A Methodology for Accurate and Computationally Efficient Thermal Design-Analysis of Electrical Machines

This work presents a methodology enabling accurate and computationally efficient thermal design-analysis of electrical machines. The proposed approach addresses three key challenges associated with the process: - intuitive and flexible model definition, - computational efficiency and accuracy, and - cost and time effective hardware calibration. The model makes use of a lumped-parameter thermal network (LPTN), which is automatically generated using a single three-dimensional static finite element (FE) simulation. Here, the dynamic FE-generated LPTN model allows for both static and transient thermal analysis to be performed seamlessly, accounting for a parametric input, i.e. external and interregional boundary conditions and temperature dependent heat sources. Further to these, the LPTN model retains all the information associated with the 3D FE model regions, with an adjustable model resolution. Moreover, to account for the manufacturing and assembly thermal factors, a motorette-based hardware calibration was employed here. The use of motorettes allows for cost and time effective approach of informing the design process. To demonstrate applicability and limitations of the proposed methodology, a representative case study permanent magnet axial-flux machine (AFPM) has been analysed. The theoretical body of work has been supplemented with experimental results from tests on AFPM motorettes to illustrate the complete workflow. Results show that the proposed approach has a significantly reduced computational effort, which is of particular importance when generating efficiency maps or analysing specific operating cycles. In the analysed case study, the solving time has been reduced 12000 times, as compared with a direct 3D FE solution when assessing the analysed motor performance over a one-hour driving cycle, with a maximum over-temperature deviation of 0.5%.