BIG LEAP Project: RU1L Model Results

These four CSV files contain cell-level **State of Health (SOH)** reference values and model predictions for the **Remaining Useful First Life (RU1L)** estimator developed in BIG LEAP (Grant Agreement 101137815). The approach is described in deliverable **D3.2**, particularly Sections **2.1.5** (baseline) and **2.1.7** (transfer learning). RU1L is inferred from the predicted SOH degradation trajectory rather than provided as a direct output column. ## Files | File | Description | Records | Cell/test IDs || --- | --- | ---: | ---: || `Baseline_training_cells_data.csv` | Baseline model predictions for the University of Michigan **NMC** cells selected for training (including the pool used for internal training/validation). | 1,031 | 15 || `Baseline_testing_cells_data.csv` | Baseline model predictions on **unseen** University of Michigan NMC testing cells. These correspond to the selected **Data Case B** baseline model in D3.2 Section 2.1.5. | 427 | 6 || `TL_NMC_testing_cells_data.csv` | Predictions on unseen **SANDIA NMC** cells after transfer learning from the NMC baseline to a different NMC cell reference (**TL5**, selected in D3.2). | 138 | 8 || `TL_LFP_testing_cells_data.csv` | Predictions on unseen **SANDIA LFP** cells after transfer learning across chemistries, from NMC to LFP (**TL4**, selected in D3.2). | 187 | 7 | ## File format and column definitions The files are **semicolon-delimited (`;`)** and include an initial, unnamed row-index column. Each data row corresponds to one point along the degradation history of a cell/test, identified by `id_cell` (IDs are specific to each source dataset). The following columns are shared across all four files: - `id_cell`: cell/degradation-test identifier.- `FEC`: cumulative **Full Equivalent Cycles** since beginning of life.- `tabs_day`: elapsed time since beginning of life, in **days**.- `T_*_perc`: fractions of elapsed time spent in temperature ranges **<10, 10-20, 20-30, 30-45 and >45 deg C**.- `Cdc_*_perc`, `C0_perc`, `Cch_*_perc`: fractions of elapsed time in **discharge C-rate**, **zero-current** and **charge C-rate** ranges (0-1C, 1-2C and >2C for charge/discharge).- `soc_*_perc`: fractions of elapsed time spent in SOC ranges **0-10, 10-30, 30-50, 50-70, 70-90 and 90-100%**.- `SOH`: reference **State of Health**, in percent; values between measured check-ups may be interpolated as explained in D3.2 Section 2.1.4.- `SOH_pred`: model-predicted **State of Health**, in percent. The `*_perc` fields are stored as **fractions (approximately 0 to 1)**, not 0-100 percentages. The baseline training file additionally contains `Timestamp`, `id_test`, and `*_avg` auxiliary per-window fields. Some `*_avg` entries use `-1000` as an apparent missing/not-applicable placeholder; these should not be interpreted as physical measurements. These additional fields are not present in the three testing files. ## How to interpret the results Compare `SOH_pred` with `SOH` within each `id_cell`, preferably plotting both against `FEC` to inspect the fitted degradation trajectory. The baseline was trained using the **University of Michigan NMC** dataset; the transfer-learning experiments adapt that baseline to **SANDIA NMC** (same chemistry) or **SANDIA LFP** (different chemistry). D3.2 identifies **TL5** as the preferred NMC adaptation and **TL4** as the preferred LFP adaptation. The CSVs contain **SOH results only**: they do not include model weights, a direct RU1L column, a separate full-retraining benchmark prediction, or labels identifying the internal 80/20 training-validation split. In D3.2, RU1L is obtained by identifying when the predicted SOH reaches an end-of-first-life threshold (nominally **80% SOH**). Reported performance metrics in the deliverable should be referenced with their documented evaluation procedure rather than assumed to equal a simple average over every CSV row. ## Source dataset references The four CSV files contain **processed model results** derived from experimental cycling data, rather than the original raw datasets. The source datasets and associated publications are: 1. **University of Michigan (UofM) - Baseline, NMC:** P. Mohtat, J. B. Siegel, A. G. Stefanopoulou, and S. Lee, *UofM Pouch Cell Voltage and Expansion Cyclic Aging Dataset*, University of Michigan, Deep Blue Data. **Dataset:** [https://doi.org/10.7302/7tw1-kc35](https://doi.org/10.7302/7tw1-kc35) (also [Deep Blue Data](https://deepblue.lib.umich.edu/data/concern/data_sets/5d86p0488)). Cited in BIG LEAP D3.2 as reference **[17]**. 2. **Sandia National Laboratories - Transfer learning, NMC and LFP:** Y. Preger et al., “Degradation of Commercial Lithium-Ion Cells as a Function of Chemistry and Cycling Conditions,” *Journal of The Electrochemical Society*, vol. 167, no. 12, article 120532, 2020. **Publication:** [https://doi.org/10.1149/1945-7111/abae37](https://doi.org/10.1149/1945-7111/abae37). **Experimental data repository:** [Battery Archive](https://www.batteryarchive.org/). Cited in BIG LEAP D3.2 as reference **[2]**. **BIG LEAP deliverable reference:** BIG LEAP, *D3.2 - Development of data-driven SoX and RUL algorithms for increase BMS interoperability (1st period report)*, Section 2.1 (especially Sections 2.1.3-2.1.7), 30 June 2025.