Open Collections

Dey, Anshuman

University of British Columbia

Increasing power densities in modern-day power electronic systems is pushing electronic components to their thermal limits, warranting the need for accurate thermal modelling. Electromagnetic components like transformers and inductors are key components in such converter systems. Despite this, thermal modelling of passive components like transformers has not received as much attention as active components. The high-frequency AC operation of such electromagnetic components leads to complex losses that are difficult to accurately predict with empirical models. Further, the temperature dependence of electrical properties leads to non-linearity between the component’s temperature and losses. Hence, bi-directional multiphysics coupling between the electromagnetic and thermal domain using numerical analysis is required to model such components accurately. Although such coupled simulations are generally accurate, they have a high computational cost. Such models are often classified as Detailed Thermal Models (DTMs). Due to the high computational cost of DTMs, low-cost models are required. Most low-cost thermal models in the literature have not been evaluated for Boundary Condition Independence (BCI). Hence, they cannot be classified as true Compact Thermal Models (CTMs). In this thesis, three novel CTM modelling methodologies are evaluated. A BCI model is developed using the Lumped Parameter Thermal Network (LPTN) modelling approach. A thermal network model is developed by discretizing the transformer geometry into smaller volumes and evaluating the thermal resistances by approximating the three-dimensional heat transfer problem as piecewise one-dimensional. The BCI LPTNM results were comparable to numerical and experimental results. Next, CTMs of an inductor were developed using a real coded Genetic Algorithm (GA). First, a DTM of the inductor under DC excitation was developed and validated using experimental test results. The real coded GA optimizes resistance network values from the DTM results for varied boundary conditions. The resulting CTM can predict junction (winding) temperatures to within ±5 % of the DTM results for a varied set of boundary conditions. Finally, CTMs were developed using an admittance matrix approach. This was observed to be the optimum compact thermal modelling approach, with the ability to predict all nodal temperatures within ±7 % of DTM results for multiple heat sources.

Text

2022-07-27

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/82210

Mechanical Engineering

University of British Columbia

UBCO

http://creativecommons.org/licenses/by-nc-nd/4.0/