In this thesis, we present an algorithm suited for constrained node-based shape optimization. It is a trust region algorithm using a first-order model, combined with vertex morphing so as to obtain smooth gradients. The choice of the step length and of the step direction is decoupled. The rules regarding the choice of the step length are similar to the rules of any trust region algorithmand are not modified in this thesis. Instead of leveraging the usual functional analysis description of subproblems, we use a geometric description that gives rise to meaningful and intuitive variables: the step-length-shares. The choice of the step direction is driven by the step-lengthshares, that give a lot of flexibility to develop ad-hoc strategies suited to practical problems encountered in industrial applications. The step direction rule presented in this thesis is a consistent way to integrate all the constraints without any user parameter. We introduce optional parameters that give more control to the user, and are an example of possible refinements of the step direction rule. The geometric description allows the formulation of response functions for packaging constraint and mesh quality control that can be used with any algorithm. The algorithm and the response functions are demonstrated on a 2Dexample and several 3D structural problems such as the weight optimization of a hook     «

In this thesis, we present an algorithm suited for constrained node-based shape optimization. It is a trust region algorithm using a first-order model, combined with vertex morphing so as to obtain smooth gradients. The choice of the step length and of the step direction is decoupled. The rules regarding the choice of the step length are similar to the rules of any trust region algorithmand are not modified in this thesis. Instead of leveraging the usual functional analysis description of subpro...     »