Molicel P45B Electrochemical Model: Open-Source DFN Parameter Set
Updated: 8 hours ago
About:Energy believes battery modelling is held back by how much of it stays hidden. Most modelling vendors treat their methodology as proprietary, leaving engineers to choose models on trust rather than evidence. We publish our testing methods, model structure, assumptions and limitations, and the data behind them, so anyone can check our working. This paper is the second in that series. Having already released an industry-standard equivalent circuit model for the Molicel P45B, we are now releasing the physics-based model for the same cell, parameterised from a full teardown and validated against the same open dataset.
Molicel P45B Physics-Based Battery Model
About:Energy has published an open parameterisation of the Molicel P45B, a 4.5 Ah high-power 21700 cell used across eVTOL, motorsport and drone platforms. The paper describes the cell teardown and physical content, the characterisation measurements, the model mathematics and parameterisation approach, and the validation against full cell data released previously on Zenodo. Parameters follow the Battery Parameter eXchange (BPX) standard, an open JSON format, so they load into DFN or SPMe models in simulators such as PyBaMM in a few lines of Python. Most published parameter sets still describe cells designed in the 2010s. This one covers a contemporary silicon-containing power cell. Full paper here.
Molicel P45B Resolving Silicon and Graphite in the Negative Electrode
The P45B negative electrode is a blend, and more than 20% of usable capacity comes from the SiOx content. Lumping that into a single open-circuit potential hides where the capacity is coming from and mis-predicts voltage at low state of charge. Instead, the model resolves graphite and silicon as parallel contributors, each with its own open-circuit potential, kinetics and transport properties, and gives silicon a one-state hysteresis model adapted from the Plett and Wycisk formulation. That improves voltage tracking under both continuous and pulse conditions, and shows when each material is doing the work: under high power, silicon becomes active at higher state of charge than you would expect from the quasi-equilibrium picture.
Figure: Charge response against experiment at 1C and 3C, with the predicted negative electrode potential from the open parameter set
Molicel P45B Validation Against Realistic Data
Because the validation dataset was already published, anyone can check the result rather than take it on trust, and can compare the physics-based model against the equivalent circuit model for the same cell on the same profiles. Voltage error stays below 20 mV RMSE across many industry-relevant validation cases, including aviation and automotive duty profiles. The paper is also explicit about the trade-offs: where teardown-led parameterisation needs tuning afterwards, how diffusivity was parameterised, and what that means for anyone repeating the approach on their own cell.
Figure: Model against experiment on an eVTOL mission and a WLTP drive cycle, with voltage error on the same time base
References
Validation dataset: https://zenodo.org/records/19052626
Companion ECM post: https://www.aboutenergy.io/post/molicel-p45b-battery-model-ecm-open-source-validation
BPX and PyBaMM: [Coming Soon]
FAQs
What is a DFN battery model?
The Doyle-Fuller-Newman model is a physics-based battery model that resolves lithium transport through the electrodes and electrolyte rather than fitting terminal behaviour empirically. It predicts internal states such as electrode potentials, concentration gradients and lithium plating risk, which makes it the standard choice for fast-charge design, cell selection and understanding why a cell behaves the way it does.
What is the difference between a DFN model and an SPMe?
The single particle model with electrolyte is a reduced form of the DFN. It assumes uniform reaction distribution through the electrode thickness, which cuts solve time substantially but loses accuracy at high C-rate where that assumption breaks down. For high-power cells under aggressive duty, the full DFN is the safer starting point. The same BPX parameter set runs in both.
What is the BPX format?
Battery Parameter eXchange is an open JSON standard for sharing physics-based battery model parameters. It fixes the model structure and units so a parameter set can move between simulators without hand translation. The Molicel P45B set is published in BPX and loads into PyBaMM in a few lines of Python.
How do I load a BPX parameter set into PyBaMM?
Install the BPX package, read the JSON file, and pass the result into a PyBaMM parameter values object. From there it drives a DFN or SPMe model directly. No manual mapping of parameter names is needed, which is the point of the standard.
How is a physics-based battery model parameterised?
Through a combination of cell teardown and electrical characterisation. Teardown gives the physical content: electrode thicknesses, porosities, active material loadings and the bill of materials. Characterisation gives the electrochemical behaviour: open-circuit potentials, kinetics and transport. Teardown-led parameterisation still needs tuning afterwards, and the P45B paper is explicit about where and why.
Can a DFN model handle a silicon-graphite blended electrode?
Yes, but not if silicon and graphite are lumped into a single open-circuit potential. In the P45B the silicon oxide content contributes more than 20% of usable capacity, and lumping it mis-predicts voltage at low state of charge. Resolving the two materials in parallel, each with its own potential, kinetics and transport, tracks voltage correctly under both continuous and pulse loads.
How accurate is a DFN model?
The P45B parameter set holds voltage error below 20 mV RMSE across a broad set of validation cases, including aviation mission profiles and automotive drive cycles. Accuracy depends heavily on parameterisation quality and on whether the validation covers the conditions you actually operate at, so always check what a published model was tested against.
Should I use a DFN model or an equivalent circuit model?
Use an equivalent circuit model when you need speed at pack and system level, real-time control, or long duty-cycle simulations. Use a DFN when you need internal states: plating margin during fast charge, the effect of electrode design, or an explanation for behaviour an equivalent circuit model cannot reproduce. Many teams run both, parameterised from the same cell.
Where can I find open-source lithium-ion parameter sets?
Most published sets describe cells designed in the 2010s, which limits their usefulness for modern high-power designs. The Molicel P45B parameter set, the validation data and the full methodology are on Zenodo under CC-BY 4.0, alongside the peer-reviewed paper, so results can be reproduced rather than taken on trust.
