1
Truss Layout Optimization
This chapter presents the general problem of truss layout optimization. After a brief introduction to the standard theory of mathematical programming in section 1.1, section 1.2 derives the governing equations of truss structures. Then, section 1.3 states the basic problem of topology optimization. The equivalence between volume and compliance minimization problems is also studied by means of the necessary conditions of optimality. On this basis, section 1.4 progressively builds up a general formulation by adding different design settings. At each step, the numerical difficulties associated with these building blocks are explained. Finally, the optimization of nodal positions is considered in section 1.5, leading to the general design problem of truss geometry and topology optimization, which remains unsolved in the literature.
1.1. Standard theory of mathematical programming
Consider a general nonlinear optimization problem consisting of the minimization of an objective function subject to inequality and equality constraints [MOR 03]:
where
f :
are smooth functions and z â
Nz is a vector of continuous variables. Smoothness of the objective function and the constraints is important to allow for a good prediction of the search direction by optimization algorithms. The feasible set of the optimization problem [1.1] is defined as:
In the feasible region, the inequality constraint
gi (z) †0 is said to be
active if
gi (z) = 0 and
inactive if
gi (z) < 0. To solve problem [1.1], we first transform it into an unconstrained optimization problem by introducing Lagrange multipliers
such that:
Thus, solving problem [1.1] amounts now to finding a stationary point to [1.3]. If gi (z) is active, we ensure that the search direction points toward the feasible region by enforcing the dual feasibility λg,i â„ 0. If gi is inactive, we can remove the constraint by setting the complementary slackness λg,igi (z) = 0. These additional constraints are parts of the KarushâKuhnâTucker (KKT) optimality conditions [MOR 03]. Let z* be a local minimizer of problem [1.1]. Provided that some regularity conditions hold, then there exists λg,i and λh,j such that the first-order necessary conditions of optimality, or KKT conditions, are satisfied:
The regularity conditions or constraint qualifications of problem [1.1] are necessary conditions that enable a numerical treatment by standard algorithms of mathematical programming. There are many constraint qualifications in the literature (see [PET 73] for a comprehensive survey). Hereafter, the following three prominent conditions are listed:
â the linear constraint qualification implies that if gi and hj are affine functions, then all subsequent constraint qualifications are satisfied;
â the linear-independence constraint qualification holds if the gradients of active inequality constraints âgi (z*) and equality constraints âhj (z*) are linearly independent at z*;
â the MangasarianâFromovitz constraint qualification holds if the gradients of active inequality constraints âgi (z*) and equality constraints âhj (z*) are positive linearly independent at z*.
For the remainder, the special case of linear programming should be mentioned. Such an optimization problem minimizes a linear objective function subject to linear equality and inequality constraints with non-negative variables. The convexity of the problem implies that a local optimum is also a global optimum and authorizes an efficient treatment by optimization algorithms [ALE 01]. We will see in section 1.3 that, very often, it is possible to reformulate topology optimization problems so that linear programming applies.
1.2. Governing equations of truss structures
Before stating the structural optimization problem, let us start with some basic notations for a linear elastic truss structure as depicted in
Figure 1.1. Using standard finite element concepts, we consider a pin-jointed structure composed of
Nn nodes interconnected by truss elements
e â {1,âŠ,
Nb}. With
d â {2, 3} being the spatial dimension and
Ns being the number of support reactions, the number of degrees of freedom is
Nd =
d.
Nn â
Ns. The vector of nodal coordinates is denoted by x â
d.Nn, the vector of nodal displacements is denoted by u â
Nd and the vector of external forces is denoted by f â
Nd (excluding support reactions). The member force is
te â
. The design parameters associated with every truss element are the...