Contribuciones al diseño de un "Dual Active Bridge" usando técnicas de optimización

Laburpena

Transformers play a key role in the transmission of energy since they are able to increase or reduce the voltage at the same time as they introduce galvanic isolation when connecting different grids, generators and loads. Thanks to the new semiconductors based on wide bandgap, capable of switching at higher frequency without increasing power losses and all the advances and studies that have been made in the design of Medium Frequency Transformers (MFT), Dual Active Bridge (DAB) converters have started to gain on importance. The employment of power switches allows to increase the fundamental frequency in the transformer up to several tens of kilohertz (depending on the power level), decreasing significantly its volume, weight and, when the switching frequency is above 20 kHz, its acoustic noise as well. The reduction in weight and volume allows to take advantage of the great capillarity of the Medium Voltage DC, MVDC, railway distribution network allows the punctual connection of loads or generation systems such as renewable energies. Having a very useful application in the generation of a very wide grid of electric vehicle chargers distributed without the need for major building works due to the low weight and volume provided by the SSTs compared to the connection with traditional LFTs. Nevertheless, due to the increasing need to design MFTs and power converters with higher power density and a wider range of input voltages operating at the frontier of the technology limit, a precise thermal study becomes a crucial parameter. The MFT is one of the most critical components in terms of temperature, since increasing the frequency exponentially decreases the size with respect to the traditional Low Frequency Transformer (LFT). Therefore, it will also be critical to reduce, or manage properly, the losses and temperature distribution in the MFT, in order to avoid the loss of insulation between the litz wire strands due to high temperatures and also prevent the thermal runaway of the magnetic core. To model the thermal behavior of the MFT there are theoretical models that allow estimating the temperature, but in some cases, such as the one proposed, these models cANNot be applied due to the lack of knowledge of some variables, such as the air velocity on the MFT surfaces when forced convection is used. To solve the problem with theoretical models, Finite Element Simulations (FEM) can be used. The main problem with FEM simulations is that they are very slow compared to theoretical models and therefore they can be used only to validate designs. In order to use these FEM simulations in the design process without slowing it down, the use of Artificial Neural Networks (ANN) is proposed, allowing the generation of an equivalent thermal model from a dataset of FEM simulations. This work shows the process of dataset generation, the training of the ANN, as well as its use for an optimization of a brute force design of a DAB converter with a very wide range of input voltages, operating with different modulations, which generate very different losses in the core and windings of the MFT. Therefore, generating MFT designs with highly unequal thermal behavior depending on the operating point and the modulation applied. The brute force optimization process generates 1000 possible converter designs with 40x40 possible combinations of input and output voltages. Reducing the generation time from almost 1 year (if FEM simulations are used directly) to only 2 days, using the ANN trained with FEM simulations. Given the use of SiC devices, FEM simulations have been used in other design steps, such as the optimization of busbars. Finally, the results obtained with the ANN have been compared with experimental results of a DAB working with different modulations and at different operating points, obtaining in all cases errors lower than 2.5 ºC in the surface temperatures for a MFT cooled by forced convection.