Medium frequency transformer leakage inductance modeling and experimental verification

100 kW, 10 kHz Medium-Frequency Transformer Design Optimization and Experimental Verification

“…Moreover, proper design of the MFT electric parameters, magnetization and leakage inductance, is paramount for the correct operation of the converter, especially in case of SRC based topologies where the MFT is a part of a resonant circuit. Therefore, the modeling and manufacturing margins associated with parameter control are more strict [20]. Furthermore, the desired increase of transformer power density is tied to the increase of the heat loss density, resulting in higher temperature gradients unless correctly addressed.…”

“…Calculation of the frequency dependent winding losses and leakage inductance is based on the developed hybrid model [20], which is a generalization of the original Dowell's model [25]. Developed model is capable of properly taking into account both frequency and geometry effects characteristic for the given frequency range and increased clearance distances of high power MFTs, necessary to properly accommodate the high dielectric withstand requirements for MV applications.…”

“…Compared to the state of the art, [22]- [24], this work introduces two novel analytical models, providing improved accuracy of hot-spot temperature rise [21] and leakage inductance [20] estimation without added computational cost. Moreover, in contrast to direct brute force approach, as used in [23], [24], the presented optimization algorithm itself uses a set of heuristics allowing pre-detection and avoidance of infeasible design domains, similar to branch and bound logic, thus saving the processing power and allowing to test much larger and more dense optimization variable sets.…”

“…According to [20], the equivalent length of the leakage flux path through the air is, in general, not equal to the core window height, as assumed by the original Dowell's geometry equivalence. As can be seen in Fig.…”

“…As shown in [20], the modeling correction is performed by a simple substitution of H w with h eq , from (19), within (7), (13) and (15) resulting in accuracy of above 94% over the entire geometry range, as shown in Fig. 4.…”