LiBforSecUse processed data for machine learning modelling and uncertainty propagation

Original data 

Original data source

The raw data have been downloaded from Zenodo:

LiBforSecUse Data Release - Impedance spectra of life cycle tests of commercial 18650 cells (v2)

DOI link to the data : https://doi.org/10.5281/zenodo.6418665

CC BY 4.0 license (allows re-distribution and re-use on the condition that the creators of the dataset are appropriately credited)

Users are asked to pay tribute to the original project and the researchers providing the data by citing the following data source.

Citation : Chan, H. S., Dickinson, E., Fedorovskaya, E., Gaberšček, M., Heins, T. P., Meddings, N., Lee, Y. Y., Moškon, J., Park, J., Ruiz, V., & Seitz, S. (2022). LiBforSecUse Data Release - Impedance spectra of life cycle tests of commercial 18650 cells (Version V2) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.6418665

Description of the original dataset

Dataset release from the EMPIR project LiBforSecUse. It includes regular impedance spectroscopy (EIS) and capacity measurements made during life cycle tests of commercial 18650 (graphite/NMC) cells. 

Measurements were performed by 6 separate institutions using different setups (hardware and software). However, during the LiBforSecUse project, the analysis of the data revealed some discrepancies for Institution 5 and 6.

In total, 34 cells were measured, at different fixed SOC (range 20-100%) after several charge-discharge cycles (up to 5900 cycles), resulting in 5479 experiments. For each experiment, EIS measurements are performed for different frequencies ranging from $10^{-2}$ to $10^{4}$ Hz. From an experiment to another, the number of discretisation points of the frequency spectrum can vary form 39 to 61. However, a vast majority of experiments are recorded at 49, 58 and 61 points.

For details on BUTs, setups, experimental design, description of measured variables and original file format, please refer to the documentation of the original dataset. 

Data processing by METAS

Exploratory data analysis and data cleaning

According to the original documentation, Institution 5 and 6 were removed from the dataset (corresponding to 8 cells and 1146 experiments).

Some SOH measurements are missing. As it is one of the principal quantities of interest, these were removed (corresponding to 33 experiments)

NFRA measurements were excluded.

Constant variables, such as temperature during EIS tests (23 °C) and cycling current (4 Amperes).

3 experiments have no data at low frequencies. They have been removed.

Potential outliers disclaimer

Some data points exhibit physically unrealistic behavior (e.g., increasing SOH with more charge-discharge cycles). However, they have been retained, as an appropriate handling (correction/removal) of these points depends on the context of use.

Interpolation of the frequency spectrum

After data cleaning, the majority of experiments are measured at 61 points descretizing the frequency spectrum from $10^{-2}$ to $10^{4}$ Hz, (10 points per decade, with logarithmic spacing).All data were aligned on the 61 frequencies measured for the experiments conducted by Institution 1, Cell 4, at SOC 100 before cycling. This is done by interpolating the measurements with Cubic Splines on the log-transformed frequency axis, with the constraint that the second derivative of each Spline curve vanishes at its bounds. When extrapolations were required, missing values (Nan) were introduced. Processing was done in Python using scipy.interpolate.CubicSpline. 

Data wrangling

Original tabular data contained in 3 separate files were merged into a unified dataset. EIS data, originally stored as triplets (frequency, resistance, reactance) were expanded into 122 variables : 61 log-regular points on the frequency axis between $10^{-2}$ to $10^{4}$ Hz, multiplied by the 2 dimensions of the couple (resistance, reactance). In this format, each line corresponds to an experiment for given cell, SOC, and number of charge-discharge cycle. A unique experiment identifier was defined as "exp_ID" = "Cell_ID"_"nCycle"_"SOC".Variable names for EIS data follow the format {Z_real, Z_imag}@{10^x_Hz}, where 

the first part of the name can represent the real part of impedance (i.e. resistance) or the imaginary part of the impedance (i.e. reactance),  

the last part is the corresponding frequency, $10^x$ Hz, for $x = -2, -1.9, ... 3.9, 4$.

The final dataset after this processing has the following properties :

26 cells

4297 experiments

133 variables (incl. 122 for EIS)