Battery Modeling for AI

The choices that decide a PyBaMM fade curve never raise an error: OKane2022’s SEI grows with time alone, so rests change cycle count but not the end-of-life date; the SEI rate constants are generic placeholders that differ 450-fold between options; activation energy is zero in every other set; a cold plating clog is overstated 5-fold on the default mesh; and ‘irreversible’ plating deposits lithium at rest. Measured in PyBaMM 26.5, with a pre-flight check.

PyBaMM has several different things called capacity, and degradation studies quietly mix them: termination=”80% capacity” compares a zero-current eSOH capacity, not a measured discharge; summary variables are read at the end of each cycle; ‘Loss of capacity to X’ is lithium, not capacity; and the default parameter set gains capacity as it loses lithium. Measured in PyBaMM 26.5, with an audit function.

PyBaMM’s thermal options run without complaint, and each of these set-ups returns a plausible temperature curve: isothermal heating variables are zero by construction, four parameter sets have no reversible heat, four describe a single electrode sheet cooled on both faces, and a cold-start run that sets only the ambient temperature begins at 25 °C. Measured on eight parameter sets, with an audit function.

PyBaMM’s Experiment API almost never raises, and each of these five mistakes returns a normal-looking solution: steps in amps stop silently at 24 hours, a flat step list overstates cycle life fourfold, save_at_cycles leaves None in the list, period never changes PyBaMM’s answers but can break yours, and initial_soc=1 is not 4.2 V. Includes a completion check.

The same experiments run on all 18 built-in parameter sets in PyBaMM 26.5. The sets that error are not the problem; the ones that finish and are wrong are. “1C” on the default set is really 0.78C, Chen2020 loses 0.6% of its capacity at −10°C, an experiment string charges an LFP cell to 4.2 V with zero warnings, and SEI runs on borrowed example parameters. Includes a check you can run.

The top search result, pybamm-param, is deprecated; PyBOP just restructured its API; and identifiability means an optimiser handed fifty free parameters returns fifty meaningless numbers. Includes a protocol separating a fit from a decoration.

When a long degradation run dies at cycle 300, reaching for tolerances is usually wrong. Tell numerics from physics with an SPMe cross-check, plus every IDAKLU option default and one trap that fails silently.