Dual Active Bridge transformer allows the bi-directional transfer of electrical energy and provides isolation between windings. For this reason, they are used in applications involving high-power batteries, such as electric car chargers. They also offer the capability to transfer electric energy back from a car to a grid, e.g., powering a home during an outage.
We are comparing results for the 10kW DAB transformer from the Frenetic report with FEM simulation results by TRAFOLO. The report includes information about the core, winding, and circuit parameters.
Disclaimer. This is not a complete verification case or the final version of the case. Results can change while TRAFOLO introduces more accurate models. For up-to-date simulation setup, please refer to a template in the software.
Simulation setup
Core
The core analyzed is EE-type, made of ferrite with a gap on the central leg. TRAFOLO core templates cover the most common core geometries, and multiple gaps can separate the core into EE, EI, LL, distributed gap, and other configurations.

Core set-up
Windings

Coil group overview

Coil group builder
The winding has a PSP arrangement made of litz wire. Although it is possible to make and run a model with individual wires, grouping them into blocks saves time and provides adequate accuracy. To achieve the right amount of Rac resistance, we set an appropriate fill factor for each coil. The correction due to the proximity effect is computed using Dowell’s equation separately for each coil and harmonic.
Materials

Generalized Steinmetz equation
For harmonic simulations, core losses are computed using the Steinmetz equation, while the generalized Steinmetz equation (GSE) is used for transient simulations.
Where appropriate, materials include such data as temperature-dependent electrical conductivity, B-H curve, coefficients for Steinmetz equation, or laminate thickness (e.g., for electrical steels).

A wide range of material properties are specified for both electromagnetism and heat transfer
Computational domains

Assembly view

Excitation set up
In assembly, geometry is checked and processed. Half of the transformer will be simulated because the problem is symmetric. The results are scaled accordingly where necessary.
In the next step, coil groups are assigned to the Primary and Secondary windings. The excitation type is selected as the current obtained from the circuit simulator.
Waveforms from circuits
Separate magnetostatic simulations are run to calculate inductance values.
To calculate the power losses in the transformer, it is necessary to know the current in windings. This can be obtained by simulating the circuit. In our case, we used Simba.io, but any tool that allows you to model the circuit will do.

Circuit scheme in Simba.io used to obtain waveforms
Leakage inductance was taken as one used by Frenetic. Although in the simulation, we got a smaller value, for the cleanliness of the experiment, we tried to obtain and use the identical waveform.

Waveform import

Discrete Fourier transform of a waveform
Last but not least, the software does DFT for transient waveforms. The Rac/Rdc coefficient is computed using Dowell’s equation for each harmonic and coil.
We take Rac/Rdc coefficients for different harmonics and compute a single one using amplitudes as weights. This is a dirty way of computing effective coil resistance and losses for transient waveform, but it does the job (for now).
Heat transfer

Heat transfer boundary condition set up
Heat transfer computation using boundary conditions with a heat transfer coefficient and external temperature is a simple numerical task. Still, it is often the most ambiguous selection in the entire simulation setup. Should we use 5 or 10? ..it depends.
Only the engineer who knows nuances about the environment the transformer is placed can set an accurate heat transfer model.
Mesh

Automatically generated mesh
Mesh is among many factors that can spoil an entire simulation. Starting with numerical error or divergence and up to exhausted RAM reserves on the computer. More mesh elements are only sometimes better. It affects the convergence speed of numerical solvers and affects computation time. Manual meshing might be a time killer if you are a pedant.
Results
Let’s begin with numbers.
Parameters
Inductances and DC resistances are within range except for leakage inductance. In our case, we computed it by shorting the secondary winding. If you know the reason for the deviation, let us know.
A similar way of magnetizing inductance was computed by setting a current in one winding while keeping another one open.

Comparison of DAB results. Original picture by Frenetic, TRAFOLO results on top.
Losses

Loss distribution. Source: Frenetic
Losses predicted by Frenetic are shown in Figure. Total losses computed by TRAFOLO were 35.6W, core loss accounted for 14.7W, and coil losses for 20.9W.
The corresponding steady-state temperature distribution shows higher temperatures than 110 °C, but it is a matter of cooling. We used a thermal heat transfer coefficient of 15 W/m2/K for all outer surfaces except the bottom, which had an increased value of 100 W/m2/K. Without knowing more information, we can mostly guess.

Temperature distribution