betaFeedback

WorksEngineering

Parameterisation of NMC and LFP 4680 cells for electrochemical and thermal modelling

Abstract

An electrochemical-thermal model of the 4680 cylindrical cell is presented: a Doyle-Fuller-Newman model of the electrode stack coupled to a two-node thermal balance of the roll and the can (a thermal DFN, TDFN), solved by an implicit finite-volume scheme. Two cells are parameterised from teardown geometry and open literature, with every value tagged by provenance: the Tesla NMC811/graphite tabless cell and the BYD FC4680 LiFePO4 multi-tab cell. The reaction heat is evaluated at the open-circuit potential of the particle-mean concentration, which restores the heat of mixing. Both sets are calibrated in the same way, with the test fixture identified from the measurement alone. Against the 1C and 2C charge test of Ank et al., the calibrated NMC set follows the mean can temperature within 0.52 and 1.46 K, with a contact heat of 0.94 mOhm matching the authors' estimate for their holder. For the LFP cell, 1C and 2C discharges to 2.0 V were measured on the authors' bench; the fixture conductance (0.337 W/K) follows from the energy balance and the heat capacity (277 J/K) from the cooling. With the end-of-lithiation drop of LiFePO4 restored in the cathode open-circuit potential, the calibrated set follows the can temperature within 0.44 K (1C) and 0.76 K (2C), reproduces the peaks and the relaxation after cut-off. What remains is what a single can node and a single representative particle do not describe: the axial gradient of the NMC can under contact heating, and a gradual decline of the LFP voltage before the knee at 2C, which points to the distribution of particle states in the electrode. The parameter sets (BPX, CSV) and the bench data are openly available.

Nomenclature

SymbolUnitDescriptionSymbolUnitDescription
asa_sm−1\mathrm{m}^{-1}interfacial area per volumeQQA hcharge passed
cec_emol m−3\mathrm{mol\,m}^{-3}electrolyte concentrationQs,r,Qs,cQ_{s,r}, Q_{s,c}Wseries-resistance heat in roll, can
cˉs,cse\bar{c}_s, c_{s\mathrm{e}}mol m−3\mathrm{mol\,m}^{-3}solid: particle mean, surfacersr_smparticle radius
cs,max⁡c_{s,\max}mol m−3\mathrm{mol\,m}^{-3}maximum solid concentrationri,ror_i, r_omroll core and outer radius
cpc_pJ kg−1 K−1\mathrm{J\,kg}^{-1}\,\mathrm{K}^{-1}specific heatRintR_{\mathrm{int}}K W−1\mathrm{K\,W}^{-1}thermal resistance roll–can
Cr,CcC_r, C_cJ K−1\mathrm{J\,K}^{-1}heat capacity of roll, rest of cellRseriesR_{\mathrm{series}}Ω\Omegaseries resistance
DeeffD_{\mathrm{e}}^{\mathrm{eff}}m2 s−1\mathrm{m}^2\,\mathrm{s}^{-1}effective electrolyte diffusivityttstime
DsD_sm2 s−1\mathrm{m}^2\,\mathrm{s}^{-1}solid diffusivityt+0t_+^0–transference number
EaE_aJ mol−1\mathrm{J\,mol}^{-1}activation energyTr,Tc,T∞T_r, T_c, T_\inftyKroll, can, ambient temperature
FFC mol−1\mathrm{C\,mol}^{-1}Faraday constantUUVopen-circuit potential
hhW m−2 K−1\mathrm{W\,m}^{-2}\,\mathrm{K}^{-1}can-surface heat-transfer coefficientVVVterminal voltage
hAhAW K−1\mathrm{W\,K}^{-1}can–ambient conductancew,nw,n–end-of-lithiation drop: width, exponent
HeH_emelectrode heightxxmcoordinate through the stack
ini_nA m−2\mathrm{A\,m}^{-2}interfacial current densityεe,εs\varepsilon_e,\varepsilon_s–electrolyte, solid volume fraction
IIAapplied currentθ,θ0\theta,\theta_0–stoichiometry; at 100 % SOC
jLij^{\mathrm{Li}}A m−3\mathrm{A\,m}^{-3}volumetric reaction currentθmin⁡,θmax⁡\theta_{\min},\theta_{\max}–stoichiometry at 0 % SOC
JJ–calibration costκeeff,κDeff\kappa_{\mathrm{e}}^{\mathrm{eff}},\kappa_{\mathrm{D}}^{\mathrm{eff}}S m−1,A m−1\mathrm{S\,m}^{-1},\mathrm{A\,m}^{-1}electrolyte, diffusional conductivity
L,pL,pmwinding length; layer pitchλ,λr\lambda,\lambda_rW m−1 K−1\mathrm{W\,m}^{-1}\,\mathrm{K}^{-1}thermal conductivity; of the roll
L±,LsepL_\pm,L_{\mathrm{sep}}mcoating, separator thicknessρ\rhokg m−3\mathrm{kg\,m}^{-3}density
mrefm_{\mathrm{ref}}see Table 1reaction rate constantσeff\sigma^{\mathrm{eff}}S m−1\mathrm{S\,m}^{-1}effective solid conductivity
NNmol m−2 s−1\mathrm{mol\,m}^{-2}\,\mathrm{s}^{-1}molar flux at particle surfaceτc\tau_cscathode diffusion time rs2/Dsr_s^2/D_s
qqW m−3\mathrm{W\,m}^{-3}volumetric heat sourceφs,φe\varphi_s,\varphi_eVsolid, electrolyte potential
qˉ\bar{q}mol m−4\mathrm{mol\,m}^{-4}mean concentration gradient in particleΔT,ΔQ,ΔV\Delta T,\Delta Q,\Delta VK, A h, Vmodel minus measurement

Table 5.

AbbreviationMeaningAbbreviationMeaning
BPXBattery Parameter eXchange formatNMCLiNixMnyCozO2\mathrm{LiNi_xMn_yCo_zO_2} (NMC811)
DFNDoyle–Fuller–Newman modelOCPopen-circuit potential
LFPLiFePO4\mathrm{LiFePO_4}SOCstate of charge
TDFNthermal DFN[M], [L], [D], [E]provenance tags (§ 3.5)

Table 6.

Introduction

Physics-based models of lithium-ion cells, from the pseudo-two-dimensional model of Doyle, Fuller and Newman (DFN) ([15, 23]) to multi-scale thermal and electrochemical formulations ([25]), are only as useful as their parameters. A DFN model of a commercial cell needs several tens of geometric, transport, kinetic, thermodynamic and thermal parameters, and most of them cannot be measured on the assembled cell. They are obtained from teardown and material characterisation ([19, 38, 13, 33]), transferred from the literature ([46]) or estimated from current and voltage data ([4]), and the resulting set is then validated against the voltage and temperature of the cell ([18]). On the thermal side, lumped models with a core and a surface temperature are established for cylindrical cells ([20, 29]).

Two further strands bear on what such a set can be tested against. The heat generated in a cell follows from an energy balance [9] whose local form contains a heat of mixing [35, 43]; Richardson and Korotkin [36] derive the exact conservation law for the DFN model and note that thermal DFN codes commonly omit that term. And terminal measurements determine only some parameter combinations: the solid-diffusion parameters enter as rs2/Dsr_s^2/D_s, and which groups a given set of pulses, capacities and curves can pin is a question of identifiability [21, 4]. For the 4680 format the measurements are few. Ank et al. [5] publish a teardown with open characterisation data, and Allgäuer et al. [3] report the thermal behaviour of a 4680 cell in cooled fixtures, including the contact resistance of the cell holder.

Parameter sets are increasingly exchanged between simulation codes in standard formats such as the Battery Parameter eXchange [42]. This makes a set easier to reuse, and also easier to reuse without knowing where its values came from. Ward et al. [47] describe the general problem for battery data in the Principles of the Battery Data Genome. In their layered metadata model, simulation model parameters belong to the tertiary tier, and their quality principle requires that all data and metadata be “cleaned, curated, and quality controlled” before they are shared. For measured time series, existing tools can do this. For a parameter set, in which most values are not measured directly, it is less clear what quality control means.

A cell-level electrochemical–thermal model needs more than the electrochemical parameters of a DFN: it needs the geometry of the winding, the material inventory, the current collection and the thermal network of the cell, because these set the capacity, the series resistance and the heat path. No parameter set of this scope has been published for a 4680 cell. Electrochemical parameters have been published for the NMC811/graphite couple in the 2170 format [13], and teardown studies are now available for a Tesla 4680 cell with an NMC811/graphite couple [5] and for the BYD 4680 cell with an LFP/graphite couple [30], which allows both cells to be parameterised on the same basis. This paper has three aims.

First, to identify complete parameter sets for both cells with traceable provenance. Every value carries a tag that separates measured data, literature values transferred from the same chemistry, quantities derived in this work and our own estimates. The coupled model needs the same 78 parameters for both cells, so they are given in one set of two-column tables with one row per parameter, and rows whose status is unresolved are marked in red. The Supplementary Material gives the complete parameterisation of both cells, with the source and the reasoning for each value.

Second, to check each set against quantities that were not used to build it. The mass of the assembled cell, its delivered capacity and the length of the unwound electrode are compared with the values implied by the parameter set, and the residuals are reported. The mass residual is +6.5% for the NMC cell and −2.5% for the LFP cell. The opposite signs point to hardware that the layer model of the NMC cell does not contain, rather than to an error common to both identifications (§ 3.6).

Third, to calibrate the parameter sets against measurement. The model is a DFN of the wound cell coupled to its temperature field. Both sets are taken through the same procedure: the transferred set is compared with a measurement, the fixture is identified from the measurement alone, and the effective parameters are calibrated on the voltage and the can temperature at 1C and 2C. For the NMC cell the measurement is the open charge test of Ank et al. [5], with the heat-transfer coefficient of their fixture taken from the measured cooling (§ 6). For the LFP cell, for which no comparable measurements are published, the discharge behaviour was predicted before a bench test and then measured at 1C and 2C, and the set is calibrated on that test (§ 6.2).

The DFN describes solid diffusion by Fick’s law in a single spherical particle. This is appropriate for NMC811 but not for the two-phase transformation of LiFePO4_4, so the solid diffusivity of the LFP cathode is reported as an effective parameter (§ 3.3). The bench test shows that the end of a fast discharge is set by the end-of-lithiation potential of the cathode, which lies outside the window of the transferred fit, and where the single-particle description still falls short (Section 7).

The contributions of this paper are:

  1. a DFN model of a 4680 cell coupled to a two-node thermal balance whose heat capacities and internal resistance follow from the parameter set, with the fixture boundary taken from measurement;

  2. two provenance-tagged parameter sets for real 4680 cells, a tabless NMC cell and a multi-tab LFP cell, built from teardown data;

  3. the completion of the heat source: a virtual experiment showing that the printed source of Gu and Wang (2000) under-predicts the heat when solid diffusion is slow, and the evaluation of the reaction heat at the open-circuit potential of the particle-mean concentration, consistent with Rao and Newman [35], Thomas and Newman [43] and Richardson and Korotkin [36];

  4. a calibration procedure, the same for both cells, that separates artefacts of the measurement from the model parameters and ends in a Nelder–Mead minimisation of a normalised cost on the voltage and the can temperature at 1C and 2C;

  5. the calibration of the NMC set on the open charge test of Ank et al. [5], with a measured boundary condition, which identifies the contact heat of the holder and restores the transferred cathode diffusivity that a resistance-based heat value had appeared to exclude;

  6. a bench test of the LFP cell at 1C and 2C with the fixture conductance and heat capacity obtained from the measurement, the calibration of the LFP set on it, and the finding that the end of a fast discharge, where the cell is hottest, depends on the end-of-lithiation potential of the cathode that a transferred fit does not cover;

  7. a statement of what remains after calibration, for both cells, and of what the open data can and cannot identify.

The two cells

The 4680 format

The 4680 cylindrical format, 46 mm in diameter and 80 mm in height, was introduced by Tesla in 2020 ([41]). Compared with the 2170 cell it succeeds, Tesla stated a fivefold increase in energy and a sixfold increase in power per cell, a 16% increase in vehicle range and a 14% reduction in cost per kilowatt-hour from the format alone ([41]). A larger cell means fewer cells per pack, and therefore fewer cans, welds and connections per kilowatt-hour. The larger diameter was made possible by tabless current collection. In a conventional cylindrical cell the current leaves each electrode through one or a few tabs, so the electronic path and the ohmic heat grow with the length of the electrode. In the tabless design one uncoated edge of each foil is welded to an end plate along the whole winding, and the current path is reduced to the width of the electrode ([41]; [44]). Tesla presented three cathode families for the format: iron-based for long cycle life, nickel and manganese based, and high nickel for long range [41]. The first-generation cell characterised by Ank et al. [5] uses NMC811.

The 46 mm diameter has since been adopted by other cell and vehicle manufacturers. BMW has adopted 46 mm cells of 95 mm and 120 mm height, supplied by CATL and EVE, for its Neue Klasse vehicles [11]. LG Energy Solution supplies 4695 cells to Rivian [27], and Samsung SDI started production of 46-series cells in 2025 [37]. The manufacturers keep the diameter and vary the height: a taller cell stores more energy without lengthening the paths for current and heat in the radial direction, and for tabless cells both the gravimetric and the volumetric energy density increase with height [34].

Heat removal is the main constraint on the diameter. The surface-to-volume ratio of a 4680 cell is about half that of a 2170 cell, 0.11 against 0.22 mm−1\mathrm{mm}^{-1}, and its electrode is almost five times longer (Li et al., 2023). For a given layer pitch, the electrode length grows with the square of the diameter [7]. Heat generated near the core has to be conducted either radially through the wound layers, whose through-plane conductivity is about two orders of magnitude below the axial conductivity of wound cells [17], or axially along the foils to the end faces. Tabless collection helps in both respects: it reduces the ohmic heat and provides an axial path from the roll to the end plates, which makes cooling through the base of the cell effective (Tranter et al., 2020; Li et al., 2023). A larger diameter would again lengthen the radial path from the core to the cooled surfaces [44], and further increases in energy per cell have instead been obtained by increasing the height.

The format does not prescribe the chemistry. BYD produces a 4680 cell with an LFP cathode for both vehicle and stationary storage applications, with safety and cost as the main motivations [30]. LFP has a lower specific energy than NMC811 but contains no nickel or cobalt. The BYD cell reaches 150.5 W h kg−1^{-1} and 374.6 W h L−1^{-1} [30], against 232.5 W h kg−1^{-1} and 622.4 W h L−1^{-1} for the Tesla NMC811 cell [5]. It also uses a different current collection, with several tabs on each electrode instead of a tabless design.

The two cells studied

The first cell is a Tesla 4680 cell with an NMC811/graphite couple, a liquid carbonate electrolyte and tabless current collection. The teardown found no silicon in the anode [5]. Each foil is left uncoated along one edge and welded to an end disc along the whole winding length, so that the current leaves the electrode sideways over tens of millimetres instead of travelling along its length. The jelly roll is 46 mm in diameter, the electrode is approximately 3.3 m long and 70 mm wide, and the capacity measured at C/3C/3 is 22.078 Ah (σ=0.186\sigma= 0.186 Ah) [5]. The DFN model describes this chemistry without qualification.

The second cell is the BYD FC4680, 46.3 × 80.2 mm and 332 g, with an LFP/graphite couple and a nominal voltage of 3.20 V, against 3.6 V for the NMC cell. It stores 15.41 Ah where the NMC cell stores 22.4 Ah, about 70 % of the charge in the same volume. Its current collection is multi-tab, with four aluminium tabs on the cathode and three nickel tabs on the anode, one of which is not connected [30]. Its geometry, inventory, thermal network and electrochemistry are identified in the same way as those of the NMC cell, with the qualification given in § 3.3.

Both cells are large, so neither the current nor the temperature can be assumed uniform, and both collect current over a distributed interface rather than at a point, so the collection topology has to be modelled. § 5.4 shows that it changes the collection resistance by a factor of forty.

What transfers between chemistries

Geometry, material inventory and the thermal and electrical network are described by the same physics in both cells. A layer stack has a thickness, a density and a thermal conductivity whatever its active material, the resistance of a can wall depends only on the steel and the dimensions, and a collector disk conducts heat to the cap in the same way in both cells. These quantities are identified and compared directly. The paths of current and heat differ, however. The model resolves one repeating unit through its thickness and takes the current as uniform along the winding. This is close to the tabless cell, whose current leaves each foil over its whole edge, and it is an approximation for the multi-tab cell, whose current converges on the tabs; the gradients along the winding are then represented only through the winding-averaged collection resistance (§ 5.4) and are not resolved.

The electrochemistry requires one qualification. LiFePO4\mathrm{LiFePO}_{4} discharges through a two-phase transformation at nearly constant potential, and the moving phase boundary inside each particle is not described by the Fickian diffusion in a single spherical particle that Equation (7) assumes. The solid diffusivity of the LFP cathode is therefore reported as an effective parameter that absorbs the phase-boundary kinetics into a Fickian form, and no physical meaning is attached to it. With this qualification the DFN needs the same parameters for both cells, and both sets are given in full.

Three limitations follow for the LFP cell, and each is marked in the tables. The solid diffusivity is effective, as described above. The model does not represent the voltage hysteresis between charge and discharge, which is of the order of tens of millivolts in LFP cells; test T2 of the bench test measures it. Finally, About:Energy Limited [2] report that their parameter set loses accuracy in voltage and capacity at high rate and low state of charge, which is the region probed by the rate-retention data of Liu et al. [30]; the bench test finds the largest deviation in that region (§ 6.2). Models that represent the cathode as an ensemble of phase-separating particles reproduce the hysteresis (Foster et al., [22]; Clarke et al., [14]), and are outside the scope of this work.

Geometry and inventory of both cells

Table 1Table 1 gives the geometry of both cells, with each value carrying the tag it has in its own parameter set, and Supplementary tbl:S5 compares the two cells in full. The geometry of the LFP cell is taken from the teardown of Liu et al. [30], supplemented by quantities derived in this work where that source gives no value.

QuantityUnitNMCLFPTags
Cell
Winding lengthm3.275.37[M] / [M]
Electrode heightmm70.069.0[M] / [M]
Roll core radiusmm2.503.16[M] / [M]
Roll outer radiusmm22.222.7[M] / [D]
Cu foil thicknessμm\mu\mathrm{m}8.0010.0[M] / [M]
Al foil thicknessμm\mu\mathrm{m}10.010.0[M] / [M]
CapacityA h22.415.5[D] / [D]
N/P ratio, full stoichiometry ranges—0.9291.18[D] / [D]
Hardware series resistancemΩ\Omega0.5000.500[E] / [E]
Holder contact resistance, heat onlymΩ\Omega0.94†1.34†[E] / [E]
Negative electrode (graphite)
Coating thickness, L−L_{-}μm\mu\mathrm{m}12654.5[M] / [M]
Porosity, εe\varepsilon_{\mathrm{e}}—0.2970.319[D] / [D]
Particle radius, rsr_{\mathrm{s}}μm\mu\mathrm{m}5.864.80[L] / [L]
Maximum concentration, cs,maxc_{\mathrm{s,max}}mol m−3^{-3}30 55531 400[L] / [L]
Stoichiometry at 100 % SOC, θ0\theta_{0}—0.700†0.796†[E] / [E]
Stoichiometry at 0 % SOC, θmin⁡\theta_{\min}—0.02740.00163[D] / [L]
Solid diffusivity, DsD_{\mathrm{s}}m2^{2} s−1^{-1}3.30×10−143.30 \times10^{-14}9.60×10−159.60 \times10^{-15}[L] / [L]
Rate constant, mrefm_{\mathrm{ref}}A m2.5^{2.5} mol−1.5^{-1.5}6.48×10−76.48 \times10^{-7}6.68×10−76.68 \times10^{-7}[L] / [L]
Solid conductivity, σ\sigmaS m−1^{-1}2157.46[L] / [L]
Separator
Thickness, LsepL_{\mathrm{sep}}μm\mu\mathrm{m}12.013.7[L] / [D]
Porosity, εe\varepsilon_{\mathrm{e}}—0.4700.470[L] / [L]
Positive electrode (NMC811 / LiFePO4_4)
Coating thickness, L+L_{+}μm\mu\mathrm{m}85.070.0[M] / [D]
Porosity, εe\varepsilon_{\mathrm{e}}—0.3350.331[L] / [D]
Particle radius, rsr_{\mathrm{s}}μm\mu\mathrm{m}5.220.500[L] / [L]
Maximum concentration, cs,maxc_{\mathrm{s,max}}mol m−3^{-3}51 55521 200[L] / [L]
Stoichiometry at 100 % SOC, θ0\theta_{0}—0.2640.0875[E] / [L]
Stoichiometry at 0 % SOC, θmax⁡\theta_{\max}—0.8890.950[D] / [L]
Diffusion time, τc=rs2/Ds\tau_{\mathrm{c}} = r_{\mathrm{s}}^{2}/D_{\mathrm{s}}s8 031†4 740†[L] / [E]
End-of-lithiation drop, width ww / exponent nn——0.131 / 2.6†— / [E]
Rate constant, mrefm_{\mathrm{ref}}A m2.5^{2.5} mol−1.5^{-1.5}3.42×10−63.42 \times10^{-6}1.40×10−71.40 \times10^{-7}[L] / [L]
Solid conductivity, σ\sigmaS m−1^{-1}0.1800.800[L] / [L]
Electrolyte
Initial concentration, ce,0c_{\mathrm{e},0}mol m−3^{-3}1 0001 000[L] / [L]
Transference number, t+0t_{+}^{0}—0.2590.259[L] / [L]

Table 1. Table 1. Geometric and electrochemical parameters of the two cells. Tags as in § 3.5, NMC / LFP. The complete sets, with sources and tolerances, are in Supplementary Sections S3, S7 and S8. Open-circuit potentials: Chen et al. [13] (NMC) and About:Energy Limited [2] (LFP); electrolyte transport: Nyman et al. [32]. Values marked † are calibrated, the NMC set on the open charge test of Ank et al. [5] (§ 6.1) and the LFP set on the bench test (§ 6.2); the values of the transferred sets are 0.701, 700 s and 25 kJ mol−1^{-1} (NMC) and 0.823, 3637 s and 80 kJ mol−1^{-1} (LFP); the end-of-lithiation drop is the term added to the transferred LFP potential in § 6.2.

Three quantities are not reported by any source: the cathode porosity, the can wall thickness and the separator thickness. None of them can be obtained from a single relation. They are determined together from an over-determined system of mass, capacity and volume balances, described in § 4.2. The outer radius of the roll follows from Equation (2) and agrees with the can bore to within 0.99% (§ 3.6).

Provenance of the parameters

The complete parameter sets of both cells are given in Supplementary Sections S7 and S8: all 78 values for each cell, each with its unit, its tag and the reasoning behind it. Every parameter carries one of four tags:

  • [M]: measured or reported for the cell itself, from teardown or manufacturer data (Ank et al., [5]; Liu et al., [30]).

  • [L]: a literature value for the same chemistry, transferred on the assumption that a 4680 NMC811/graphite electrode behaves like the LG M50 electrode characterised by Chen et al. [13], with the electrolyte correlations of Nyman et al. [32].

  • [D]: derived in this work from tagged quantities by a stated relation, and therefore carrying their uncertainty.

  • [E]: estimated by the author, with the reasoning recorded. These values should be treated as fitting parameters.

Of the 78 parameters, 16 are [M], 37 are [L] and 21 are [E]. About a third of each set is therefore estimated, which is to be expected for cells without a published parameter set, and which is visible only because the tags are printed.

The tags also show where the results are sensitive. The heat-transfer coefficient at the can surface is [E] and, in the thermal balance of § 5.3, accounts for 76 % of the thermal resistance between the roll and ambient at h=50 W m−2 K−1h = 50\ \mathrm{W}\,\mathrm{m}^{-2}\,\mathrm{K}^{-1} and 93 % in the test holder of § 6, so the temperature rise scales almost linearly with it. The thermal properties of the two coatings and the separator are also [E]. They set the through-plane conductivity and therefore the radial temperature gradient, so the shape of the temperature profile depends on estimates even where its magnitude is constrained by measurement.

The tables are generated from the solver’s own objects, and a build-time assertion fails if any field of any layer lacks a declared source. The tables, the released metadata and the model therefore remain consistent.

Three choices in the parameter sets require comment.

The stoichiometry windows are calibrated. Transferred with the open-circuit potentials of Chen et al. [13], the stoichiometries at 100 % state of charge let the NMC model discharge well past the capacity of the cell: to 2.5 V it delivered 26.1 A h at C/3, 18 % above the 22.078 Ah measured by Ank et al. [5]. A closure check of this kind is what the parameter set is built to expose. The windows are therefore calibrated to the cut-off voltages, a standard step [10, 38]: with the potentials kept and the electrode capacities fixed by the loadings, the two 100 % stoichiometries are solved so that the open-circuit voltage at 100 % is 4.2 V and a C/20 discharge to 2.5 V delivers 22.31 A h, the C/20 capacity of Ank et al. [5]. This gives θ0=0.7006\theta_0 = 0.7006 for the anode, against 0.9014 transferred, and 0.2638 for the cathode, against 0.2661; both are tagged [E]. The anode capacity entering this step is that of the teardown: the loading of 19.5 mg cm−219.5\ \mathrm{mg}\,\mathrm{cm}^{-2} at 1.55 g cm−31.55\ \mathrm{g}\,\mathrm{cm}^{-3} gives a coating of 126 μm126\ \mu\mathrm{m} per side, tagged [M], and a solid fraction of 0.70; the set carried an estimated 110 μm110\ \mu\mathrm{m} before. Over the full stoichiometry ranges the N/P ratio is then 0.93, an output of the loadings. Two measurements not used in the calibration test it. The C/3 capacity follows as 22.083 A h, against 22.078 Ah measured. The central graphite peak of the differential voltage, which Ank et al. [5] place at about 14 A h of the C/20 charge, falls at 19.6 A h in the model. The measured peak corresponds to an anode stoichiometry of 0.44, whereas the transferred potential places its central peak at 0.61. That potential is of a graphite–SiOx_x electrode, and Ank et al. [5] find no silicon in this anode; the shape of the potential, not the window, is the cause, and it is a finding about the transferred open-circuit potentials on this cell. The capacity agreement is therefore evidence only at C/3.

Two variants of the maximum solid concentration. Two pairs of values are in use for this chemistry, one measured and one obtained by fitting. We use the measured values for everything that determines capacity, and the fitted values for the kinetic group, whose members were fitted together and cannot be separated. This mixed use is marked in the tables.

Phase-consistent layer properties. Density and heat capacity are averaged over the same inventory, with solid and electrolyte each counted once,

ρeff=ρs(1−εe)+ρeεe,(ρcp)eff=ρscp,s(1−εe)+ρecp,eεe,(1)\rho_{\mathrm{eff}} = \rho_s (1-\varepsilon_e) + \rho_e\varepsilon_e,\qquad(\rho c_p)_{\mathrm{eff}} = \rho_s c_{p,s}(1-\varepsilon_e) + \rho_e c_{p,e}\varepsilon_e, \tag*{(1)}

where εe\varepsilon_e is the porosity used in the electrochemical model. Averaging the two quantities over different inventories, for example a solid-only density combined with specific heats that include the electrolyte, changes the thermal mass of the roll by +12.8 %. Such an error does not affect the voltage, appears directly in the temperature rise, and is not detected by a mass balance.

Closure checks: consistency of a parameter set

The quality of a measured time series can be checked by inspection, because a dropped channel or a clock error leaves a visible trace. A parameter set has no equivalent. A single value can be plausible, correctly tagged and still wrong, and the model output it produces gives no independent signal.

A cell that has been torn down has, however, been measured in several redundant ways that the parameter set must satisfy together. A layer stack implies a mass; a coating thickness and an areal capacity imply a charge; a winding pitch and an electrode length imply an outer radius, which must fit inside a can whose bore was measured separately. None of these quantities is used to build the set. Each can be computed from the set and compared with the measurement, and the residual is a measure of quality of the kind that Ward et al. [47] require for their tertiary tier. We report three such checks for each cell. For the LFP cell the residuals are those of the initial estimates; § 4.2 then uses the same three relations to determine the quantities that no source reports, after which they close by construction.

Mass. Summing all layers over the electrode area, with solid and electrolyte each counted once through Equation (1), and adding the can, the collector disks and the insulators gives the mass of the assembled cell. For the NMC cell this is 332.8 g against 355.9 g weighed, a shortfall of +6.5 %. For the LFP cell it is 340.3 g against 332 g, an excess of −2.5 %.

The two residuals have opposite signs. A systematic error in the method, such as a wrong electrolyte density, a separator counted twice or an error in the electrode area, would shift both cells in the same direction, so the opposite signs make such an error unlikely. The shortfall of about 23 g in the NMC cell corresponds to hardware that the layer model does not contain: the mandrel, the tab welds, the vent and burst disc assembly, the crimped cap, and the electrolyte held outside the roll in the headspace and in the separator overhang. For the LFP cell, whose teardown reports the mandrel and disk dimensions, this hardware is largely included. Its excess of about 8 g comes from the can wall and separator carried over from the NMC cell (0.50 mm and 12 μ\mum); § 4.2 determines them as 0.418 mm and 13.7 μ\mum, after which the mass closes.

Capacity. For the LFP cell, the reported areal capacity of 2.09 mAh cm−2^{-2} over the double-sided cathode gives 15.49 Ah against 15.41 Ah measured, a residual of +0.51 %. This is an independent prediction. For the NMC cell the transferred windows delivered 18 % more than the measured C/3 capacity; the windows were calibrated to the cut-off voltages and the C/20 capacity (§ 3.5), after which the C/3 capacity, not used in the calibration, closes to 0.02 %.

Winding. The winding does not set the current path of the tabless cell, which runs across the electrode height, but it sets the outer radius of the roll and the radial path of the heat, so the check applies to both cells. An Archimedean spiral of given pitch, core radius and length has a single outer radius,

L=π(rout2−rcore2)p,rout=rcore2+Lpπ(2)L = \frac{\pi\left(r_{\mathrm{out}}^2-r_{\mathrm{core}}^2\right)}{p}, \qquad r_{\mathrm{out}} = \sqrt{r_{\mathrm{core}}^2+\frac{Lp}{\pi}} \tag*{(2)}

and this radius must fit inside the can bore. For the LFP cell, the 5.37 m cathode at its measured layer pitch gives 22.96 mm against a bore of 22.73 mm, an overshoot of 0.99 %. For the NMC cell the same calculation gives +1.3 %.

With the anode coating estimated at 110 µm, as the set carried before, the NMC winding overshot by 8.8 %, the largest residual of the set. Taking the anode thickness from the loading reported in the teardown, 126 µm per side (§ 3.5), closes it to +1.3 %. The check located the error in the one geometric quantity that had been estimated rather than measured. What remains is within the variation of layer thickness between the core and the outer wraps that the micrographs of the teardown show, which a single pitch, as assumed in Equation (2), does not describe.

Scope of the checks. The closure checks do not validate the model, since no dynamics enter them. They test whether the parameter set describes a physically realisable cell. Their value lies in their ability to discriminate: checks that were passed equally by both cells would also be consistent with a residual absorbed into the least constrained parameter. Here the mass check separates the two cells by sign, and the winding check identified an estimated anode thickness in the NMC set; with the measured value both cells close the winding to about one per cent (+1.3 and 0.99 %).

The thermal network of both cells

The end hardware of the two cells differs, and the differences partly cancel. The insulator disk of the LFP cell is 0.29 mm thick, against 0.65 mm in the NMC cell, which increases the conductance from the positive collector disk to the can by a factor of 2.2. The multi-tab construction reduces the conductance from the roll to the collector disk by a factor of about fifty. The first improves the thermal coupling of the cap, and the second reduces that of the roll. Both factors are computed from the tagged geometry of the end hardware and are not adjusted. They are not inputs of the two-node balance of § 5.3, which lumps the end hardware into the can node; they describe how a distributed model would divide the heat between the two ends.

As a result the LFP cell is thermally less asymmetric than the NMC cell. The decomposition of the LFP network, computed directly rather than scaled from the NMC result, gives an end-conductance asymmetry ratio of 10.4 against 19.2 and a fixture response coefficient of 2.21 against 2.87 K W−1^{-1}. Both differences are large enough to matter in an experiment, and neither could be estimated reliably by hand.

Parameter extraction

The parameters were obtained by three routes: measurement on the cell itself, transfer from a characterised cell of the same chemistry, and determination from constraints that several measured quantities impose together. This section summarises the first two and describes the third, which is used for the three LFP quantities that no source reports. The group-by-group account, with the tolerance of every parameter, is given in Supplementary Section S3 and Supplementary tbl:S1.

Measured and transferred parameters

The tolerance attached to a parameter depends on its tag. Measured [M] values carry the resolution or scatter of the measurement, derived [D] values the uncertainty propagated from their inputs, and estimated [E] values a declared range. Transferred [L] values carry a transfer band, which is not the precision of the source but the spread between independent parameterisations of the same material where two exist.

Four groups determine most of the uncertainty in the predictions. The coating thicknesses set the capacity linearly,

Q=AεsLcs,maxΔθF,(3)Q = A \varepsilon_s L c_{\mathrm{s,max}} \Delta\theta F, \tag*{(3)}

with AA the coated area and Δθ\Delta\theta the stoichiometry window. Inverting the areal capacity reported by Lacey [26] through Equation (3) gives a cathode thickness of 82 µm against the 85 µm measured. The winding length of the NMC cell has two determinations that differ by 10.7%, and it is carried as a one-sided band. The heat-transfer coefficient at the can surface accounts for 76 to 93% of the thermal resistance between the roll and ambient, from h=50 W m−2 K−1h = 50\ \mathrm{W\,m^{-2}\,K^{-1}} to the test holder of § 6 (Equation (10)). The electrolyte concentration, in contrast, has little influence: at 1 mol L−1^{-1} both cells lie close to the conductivity maximum of the Nyman et al. [32] correlation.

Determination of unmeasured quantities

For the LFP cell, Liu et al. [30] report neither the cathode porosity ε+\varepsilon_+, nor the can wall thickness ww, nor the separator thickness LsepL_{\mathrm{sep}}. Three relations constrain them. The mass of the assembled cell is the sum of the can, the end hardware and the layers of the roll,

mcell=mcan(w)+mhw+A∑kLk[ρs,k(1−εk)+ρeεk],(4)m_{\mathrm{cell}} = m_{\mathrm{can}}(w) + m_{\mathrm{hw}} + A \sum_k L_k \left[\rho_{\mathrm{s},k}(1-\varepsilon_k) + \rho_{\mathrm{e}}\varepsilon_k\right], \tag*{(4)}

where the sum runs over the coatings, the separator and the foils. The measured areal capacity of the cathode fixes its solid fraction through Equation (3). The wound roll must fit the can bore, which for an Archimedean spiral of length LwL_w and period pp gives

rcore2+Lwpπ=Dcan2−w,p=LCu+LAl+2(L−+L++Lsep).(5)\sqrt{r_{\mathrm{core}}^2+\frac{L_w p}{\pi}} = \frac{D_{\mathrm{can}}}{2} - w,\qquad p = L_{\mathrm{Cu}} + L_{\mathrm{Al}} + 2(L_- + L_+ + L_{\mathrm{sep}}). \tag*{(5)}

Equation (5) contains the wall and the separator but no chemistry, the capacity relation contains the porosity but no hardware, and the mass balance contains all three. The system is square and has a unique solution: ε+=0.331\varepsilon_+ = 0.331, w=0.418w = 0.418 mm and Lsep=13.7L_{\mathrm{sep}} = 13.7 µm. The anode porosity is an input to the mass balance, not an unknown: inverting the measured anode areal capacity through Equation (3) gives 0.319. Before the determination the cathode porosity had been obtained from the mass balance alone, with a wall of 0.50 mm taken from the NMC cell, and was 0.419. Re-solving the system for every input perturbed across its plausible range keeps the wall between 0.35 and 0.48 mm, the separator between 11.0 and 16.5 µm and the porosity between 0.29 and 0.37 (Supplementary Table S2).

Two consequences can be checked independently. The anode porosity of 0.319, obtained from the anode alone, is close to the 0.331 the three relations give for the cathode. And the determined wall implies an empty can of 50 g, against about 65 g [D], computed from the measured can dimensions for a 0.55 mm wall, the middle of the 0.5 to 0.6 mm case of a structural 4680 can (Battery Design [8]); a single weighing can confirm it. Because the three relations are used for the determination, they are no longer independent checks for the LFP cell.

The method applies to quantities that enter a conservation statement: masses, volumes, charges and lengths. Kinetic, transport and thermal properties enter none of these relations and have to be measured or transferred.

Range of validity of the transferred potentials

The open-circuit potentials of the LFP cell are polynomial and exponential fits ([2]), valid inside the stoichiometry window over which they were fitted and divergent outside it; the cathode fit returns 9.06×105 V9.06 \times10^{5}\ \mathrm{V} at θ=0.05\theta= 0.05. A Newton iterate outside the window therefore produces an unphysical residual. The released code evaluates each fit inside its window [θa,θb][\theta_a,\theta_b] and extends it linearly outside,

U(θ)=U(θb)+U′(θb) (θ−θb),θ>θb,(6)U(\theta) = U(\theta_b) + U'(\theta_b)\,(\theta-\theta_b), \qquad\theta> \theta_b, \tag*{(6)}

and correspondingly below θa\theta_a, which leaves the potential unchanged inside the window and keeps a gradient outside it. The tanh forms of Chen et al. [13] used for the NMC cell are bounded and need no such extension.

The model

The model is the porous-electrode formulation of Doyle et al. [15] and Fuller et al. [23], extended to non-isothermal operation by Gu and Wang (2000), applied to one repeating unit of the wound electrode and coupled to a two-node thermal balance of the cell (Figure 1). We refer to it as a thermal DFN (TDFN). The transport and charge balances and the Butler–Volmer kinetics are those of the textbook DFN; the complete set of equations, labelled (E1) to (E9), is given in Appendix § 10 (Table 2), with their coupling (Figure 2). This section describes only what differs from the textbook form: the domain of the wound cell, the particle closure, the heat source, the two-node thermal balance and the current collection.

Figure 1. The model. (A) Cross-section of the 4680 cell: the jelly roll around its hollow core and the can. The roll node carries the heat capacity of the layers of the roll at their mean temperature TrT_r; the can node carries the rest of the cell, the can, end discs, tabs and electrolyte outside the roll, at TcT_c, and exchanges heat with the surroundings through hAhA over the whole exterior. (B) The repeating unit of the DFN model, resolved through its thickness from the centreline of the copper foil to that of the aluminium foil; both boundaries are symmetry planes because the electrodes are coated on both faces. The particle concentration follows the quartic profile of Subramanian et al. [39]. (C) The two-node thermal balance of Equation (9), with the heat of the electrode stack and the I2RseriesI^2 R_{\mathrm{series}} heat entering the two nodes and the internal resistance of Equation (10).

Figure 2. Coupling of the equations of the TDFN, labelled as in Table 2. Boxes are the equations and their unknowns, arrows the quantities passed between them, and the numbers the steps described in Appendix § 10.§ 2.

No.ElementEquationsText
(E1)Electrolyte∂(εece)∂t=∂∂x(Deeff∂ce∂x)+1−t+0FjLi\frac{\partial(\varepsilon_e c_e)}{\partial t}=\frac{\partial}{\partial x}\left(D_e^{\mathrm{eff}}\frac{\partial c_e}{\partial x}\right)+\frac{1-t_+^0}{F}j^{\mathrm{Li}}<br>∂∂x(κeff∂ϕe∂x+κDeff∂ln⁡ce∂x)+jLi=0\frac{\partial}{\partial x}\left(\kappa^{\mathrm{eff}}\frac{\partial\phi_e}{\partial x}+\kappa_D^{\mathrm{eff}}\frac{\partial\ln c_e}{\partial x}\right)+j^{\mathrm{Li}}=0–<br>–
(E2)Solid∂∂x(σeff∂ϕs∂x)−jLi=0,∂(εscˉs)∂t=−jLiF\frac{\partial}{\partial x}\left(\sigma^{\mathrm{eff}}\frac{\partial\phi_s}{\partial x}\right)-j^{\mathrm{Li}}=0,\qquad\frac{\partial(\varepsilon_s\bar{c}_s)}{\partial t}=-\frac{j^{\mathrm{Li}}}{F}–, –
(E3)Particle (quartic)dqˉdt=30Dsrs2qˉ−45N2rs2,35Dsrs(cse−cˉs)−8Dsqˉ=−N\frac{d\bar{q}}{dt}=\frac{30D_s}{r_s^2}\bar{q}-\frac{45N}{2r_s^2},\qquad\frac{35D_s}{r_s}(c_{se}-\bar{c}_s)-8D_s\bar{q}=-N<br>N=±jLi/(asF),as=3εs/rsN=\pm j^{\mathrm{Li}}/(a_sF),\qquad a_s=3\varepsilon_s/r_s(7)
(E4)InterfacejLi=asi0[exp⁡(αFηRT)−exp⁡(−αFηRT)],η=ϕs−ϕe−Uj^{\mathrm{Li}}=a_si_0\left[\exp\left(\frac{\alpha F\eta}{RT}\right)-\exp\left(-\frac{\alpha F\eta}{RT}\right)\right],\qquad\eta=\phi_s-\phi_e-U<br>i0=k ce1/2[cse(cs,max⁡−cse)]1/2i_0=k\,c_e^{1/2}\left[c_{se}(c_{s,\max}-c_{se})\right]^{1/2}–
(E5)Anode potentialU=U−(θ−)U=U_-(\theta_-), graphite; open-circuit potentials of both cells as in Table 1Suppl.
(E6)Cathode potentialU=U+(θ)U=U_+(\theta), NMC or LFP; for LFP with the end-of-lithiation drop of § 6.2Suppl.
(E7)Heatq=jLi(ϕs−ϕe−U(θˉ)+T∂U∂T)+σeff(∂ϕs∂x)2q=j^{\mathrm{Li}}\left(\phi_s-\phi_e-U(\bar{\theta})+T\frac{\partial U}{\partial T}\right)+\sigma^{\mathrm{eff}}\left(\frac{\partial\phi_s}{\partial x}\right)^2<br>+κeff(∂ϕe∂x)2+κDeff∂ln⁡ce∂x∂ϕe∂x+\kappa^{\mathrm{eff}}\left(\frac{\partial\phi_e}{\partial x}\right)^2+\kappa_D^{\mathrm{eff}}\frac{\partial\ln c_e}{\partial x}\frac{\partial\phi_e}{\partial x}(8)
(E8)Two-node thermalCrdTrdt=∫rollq dV+Qs,r−Tr−TcRintC_r\frac{dT_r}{dt}=\int_{\mathrm{roll}}q\,dV+Q_{s,r}-\frac{T_r-T_c}{R_{\mathrm{int}}}<br>CcdTcdt=Tr−TcRint+Qs,c−hA(Tc−T∞)C_c\frac{dT_c}{dt}=\frac{T_r-T_c}{R_{\mathrm{int}}}+Q_{s,c}-hA(T_c-T_\infty)<br>Qs,r+Qs,c=I2RseriesQ_{s,r}+Q_{s,c}=I^2R_{\mathrm{series}}, RintR_{\mathrm{int}} of Equation (10)(9)<br>(10)
(E9)TerminalV=ϕs(xAl)−ϕs(xCu)−IRseriesV=\phi_s(x_{\mathrm{Al}})-\phi_s(x_{\mathrm{Cu}})-IR_{\mathrm{series}}; DeD_e, κ\kappa, DsD_s and kk Arrhenius in TrT_r§ 5.4

Table 2. Table 2. Equations of the TDFN, labelled as in Figure 2. xx runs through the repeating unit, rr through a particle; θ=cse/cs,max⁡\theta=c_{se}/c_{s,\max} and θˉ=cˉs/cs,max⁡\bar{\theta}=\bar{c}_s/c_{s,\max} are the surface and mean stoichiometries. Boundary conditions: no flux of lithium through the foils, no electrolyte current into the foils, the applied current through the aluminium foil, and continuity of all fluxes at the electrode–separator interfaces. The last column gives the equation of the text.

Domain and particle closure

The repeating unit is resolved through its thickness xx, from the centreline of the copper foil to that of the aluminium foil; the electrode is coated on both faces, so the boundaries are symmetry planes and the unit stands for the whole winding with the current taken as uniform along it. Each control volume of an electrode carries one representative particle of radius rsr_s and mean concentration cˉs\bar{c}_s. Instead of resolving the diffusion inside it, its surface concentration csec_{se} follows the quartic profile of Subramanian et al. [39], which adds the volume-averaged concentration gradient qˉ\bar{q} as one state per control volume,

dqˉdt=−30Dsrs2qˉ−452Nrs2,35Dsrs(cse−cˉs)−8Dsqˉ=−N,as=3εsrs,(7)\frac{\mathrm{d}\bar{q}}{\mathrm{d}t} = -\frac{30D_s}{r_s^2}\bar{q} - \frac{45}{2}\frac{N}{r_s^2}, \qquad\frac{35D_s}{r_s}(c_{se}-\bar{c}_s)-8D_s\bar{q}=-N, \qquad a_s=\frac{3\varepsilon_s}{r_s}, \tag*{(7)}

where N=±jLi/(asF)N=\pm j^{\mathrm{Li}}/(a_sF) is the molar flux out of the particle surface, so that the surface responds to a step in current partly at once and partly with the time constant rs2/(30Ds)r_s^2/(30D_s), and reduces at constant current to the parabolic closure cse−cˉs=−Nrs/(5Ds)c_{se}-\bar{c}_s=-Nr_s/(5D_s) of Gu and Wang (2000).

Heat source

With the reaction current jLi=asinj^{\mathrm{Li}}=a_{\mathrm{s}}i_{\mathrm{n}} of Butler–Volmer kinetics (Appendix A), heat is generated at the volumetric rate

q=asin(φs−φe−U+T∂U∂T)+σeff(∂φs∂x)2+κeff(∂φe∂x)2+κDeff∂ln⁡ce∂x∂φe∂x.(8)q=a_{\mathrm{s}}i_{\mathrm{n}}\left(\varphi_{\mathrm{s}}-\varphi_{\mathrm{e}}-U+T\frac{\partial U}{\partial T}\right)+\sigma^{\mathrm{eff}}\left(\frac{\partial\varphi_{\mathrm{s}}}{\partial x}\right)^2+\kappa^{\mathrm{eff}}\left(\frac{\partial\varphi_{\mathrm{e}}}{\partial x}\right)^2+\kappa_{\mathrm{D}}^{\mathrm{eff}}\frac{\partial\ln c_{\mathrm{e}}}{\partial x}\frac{\partial\varphi_{\mathrm{e}}}{\partial x}. \tag*{(8)}

Equation (8) is the volumetric heat source of Gu and Wang (2000), with one completion. As printed, it evaluates UU at the particle surface and neglects the heat of mixing released as concentration gradients inside the particles relax [35, 43, 36]. The model evaluates the reaction heat at the open-circuit potential of the particle-mean concentration, which restores this term; Butler–Volmer kinetics keep UU at the surface. This is the distributed counterpart of the cell-level irreversible heat I(Uavg−V)I(U_{\mathrm{avg}}-V) of Bernardi et al. [9], not a new heat model. It is exact for the heat integrated until the particles have relaxed; while a gradient is building up, as after a step in current, it counts as heat the energy still stored in that gradient, which the mixing term releases only later. The entropic term uses the coefficients of O’Regan et al. [33]. For the NMC cell at 1C and 50 % state of charge the completed source agrees with a resolved-particle model including the heat of mixing [40] to within 0.26 W for diffusion times between 680 and 6800 s. With the surface form, the heat is nearly independent of the solid diffusivity (3.82 to 3.95 W over the same range, against 4.13 to 5.86 W completed), so it would appear to agree with a measured heat by omission (Supplementary Table S6). The electrolyte diffusivity and conductivity, the solid diffusivity and the rate constant follow Arrhenius laws [16] in the temperature of the roll. The open-circuit potentials and the electrolyte correlations are given in Supplementary Section S2.

Thermal balance

The temperature of the cell is carried by two nodes [20, 29]: the roll, at its heat-capacity-weighted mean temperature TrT_{\mathrm{r}}, and the can, at TcT_{\mathrm{c}},

CrdTrdt=∫rollq dV+Qs,r−Tr−TcRint,CcdTcdt=Tr−TcRint+Qs,c−hA(Tc−T∞).(9)C_{\mathrm{r}}\frac{\mathrm{d}T_{\mathrm{r}}}{\mathrm{d}t}=\int_{\mathrm{roll}}q\,\mathrm{d}V+Q_{\mathrm{s,r}}-\frac{T_{\mathrm{r}}-T_{\mathrm{c}}}{R_{\mathrm{int}}},\qquad C_{\mathrm{c}}\frac{\mathrm{d}T_{\mathrm{c}}}{\mathrm{d}t}=\frac{T_{\mathrm{r}}-T_{\mathrm{c}}}{R_{\mathrm{int}}}+Q_{\mathrm{s,c}}-hA\left(T_{\mathrm{c}}-T_{\infty}\right). \tag*{(9)}

CrC_{\mathrm{r}} is the heat capacity of the layers of the roll and CcC_{\mathrm{c}} the rest of the cell, the can, the end discs, the tabs and the electrolyte outside the roll, so that Cr+Cc=mcpC_{\mathrm{r}}+C_{\mathrm{c}}=mc_{\mathrm{p}}. The series resistance of Section 4.4 is part of the cell’s loss: its heat I2RseriesI^2R_{\mathrm{series}} enters as Qs,rQ_{\mathrm{s,r}} for the share of the foils and as Qs,cQ_{\mathrm{s,c}} for the tabs, welds and end discs. The internal resistance is the exact mean-to-surface resistance of a uniformly heated annulus of height HeH_{\mathrm{e}} between the adiabatic core rir_{\mathrm{i}} and the roll surface ror_{\mathrm{o}},

Rint=12πλrHe14(ro2−ri2)2−2ri2[14(ro2−ri2)−12ri2ln⁡(ro/ri)](ro2−ri2)2,(10)R_{\mathrm{int}}=\frac{1}{2\pi\lambda_{\mathrm{r}}H_{\mathrm{e}}}\frac{\frac{1}{4}\left(r_{\mathrm{o}}^2-r_{\mathrm{i}}^2\right)^2-2r_{\mathrm{i}}^2\left[\frac{1}{4}\left(r_{\mathrm{o}}^2-r_{\mathrm{i}}^2\right)-\frac{1}{2}r_{\mathrm{i}}^2\ln\left(r_{\mathrm{o}}/r_{\mathrm{i}}\right)\right]}{\left(r_{\mathrm{o}}^2-r_{\mathrm{i}}^2\right)^2}, \tag*{(10)}

which reduces to 1/(8πλrHe)1/(8\pi\lambda_{\mathrm{r}}H_{\mathrm{e}}) for a solid cylinder; λr\lambda_{\mathrm{r}} is the through-stack conductivity of the roll, the layers in series (1.31 W m−1^{-1} K−1^{-1} for the NMC cell). The can exchanges heat with its surroundings through hAhA, over the whole exterior. The electrochemistry sees TrT_{\mathrm{r}}; a sensor on the can reads TcT_{\mathrm{c}}.

Current collection

Current collection enters the model through the series resistance. The NMC cell collects tablessly, over the whole edge of each foil into an end disc; the LFP cell through four aluminium tabs on the cathode, at radii of 0.4, 13.1, 16.0 and 18.3 mm, and three nickel tabs on the anode, at 9.0, 15.6 and 20.0 mm, one of which carries no current [30]. Averaged over the winding, the collection resistance is 0.020 mΩ0.020\ \mathrm{m}\Omega for the tabless cell and 0.792 mΩ0.792\ \mathrm{m}\Omega for the multi-tab cell.1 The can walls of the two cells have nearly equal resistances, 0.7910.791 and 0.942 mΩ0.942\ \mathrm{m}\Omega. In the tabless cell the collection resistance is two orders of magnitude below the can wall and can be neglected in a lumped description; in the multi-tab cell the two are of the same size.

Discretisation and verification

The repeating unit is discretised by the cell-centred finite-volume method on a grid that conforms to the layers, so that no control volume contains two materials. Face coefficients are harmonic means, which is required because the foils are seven orders of magnitude more conductive than the adjacent electrodes, and each foil is resolved by one control volume. Time is advanced by backward Euler with one Newton solve over all unknowns per step, and each linear system is equilibrated before a sparse LU factorisation; the can node of Equation (9) is eliminated exactly inside each step. Supplementary Section S1 gives the details. The grid used for all results has one control volume in each foil, 20 in the negative electrode, 8 in the separator and 20 in the positive electrode (Figure 1B). Doubling the number of control volumes in each layer and halving the voltage step that controls the time step changes the delivered charge of the LFP model of the diagnostic calibration (§ 6.2) by less than 0.01 % and the peak can temperature by 0.01 K at 2C; halving them changes the same quantities by less than 0.01 % and 0.02 K. The identified parameters therefore do not depend on the grid at this resolution.

The integral of the interfacial current over each electrode matches the applied current to 9.6×10−139.6 \times10^{-13} in relative terms, and the integral of Equation (8) matches the electrical dissipation to 4.3×10−14.3 \times10^{-1}. Over a 2C charge in the test holder of § 6, the heat generated equals the heat stored in the two nodes plus the heat lost to the surroundings to within 0.3 %. The solver reproduces the delivered capacity reported by Gu and Wang (2000) for their cell to within two percentage points of depth of discharge in the coupled, decoupled and isothermal cases.

Comparison with measurement

Both parameter sets are compared with measurement in the same way, in the five steps of Figure 3: (1) the set is transferred from published values, each tagged, and its prediction is stated; (2) the measurement is cross-checked against other published data on quantities the model does not enter, to separate artefacts of either; (3) the fixture is identified from the measurement alone; (4) with the fixture fixed, the [E] parameters are calibrated in a loop of prediction, comparison, diagnosis and update, ending in a Nelder–Mead minimisation [31] of the cost

Figure 3. Parameterisation and calibration against measurement. Published data and the measurement feed the identification of the parameter set, in which every value carries a provenance tag; the fixture quantities are identified from the measurement alone. The transferred set gives a prediction that is stated before the test. In the calibration loop the model is simulated and compared with the measured curves, each residual is attributed to an artefact of the measurement or to the estimated [E] values, and the [E] values are adjusted within their bands, the final passes by Nelder–Mead minimisation of a misfit cost. The loop ends either in a calibrated set or in a residual that no adjustment removes, which is attributed to the model form. § 6 applies the procedure to both cells with the cost of Equation (11): to the NMC cell on the open data of Ank et al. [5], to the LFP cell on the bench test.

J=∑1C, 2C[(ΔTrms0.5 K)2+(ΔTpeak0.5 K)2+(ΔQ0.1 A h)2+(ΔVrms20 mV)2],(11)J = \sum_{1\mathrm{C},\,2\mathrm{C}} \left[ \left(\frac{\Delta T_{\mathrm{rms}}}{0.5\,\mathrm{K}}\right)^2 + \left(\frac{\Delta T_{\mathrm{peak}}}{0.5\,\mathrm{K}}\right)^2 + \left(\frac{\Delta Q}{0.1\,\mathrm{A\,h}}\right)^2 + \left(\frac{\Delta V_{\mathrm{rms}}}{20\,\mathrm{mV}}\right)^2 \right], \tag*{(11)}

where ΔTrms\Delta T_{\mathrm{rms}} and ΔTpeak\Delta T_{\mathrm{peak}} are the root-mean-square and the peak differences of the can temperature, ΔQ\Delta Q the difference in the charge passed at the cut-off voltage, ΔVrms\Delta V_{\mathrm{rms}} the root-mean-square difference of the voltage where a voltage trace is measured, and the scales are the resolution of the measurements; (5) the calibrated set is compared with the measurement and the residual it leaves is examined. Each evaluation of JJ runs the model at both rates. For the NMC cell the measurements are the open data of Ank et al. [5]; for the LFP cell, for which no comparable data are published, the author’s bench test (§ 6.2). The model is that of § 5 with the thermal balance of § 5.3 in both cases.

The NMC cell against the open data

Step 1: transfer and prediction. The set of Table 1 and Table 3 was built from the teardown of Ank et al. [5] and the electrochemistry of Chen et al. [13], with the windows calibrated on the C/20 capacity (§ 3.5). The open data used for the comparison are the can temperature at the cap, centre and bottom during 1C and 2C charges from the discharged state to 4.2 V followed by 1800 s of rest, and the C/3 capacity. The pulse resistances and the impedance of the same source are not used: they test the kinetic parameters, which the thermal data do not constrain, and their use is left to the identification follow-up. No voltage trace of these charges is published; the duration of each constant-current charge, read from the published current profile, gives the charge passed when 4.2 V was reached, 18.55 and 15.91 Ah at 1C and 2C (84 and 72 % of the C/3 capacity). The transferred set’s prediction for these tests is the dotted curve of Figure 4: it reaches 4.2 V at 19.44 and 17.76 Ah and raises the can by 13.5 and 29.5 K against 21.2 and 44.2 K measured, mean of the three positions.

Figure 4. NMC cell against the open measurements* of Ank et al. [5], with the transferred set (step 1 of Figure 3) and the calibrated set (step 4). (A,B) Terminal voltage during the constant-current charge at 1C and 2C from the discharged state, against the charge passed; no voltage trace is published, and the measured* point is the charge at which 4.2 V was reached, read from the published current profile. (C,D) Rise of the can temperature through the charge and the following 1800 s of rest: measured* mean of the cap, centre and bottom (their range shaded; it reaches 9 K during the 2C charge), and the can node of the two-node balance with hA=0.18 W K−1h_A = 0.18\ \mathrm{W\,K^{-1}} from the measured cooling and the whole-cell heat capacity. Line styles as in (A) in all panels. *Measured by Ank et al. [5], digitised from their Figure 16.

QuantityUnitNMCLFPTags
Layers
Negative electrode density, ρ\rhokg m−3^{-3}1 9091 887[E] / [D]
Negative electrode specific heat, cpc_pJ kg−1^{-1} K−1^{-1}896908[E] / [D]
Negative electrode thermal conductivity, λ\lambdaW m−1^{-1} K−1^{-1}1.501.04[E] / [E]
Separator density, ρ\rhokg m−3^{-3}1 0561 056[E] / [D]
Separator specific heat, cpc_pJ kg−1^{-1} K−1^{-1}1 6931 693[E] / [D]
Separator thermal conductivity, λ\lambdaW m−1^{-1} K−1^{-1}0.3300.330[E] / [E]
Positive electrode density, ρ\rhokg m−3^{-3}3 4012 678[E] / [D]
Positive electrode specific heat, cpc_pJ kg−1^{-1} K−1^{-1}921993[E] / [D]
Positive electrode thermal conductivity, λ\lambdaW m−1^{-1} K−1^{-1}1.501.50[E] / [E]
Cu foil density, ρ\rhokg m−3^{-3}8 9608 960[L] / [L]
Cu foil specific heat, cpc_pJ kg−1^{-1} K−1^{-1}385385[L] / [L]
Cu foil thermal conductivity, λ\lambdaW m−1^{-1} K−1^{-1}400401[L] / [L]
Al foil density, ρ\rhokg m−3^{-3}2 7002 700[L] / [L]
Al foil specific heat, cpc_pJ kg−1^{-1} K−1^{-1}900900[L] / [L]
Al foil thermal conductivity, λ\lambdaW m−1^{-1} K−1^{-1}238237[L] / [L]
Jelly roll (derived)
Through-plane conductivityW m−1^{-1} K−1^{-1}1.311.05[D] / [D]
In-plane conductivityW m−1^{-1} K−1^{-1}13.422.6[D] / [D]
Heat capacity of the rollJ K−1^{-1}242247[D] / [D]
Heat capacity of the cellJ K−1^{-1}315277[D] / [M]
Boundary and temperature dependence
Can-surface heat-transfer coefficientW m−2^{-2} K−1^{-1}50.022.4[E] / [M]
Entropic coefficient, negative, ∂U/∂T\partial U/\partial TmV K−1^{-1}0.04450.0331[L] / [L]
Entropic coefficient, positive, ∂U/∂T\partial U/\partial TmV K−1^{-1}−0.0193−0.0275[L] / [L]
Activation energy, rate constant −, Ea,i0E_{a,i_0}kJ mol−1^{-1}35.055.0[E] / [L]
Activation energy, rate constant +, Ea,i0E_{a,i_0}kJ mol−1^{-1}17.835.0[E] / [L]
Activation energy, diffusivity −, Ea,DsE_{a,D_s}kJ mol−1^{-1}30.030.0[E] / [L]
Activation energy, diffusivity +, Ea,DsE_{a,D_s}kJ mol−1^{-1}11.0†18.8†[E] / [E]
Activation energy, electrolyte diffusivity, Ea,DeE_{a,D_e}kJ mol−1^{-1}17.117.1[E] / [L]
Activation energy, electrolyte conductivity, Ea,κE_{a,\kappa}kJ mol−1^{-1}17.117.1[E] / [L]

Table 3. Thermal parameters of the two cells. Layer properties are phase-consistent averages (Equation (1)); the effective conductivities of the roll and its heat capacity are derived from them. Tags as in Table 1; †as there. The LFP heat-transfer coefficient and heat capacity are measured in the bench fixture (§ 6.2), the NMC coefficient in the fixture of Ank et al. [5] gives hA=0.18h_A = 0.18 W K−1^{-1} (§ 6).

Step 2: cross-check. The C/3 capacity, which the window calibration did not use, is reproduced to 0.02 % (Table 4). The cross-check between the published references concerns the heat: About:Energy [1] report 3.25 W at 1C and 50 % state of charge, computed from a resistance with the reversible heat excluded and for a cell generation that is not stated, whereas inverting the temperature measured by Ank et al. [5] during the 1C charge through the lumped balance gives 4.2 to 5.7 W. The two are not consistent, and the working value τC=700\tau_C = 700 s of the transferred set, chosen so that the model’s heat and DC resistance stayed near the first, is the choice the calibration revises.

CellQuantityModelReferenceDifferenceReference type
NMCCharge at 4.2 V, 1C / 2C, transferred19.44 / 17.76 A h18.55 / 15.91 A h+0.89 / +1.85 A hmeasured
NMCCharge at 4.2 V, 1C / 2C, calibrated18.51 / 15.92 A h18.55 / 15.91 A h−0.04 / +0.01 A hmeasured
NMCCan temperature, RMS, charge and rest, transferred——5.07 / 8.99 Kmeasured
NMCCan temperature, RMS, charge and rest, calibrated——0.52 / 1.46 Kmeasured
NMCPeak can rise, 1C / 2C, transferred13.5 / 29.5 K21.2 / 44.2 K−7.8 / −14.7 Kmeasured
NMCPeak can rise, 1C / 2C, calibrated21.1 / 44.3 K21.2 / 44.2 K−0.1 / +0.1 Kmeasured
NMCMean heat, 1C / 2C charge, calibrated, contacts included4.5 / 15.0 W4.2–5.7 / 13–17 W—inverted
NMCCapacity, C/3 to 2.5 V22.1 A h22.08 A h+0.02 %measured (check)
LFPDC resistance, 1 s, full charge6.18–8.10 mΩ7.9 mΩ—measured; model: inference
LFPCharge to 2.8 V, 1C / 2C, transferred15.9 / 15.9 A h14.24 / 13.41 A h+12 / +19 %measured
LFPCan temperature, RMS to 2.8 V, transferred——0.97 / 2.35 Kmeasured
LFPCharge to 2.8 V, 1C / 2C, calibrated14.10 / 13.47 A h14.24 / 13.41 A h−1.0 / +0.4 %measured
LFPCharge to 2.0 V, 1C / 2C, calibrated15.08 / 14.55 A h15.30 / 14.44 A h−1.4 / +0.8 %measured
LFPCan temperature, RMS over the discharge, calibrated——0.44 / 0.76 Kmeasured
LFPPeak can rise, 1C / 2C, calibrated14.1 / 26.6 K15.0 / 26.6 K−0.9 / 0.0 Kmeasured
LFPVoltage 1 h after cut-off, 1C / 2C, calibrated3.01 / 3.17 V2.98 / 3.21 V+0.03 / −0.04 Vmeasured, not fitted

Table 4. Table 4. Comparison with measurement. Both cells: the DFN with the two-node thermal balance, quartic particle profile, reaction heat at the open-circuit potential of the particle-mean concentration; transferred, the set as predicted before the comparison; calibrated, the set after the calibration of Section 5, whose cost contained the can temperature and the charge at cut-off at both rates (and the voltage for the LFP cell). NMC: fixture hA=0.18 W K−1hA = 0.18\ \mathrm{W\,K^{-1}} from the measured cooling tails, whole-cell heat capacity, can temperature as the mean of the three measured positions. Reference type: measured, a direct measurement; inverted, heat obtained from a measured temperature trace through a lumped balance (±25 %); check, the C/3 capacity, not used in the window calibration at C/20. Sources: Ank et al. [5] for the charge temperatures and the C/3 capacity; Liu et al. [30] for the LFP rate retention. LFP: the author’s bench test (§ 6.2), cell c1 at 23.2 °C effective ambient with the measured hAhA and heat capacity.

Step 3: the fixture. The roll node carries 242 J K−1^{-1} and the can node 73 J K−1^{-1}, so that the two together equal mcpm c_p with the measured mass of 355.9 g (Ank et al., [5]) and cp=884.6c_p = 884.6 J kg−1^{-1} K−1^{-1} transferred with the set. The can node exchanges heat with the chamber through hAhA, which is not chosen but obtained from the measured cooling after the two charges: the excess temperature decays with time constants of 1708 and 1764 s, and hA=mcp/τhA = m c_p/\tau with τ=1735\tau= 1735 s gives 0.18 W K−1^{-1}, radiation included. This determination fixes the ratio C/hAC/hA; the level of the temperature rise then tests the heat. Fitted in the same way, the model’s own cooling gives 1876 s, longer because the roll cools through the can. The lumped still-air estimate used before these data were available, C/hA=1812C/hA = 1812 s with an estimated h=10h = 10 W m−2^{-2} K−1^{-1}, now has this measured counterpart. The second property of the fixture is the heat of the terminal contacts: during the 2C charge the cap runs 9 K above the bottom and the gradient collapses within minutes of the cut-off (Figure 4D, shaded), which Ank et al. [5] attribute to a contact resistance of about 0.5 mΩ\Omega per terminal, 2 W at 44 A. This heat does not appear in a voltage sensed at the terminals. It is represented, as for the LFP cell, by I2RcI^{2}R_{\mathrm{c}} in the can node, with RcR_{\mathrm{c}} left to the calibration.

Step 4: calibration. With the fixture fixed, the transferred set under-predicts the mean can temperature by 5.07 and 8.99 K root-mean-square and reaches 4.2 V 0.89 and 1.85 Ah late, at 1C and 2C. Single-parameter runs showed which [E] parameters act on these residuals: the diffusion time τc\tau_{\mathrm{c}} of the positive electrode raises the heat and brings the cut-off forward at both rates; RcR_{\mathrm{c}} raises the 2C temperature four times as much as the 1C one; the activation energy of the positive-electrode diffusivity sets how much of the polarisation survives the 45 K rise of the 2C charge; the anode window moves the cut-off. The entropic coefficients, with the magnitude of O’Regan et al. [33], change the 2C peak by about 1 K and were left transferred, as were the kinetic parameters, which the thermal data do not separate from τc\tau_{\mathrm{c}}. The four parameters were then minimised on the cost of Equation (11) without its voltage term, ΔQ\Delta Q taken at 4.2 V. JJ fell from 1957 for the transferred set to 9.8 in 53 evaluations. The calibrated set has τc=8 031\tau_{\mathrm{c}} = 8\,031 s (Ds=3.39×10−15 m2 s−1D_{\mathrm{s}} = 3.39 \times10^{-15}\ \mathrm{m}^{2}\,\mathrm{s}^{-1}), Ea,Ds=11.0 kJ mol−1E_{\mathrm{a},D_{\mathrm{s}}} = 11.0\ \mathrm{kJ}\,\mathrm{mol}^{-1}, Rc=0.94 mΩR_{\mathrm{c}} = 0.94\ \mathrm{m}\Omega and an anode stoichiometry of 0.700 at full charge. Two of these values meet an independent number. RcR_{\mathrm{c}} corresponds to 0.5 mΩ\Omega per terminal, the value Ank et al. [5] estimated for their holder and Allgäuer et al. [3] measured in a holder of the same type. τc\tau_{\mathrm{c}} is within a factor of 1.2 of the value transferred from Chen et al. [13], 6800 s, which the resistance-based heat of About:Energy (2024) had appeared to exclude; the open thermal data, which are a direct measurement with a measured boundary, carry more weight than a figure-only value for an unstated cell generation, and the transferred diffusivity is restored.

Step 5: comparison and residual. The calibrated set reaches 4.2 V at 18.51 and 15.92 Ah against 18.55 and 15.91 Ah measured, and raises the can by 21.1 and 44.3 K against 21.2 and 44.2 K; it follows the mean can temperature to within 0.52 and 1.46 K root-mean-square through the charge and the rest (Figure 4). Its mean heat is 4.09 W over the 1C charge and 13.2 W over the 2C charge, plus 0.46 and 1.8 W at the contacts, against 4.2 to 5.7 W and 13 to 17 W inverted from the measurement. What remains is of two kinds. The 2C residual is largest in the first ten minutes, where the measured mean rises faster than the model: the cap, heated by the contacts, leads the mean, and a single can node cannot carry the 9 K axial gradient. Both rates entered the cost; no test was held out.

The LFP cell on the bench

Test. One FC4680 cell (c1) was tested in a horizontal toggle clamp (Figure 5) with four spring contacts on the positive terminal and a single pointed contact on the negative terminal, in a chamber held at 25 °C by fans. Three type T thermocouples were glued with epoxy to the can, 1 mm from each end and at mid-height. After two preconditioning cycles at 1C, the cell was charged at C/3C/3 to 3.65 V with a constant-voltage hold to C/20C/20, rested for 2 h, discharged at 1C (15.29 A) to 2.0 V and rested for 1 h; the sequence was repeated with a 2C discharge (30.59 A). The discharges and the rests after them were recorded at 10 Hz, the rest at 1 Hz. The predictions of Supplementary Section S6 were stated before the test. The 10 and 40 °C runs of the plan were not carried out.

Figure 5. The bench setup of the LFP cell: cell c1 lying horizontally in the foam cradle of the clamp fixture, with the spring-pin current and voltage-sense contacts at one end, inside the fan-ventilated 25 ∘C25\,^{\circ}\mathrm{C} chamber. For the discharges of this paper three type T thermocouples were glued to the can, 1 mm from each end and at mid-height; the photograph shows the one at mid-height.

Procedure. The LFP set was built and calibrated in the five steps of Figure 3, as the NMC set: transferred from the teardown of Liu et al. [30] and the electrochemistry of About:Energy Limited [2], each value tagged, with its predictions for the test stated before the test (Supplementary Section S6); cross-checked against the published data; the fixture identified from the bench data alone; the [E] parameters calibrated on the cost of Equation (11), here with its voltage term, since the voltage is measured; and the residual examined. Figure 6 shows the transferred set of step 1 and the calibrated set of step 4 against the measurement.

Figure 6. LFP cell c1 on the bench, discharges at 1C and 2C from full charge to 2.0 V2.0\ \mathrm{V} in the 25 ∘C25\,^\circ\mathrm{C} chamber (effective ambient 23.2 ∘C23.2\,^\circ\mathrm{C}), against the model with the transferred set (step 1 of Figure 3) and the calibrated set (step 4), both with the measured hA=0.337 W K−1hA = 0.337\ \mathrm{W\,K^{-1}} and heat capacity 277 J K−1277\ \mathrm{J\,K^{-1}}. (A,B) Terminal voltage at 1C and 2C; the transferred set stops at 2.8 V2.8\ \mathrm{V} with 15.9 A h15.9\ \mathrm{A\,h} delivered. (C,D) Rise of the can temperature over the rest temperature, mean of three thermocouples (their range, typically 0.6 K0.6\ \mathrm{K} at 1C and 1.8 K1.8\ \mathrm{K} at 2C, is shaded), and the can node of the model, through the discharge and the following hour of rest. Line styles as in (A) in all panels. The charge delivered to 2.8 V2.8\ \mathrm{V} and 2.0 V2.0\ \mathrm{V} is compared in Table 4.

Step 2: measurement and cross-check. The cell delivered 15.30 Ah at 1C and 14.44 Ah at 2C, and reached 2.8 V at 14.24 and 13.41 Ah. Liu et al. [30] report 94.8 and 88.1 % of the C/2 capacity to 2.8 V at the same rates, 14.2 and 13.2 A h for their C/2 capacity of about 15.0 A h; the measured values lie within 0.2 A h of these. The can rose by 10.2 and 21.9 K over the rest temperature by 2.8 V and peaked 11 to 12 s after cut-off at 15.0 and 26.6 K, mean of the three sensors, with a maximum of 51.3 °C at 2C. The sensor near the positive terminal ran 2 to 4 K cooler than the other two at the peak, which is consistent with the four spring contacts conducting heat into the clamp. The 1C discharge repeats the three preconditioning discharges, made at 15.0 A, to within 0.06 A h in charge and 0.5 K in rise at 2.8 V, of which about 0.4 K follows from the 2 % higher current. The cell settles at 23.2 °C in the 25 °C chamber after every rest, including the 2 h rests, so this is the effective ambient of the fixture and is used as such. The voltage steps at the start of the 1C discharge give 4.7 mΩ, 7.9 mΩ and 11.7 mΩ after 0.1, 1 and 10 s, nearly the same at 2C; the same steps at the end of the 1C discharge give 7.7 mΩ, 16.5 mΩ and 35 mΩ.

The published and the bench data agree on the charge to 2.8 V within 0.2 A h at both rates, so the capacity of the bench cell is not in doubt. They disagree on the resistance by a factor of 5 to 10: the area-specific values of Liu et al. [30], with the electrode area of Table 1, give 0.8 to 2.3 mΩ against the pulse resistances above. The normalisation area of the published values is not stated, and they are not used.

Step 3: the fixture. Over a full cycle the cell returns to its initial state, so the charge energy less the discharge energy, 4.78 W h for the 1C cycle, is the heat generated, and all of it leaves through the fixture. Dividing by the time integral of the measured excess temperature gives hA=0.337 W K−1hA = 0.337\,\mathrm{W\,K^{-1}}, and the 2C cycle, with the charge left in the cell accounted for, gives 0.33 W K−10.33\,\mathrm{W\,K^{-1}}. This determination needs neither the open-circuit voltage nor the heat capacity. It corresponds to h=22.4 W m−2 K−1h = 22.4\,\mathrm{W\,m^{-2}\,K^{-1}} over the whole can, at the lower edge of the band assumed for the predictions. The cooling after both discharges decays with a time constant of 935 s. With the internal resistance of Equation (10) the two-node balance reproduces it for a heat capacity of 277 J K−1277\,\mathrm{J\,K^{-1}}; an isothermal roll would require 315 J K−1315\,\mathrm{J\,K^{-1}}. Surface sensors cannot separate the two, and the transferred mcp=305 J K−1mc_p = 305\,\mathrm{J\,K^{-1}} lies between them.

Step 4: calibration. At the measured ambient, hAhA and heat capacity, the transferred set follows the plateau voltage to within 10 mV at both rates but places 2.8 V at 15.9 A h at both, and under-predicts the can temperature: 0.97 and 2.35 K root-mean-square up to 2.8 V, at 1C and 2C. It stops at 2.8 V, where the particle surfaces saturate and the solver loses its time step, so it gives no peak.

The calibration is a loop (Figure 3): the model is run, its residuals in voltage, temperature and charge are compared at both rates, each residual is attributed either to an artefact of a measurement or to the model, and the treatment of the measurement or the [E] parameters of the model are updated. The first passes were diagnostic: 14 parameter directions, each at 1C and most at 2C (Supplementary Table S8), identified which parameters act on which residual. They led to four changes.

(a) A heat I2RcI^2R_c in the can node that does not enter the terminal voltage. With the plateau voltage already matched, the measured heat exceeds the model’s by about 0.5 W at 1C and 2 W at 2C, a ratio of four for a ratio of two in current. The heat of the holder contacts outside the voltage sense points has this signature; for the NMC cell the same losses were estimated at 1 to 3 W at 2C (§ 6).

(b) A lower effective diffusivity of the LFP cathode and a lower activation energy of it. With the transferred 80 kJ mol−1^{-1} no single diffusivity places the knee at both rates, because the 2C discharge runs about 15 K hotter and diffuses several times faster in the model, whereas the cell reaches its knee earlier at 2C.

  • (c) A lower anode stoichiometry at full charge, which corresponds to a loss of cyclable lithium relative to the transferred window and sets where the graphite is emptied.

  • (d) A floor on the exchange-current density of 2×10−4cs,max2 \times10^{-4} c_{\mathrm{s,max}} in its concentration factors, which lets the solver follow the voltage below the knee to 2.0 V. It is numerical and has no effect above the knee.

The parameters of (a) to (c) were then minimised, with the Nelder–Mead simplex method, on the cost of Equation (11) without its peak term, ΔVrms\Delta V_{\mathrm{rms}} and ΔTrms\Delta T_{\mathrm{rms}} taken up to the measured 2.8 V point and ΔQ\Delta Q at that point. Each evaluation runs the model at both rates. JJ fell from 11.3 to 3.7 in 50 evaluations. The minimum, at a factor of 0.37 on the diffusivity, an activation energy of 25 kJ mol−1^{-1}, an anode stoichiometry of 0.731 and Rc=1.3R_{\mathrm{c}} = 1.3 mΩ\Omega, places 2.8 V at 14.20 and 13.39 A h and follows the can temperature to within 0.31 and 0.53 K up to that point. Below 2.8 V it does not (Figure 6, dashed): the cell delivers 1.06 and 1.03 A h between 2.8 V and 2.0 V at 1C and 2C, this set 0.72 and 1.48 A h, so the 1C peak is 2.1 K too low and the 2C peak 1.5 K too high, and after cut-off it recovers 0.7 V at once where the cell recovers 0.12 V in the first 0.1 s and the rest over 10 to 600 s. Five further parameter directions around this set, the solid diffusivity of the anode, the electrolyte diffusivity, the floor on the exchange current, the anode window and a Nernstian term at the end of the cathode potential, did not remove this residual (Supplementary Table S8).

The end-of-lithiation potential. The residual below the knee was traced to the open-circuit potential of the cathode. The transferred fit of About:Energy Limited [2] is valid up to θ=0.950\theta= 0.950, and the set extrapolates it linearly beyond that (Supplementary Section S1), which is nearly flat. The fit itself falls from 3.39 V at θ=0.95\theta= 0.95 to 2.91 V at 0.99 and 1.94 V at 1.0, which is the end of lithiation of LiFePO4\mathrm{LiFePO_4}. The guard removes the feature that ends the discharge, so in the model the end was set by the anode emptying, which does not move with the rate, whereas in the cell it is set by the cathode and moves with the rate. With the fit used to θ=1\theta= 1, and the anode window reopened, the model ends 1C and 2C 0.84 A h apart, as measured (0.86 A h), but falls from 2.8 V to 2.0 V within 0.2 A h, where the cell takes 1 A h. The drop was therefore given a width: the guarded fit is kept and a term −A[(θ−1+w)/w]n-A [(\theta- 1 + w)/w]^n for θ>1−w\theta> 1 - w is added, with A=1.4A = 1.4 V fixed by the fit’s value at θ=1\theta= 1, and the width ww and exponent nn left free. This is change (e); it is an effective description of a drop whose shape, in a real electrode, follows from the distribution of particle states, and it is tagged [E].

The cost was then extended to the full Equation (11), with the peak term and the charge at 2.0 V added to the charge at 2.8 V, and minimised over the six parameters of (a) to (c) and (e), each evaluation now running to 2.0 V. On this cost the set above scores 64; the minimum, reached in 45 evaluations, scores 21.4 and is the calibrated set: a factor of 0.77 on the diffusivity, τc=4740\tau_{\mathrm{c}} = 4740 s, an activation energy of 18.8 kJ mol−1^{-1}, an anode stoichiometry of 0.796, Rc=1.34R_{\mathrm{c}} = 1.34 mΩ\Omega, w=0.131w = 0.131 and n=2.6n = 2.6.

Step 5: comparison and residual. The calibrated set places 2.8 V at 14.10 and 13.47 Ah against 14.24 and 13.41 Ah measured, ends the discharge at 15.08 and 14.55 Ah against 15.30 and 14.44 Ah, and peaks at 14.1 and 26.6 K against 15.0 and 26.6 K (Figure 6). The can temperature is followed to within 0.44 and 0.76 K root-mean-square over the whole discharge, and the voltage to within 27 and 46 mV up to 2.8 V. After cut-off the voltage now recovers over the measured range of time, 2.81 V against 2.83 V after 100 s at 1C, and settles at 3.17 V against 3.21 V after 2C (Table 4). The shape of the recovery differs: the model recovers 0.33 and 0.43 V at once at 1C and 2C, then little until about 10 s, then most of the rest between 10 and 300 s; the cell recovers 0.15 and 0.21 V within 0.1 s and the rest at a nearly constant rate per decade of time from 0.1 to 1000 s. What remains is a gradual decline of the measured voltage before the knee that the model does not follow: at 2C the cell is at 3.0 V after 11.0 A h, the model after 12.8 A h, and the model is up to 90 mV too high between 10 and 13 A h; at 1C the discharge ends 0.2 A h early and the peak is 0.9 K low. Section 7 discusses this residual. The details, the acceptance criteria stated before the test and the calibration runs are in Supplementary Section S6.

Estimation methods. The time constants of the cooling are linear least-squares fits of the logarithm of the excess temperature, 300 to 3300 s after cut-off, for each sensor and for their mean. hAhA is the ratio of the heat over a full cycle to the time integral of the excess temperature, evaluated by the trapezoidal rule, and the two cycles give values within 5 % of each other. The model parameters are the minimiser of the extended cost of Equation (11) found by the Nelder–Mead simplex method (Nelder and Mead [31]), as implemented in SciPy ([45]); it needs no derivatives of the model and is a local search, started from the diagnostic set. No confidence intervals are given for them: the repeatability of four 1C discharges of the same cell, 0.06 A h in charge and 0.5 K in rise at 2.8 V, sets the scale below which differences are not resolved, and a statistical estimation with uncertainties, of the kind outlined for the NMC cell in § 6, requires more than one cell and is left to the identification follow-up.

The comparison tests the cell-level behaviour of the parameter sets. The distribution of temperature inside the roll is not resolved by the two-node balance and is not tested here.

Results

Table 1 and Table 3 give the parameters of both cells, grouped into geometric and electrochemical parameters and thermal parameters. The complete sets, with the source and tolerance of every value, are in the Supplementary Material.

Electrochemical parameters. The two cells share the graphite anode and the electrolyte, and their anode particle radii, rate constants and electrolyte properties are close. They differ in the cathode. The LiFePO4\mathrm{LiFePO}_{4} particles are ten times smaller than the NMC811 particles, and the diffusion times of the two cathodes, both calibrated, are of the same order: 4740 s for the LFP cathode, an effective quantity for a two-phase material, and 8031 s for the NMC811 cathode, close to the value transferred from Chen et al. [13]. The LFP coatings are thinner and its winding is longer, which gives the LFP cell 62 % more electrode area and 70 % of the capacity of the NMC cell.

Thermal parameters. The layer properties of the two cells are similar. The through-plane conductivity of the roll, which sets the difference between core and surface temperature, is 1.30 and 1.05 W m−1 K−1\mathrm{W\,m^{-1}\,K^{-1}}, and the heat capacities of the rolls differ by 8 %. The thermal behaviour of the format is dominated by quantities that do not depend on the chemistry: the can, the end hardware and the heat-transfer coefficient at the can surface. For the LFP cell this coefficient and the heat capacity are measured in the bench fixture; for the NMC cell the coefficient is taken from the cooling tails of Ank et al. [5]. In both fixtures a contact heat outside the voltage path, 0.94 and 1.34 mΩ\mathrm{m\Omega}, is identified from the temperature at two rates. The chemistry changes the heat that is generated, and the current collection changes where it is generated.

Release. Both parameter sets are released in the Battery Parameter eXchange format [42], as version 1.2, which carries the calibrated values of both cells (§ 6) and the corrections of version 1.1, verified by loading each file into PyBaMM (Sulzer et al. [40]), and as a table with one row per parameter and the provenance tag of each value, organised in the three metadata tiers of Ward et al. [47] (Supplementary Section S4).

Discussion

The end of a fast discharge

The end of discharge is where the LFP cell reaches its highest temperature, and it is the part of the test that the transferred set described worst. Two findings follow from the calibration.

First, most of the residual was an artefact of the transfer, not of the model form. The transferred open-circuit potential of the cathode is a fit whose validity window ends at θ=0.950\theta= 0.950, and the parameter set extrapolated it linearly beyond that, as it does for every fitted potential. For LiFePO4_4 the end of lithiation lies just beyond that window, and the extrapolation removed the 1.4 V drop that ends the discharge. With the drop in place, the end moves with the rate as measured, the peaks are reproduced within 0.9 K, and the relaxation after cut-off, which the set without the drop recovered at once, follows the measured time scale (§ 6.2). Parameter changes in five other directions had not found this, because the error was in a function, not in a value. The same guard protects the anode potential at θ=0.823\theta= 0.823, and the NMC potentials at the edges of their windows; a potential that is transferred as a fit should be checked at the ends of the stoichiometry range the model will reach.

Second, the drop had to be given a width. The fit falls from 2.8 V to 2.0 V within Δθ=0.01\Delta\theta= 0.01, which is 0.2 A h of this cell, and the cell takes 1 A h; the calibrated width is w=0.131w = 0.131 with exponent 2.6. In a model with one representative particle per control volume, the whole electrode reaches the end of lithiation at once. In a porous LiFePO4_4 electrode, particles of different size fill in sequence at nearly the same potential, and the reaction concentrates in a front that moves through the electrode ([12]); the end of discharge is then spread over the charge that the last particles still accept. Models that describe the cathode as an ensemble of phase-separating particles reproduce the hysteresis and the path dependence that follow from this ([22, 14]). The width identified here is an effective description of that spread, and the residual that remains has the same signature: the measured voltage declines gradually from about 10 A h, by up to 90 mV more than the model at 2C, before the knee. The calibrated activation energy of the effective diffusivity, 18.8 kJ mol−1^{-1} against the transferred 80, is a further sign that the single diffusivity absorbs a mechanism that is not diffusion. Whether an ensemble description removes the remaining residual with the parameters found here above the knee can be tested on the same data. The quartic particle profile already used here changes the response to steps in current, not the particle ensemble.

Heat outside the jelly roll

For both cells a part of the heat is generated at the terminal contacts of the holder: 0.94 mΩ\Omega for the NMC cell in the holder of Ank et al. [5], 1.34 mΩ\Omega for the LFP cell in the bench clamp. It does not appear in a voltage sensed at the cell terminals but raises the can temperature by several kelvin at 2C. Both calibrations identified it from the ratio of the heat at two currents, which requires hAhA to be known independently: from the cooling tails and the heat capacity for the NMC cell, from the energy balance over a cycle for the LFP cell. For the NMC cell the value agrees with the 0.5 mΩ\Omega per terminal that Ank et al. [5] estimated and Allgäuer et al. [3] measured in a holder of the same type. The contact heat enters at one end of the can, and the axial gradient it produces, 9 K at 2C on the NMC can, is the largest residual a single can node leaves. Measuring the voltage drop across each contact would give RcR_c directly.

Conduction and relaxation

The model uses one diffusion time per electrode, identified from the knee under current. With the end-of-lithiation drop in place the same diffusion time also describes the relaxation after the discharges in its range and its end point, to within 0.04 V after an hour at both rates (Table 4); without it the model relaxed at once and to the wrong voltage. The relaxation data were not used in the calibration, so this is a check of the diffusion time outside the data that set it. The shape of the recovery is where the two still differ, and the difference has a mechanism. At cut-off the model’s overpotential has three parts: an ohmic part, a kinetic part, and a concentration part. The first two vanish at once when the current stops, and the kinetic part is large at the end of discharge because the exchange-current density falls with (cs,max−cse)1/2(c_{\mathrm{s,max}}-c_{\mathrm{se}})^{1/2} as the surface saturates; together they are the 0.33 and 0.43 V that the model recovers immediately. The concentration part then relaxes with the single time constant of the quartic profile, rs2/(30Ds)r_{\mathrm{s}}^{2}/(30D_{\mathrm{s}}), about 160 s at the calibrated diffusivity, which is the step between 10 and 300 s. The cell recovers only its ohmic drop at once, 0.15 and 0.21 V, close to II times the 1 s resistance, and the rest at a nearly constant rate per decade of time over four decades, which is the signature of a distribution of relaxation times rather than one.

Both differences follow from the representative particle, and the quartic profile does not remove them. The profile adds one state per control volume, the volume-averaged concentration gradient, and so changes how the surface of a particle responds to a step in current; it still gives every particle in a control volume the same radius, the same surface concentration and the same relaxation time. At cut-off, that one surface is saturated and the kinetic overpotential of the whole electrode collapses when the current stops. In an electrode with a distribution of particle sizes, only the smallest particles are saturated at cut-off; the larger ones still accept lithium at a lower overpotential, so the instantaneous recovery of the cell is smaller and the rest is spread over the relaxation times of the distribution. The same distribution widens the end-of-lithiation drop (§ 8.1). A TDFN whose cathode is an ensemble of phase-separating particles ([22, 14]), which we refer to as EP-TDFN, carries that distribution and is expected to reproduce the instantaneous recovery, the spread of the relaxation and the width of the drop with the parameters found above the knee; this is a hypothesis to be tested on the same data, not a result of this work. The parameters that govern conduction and those that govern relaxation are not in general the same, and neither needs to be constant over the state of charge; here they are consistent at the resolution of one cell. The same applies to the NMC cell, whose cathode consists of polycrystalline secondary particles of 3 to 16 μm ([5]), so a single particle radius is also an approximation there.

Transferability of the parameter sets

The parameter sets combine the electrochemistry with the geometry, the inventory and the current collection of each cell, and several values are effective: the cathode diffusion times and their activation energies, the stoichiometry windows, the contact heats and the heat-transfer coefficients of the fixtures. These absorb effects of the cell design, the test fixture and the one-dimensional description of the electrode. The sets are therefore parameter sets of these two cells in this model, not of the materials, and they should not be transferred to another format, another fixture or another model without re-identification of the values tagged [E]. The tags show which values these are. The identified values do not depend on the grid at the resolution used (§ 5.5).

Conclusion

Complete parameter sets have been identified for two 4680 cells, a tabless NMC811/graphite cell and the multi-tab BYD LFP/graphite cell, with the provenance of every value recorded. The sets are checked against the mass, the capacity and the winding of the assembled cells. The mass residuals, +6.5 % and −2.5 %, have opposite signs and locate the NMC shortfall in hardware that the layer model does not contain. For the LFP cell, three quantities that no source reports, the cathode porosity, the can wall and the separator thickness, are determined from the mass, capacity and volume relations together.

The model couples a DFN of the electrode stack to a two-node thermal balance whose heat capacities and internal resistance follow from the parameter set, and whose boundary is taken from the measured cooling of the test fixture. The current collection enters through the series resistance: between the tabless and the multi-tab cell the collection resistance changes by a factor of forty, from negligible to comparable with the can wall.

The heat source of Gu and Wang (2000) is completed by evaluating the reaction heat at the open-circuit potential of the particle-mean concentration, which restores the heat of mixing; with the printed form the heat is nearly independent of the solid diffusivity.

Both sets are calibrated by the same procedure: the fixture is identified from the measurement alone, and the effective parameters, the cathode diffusion time, its activation energy, the anode window and a contact heat outside the voltage path, are adjusted by Nelder–Mead minimisation of a normalised least-squares cost on the voltage and the can temperature at 1C and 2C. For the NMC cell, against the open charge test of Ank et al. [5] with the fixture coefficient taken from the measured cooling, the transferred set under-predicts the can temperature by 8 and 15 K at the end of the 1C and 2C charges and reaches 4.2 V late. The calibrated set follows the mean can temperature to within 0.52 and 1.46 K and the charge at 4.2 V to within 0.1 A h. Its contact heat, 0.94 mΩ\Omega, equals the value the authors estimated for their holder, and its diffusion time, 8 031 s, is within a factor of 1.2 of the value transferred from Chen et al. [13], which a resistance-based heat value had appeared to exclude; a direct thermal measurement with a measured boundary carries more weight than that value. The kinetic parameters, which the thermal data do not constrain, are left to be identified from the pulse and impedance data of the same source in a follow-up.

For the LFP cell, the bench test at 1C and 2C gives the fixture conductance from the energy balance over a cycle, 0.337 W K−1^{-1}, and the heat capacity from the cooling, 277 J K−1^{-1}. The calibration identifies a contact heat of 1.34 mΩ\Omega outside the voltage path, an effective cathode diffusivity with a low activation energy, and the anode window. With these the model follows the can temperature to within 0.3 and 0.5 K up to the knee but misses the end of discharge, and the cause is the transferred open-circuit potential of the cathode: its fit ends at a stoichiometry of 0.95 and the set extrapolated it flat, removing the end-of-lithiation drop. With the drop restored and its width identified, the calibrated set follows the can temperature to within 0.44 and 0.76 K over the whole discharge, reproduces the peaks to within 0.9 K and the relaxation after cut-off. The residual that remains, a gradual decline of the voltage before the knee at 2C, is of the kind that a distribution of particle states produces and a single representative particle does not; whether an ensemble description removes it can be tested on the same data.

The Complete Model

Table 2 lists all equations of the TDFN. Those not discussed in Section 4 are the textbook balances of the DFN model [15]: conservation of lithium and of charge in the electrolyte (E1) and in the solid (E2), coupled by the interfacial reaction current jLi=asinj^{\mathrm{Li}} = a_{\mathrm{s}}i_{\mathrm{n}} with Butler–Volmer kinetics (E4). The boundary conditions are no flux of lithium through the foils, no electrolyte current into the foils, the applied current through the aluminium foil, and continuity of all fluxes at the electrode–separator interfaces.

Coupling of the equations

Figure 2 can be read in eight steps:

  1. The applied current enters the solid (E2) at the aluminium foil; it also sets the series drop at the terminal (E9).

  2. The solid and the electrolyte (E1, E2) pass the potentials φs\varphi_{\mathrm{s}}, φe\varphi_{\mathrm{e}} and the electrolyte concentration cec_{\mathrm{e}} to the interface (E4).

  3. The interface turns the overpotential into the reaction current jLij^{\mathrm{Li}}, which it returns to E1 and E2 as the source of their charge and lithium balances.

  4. The reaction flux NN and the mean concentration cˉs\bar{c}_{\mathrm{s}} from E2 drive the particle (E3), which returns the surface stoichiometry θ\theta.

  5. The stoichiometry sets the open-circuit potentials (E5, E6), which return UU to the overpotential of E4. This closes the electrochemical loop.

  6. The terminal voltage (E9) is the difference of the solid potentials at the two foils minus the series drop.

  7. The heat (E7) is collected from the reaction current and overpotential of E4, the ohmic losses of E1 and E2, and the open-circuit potential at the mean stoichiometry with its temperature coefficient.

  8. The two-node balance (E8) adds the I2RseriesI^{2}R_{\mathrm{series}} heat, turns the heat into the roll and can temperatures, and the roll temperature sets the transport and kinetic coefficients of E1 to E4.

The steps give the direction of each dependence, not an order of computation: all equations are solved together within each time step (§ 5.5), so the loops of steps 3, 5 and 8 close within the step.

CONFLICT OF INTEREST STATEMENT

WP is employed by NXP Semiconductors Austria GmbH & Co KG. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

AUTHOR CONTRIBUTIONS

RT: conceptualisation, methodology, software, validation, formal analysis, data curation, writing (original draft), visualisation. NAW, BB and YZ: bench test of the LFP cell (investigation, resources, data curation). WP: review of the model and the manuscript (writing, review and editing). All authors read and approved the submitted version.

FUNDING

This work was supported by the Villum Investigator Smart Battery Project (grant 22860) at Aalborg University.

Generative AI Statement

Generative AI (Claude, Anthropic) was used to assist with writing and editing the text, writing and checking code, and preparing figures. The authors reviewed and edited all generated content and take full responsibility for the manuscript.

Data Availability Statement

The parameter sets for both cells and the closure checks are openly available. Parameters are released as Battery Parameter eXchange JSON [42] for direct loading into PYBAMM and other BPX-aware tools, and as tidy CSV and JSON with provenance tags. The CSV is organised into the primary, secondary and tertiary metadata tiers of Ward et al. [47], with the closure checks released as quality control on the tertiary tier. The LFP cell’s BPX file is released under CC-BY-SA 4.0 because it incorporates electrochemistry from About:Energy Limited [2]; every other artefact is CC BY 4.0. The geometric and inventory data for the LFP cell derive from the published teardown of Liu et al. [30]; no manufacturer-confidential material is reproduced or relied upon. The measurements used for the NMC comparison are those of Ank et al. [5], openly available from mediaTUM [6] (https://doi.org/10.14459/2023mp1725661, CC BY 4.0); the values used here were digitised from the published figures and are included in the release. The bench data of the LFP cell (cell c1, 1C and 2C discharges at 25 °C, voltage, current and three surface temperatures) and the calibration logs are also included.

The data of this paper are available at https://github.com/RemusTeodorescu/4680-cell-models; a versioned archive with a DOI will be deposited on Zenodo on acceptance.

Supplementary Material

The Supplementary Material gives the numerical implementation of the solver, the constitutive relations, the source and tolerance of every parameter, the machine-readable release, the validation calculations, the bench test of the LFP cell with its predictions, results and calibration runs, and the complete parameter sets of both cells. Every supplementary table is generated from the parameter objects of the solvers.

Figures

TABLES

References

  1. [1][Dataset] About:Energy (2024). Tesla 4680 battery cell: performance analysis with lab data. https://www.aboutenergy.io/post/tesla-4680-battery-cell-performance-analysis. DC resistance and calculated heat generation at 50 % SOC, 25 °C; accessed 27 September 2026
  2. [2][Dataset] About:Energy Limited (2023). BPX parameterisation of a 2ah LFP/graphite cylindrical 18650 cell. https://github.com/About-Energy-OpenSource/About-Energy-BPX-Parameterisation. CC-BY-SA 4.0. Doyle–Fuller–Newman parameter set released under the Battery Parameter eXchange standard
  3. [3]Allgäuer, C., Grenz, F., Abo Gamra, K., and Lienkamp, M. (2025). Thermal management of 4680 battery cells: Cell design and cooling concept. Journal of Energy Storage 131, 117318. doi:10.1016/j.est.2025.117318DOI
  4. [4]Andersson, M., Streb, M., Ko, J. Y., Löfqvist Klass, V., Klett, M., Ekström, H., et al. (2022). Parametrization of physics-based battery models from input–output data: A review of methodology and current research. Journal of Power Sources 521, 230859. doi:10.1016/j.jpowsour.2021.230859DOI
  5. [5]Ank, M., Sommer, A., Abo Gamra, K., Schöberl, J., Leeb, M., Schachtl, J., et al. (2023a). Lithium-ion cells in automotive applications: Tesla 4680 cylindrical cell teardown and characterization. Journal of The Electrochemical Society 170, 120536. doi:10.1149/1945-7111/ad14d0DOI
  6. [6][Dataset] Ank, M. et al. (2023b). Lithium-ion cells in automotive applications: Tesla 4680 cylindrical cell teardown and characterization, measurement data. mediaTUM, Technical University of Munich. doi:10.14459/2023mp1725661. Dataset, CC BY 4.0
  7. [7]Baazouzi, S., Feistel, N., Wanner, J., Landwehr, I., Fill, A., and Birke, K. P. (2023). Design, properties, and manufacturing of cylindrical Li-ion battery cells—A generic overview. Batteries 9, 309. doi:10.3390/batteries9060309DOI
  8. [8]Battery Design (2024). Tesla 4680 cell. batterydesign.net https://www.batterydesign.net/tesla-4680-cell/, accessed 27 September 2026
  9. [9]Bernardi, D., Pawlikowski, E., and Newman, J. (1985). A general energy balance for battery systems. Journal of The Electrochemical Society 132, 5–12. doi:10.1149/1.2113792DOI
  10. [10]Birkl, C. R., McTurk, E., Roberts, M. R., Bruce, P. G., and Howey, D. A. (2015). A parametric open circuit voltage model for lithium ion batteries. Journal of The Electrochemical Society 162, A2271–A2280. doi:10.1149/2.0331512jes
  11. [11]BMW Group (2022). BMW Group to use innovative round BMW battery cells in Neue Klasse from 2025. BMW Group PressClub Press release T0403470EN, 9 September 2022, https://www.press.bmwgroup.com/global/article/detail/T0403470EN, accessed 27 September 2026
  12. [12]Castle, M., Richardson, G., and Foster, J. M. (2022). Understanding rapid charge and discharge in nanostructured lithium iron phosphate cathodes. European Journal of Applied Mathematics 33, 328–368. doi:10.1017/S0956792521000036DOI
  13. [13]Chen, C.-H., Brosa Planella, F., O’Regan, K., Gastol, D., Widanage, W. D., and Kendrick, E. (2020). Development of experimental techniques for parameterization of multi-scale lithium-ion battery models. Journal of The Electrochemical Society 167, 080534. doi:10.1149/1945-7111/ab9050DOI
  14. [14]Clarke, W., Sugunan, D., Marinescu, M., Dickinson, E. J. F., Richardson, G., and Foster, J. M. (2026). A continuum approach to modelling hysteresis in phase-change lithium-ion battery electrodes. Journal of Energy Storage 154, 121252. doi:10.1016/j.est.2026.121252DOI
  15. [15]Doyle, M., Fuller, T. F., and Newman, J. (1993). Modeling of galvanostatic charge and discharge of the lithium/polymer/insertion cell. Journal of The Electrochemical Society 140, 1526–1533. doi:10.1149/1.2221597DOI
  16. [16]Doyle, M., Newman, J., Gozdz, A. S., Schmutz, C. N., and Tarascon, J.-M. (1996). Comparison of modeling predictions with experimental data from plastic lithium ion cells. Journal of The Electrochemical Society 143, 1890–1903. doi:10.1149/1.1836921DOI
  17. [17]Drake, S. J., WetZ, D. A., Ostanek, J. K., Miller, S. P., Heinzel, J. M., and Jain, A. (2014). Measurement of anisotropic thermophysical properties of cylindrical Li-ion cells. Journal of Power Sources 252, 298–304DOI
  18. [18]Ecker, M., Käbitz, S., Laresgoiti, I., and Sauer, D. U. (2015a). Parameterization of a physico-chemical model of a lithium-ion battery: II. Model validation. Journal of The Electrochemical Society 162, A1849–A1857. doi:10.1149/2.0541509jesDOI
  19. [19]Ecker, M., Tran, T. K. D., Dechent, P., Käbitz, S., Warnecke, A., and Sauer, D. U. (2015b). Parameterization of a physico-chemical model of a lithium-ion battery: I. Determination of parameters. Journal of The Electrochemical Society 162, A1836–A1848. doi:10.1149/2.0551509jesDOI
  20. [20]Forgez, C., Do, D. V., Friedrich, G., Morcrette, M., and Delacourt, C. (2010). Thermal modeling of a cylindrical LiFePO₄/graphite lithium-ion battery. Journal of Power Sources 195, 2961–2968. doi:10.1016/j.jpowsour.2009.10.105DOI
  21. [21]Forman, J. C., Moura, S. J., Stein, J. L., and Fathy, H. K. (2012). Genetic identification and Fisher identifiability analysis of the Doyle–Fuller–Newman model from experimental cycling of a LiFePO₄ cell. Journal of Power Sources 210, 263–275. doi:10.1016/j.jpowsour.2012.03.009DOI
  22. [22]Foster, J. M., Grudeva, Y., Korotkin, I., Dickinson, E. J. F., Offer, G., and Richardson, G. (2025). The Newman model for phase-change electrodes: Physics-based hysteresis. Journal of The Electrochemical Society 172, 040501. doi:10.1149/1945-7111/adb219DOI
  23. [23]Fuller, T. F., Doyle, M., and Newman, J. (1994). Simulation and optimization of the dual lithium ion insertion cell. Journal of The Electrochemical Society 141, 1–10. doi:10.1149/1.2054684DOI
  24. [25]Kim, G.-H., Smith, K., Lee, K.-J., Santhanagopalan, S., and Pesaran, A. (2011). Multi-domain modeling of lithium-ion batteries encompassing multi-physics in varied length scales. Journal of The Electrochemical Society 158, A955–A969. doi:10.1149/1.3597614DOI
  25. [26]Lacey, M. J. (2022). Commentary on the teardown of the Tesla 4680 cell. lacey.se https://lacey.se/2022/07/18/commentary-tesla-4680-teardown, accessed 27 September 2026
  26. [27]LG Energy Solution (2024). LG Energy Solution to supply next-generation 4695 cylindrical batteries to Rivian. LG Energy Solution Newsroom Press release, 8 November 2024, https://news.lgensol.com/company-news/press-releases/3351/, accessed 27 September 2026
  27. [29]Lin, X., Perez, H. E., Mohan, S., Siegel, J. B., Stefanopoulou, A. G., Ding, Y., et al. (2014). A lumped-parameter electro-thermal model for cylindrical batteries. Journal of Power Sources 257, 1–11. doi:10.1016/j.jpowsour.2014.01.097DOI
  28. [30]Liu, A., Yao, W., Yang, S., Gonsoulin, D., Larson, A., Condon, A., et al. (2025). Design and performance of the BYD LFP/graphite 4680 cylindrical cell. Journal of The Electrochemical Society 172, 110536. doi:10.1149/1945-7111/aele33. Open access, CC BY 4.0DOI
  29. [31]Nelder, J. A. and Mead, R. (1965). A simplex method for function minimization. The Computer Journal 7, 308–313. doi:10.1093/comjnl/7.4.308DOI
  30. [32]Nyman, A., Behm, M., and Lindbergh, G. (2008). Electrochemical characterisation and modelling of the mass transport phenomena in LiPF6–EC–EMC electrolyte. Electrochimica Acta 53, 6356–6365. doi:10.1016/j.electacta.2008.04.023DOI
  31. [33]O’Regan, K., Brosa Planella, F., Widanage, W. D., and Kendrick, E. (2022). Thermal-electrochemical parameters of a high energy lithium-ion cylindrical battery. Electrochimica Acta 425, 140700. doi:10.1016/j.electacta.2022.140700
  32. [34]Pegel, H., Wycisk, D., and Sauer, D. U. (2023). Influence of cell dimensions and housing material on the energy density and fast-charging performance of tabless cylindrical lithium-ion cells. Energy Storage Materials 60, 102796. doi:10.1016/j.ensm.2023.102796DOI
  33. [35]Rao, L. and Newman, J. (1997). Heat-generation rate and general energy balance for insertion battery systems. Journal of The Electrochemical Society 144, 2697–2704. doi:10.1149/1.1837884DOI
  34. [36]Richardson, G. and Korotkin, I. (2021). Heat generation and a conservation law for chemical energy in Li-ion batteries. Electrochimica Acta 392, 138909. doi:10.1016/j.electacta.2021.138909DOI
  35. [37]Samsung SDI (2025). Samsung SDI begins production of 46-series cylindrical batteries and initial supply for U.S. customer. Samsung SDI News Press release, 31 March 2025, https://samsungsdi.com/sdi-now/sdi-news/4322.html, accessed 27 September 2026
  36. [38]Schmalstieg, J., Rahe, C., Ecker, M., and Sauer, D. U. (2018). Full cell parameterization of a high-power lithium-ion battery for a physico-chemical model: Part I. Physical and electrochemical parameters. Journal of The Electrochemical Society 165, A3799–A3810. doi:10.1149/2.0321816jesDOI
  37. [39]Subramanian, V. R., Diwakar, V. D., and Tapriyal, D. (2005). Efficient macro-micro scale coupled modeling of batteries. Journal of The Electrochemical Society 152, A2002–A2008. doi:10.1149/1.2032427DOI
  38. [40]Sulzer, V., Marquis, S. G., Timms, R., Robinson, M., and Chapman, S. J. (2021). Python battery mathematical modelling (PyBaMM). Journal of Open Research Software 9, 14. doi:10.5334/jors.309DOI
  39. [41]Tesla, Inc. (2020). Battery day presentation deck. Tesla, Inc. 22 September 2020, https://digitalassets.tesla.com/tesla-contents/image/upload/IR/2020-battery-day-presentation-deck, accessed 27 September 2026
  40. [42][Dataset] The Faraday Institution (2022). BPX: an open standard for physics-based battery models. https://bpxstandard.com/. Battery Parameter eXchange, developed in the Faraday Institution Multi-Scale Modelling project
  41. [43]Thomas, K. E. and Newman, J. (2003). Thermal modeling of porous insertion electrodes. Journal of The Electrochemical Society 150, A176–A192. doi:10.1149/1.1531194
  42. [44]Tranter, T. G., Timms, R., Shearing, P. R., and Brett, D. J. L. (2020). Communication—Prediction of thermal issues for larger format 4680 cylindrical cells and their mitigation with enhanced current collection. Journal of The Electrochemical Society 167, 160544. doi:10.1149/1945-7111/abad44fDOI
  43. [45]Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., et al. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature Methods 17, 261–272. doi:10.1038/s41592-019-0686-2DOI
  44. [46]Wang, A. A., O’Kane, S. E. J., Brosa Planella, F., Le Houx, J., O’Regan, K., Zyskin, M., et al. (2022). Review of parameterisation and a novel database (LiionDB) for continuum Li-ion battery models. Progress in Energy 4, 032004. doi:10.1088/2516-1083/ac692cDOI
  45. [47]Ward, L., Babinec, S., Dufek, E. J., Howey, D. A., Viswanathan, V., Aykol, M., et al. (2022). Principles of the Battery Data Genome. Joule 6, 2253–2271. doi:10.1016/j.joule.2022.08.008DOI

Paper details

vlnc:W_f5vafs3k
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Claude

View statement
Looking for related papers…
Contents