Comparison of Surface Mounted Permanent Magnet Coaxial Radial Flux Magnetic Gears Independently Optimized for Volume, Cost, and Mass

• Published in: IEEE transactions on industry applications 2018-05, Vol.54 (3), p.2237-2245
• Main Authors: Gardner, Matthew C., Jack, Benjamin E., Johnson, Matthew, Toliyat, Hamid A.
• Format: Article
• Language: English
• Subjects: Direct drive; finite element analysis (FEA); Genetic algorithms; genetic algorithms (GAs); magnetic gear; Magnetic gears; optimization; permanent magnet; Solid modeling; Three-dimensional displays; Torque; torque density; Two dimensional displays
• ISSN: 0093-9994; 1939-9367
• DOI: 10.1109/TIA.2018.2803039

This study employs a genetic algorithm (GA) to optimize surface mounted permanent magnet coaxial radial flux magnetic gear designs using both 2-D finite element analysis (FEA) and 3-D FEA. Specifically, the GA optimizes different designs, which are all rated for a stall torque of 500 N·m and a gear ratio of approximately 5, to independently maximize volumetric torque density (VTD), torque per dollar (TPD), and gravimetric torque density (GTD). Maximum VTDs of 274 and 210 kN·m/m 3 were obtained with 2-D and 3-D simulations, respectively. Including the space required to provide an axial buffer for leakage flux resulted in a maximum leakage adjusted VTD of 162 kN·m/m 3 . Maximum TPDs of 5.86 and 5.47 N·m/ were obtained with 2-D and 3-D simulations, respectively. Maximum GTDs of 102.8 and 86.8 N·m/kg were obtained with 2-D and 3-D simulations, respectively. The results demonstrate that independently maximizing these three metrics leads to markedly different designs with widely varying performance characteristics. The most significant differences occur between the maximum VTD and maximum TPD designs, and the analysis includes a thorough discussion of the dominant design parameters driving this phenomenon. Finally, the impacts of end effects on the optimal design parameters are also illustrated to demonstrate that consideration of these 3-D effects leads to significantly different performance predictions and to different optimal design selections.