To start, parametrize your project. During project inquiry, the AEDT integration scans every model in the project for variable definitions, and it also scans for project variables. Dependent parameters are skipped. If a parameter is marked as hidden in AEDT, that information is not sent to optiSLang, so hidden parameters are always included. You must make a decision on whether to treat parameters from different models in the project as independent or as communal and to be always kept synchronized. In optiSLang, the AEDT integration has a shared parametrization mode with an additional include non-shared option. in AEDT, there are project variables you can use to steer model parameters. So you can leverage features on both sides for managing overlapping, (partially synchronized), parameter sets. You can also use optiSLang's standard parametrization management tools, using the dependent parameter type for aligning or otherwise controlling parameters.

When bundling optiSLang-demanded designs for simultaneous execution as DSO jobs, DSO design variation tables are hosted in the Optimetrics section of each model, and that there is no Optimetrics branch on the project level. Even when varying solely project variables, this will still yield separate DSO tables and separate jobs for each model in the project.

Also ensure that your parametrization scheme is connected with the goals you are pursuing with the optiSLang analysis you are about to undertake. Bad parametrization might hide the design optimum from the best optimizer or might destroy the connection of a robustness sampling with reality. A thoughtfully improved parametrization can achieve your goals with a fraction of designs that are required for the task within the most simple ad hoc parametrization scheme. Use as an example, a 2D Maxwell model of a permanent magnet synchronous motor. For design optimization, the magnet positions and angles can be the crucial parameters and due to symmetry, a quarter model can represent the entire machine. For questions on robustness, reliability, tolerances, and achieving specifications, it may be necessary to parametrize eccentricity and ellipticity. If magnet quality scatter is important, this may require a full 360 degree model because it would be unrealistic to assume that 16 worst-case magnets can find their way into one single machine.

Lastly, approach this task from the perspective of the algorithms to be applied. It is in your interest to give a robustness sampling algorithm the chance to examine a parameter space where the simulation can represent the relevant physical effects happening in the real world (for example, gearing the parameter space at being representative for the scatter generated by the real factory line). And it is in your interest to give an optimizer algorithm the chance to hit successfully feasible designs frequently. Particularly on the question of how the parametrization affects the stability of feasible geometries, a lot of influence can be leveraged on your side. If you devise a stable and meaningful parametrization, are attentive to the kind of errors encountered by failing designs, are quick in eliminating the origins of the errors appearing most frequently in error logs, and if you learn to use optiSLang postprocessing as a diagnostic toolbox, you will quickly become an efficient user of design variation algorithms for robust design optimization purposes.