Electric Motor Optimal Design based on Multi-physics Modelling and Artificial Intelligence

Multi-physics modelling and Artificial Intelligence have been exploited to perform the optimal electromechanical design of a permanent magnet (PM) motor. Multiobjective programming together with machine learning/deep learning algorithms have been used, so that the designer has been enabled to find the preferred compromise among conflicting performance indices of the PM motor. Sixteen design variables were used to define the geometry f stator and rotor, pole pieces and permanent magnets. Three objective functions were selected and optimized as function of the motor's design variables, namely, the peak torque was maximized, the rotor inertia and the dissipated power (power consumption) were minimized. Design constraints were derived as function of the design variables and referred to structural safety and thermal status. Multi-physics modelling was undertaken to perform: the choice of architecture and topology of the motor, the electromagnetic field and related currents computation, the stresses and strains evaluation, the rotor oscillation and vibration, the thermo-fluid dynamic simulation. Artificial Intelligence has allowed the computation of objective functions in 1/105 of the time required for multi-physics simulations based finite element models. Pareto-optimal sets could be found in a very accurate way. The proposed optimization method provided an insight into the Pareto-optimal design solutions. Actually, the non-linear relationships between Pareto-optimal parameters and Pareto-optimal objective functions were investigated. The combination of such an insight with multi-physics modelling and with Artificial Intelligence has led to the derivation of a PM motor with very high performance.