Considerations on matrix based analytical temperature prediction tools for electrical machines

Abstract

Thermal analysis is an important part of electrical machine design. Improving the accuracy and speed of the thermal analysis model is an important step in optimizing the design procedure, considering the need for temperature prediction of high-performance electrical machines at both steady-state and transient operation conditions. In this paper, a fast thermal modeling technique with lower computational cost and runtime is implemented. This model applies matrix arithmetic as an advanced variant of the classical lumped parameter thermal network (LPTN). This allows flexible adjustment of the model parameters in cases such as sensitivity analyses. A 24-slot 8-pole high-speed PMSM is used as a test case. An equivalent five-node thermal model is built for one of its stator slot sections, thus finding the temperature at the copper winding and stator iron. The steadystate temperature at the five nodes derived using the classical LPTN and the proposed approach are compared, along with the calculation time taken. The error values of the temperature results obtained by the two methods are also presented.