Genetic Optimization of Heat Transfer Coefficients for LPTN Models

Nowadays, the main focus of the design process of electrical machines is typically electromagnetic analysis. However, as requirements have increased for electrical machines and drives with specific uses and higher efficiency, thermal analysis has become a more significant part of the design process. The temperature rise within a machine influences its output power and can also lead to the thermal degradation of significant parts of the machine, in particular the winding insulation and permanent magnets, if these are used in the machine construction. A precise estimate of temperature is crucial for the protection of critical machine components, which in turn requires an accurate prediction of machine temperatures. This can be achieved through various research methods, such as finite element analysis or analytical approaches, along with appropriate analysis methodologies. In this work, Lumped Parameter Thermal Networks are used. This analytical method is very convenient due to its need for low computing time and fast optimization execution. This publication focuses on the optimization of heat transfer coefficients using a genetic algorithm, which is a key factor in achieving accurate thermal analysis. Various methods for the estimation of these coefficients are evaluated and incorporated into the optimization process. In addition, graphical outputs of the calculations, including comparisons of the calculated and measured temperatures, for different methods used to approximate the heat transfer coefficients, are also presented in this paper. The measured temperatures were obtained on a fully automated test bench under stable conditions. The paper concludes with a discussion of the deviations between individual results, highlighting their impact on overall optimization accuracy. Moreover, an improved methodology for thermal analysis is suggested, enabling real-time, sensor-less temperature predictions.