论文和技术资料

Simulations usually depend on a variety of parameters like geometric dimensions, process parameters, loads, or material properties. Typically, some of these parameters are fixed whereas others may be varied to obtain a special result. In many cases a simple sweep can be used to examine their influence on the result. However, things get much more complicated if several parameters, multiple studies and components, or even different models are involved. Let’s start with a simple example, the “busbar model” from the Application Library in COMSOL® Multiphysics: A piece of metal with an applied voltage gets hot due to Joule heating. We like to know how the maximum current depends on the width of the metal, with a boundary condition of 80°C max due to Joule heating. To calculate this curve, the applied current has to be adapted for every single width of the busbar.
The add-in ‘GoalSeeker’ was designed to solve this kind of questions. The core algorithm varies one or more parameters within a given range to find a target value or a minimum value. The optimization is based on linear extrapolation and parabolic interpolation. In case of bad convergence, a boxing algorithm is applied. To allow more complex simulations the add-in allows additional steps for initialization and analysis before and after the search algorithm, respectively. Finally, a user defined list of derived results will be calculated. The optimization algorithm may be combined with a parameter sweep to answer questions like the one above. All results are stored in a result table. Optionally, individual results may be saved as soon as the optimization of a parameter set is finished. If required, a classic parameter sweep based on the found optimized values may be calculated at the end. Sometimes simulations can get very complicated with multiple studies and even different models which depend on each other. To reduce this complexity, combined steps can be defined by the user. Besides combinations of multiple studies and / or method calls it is possible to change parameter values based on formulas or derived results, import Meshes and interpolation data, or call any export feature. All features described so far may be combined with an external model. A typical scenario combines the production process of a device with its use case. The possibility to split the simulation in two different models is very convenient since the models remain compact, and different people may work on the task at the same time. During the optimization process the add-in ‘GoalSeeker’ is able to run studies in both models and exchange all kind data between them, similar to a model containing multiple components.