: Reducing iron loss in axial flux permanent magnet (AFPM) motors is critical for improving efficiency. This study proposes a design-optimization procedure that combines 3D finiteelement analysis (FEA) data with an artificial neural network (ANN) surrogate. For four design variables—airgap length, rotor back-yoke thickness, stator slot width, and stator slot depth—the search bounds were defined to avoid tooth and back-yoke saturation, and the corresponding space was sampled to construct a dataset. Using this dataset, the ANN was trained and then used to explore low-iron loss solutions. On an independent validation set, ANN predictions showed high agreement with 3D-FEA reference values, enabling rapid evaluation of many design candidates. As a result of the optimization, total iron loss decreased relative to the baseline, and torque increased by 3 Nm. These results demonstrate that the ANN-based surrogate model can reliably perform geometrydependent
iron loss optimization in AFPM motors, providing a fast and accurate alternative to repetitive 3D-FEA evaluations.