Accelerated synthesis of electrically and thermally constrained power electronic converter systems

Every speci¯c application ¯eld of power electronics requires di®erent converter topologies, di®erent components, and di®erent considerations of speci¯c legal reg- ulations. However, the general concept of a switching circuit with passive compo- nents for electrical energy storage remains essentially the same. Moreover, a power converter is designed in such a way that the maximum processed power should be ¯t to the volume that is available for implementing the converter. Nevertheless, judging the converter only on the basis of its power density is somewhat weak. For example, a supply for consumer electronics should be low in price to withstand market competition. However, designing a converter with the highest power den- sity, which means using the best available components, will unavoidably result in an expensive product. In this simple example the objectives of keeping the highest power density and maintaining an acceptable cost are con°icting. This con°ict is ¯nally resolved by the designer, who emphasizes the importance of one or other requirement. In practice, a skilled designer explores several scenarios and chooses the one which in his/her highly subjective opinion is the best. Otherwise stated, when creating a new design, multiple con°icting requirements have to be satis¯ed and multiple possible solutions must be explored. Formally, this process is called synthesis. Designing a power converter means to characterize it with respect to the electrical, thermal and magnetic domains. As soon as synthesis is performed by means of a computer, the simulation methods and the corresponding models should integrate the above three domains. In this thesis a formal and general approach to assist power converter syn- thesis is developed. The approach is general enough to handle state-of-the-art topologies. The synthesis is performed in consecutive steps, and does not depend on a speci¯c converter topology or the design assignment itself. Synthesis is performed by the custom software tool M SIM that starts with a netlist description of the converter and handles a wide range of topolo- gies. The design objectives and constraints can be formulated in the electrical, thermal and magnetic domains. M SIM utilizes the multi-objective genetic algo- rithm NSGA-II for optimization and explores multiple tradeo®s without actually prede¯ning the preferences. In the electrical domain the power converter is modelled as a piece-wise linear system. This description simpli¯es calculations and saves simulation time, while describing the converter behavior closely. The time-domain \shooting" method for determination of the steady-state is used as an acceleration technique. In the thermal domain the behavior of the converter is described by a lin- earized thermal model, which is experimentally extracted from the power converter PCB. The method and the setup for the experimental extraction of the thermal model are developed and applied to a practical converter. The model is veri¯ed by means of independent measurements and shows a deviation in the order of 2%. The bidirectional link between the electrical and the thermal domains is established by means of temperature-dependent models and an iterative algorithm that ¯nds both electrical and thermal steady state. The proposed approach was applied to a practical example for predicting the temperatures of the designed converter and showed a 10% discrepancy with the measured results. Modeling and optimization of magnetic components is also performed by M SIM. The transformer of a representative converter was optimized, resulting in a discrepancy of 10% with the physical component. During the power converter synthesis multiple objectives and design param- eters are explored and multiple tradeo®s are found. A truly multi-objective op- timization algorithm, NSGA-II is used for this purpose. The genetic algorithm handles discontinuous search spaces and concave fronts. Three design domains, namely electrical, thermal and magnetic are handled by this single tool. The results are presented in the form of Pareto fronts of feasible designs and were ver- i¯ed by building a prototype. The resulting discrepancy is within the range of measurement error of the equipment.