An Investigation of torque density and losses in high-speed permanent magnet machines

High speed permanent magnet machines have been widely adopted for their ability to achieve high power densities while also retaining high efficiencies. However, operation at high speeds introduce several design and analysis challenges which encompass electromagnetic, thermal and mechanical considerations. Arguably the most challenging aspect of high-speed machine design is the reliable prediction of losses in the machine, particularly if the influence of the converter is accounted for. This thesis is focussed on the design and loss modelling of high speed permanent magnet machines, with a particular emphasis on establishing a detailed understanding factors that result in the torque density of machines decreasing with increasing speed. The thesis reports on a systematic investigation to establish the variation of torque density and power density with machine speed by way of series of design studies for a 250kW surface-mounted permanent magnet. The torque density is shown to reduce with machine speed rating at different rates depending on the constraints applied. In several cases, an optimum speed for achieving maximum power density is observed, beyond which the power density begins to reduce. The thesis then considers in detail the influence of ripple currents generated from hysteresis closed loop control, on machine torque output and iron losses. A novel post-processing method is developed for iron loss calculation to accommodate with the large number of data points required to fully capture the effect of high frequency current ripple. A series of analytical derivations are developed to illustrate that high frequency iron losses due to switching are largely independent of the exact nature of the switching behaviour and governed by steady-state machine parameters. The rotor eddy current losses in rotor magnets and a metallic containment sleeve are then calculated, using a novel three-dimensional analytical model for field, current, and loss prediction. Good agreement is achieved between the analytical model and finite element simulations.