技術論文とプレゼンテーション

The architecture of axial flux machines has gained strong popularity in recent years due to its enhanced torque density in comparison to conventional radial flux motors, but finding the optimal configuration for specific applications is a multidimensional problem, which is difficult to solve using traditional parameter sweeps or manual trial and error. In this project, we combined a high-fidelity electromagnetic model of an axial flux motor with parametric optimization to search for the optimal parameter configuration of the motor based on arbitrary customer demand. This process completely bypasses manual iteration, accelerating motor development cycles using tools within COMSOL Multiphysics®.

A 20-pole, 24-slot motor is represented using the Rotating Machinery, Magnetic (RMM) interface together with the moving mesh module. This configuration enables realistic representation of stator to rotor interactions, eddy-current behavior in permanent magnets, and the core losses produced by the time-varying magnetic field. To make the model tractable, we exploit the 4-fold geometric and electromagnetic periodicity of the 20-pole/24-slot configuration through periodic boundary conditions, applying a powerful reduction of complexity that concentrates numerical expense where it affects the result rather than on repetitive calculations across equivalent periodic sectors. Together with other coil and rotor mesh refinements, the total simulation time for a transient study has shortened by over an order of magnitude. To validate the digital motor, its result was compared to a physical twin. The simulation yielded a nominal dynamic torque deviation of only 1.13% in comparison to the real motor, confirming the fidelity of the model. The refined and validated model is then driven by a parametric optimization in a 13-dimensional search space consisting of 11 geometric parameters and 2 electrical parameters. To ensure the efficient search of this space, the BOBYQA algorithm was chosen over the alternatives of Nelder-Mead and COBYLA based on their performance on the Rosenbrock test function. It was noticeable that BOBYQA’s balance between stability and strong search space exploration enabled the fastest convergence of this strongly nonlinear problem while also covering the largest search space.

The viability of the COMSOL Multiphysics®-automated design process is showcased through a demonstration optimization run, targeting the motor constant K_m on a baseline geometry. The demonstration run is completed in roughly 300 stationary iterations, where BOBYQA delivers a twofold increase in motor constant. This routine can be generalized to arbitrary objective functions reflecting the customer’s priorities and supports the automated design of essentially arbitrary axial flux motor topologies as the entry point to a rapid-prototyping pipeline.