Skip to content

validation_chen2020.md

Latest commit

 

History

History
76 lines (61 loc) · 3.99 KB

validation_chen2020.md

File metadata and controls

76 lines (61 loc) · 3.99 KB
Raw

DFN validation against Chen2020 experimental data

Validation of the PyBaMM DFN against real LG M50 discharge data (Zenodo 4032561), four C-rates, isothermal 25 C, literature parameters (uncalibrated). Reproduce with python scripts/validate_chen2020.py.

Setup

  • Model: Doyle-Fuller-Newman, PyBaMM 26.6.0
  • Parameters: Chen2020 (literature, not fitted to these cells)
  • Mesh: 32 points/domain (within 2.6 mV of the 48-pt mesh; estimated absolute discretisation error ~3.7 mV, far below the physics error budget)
  • Initial state: 100% SOC, discharge to 2.5 V
  • Reference: cell02, discharges at C/10, C/2, 1C, 1.5C
  • Thermal: isothermal at 25 C (no thermal coupling)

Results

Rate Cap err Start-V err Scatter (DoD) Scatter (raw) Cell T mean / rise
C/10 +2.7% +7 mV 15 mV 32 mV 27 C / +2 C
C/2 +2.9% +2 mV 16 mV 31 mV 31 C / +8 C
1C +1.3% +28 mV 17 mV 19 mV 38 C / +17 C
1.5C -0.3% +71 mV 18 mV 19 mV 46 C / +28 C
  • Scatter (DoD): std of the voltage residual after normalising each curve to its own capacity (this forces the endpoints to coincide, so it removes the capacity error from the residual) and removing the mean bias.
  • Scatter (raw): the same residual on the native capacity axis, with the capacity offset kept in. This is the honest uncalibrated-model number.
  • Start-V err: v_sim[0] - v_exp[0]. NOT a clean IR/resistance: it carries the OCV/SOC reference offset and is single-sample noisy (the non-monotonic 7->2->28 ->71 sequence is the tell). Do not read it as ohmic resistance.
  • Cell T: the experiment self-heats; at 1C/1.5C it runs 17-28 C above the chamber, so the isothermal-25 C comparison is thermally confounded at those rates.

What this says

  1. The DFN shape is in the right ballpark, strongest at low rate. On the honest raw-capacity axis the residual is ~31 mV at C/10-C/2 and ~19 mV at the higher rates. The DoD-normalised ~15-18 mV is smaller because it removes the ~3% capacity offset; both are reported above so the choice is visible. This is directional evidence that the physics shape is reasonable. It is not a fitted-model comparison, so it should not be equated with the literature's best-fit (~18 mV), which is a calibrated number.

  2. The dominant error is a ~40 mV near-constant offset (hypothesis: largely calibratable). The mean bias is ~38-44 mV across rates. It is consistent with an OCV/reference plus SOC-alignment mismatch (the model pins SOC=1 to the 4.2 V cutoff OCV, while the cell's relaxed full-charge OCV is ~4.185 V), and it is not flat within a discharge (small near full charge, ~+60 mV mid-plateau). Whether it collapses under calibration is the test the planned PyBOP fit will settle; until then "calibratable" is a hypothesis, not a result.

  3. Capacity is within ~3% uncalibrated, shrinking at higher rate.

  4. High-rate comparison is temperature-confounded. At 1C/1.5C the real cell self-heats 17-28 C above the chamber while the model is isothermal at 25 C, so the high-rate residual mixes shape error with an uncontrolled temperature delta. The growing start-V error (to +71 mV) is consistent with this. Adding the thermal model is the fix before those rates are quoted.

Trust tier

Directional / tier 3 (uncalibrated literature parameters). This establishes that PyBaMM's DFN machinery is sound and the physics shape is reasonable. It is not design-guidance grade and makes no claim about IBC cells; calibration on a manufacturer's own data is what would convert directional into design-guidance.

Next

  • Add the lumped thermal model, feed the measured chamber temperature, and re-run 1C/1.5C before those numbers are shown externally.
  • Calibrate (PyBOP) against cell02 and test whether the ~40 mV bias collapses.
  • Extend to cells 03/04 (cell-to-cell spread) and build the C-rate x temperature validation matrix.