Optimizations using conventional genetic algorithms or comparable methods already save a lot of time in finding the optimal design for a motor or generator based on a list of requirements. However, complexity is constantly increasing, meaning that this process still takes weeks or months and is computationally intensive.
To reduce this complexity, especially in light of rapidly changing requirements and multiple optimization goals (multi-objective optimization, MOO), an optimization framework is being developed at iMEK. Mechatronic systems, in this case electrical machines, are divided into smaller segments, for example into their physical domains, and modeled. The models are examined using Design of Experiments methods and metamodels are generated that approximate the behavior of the models as precisely and predictably as possible. With the metamodels, a broad design space can be completely calculated within a very short time and evaluated using simple cost functions.
This also makes it conceivable to use the methodology in the Engineer-to-Order process, in which customers receive a data sheet for a simulated machine that is optimal for them within a few minutes after specifying their requirements. This reduces development costs, so that only a sample validation is necessary before actual use and the optimal machine can be found for each application.
