The cell-fidelity programme forecasts how much lithium a battery design loses to plating over 100 cycles from its first ten, so an electrode design search can skip the other 90 cycles of the full DFN model. The forecast needs one kinetic exponent per cell type, and the last milestone asked whether that exponent is just the textbook αF/RT, rescaled by how fast the negative electrode's potential rises at the separator. That worked on two of PyBaMM's four parameter sets. On Ai2020 it was three times too high.
A question that was already answered
This milestone first asked where Ai2020's lithium goes. An earlier step (5z) had noted that only 3% of its loss was booked to plating or SEI. A lithium ledger answers that cleanly. Over ten cycles on eight designs, the charge PyBaMM's own plating current passes equals the lithium that leaves the particles, to 0.01%. The booked figure is exactly 1/34 of it, because PyBaMM 26.10.1.1 books side reactions for one electrode pair while Ai2020 models a pouch of 34 in parallel. A doubled mesh moves the loss by 0.1%.
That answer was not new. Milestone 5p had found and documented the factor of 34, and 5za had confirmed it on 128 runs and corrected 5z. This run's plan was written from the newest log entries and missed both. The ledger is logged as a replication by a second route (integrated current against booked totals), not as a finding. The lesson for the next run: search the whole log before writing a plan.
What the separator node gives back
The question actually left open was why Ai2020's steep surface potential does not slow its plating as much as the lever says. 5z's committed records split the cycle-2-to-99 change of the plating exponent at the separator node into terms. As shares of the open-circuit-potential term:
| set | intercalation overpotential | charge-profile shape | what is left |
|---|---|---|---|
Ai2020 | 0.21 | 0.23 | 0.59 |
Chen2020 (LG M50) | 0.20 | 0.02 | 0.82 |
OKane2022 (LG M50) | 0.24 | 0.04 | 0.80 |
NCA_Kim2011 | −0.01 | −0.22 | 1.19 |
On Ai2020 the plating spreads over more of the charge, so the end of charge over-reads how fast it slows, and the intercalation overpotential grows. Rescaling the lever by what is left takes Ai2020's implied exponent from 73 to 43, against the 21 its forecast needs. That is half the gap. But the same rescaling takes Chen2020 from 33 to 27 (it needs 52), OKane2022 from 37 to 29 (needs 38) and NCA_Kim2011 from 23 to 27 (needs 23). One node's terms cannot be what sets the exponent.
What it means
Ai2020 is a plating cell like the other three, and every forecast fitted on it since milestone 5u was on a consistent scale. The exponent each set needs still has no physical reading. The next step reads the plating where it happens: the plating current's change split through the electrode's thickness, not at the separator alone.
Caveats: the term split is post hoc, on 5z's four pools rather than the 24 pools 5zi used, and it compares cycle 2 with cycle 99. The refined mesh check ran 3 cycles because the planned 10-cycle run at tight tolerances was killed for memory. Everything is modelled, in one PyBaMM version. Numbers: terms.txt and results.txt; plan and outcome in the log.