively; the elastic modulus of the whole domain
material is 25 × 106 kPa; Poisson’s ratio is 0.3; the unit weight is 2.5 × 103 kg/m3;
the weight of each additional non-structural lumped mass is 80kg, which is
evenly distributed at two nodes in the non-design domain. The weight of the
whole domain except for the additional non-structural lumped masses is 200kg.
Based on these parameters, the initial fundamental natural frequency of the
system before optimization is 344.47 rad/s. Also, β1 and β2 are of the same
values as used in the first example. Since there is only one symmetric axis in the
system, the minimum number of elements to be removed in an iteration is equal
to two so as to maintain the symmetric nature of the target structure.
Figures 6 to 9 show a series of optimal structural topologies for different
weights of a target structure. In these figures, Ws is the weight of the target
structure; Wd is the initial weight of the whole domain excluding the additional
non-structural lumped masses and is equal to 200kg in this example; ω1 is the
Natural
frequency
optimization
247
Figure 6.
Optimization of a 2D
structure with two
lumped masses
EC
14,2
248
Figure 7.
Optimization of a 2D
structure with two
lumped masses
Natural
frequency
optimization
249
Figure 8.
Optimization of a 2D
structure with two
lumped masses
EC
14,2
250
Figure 9.
Optimization of a 2D
structure with two
lumped masses
fundamental natural frequency of the target structure. As expected, the same Natural
conclusions as were obtained from the first example can be drawn from the frequency
results shown in these figures. Thus, the feasibility of the extended
evolutionary structural optimization method has been further demonstrated for
optimization
solving the natural frequency optimization problem of a structure with
additional non-structural lumped masses.
251
Conclusions
To extend the evolutionary structural optimization method to the solution for
the natural frequency optimization of a structure with additional non-structural
lumped masses, five basic criteria have been presented in this paper. These
criteria are:
(1) the material efficiency criterion;
(2) the positive definite structure criterion;
(3) the smooth change criterion;
(4) the average value criterion; and
(5) the minimum number of elements removed criterion.
Using these criteria, the evolutionary structural optimization method has been
successfully extended for solving the natural frequency optimization of
structures with additional non-structural lumped masses. The related results
from two examples have demonstrated the feasibility of the extended
evolutionary structural optimization method.
References
Bendsøe, M.P. (1989), “Optimal shape design as a material distribution problem”, Structural
Optimization, Vol. 1, pp. 193-202.
Bendsøe, M.P. and Kikuchi N. (1988), “Generating optimal topologies in structural design using a
homogenization method”, Computer Methods in Applied Mechanics and Engineering, Vol. 71,
pp. 197-224.
Diaz, A.R. and Kikuchi, N. (1992), “Solutions to shape and topology eigenvalue optimization
problems using a homogenization method”, International Journal for Numerical Methods in
Engineering, Vol. 35 pp. 1487-1502.
Eschenauer, H.A., Schumacher, A. and Vietor, T. (1993), “Design making for initial designs
made of advanced materials”, in Bendøe, M.P. and Soares, C.A.M. (Eds), NATO ARW, Topology
Design of Structures, Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 469-80.
Hinton, E. and Sienz, J. (1995), “Fully stressed topological design of structures using an
evolutionary procedure”, Engineering Computations, Vol. 12, pp. 229-44.
Rozvany, G.I.N. and Zhou, M. (1993), “Layout and generalized shape optimization by iterative
COC methods”, in Rozvany, G.I.N. (Ed.), NATO DFG ASI, Optimization of Large Structural
Systems, Vol. 1, Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 103-20.
Suzuki, K. and Kikuchi, N. (1991), “A homogenization method for shape and topology
optimization”, Computer Methods in Applied Mechanics and Engineering, Vol. 93, pp. 291-318.
Xie, Y.M. and Steven, G.P. (1993), “A simple evolutionary procedure for structural optimization”,
Computers and Structures, Vol. 49, pp. 885-96.
Xie, Y.M. and Steven, G.P. (1994), “A simple approach to structural frequency optimization”,
Computers and Structures, Vol. 53, pp. 1487-91.