Tags:automatic differentiation, computation of electric machines, electric machine and uncertainty Quantification
Abstract:
The design of electric machines is constantly improving and modern tools respect both thermal and electromagnetic boundary conditions. While these tools allow for the consideration of designs for specific parameters under ideal assumptions, the final real-world machines additionally depend on deviations due to uncertainties from material data or production processes. These deviations lead to differing results for temperatures and loss data. Considering the influences of many uncertainties at once - e.g. with Monte-Carlo-Methods - can be overly time-consuming. The paper deals with a methodology of computing machine setpoints including the uncertainty propagation through the model in a faster manner. To achieve this goal, existing models for electric machines are extended by an inherent uncertainty modeling. The error propagation throughout the models is based on derivatives computed by automatic differentiation. This approach leads to a very efficient calculation of complete characteristic maps including uncertainties with respect to large numbers of parameters. The uncertainties generated from high dimensional input parameters are of particular interest for motor calculation based on flux linkage characteristics. A novel modeling approach for these parameters is presented and verified with respect to capability and performance by numerical experiments. The benefit of this method is demonstrated by computational examples of general interest, e.g. the computation of maximum torque characteristics with confidence intervals.
Efficient Uncertainty Propagation for Electric Machine Computation