An Automated Design Environment for Micromechanical Sensors

In this paper, we present a Computer-Aided Design system for MicroMechanical Sensors (MMS). Moreover, it allows designer to improve automatically MMS response by the means of dimensions or shape modification of their design. Our system has a modular architecture centered around a widespread Finite-Element (FE) code, Ansys. The functional system scheme comprises three stages: Construction of a virtual numerical model accurately representing the real MMS. This stage is composed by the following descriptive levels: (a) Creation of a geometrical model of the MMS Specification of physical properties of materials composing the MMS, (b) Generation of a FE model for numerical analysis. This modeling stage is realized through Ansys preprocessing in interactive mode. Since our optimization approach is based on a CAD geometric model, boundary of the MMS design must be parametrically represented. We use cubic Bezier curves (patches) and cubic lines (bicubic patches). Validation of the virtual MMS model. This stage is carried out by the means of three descriptive levels: (a) Definition of operating and boundary conditions, (b) FE model simulation by numerical simulation of linear or nonlinear statical and dynamical equations, (c)Verification of simulation results in Ansys postprocessing with full graphical interaction. After a successful validation, the virtual model is used to make a design optimization. The design optimisation in an automated cycle of analysis, evaluation and modification of the MMS design. The optimum design is obtained for a given criterion, e.g. volume minimization (MMS miniaturization), fundamental frequency maximization, etc. It must also satisfy some behavior and design constraints, e.g. limitations on natural frequencies, stress, displacements, etc. For design modification, we have implemented an optimization module outside of Ansys code, which retrieves analysis results and different FE information from Ansys database. MMS are generally complicated and their response functions can only be known by the help of an expensive FE simulation. In order to reduce design cycle cost, our design modification approach is based on an approximation method called the method of moving asymptotes. This method replaces the initial and implicit optimization problem with a sequence of convex, explicit and separable subproblems having a simple algebric form. Our optimization method needs the knowledge of the derivatives of the response functions. Ansys, like most commercial codes, does not provide them, therefore we have implemented a design sensitivity. In the present state, this module calculates in analytical manner the derivatives of static (displacements, stress) and dynamical response (natural frequencies, eigenvectors). Our automated design system has been used for optimal search of frequency output and piezoresistif MMS. The MMS structures are modelized by 2d and 3d solid FE. In all cases considered, the design process always needs less than ten FE analysis or iterations. We finally note that FE mesh is automatically updated after each iteration.