Axial-flux actuator MDO (axfluxmdo)
- Python
- NumPy
- OpenMDAO
- pymoo
- scikit-learn
- Gmsh/GetDP
Problem
An axial-flux motor for a robot joint has to satisfy electrical, magnetic, thermal, and mechanical limits at once, and those limits interact. Picking pole count, air gap, and current density by intuition, one objective at a time, tends to produce a design that looks strong on torque and fails on temperature or packaging.
Approach
axfluxmdo is a layered design-exploration toolkit. An analytical workbench evaluates torque, losses, winding temperature, mass, and constraint margins fast enough for sweeps. A 2.5D annular model resolves fields by radius and adds manufacturing effects (air-gap error, coning, runout) plus a torque-ripple proxy and axial force. A pymoo and OpenMDAO layer runs Pareto trade studies over mixed continuous and discrete variables and supports system co-design with the gearbox, inverter, and structure. Gmsh and GetDP hooks check the closed-form fields against magnetostatic FEA and report the residual, and a Gaussian-process surrogate drives Bayesian optimization for expensive objectives.
Result
The toolkit turns early actuator design into fast, constraint-aware trade studies instead of spreadsheet guesses. One result the model makes concrete: at fixed size and loading, torque is set by the air-gap shear stress and is independent of pole count, so the gain from more poles is torque density through thinner yokes rather than extra torque. The write-up works through the pole-pair study, a feasible Pareto front, and the radius-resolved fields; the code is on GitHub.