ANNcoilsDataset

Electromagnetic FEA simulations of coupled coils for IMDs from the article Neural Network-Based Design of Wireless Power Transfer Systems for Implantable Medical Devices (Álvaro Rodríguez Fuentes, Miguel Jiménez Carrizosa, Regina Ramos Hortal).

Description of the dataset

The dataset includes geometric and electrical data of coupled multilayer squared planar coils. 4635 simulations were performed using the electromagnetic FEA software Ansys HFSS 2025 R1. This data was used to train an Artificial Neural Network using the Neural Net Fitting application of MATLAB. The file includes:

Simulation inputs: geometry

The coupled coils of the simulations are defined by their geometry. Thus, the inputs of the simulations are included in columns 1 to 16. These variables are represented in the following figure:

1. Freq (MHz): operating frequency (fixed at 6.78 MHz)
2. Dist (mm): distance between primary and secondary coils (fixed at 15 mm)
3. W1 (mm): trace width of primary coil
4. W2 (mm): trace width of secondary coil
5. S1 (mm): trace separation of primary coil
6. S2 (mm): trace width of secondary coil
7. N1: number of turns per layer of primary coil
8. N2: number of turns per layer of secondary coil
9. D1 (mm): external length of primary coil
10. D2 (mm): external length of secondary coil
11. Nlay1: number of layers of of primary coil
12. Nlay2: number of layers of secondary coil
13. H1 (oz): trace thickness of primary coil (fixed at 1 oz.)
14. H2 (oz): trace thickness of secondary coil (fixed at 1 oz.)
15. X1 (mm): total thickness of primary coil (fixed at 1.6 mm)
16. X2 (mm): total thickness of secondary coil (fixed at 1.6 mm) The range of each variable is defined in the following table:

Var.	Freq (MHz)	Dist (mm)	W1 (mm)	W2 (mm)	S1 (mm)	S2 (mm)	N1	N2	D1 (mm)	D2 (mm)	Nlay1	Nlay2	H1 (oz)	H2 (oz)	X1 (mm)	X2 (mm)
Min	6.78	15	0.5	0.5	0.5	0.5	1	1	10	5	2	2	2	2	1.6	1.6
Max	6.78	15	4	3	3	2	5	5	40	20	4	4	2	2	1.6	1.6

Simulation outputs: electrical variables

The outputs of the simulations are the electrical parameters of the inductive link. The FEA software resolves the system as a two-port network, yielding the complex impedance matrix of the system: $\mathbf{U} =\mathbf{Z}\cdot\mathbf{I}$;

Their complex values are included in columns 17-20. From the impedance matrix (Z-matrix), the electrical variables of the inductive link are extracted (columns 20-28):

20. L1 (uH): Self inductance of primary coil, in uH.

$L_1=\frac{1}{2\pi f_{sw}} \Im{(Z_{11})}$

21. L2 (uH): Self inductance of secondary coil, in uH.

$L_2=\frac{1}{2\pi f_{sw}} \Im{(Z_{22})}$

22. R1 (Ohm): Self resistance of primary coil, in Ohms.

$R_1=\Re{(Z_{11})}$

23. R2 (Ohm): Self resistance of secondary coil, in Ohms.

$R_2=\Re{(Z_{22})}$

24. k: Coupling factor of the inductive link.

$k=\frac{\Im{(Z_{12})}}{\sqrt{\Im{(Z_{11})}\Im{(Z_{22})}}}$

25. Q1: Quality factor of primary coil.

$Q_1=\frac{\Im{(Z_{11})}}{\Re{(Z_{11})}}$

26. Q2: Quality factor of secondary coil.

$Q_2=\frac{\Im{(Z_{22})}}{\Re{(Z_{22})}}$

27. Ro: optimal output load, in Ohms.

$R_o = R_2 \sqrt{1+k^2Q_1Q_2}$

28. Eff_max: maximum theoretical efficiency.

$\eta_{max} = \frac{k^2Q_1Q_2}{(1+\sqrt{1+k^2Q_1Q_2})^2}$

Usage

This repository includes two versions of the dataset: a .csv file and a .mat file. The .mat file includes a table (ANNCoilsFEA) and three arrays:

• In: includes the 10 geometrical variables that represent the inputs of the ANN: (W1, W2, S1, S2, N1, N2, Dext1, Dext2, Nlay1, Nlay2).
• Out: includes the 5 electrical variables that represent the outputs of the ANN: (L1, L2, R1, R2, k).
• Out_norm: includes the 5 electrical variables that represent the outputs of the ANN. These variables are preprocessed to enhance the regression capabilities of the ANN: (log10(L1), log10(L2), log10(R1), log10(R2), 10.*k)

For simple and fast tests, the ANN can be created and trained using the Neural Net Fitting application of MATLAB, either by its graphical GUI or generating the proper training code. More information about this tool can be found in:

• https://www.mathworks.com/help/deeplearning/ref/neuralnetfitting-app.html
• https://www.mathworks.com/help/deeplearning/ref/feedforwardnet.html

Additional information

More information about how to properly train the ANN and adjust their hyperparameters can be found in the article Neural Network-Based Design of Wireless Power Transfer Systems for Implantable Medical Devices (Álvaro Rodríguez Fuentes, Miguel Jiménez Carrizosa, Regina Ramos Hortal).

Credits

If you use this dataset, please consider the proper citation of the article:

[1] A. Rodriguez-Fuentes, M. J. Carrizosa and R. Ramos, "Neural Network-Based Design of Wireless Power Transfer Systems for Implantable Medical Devices," in IEEE Transactions on Power Electronics, doi: 10.1109/TPEL.2025.3614366.

Author

• Name: Álvaro Rodríguez Fuentes
• Affiliation: Centro de Electrónica Industrial, Universidad Politécnica de Madrid
• Email: alvaro.rofuentes@upm.es