Thermal Networks for Time-Variant Cooling Systems: Modeling Approach and Accuracy Requirements for Lifetime Prediction

Boundary-condition dependent thermal network models are attractive because of their easy fitting and low computational expense. In the present work, a modeling approach for time-variant cooling systems is presented, switching between such networks fitted to different boundary conditions. The accuracy of the approach, both in terms of temperature histories and predicted IGBT lifetime, is investigated for Foster and Cauer networks, and is compared to errors in the thermal-interface-material resistance. In the case of boundary-condition changes, Cauer networks behave in a physically meaningful way, while Foster networks react too fast. The errors in predicted lifetime based on Foster networks are, for sufficient network length (N = 5), however, in the same range as an error of only 15 % in the thermal resistance of the interface material. As the latter error is difficult to avoid, even Foster networks can be candidates if ´engineering´ accuracy levels are required. Cauer networks are recommended if increased accuracy is required. On account of difficult-to-avoid inaccuracies such as that from the thermal interface material, the accuracy provided by the present modeling approach using network switching is found to be sufficient.