Thermal analysis and optimization on a transformer winding based on non-uniform loss distribution

•An electromagnetic-thermo-flow bidirectional coupling model was developed.•Winding temperature was calculated based on non-uniform loss distribution.•Winding temperature distributions under two loss conditions were compared.•Multi-response optimization results under two loss conditions were compared.•Negative impacts of uniform loss assumption were quantified. Generally, thermal analyses and optimizations on the transformer winding were carried out under an assumption of uniform heat losses. But in fact, winding heat losses are distributed unevenly, which may intensify local hot-spots and result in winding performance deterioration. This work was to investigate the difference of winding temperature distributions under the two loss conditions. To realize it, first, an electromagnetic-thermo-flow bidirectional coupling numerical model was developed for a converter transformer winding to calculate actual winding losses and temperature. Then, the winding temperature distribution was compared with that under the assumed uniform loss condition, and their differences were quantitatively analyzed. The results showed that for most winding structures studied in this work, the winding hot-spot temperature rises (ΔTmax) under the non-uniform loss condition was greater than that under the assumed uniform loss condition, and their difference was greater than 5%. This difference was more obvious at a less flow rate. This suggested that ΔTmax would be underestimated based on the assumption of uniform heat losses, which posed a threat to the transformer reliability. Therefore, it was necessary to consider non-uniform loss distribution to calculate the winding temperature, so as to improve the safety margin of the transformer and thus improve its reliability. Furthermore, to eliminate the uncertainty of multi-response optimization results caused by the uniform loss assumption, a multi-response optimization on the winding structural parameters was conducted based on the non-uniform loss distribution. The optimization results under the two loss conditions were compared. The results showed that the optimum winding structural parameters under the two loss conditions were consistent. However, the hot-spot positions of the optimum winding structure under the two loss conditions were different. This meant that it was feasible to adopt the general assumption to conduct multi-response optimizations to improve efficiency, but to predict the hot-spot of the optimum winding structure