Search for a command to run...
Electric machines play an important role in a wide range of modern industrial applications, such as the automotive, aerospace, and renewable energy sectors. To meet the increasing demands for performance, reliability, and efficiency, it is necessary to develop robust designs (geometries) that largely maintain their performance even under uncertain material properties and due to manufacturing tolerances. This is precisely what this work addresses by explicitly accounting for these uncertainties and developing a method that enables the efficient determination of robust designs for electric machines. First, a framework is developed that simultaneously enables shape and parameter optimization with the help of Isogeometric Analysis. For this, gradients are calculated with respect to control points as well as with respect to design parameters, which efficiently enables the combined optimization of multiple objective functions such as cost, average torque, and variance of the torque. For the analysis of different conflicting objectives, a predictor-corrector method based on second derivatives is used. This enables an efficient exploration of the Pareto front and reduces the computational effort by exploiting the local superlinear convergence. However, since no uncertainties have been considered in this model so far, the designs found are sensitive w.r.t to parameters. Therefore, the method is extended by considering uncertainties that can be result, for example, by production tolerances. The problem is then formulated as a min-max problem in order to find solutions that guarantee a reliable performance even under the most unfavorable modeled scenarios w.r.t to a predefined uncertainty set. To solve this explicitly robust problem, an efficient algorithm was developed that uses generalized derivatives and can be directly appliedto other applications. The additional computational cost in the optimization resulting from this remains within a reasonable factor, while the robustness of the electrical machine is significantly increased because the designs are less sensitive to the considered uncertainties. Finally, this principle is then applied to topology optimization under parameter uncertainties by showing the existence of a robust topological derivative and applying it in a level-set algorithm. This robust derivative results from evaluating the classical topological derivative for the most unfavorable parameter value and thus extends the concept of the classical topological derivative for robust problems. Using the example of a synchronous machine, it is shown in numerical tests that with this method a design can be found with which the performance in the worst case scenario can be significantly increased, without the nominal performance (without consideration of the uncertainties) being substantially worsen. The additional costs that arise during the calculation of the robust design also remain, as with the parameter and shape optimization, manageable(4-10 times higher compared to the nominal optimization). Overall, the results show that a robust shape and parameter optimization as well as a robust topology optimization of electrical machines can be done efficiently, which could bring an added value for the industry. In conclusion, based on the findings, this thesis proposes several promising possibilities for further research.