Solid Phase Diffusion Modeling for Battery State Estimation
Solid phase diffusion modeling estimates active particle surface concentration, enabling aggressive fast charging without lithium plating or premature voltage cutoff.

Diffusion

Particle Gradients and Surface Concentration
Solid-state mass transport dictates how rapidly lithium ions travel from the core of an active material particle out to its electrochemical reaction interface. Inside lithium-ion cells, that movement follows concentration gradients through spherical intercalation particles, modeled via Fick’s second law of diffusion. Under heavy continuous discharge or sharp pulse-charging currents, a steep concentration profile develops across the particle radius.
While bulk state of charge tracks total lithium inventory within the particle, open-circuit voltage responds directly to lithium concentration right at the particle perimeter.
Equivalent circuit models approximate this behavior through parallel resistance-capacitance pairs. Those networks handle rapid interfacial polarization well enough, but they drift under sustained dynamic loads. Once solid-state transport hits a mass transfer bottleneck, lithium either piles up or strips away at the surface while the core remains untouched.
Terminal voltage then trips lower cutoff limits well before cell capacity is actually exhausted. Without solid-phase diffusion dynamics, an estimator simply misreads this polarization drop as sudden resistance growth or capacity fade.
The divergence between bulk particle stoichiometry and outer surface lithium concentration generates significant voltage prediction errors during multi-C dynamic cycles.
Tracking boundary concentration accurately is what keeps state of charge, state of health, and power capability estimates reliable. Running physical diffusion equations directly in onboard estimation software ties benchtop cell characterization to field pack operation. Without that boundary tracking, high-rate charging algorithms tend to throttle current too early ~ stretching out charge times ~ or too late, triggering metallic lithium plating on graphite anodes.

Electrochemical States in Battery Management
Battery management systems in stationary storage and electric mobility rely on real-time awareness of internal electrochemical states. Standard coulomb counting simply integrates terminal current over time from an assumed initial state. Over long cycles, current sensor drift, offset errors, and measurement noise inevitably erode that open-loop calculation.
Correcting the drift through static voltage lookup tables requires long rest intervals, giving internal lithium gradients enough time to fully equilibrate.
Dynamic duty cycles rarely allow the rest periods needed for thermodynamic equilibrium. Urban driving, grid frequency support, and fast charging keep cells in persistent non-equilibrium states. When estimation routines track solid-phase gradients, they calculate the overpotentials caused by surface-to-bulk concentration differences on the fly.
That calculation separates kinetic and transport bottlenecks from true, irreversible capacity loss during diagnostics.
Embedding solid diffusion physics into onboard microcontrollers requires balancing model fidelity against raw computing limits. The partial differential equations governing radial mass transport across millions of active particles cannot run directly on automotive-grade processors, especially those restricted to fixed-point math. Model reduction techniques condense spherical diffusion equations into discrete state-space forms, preserving boundary concentration outputs while respecting strict memory and cycle budgets.
Flash memory allocations and execution budgets dictate which model structure can run on a safety-critical pack microcontroller. The trade-off between low-order polynomial approximations and higher-order spectral state-space models ultimately sets how closely state of charge can be tracked during fast current pulses.
Any dynamic pulse lasting longer than ten seconds pushes graphite intercalation past surface equilibrium, requiring active diffusion tracking to prevent premature cutoff thresholds.

Particle

Governing Equations of Spherical Transport
Mass transport within an idealized spherical active material particle follows pure radial diffusion without angular dependence. Fick’s second law describes the time derivative of local lithium concentration as a function of radial position and the chemical diffusion coefficient:
The partial derivative of concentration with respect to time equals the diffusion coefficient divided by the radius squared, multiplied by the radial derivative of the radius squared times the concentration gradient. At the particle center, spatial symmetry enforces a zero-flux boundary condition, meaning the radial gradient of concentration equals zero at radius zero. At the outer particle boundary, mass flux matches the pore-wall interfacial current density divided by Faraday’s constant:
The radial derivative of concentration at the outer radius equals the negative interfacial current flux divided by the solid-state diffusion coefficient. This boundary condition directly couples the solid-phase transport equations to the Butler-Volmer electrochemical kinetics governing charge transfer across the solid-electrolyte interphase.
| Material Chemistry | Mean Particle Radius (microns) | Solid Diffusion Coefficient (cm2/s) | Theoretical Capacity (mAh/g) | Diffusion Time Constant (seconds) |
|---|---|---|---|---|
| Lithium Iron Phosphate (LFP) | 0.08 to 0.50 | 1.0e-14 to 5.0e-13 | 170 | 12.8 to 500 |
| Nickel Manganese Cobalt 811 (NMC811) | 3.50 to 8.00 | 2.0e-11 to 8.0e-10 | 200 | 153 to 3200 |
| Lithium Nickel Cobalt Aluminum (NCA) | 4.00 to 10.00 | 1.0e-11 to 5.0e-10 | 275 | 320 to 10000 |
| Mesocarbon Microbead Graphite (MCMB) | 6.00 to 12.00 | 5.0e-11 to 3.0e-09 | 372 | 120 to 2880 |
| Silicon-Graphite Composite (5wt% Si) | 2.00 to 6.00 | 1.0e-12 to 4.0e-10 | 450 | 100 to 4000 |
The diffusion time constant represents the square of the particle radius divided by the solid diffusion coefficient. Active materials with sluggish diffusion coefficients, such as olivine lithium iron phosphate, offset slow transport by reducing particle sizing down to nanoscale dimensions. Layered transition metal oxides rely instead on secondary spherical agglomerates built from smaller primary crystallites, forming multi-modal diffusion networks inside each macroscopic particle.

Discretization and State-Space Formulation
Transforming continuous spherical partial differential equations into an onboard state estimator involves spatial discretization. The finite difference method divides the particle radius into concentric spherical shells of uniform or progressive thickness. Applying spatial derivatives across multiple radial shells converts the governing partial differential equation into a set of coupled ordinary differential equations suitable for matrix operations.
Polynomial profile approximations simplify spatial modeling by assuming a predefined polynomial concentration profile across the particle radius. A parabolic approximation uses an average concentration and a surface concentration variable to track mass transfer. Higher-order approximations incorporate volume-averaged concentration, surface concentration, and volume-averaged concentration gradient as dynamic states.
These reduced formulations preserve mass conservation within the particle while avoiding high-dimensional matrix inversions in BMS microcontrollers.
State estimators employ the single particle model framework when liquid-phase electrolyte limitations are negligible, typically at rates below two C. The single particle model represents each electrode as an equivalent spherical particle subjected to area-averaged current densities. At higher C-rates, electrolyte concentration gradients along the electrode thickness induce non-uniform reaction current distributions, degrading single particle assumptions.
- Electrode thickness profiling identifies where spatial variations in current density overpower single-particle assumptions, triggering two-particle or pseudo-two-dimensional frameworks.
- Effective diffusion scaling adjusts laboratory-measured transport coefficients through Bruggeman tortuosity factors, matching physical electrode microstructure realities.
- State dimension truncation caps the internal discretization order between three and five radial shells to preserve microcontroller execution speed while limiting truncation error under two percent.
Ignoring particle size distribution introduces measurable mismatch under dynamic charging profiles. Real electrode coatings contain polydisperse particle networks where smaller particles saturate rapidly while larger particles lag, shifting local current distributions across the electrode depth.
While single-particle diffusion parameters are frequently assumed to reflect real pack performance, cell validation procedures routinely uncover premature overpotential cutoffs driven by unmodeled particle size distribution spreads.

Estimator

Observer Architectures and Convergence
Translating solid-phase transport formulations into real-time algorithms requires closed-loop observer designs. Open-loop model execution allows cumulative parameter errors and unmodeled side reactions to compound over time. A closed-loop state observer continuously adjusts internal concentration states using terminal voltage feedback, correcting discrepancies between measured cell voltage and simulated surface potential.
Extended Kalman filters linearize non-linear electrochemical output equations around local operating points at each execution interval. The non-linear relationship between surface concentration and open-circuit voltage introduces Jacobian matrix calculations that must be evaluated within the microcontroller task cycle. Unscented Kalman filters mitigate linearization inaccuracies across steep open-circuit voltage plateaus by projecting deterministically chosen sigma points through the non-linear measurement model, improving convergence stability at the expense of higher floating-point instruction counts.
Observer convergence rates hinge directly on the slope of the open-circuit voltage curve relative to surface stoichiometry, rendering flat-plateau chemistries vulnerable to gain instability.
Particle filters evaluate non-Gaussian noise distributions and extreme non-linearities by propagating hundreds of weighted state candidates. Computational load restricts particle filters from production automotive controllers, reserving their use for cloud-based telemetry analytics and pack life-cycle profiling. Sliding mode observers provide alternative state estimation schemes, utilizing discontinuous switching functions to drive estimation error trajectories onto sliding manifolds, demonstrating robust performance against parameter uncertainties and current sensor noise.
| Observer Algorithm | Computational Complexity | Memory Footprint | Non-Linear Tracking Capability | Tuning Parameter Count |
|---|---|---|---|---|
| Extended Kalman Filter (EKF) | Moderate (O(N^3) matrix inversion) | Low (4 to 8 KB RAM) | Moderate (first-order Taylor series) | 6 to 12 covariance entries |
| Unscented Kalman Filter (UKF) | High (2N+1 sigma point passes) | Medium (8 to 16 KB RAM) | High (unscented transform accuracy) | 8 to 14 covariance and scale terms |
| Sliding Mode Observer (SMO) | Low (vector operations) | Low (2 to 4 KB RAM) | High (robust boundary enforcement) | 3 to 6 boundary layer gains |
| Dual Extended Kalman Filter (DEKF) | Very High (coupled state and parameter) | High (16 to 32 KB RAM) | High (tracks concurrent aging shifts) | 12 to 24 covariance entries |
Observability conditions determine whether solid-phase concentration states can be mathematically reconstructed from terminal current and voltage measurements alone. During prolonged current rest intervals, the terminal voltage relaxes as diffusion gradients decay. In these rest regimes, the sensitivity of terminal voltage to radial concentration gradients drops toward zero, temporarily diminishing observer correction authority.

Worked Estimation Arithmetic
Evaluating observer execution timing requires precise tracking of state-space dimensions and mathematical operations. Consider an electric vehicle battery pack governed by a central management unit with an operating task frequency of ten hertz, providing one hundred milliseconds per cycle. Assume the pack estimator runs a reduced-order solid phase diffusion model implemented across an electrode pair using five radial discretization nodes per electrode, generating ten total state variables.
The state-space model executes discrete updates where the state vector at time step k plus one equals the system transition matrix multiplied by the state vector at time step k, plus the input vector multiplied by terminal current. A state dimension of ten yields a transition matrix containing one hundred floating-point coefficients. Updating the state vector requires one hundred multiplications and ninety additions, totaling one hundred ninety floating-point operations per cell node.
In a pack design comprising ninety-six series-connected cell monitoring nodes, computing the physical states independently for each cell channel scales the state prediction arithmetic to eighteen thousand two hundred forty floating-point operations per cycle. The Extended Kalman Filter observation update adds matrix inversion and covariance propagation steps, demanding approximately eight times the raw state calculation budget:
Eighteen thousand two hundred forty operations multiplied by eight equals one hundred forty-five thousand nine hundred twenty floating-point operations per measurement cycle. On an automotive microcontroller rated at one hundred fifty megahertz with a single-precision floating-point unit processing one instruction per clock cycle, executing this routine consumes approximately one millisecond of processor core time. This allocation represents exactly one percent of the total one-hundred-millisecond execution window, leaving ninety-nine percent of processor resources available for thermal controls, communications, isolation checks, and diagnostics.
Parameter divergence threatens observer execution integrity over field operating lifespans. As batteries age, cathode particle micro-cracking and anode graphite exfoliation alter effective diffusion paths, shifting true diffusion coefficients away from factory baseline values. Dual estimation architectures update diffusion coefficients concurrently with concentration states, preventing divergence at the expense of doubling matrix operations.
Deploying state estimators with static diffusion coefficients into dynamic field environments generates cumulative state of charge errors exceeding twelve percent as cells approach end-of-life resistance milestones.

Stress

Diffusion-Induced Mechanical Strain
Intercalation and de-intercalation of lithium ions induce volumetric expansion and contraction within host crystal structures. When solid phase diffusion is slow relative to the applied current, severe radial concentration gradients establish internal mechanical stress distributions within individual particles. The particle shell attempts to expand or contract against a constraining particle core, generating tensile and compressive stress profiles across the active radius.
Tangential stress at the outer particle surface shifts into tension during high-rate discharge cycles as lithium leaves the perimeter faster than internal transport can replenish it. In layered oxide cathode materials such as NMC811, continuous unit-cell volume variations reach eight percent across complete state-of-charge swings. These localized stress differentials regularly exceed the critical fracture toughness of the crystal grains, initiating intra-granular and inter-granular micro-cracks.
Mechanical grain fracture inside secondary particle agglomerates isolates active crystallites from the conductive carbon matrix, triggering sudden drops in observable cell capacity.
Micro-cracks propagate along grain boundaries, cleaving secondary agglomerates and exposing fresh, unpassivated transition-metal surfaces to liquid electrolyte decomposition reactions. Transition metal dissolution accelerates at these newly exposed fracture interfaces, leading to manganese, cobalt, and nickel migration across the porous separator toward the anode. Active material cracking directly decreases the effective solid diffusion coefficient by severing continuous crystallographic lithium pathways, permanently increasing cell transport resistance.

Fast-Charging Boundaries and Lithium Plating
High-rate charging presents the most severe operational threat to cell health through diffusion-induced overpotentials. During charge, lithium ions enter graphite anode particles, driving surface concentration toward theoretical saturation stoichiometry. If the local surface overpotential drives the anode solid-liquid interfacial potential below zero volts versus the lithium reference electrode, metallic lithium deposition occurs directly on the particle surface.
Solid phase diffusion modeling calculates anode surface potential dynamically, providing an essential software safeguard against plating conditions. When the model detects graphite surface stoichiometry approaching unity, the charging controller dynamically limits input current, throttling charge rates to match internal diffusion relaxation speeds. Without precise particle surface tracking, charging profiles rely on conservative rule-based step-charging protocols, which unnecessarily lengthen charging cycles or operate blind to localized plating risks.
- Interfacial overpotential depression forces metallic lithium nucleation when graphite surface stoichiometry reaches saturation under high-rate charging conditions.
- Dead lithium generation occurs when stripping cycles sever electrical contact between deposited lithium dendrites and the anode current collector, driving rapid capacity loss.
- Solvent consumption cascades accelerate as metallic lithium reacts with organic alkyl carbonates, producing thick, resistive secondary passivating layers.
- Thermal runaway sensitivity escalates when low-melting-point metallic lithium lowers cell self-heating onset temperatures down toward one hundred degrees Celsius.
Under low-temperature charging regimes, solid-state diffusion rates plummet by an order of magnitude for every twenty-degree Celsius reduction in ambient temperature. Anode diffusion restrictions govern charge acceptance under cold operating conditions, transforming moderate charge pulses into severe degradation events.
Should model execution fail to account for Arrhenius-driven diffusion attenuation at sub-zero operating temperatures, fast-charging controls inevitably trigger metallic dendrite growth, generating irreversible field failures within dozens of cycles.

Aging

Parameter Drift over Long-Term Cycling
Solid-state diffusion properties do not remain constant over the service life of an energy storage system. Long-term cycle and calendar aging systematically alter the transport kinetics, active surface areas, and solid diffusion coefficients within positive and negative electrodes. Progressive phase transitions, lattice disordering, and transition metal reduction degrade crystallographic tunnels, suppressing intrinsic diffusion coefficients by up to seventy percent over one thousand equivalent full cycles.
Surface film thickening further complicates parameter identification. The gradual growth of the solid-electrolyte interphase layer on carbonaceous anodes consumes cyclable lithium inventory and impedes mass transport across particle boundaries. State estimators tracking diffusion must adapt to these shifting physical values or risk substantial state estimation skew.
Tracking diffusion parameter evolution over time serves as an advanced diagnostic indicator of physical health, isolating interfacial impedance growth from deep bulk material degradation.
| Cycle Count | Retained Capacity (%) | Cathode Diffusion Coeff (cm2/s) | Anode Diffusion Coeff (cm2/s) | Solid Transport Overpotential (mV) |
|---|---|---|---|---|
| 1 (Fresh) | 100.0 | 3.8e-10 | 1.2e-09 | 18 |
| 250 | 96.2 | 3.1e-10 | 9.8e-10 | 24 |
| 500 | 91.5 | 2.2e-10 | 6.4e-10 | 35 |
| 750 | 86.1 | 1.5e-10 | 3.8e-10 | 52 |
| 1000 | 79.4 | 9.2e-11 | 1.9e-10 | 88 |
| 1250 | 71.8 | 4.5e-11 | 8.5e-11 | 145 |
Evaluating parameter sensitivity reveals that cathode diffusion attenuation dominates transport-induced impedance growth in nickel-rich layered chemistries. Lattice collapse during repetitive high-voltage extraction blocks lithium pathways, forcing localized current redistributions and accelerating secondary particle disintegration.

Electrochemical Impedance Spectroscopy Validation
Validating solid-state diffusion models requires precise parameter extraction through laboratory characterization. Electrochemical impedance spectroscopy applies small-amplitude sinusoidal current perturbations across frequency sweeps from hundreds of kilohertz down to single-digit millihertz. Solid-phase diffusion manifests in the lowest frequency domain, presenting as an inclined linear Warburg impedance tail on a Nyquist plot.
At low frequencies, where the diffusion penetration depth exceeds particle radii, the impedance response transitions from semi-infinite Warburg diffusion with a forty-five-degree phase angle to finite-length capacitive behavior characterized by a steep vertical phase trajectory. Extracting the solid diffusion coefficient involves calculating the slope of the real impedance plotted against the reciprocal square root of angular frequency within the Warburg transition zone.
Galvanostatic intermittent titration testing provides complementary transient transport tracking. Current pulses applied for discrete intervals disrupt concentration equilibrium, followed by extended relaxation intervals where open-circuit potential decay is monitored. Calculating chemical diffusion coefficients from titration curves relies on the square root of time potential response, requiring rigorous control over thermal fluctuations during multi-week test routines.
Discrepancies frequently emerge between titration-derived coefficients and impedance-derived values across identical cell lots. Titration measurements subject cells to broad concentration excursions, capturing macroscopic state shifts, whereas impedance spectroscopy probes infinitesimal perturbations around fixed equilibrium states.
Standard clauses in high-reliability energy procurement agreements stipulate that cell qualification dossiers must include low-frequency impedance verification down to ten millihertz, establishing baseline solid diffusion parameters across five distinct operating temperatures.

Execution

Microcontroller Implementation Constraints
Deploying physics-based state estimators into embedded automotive and stationary storage hardware requires strict compliance with embedded software constraints. Production battery management controllers operate real-time operating systems with deterministic task execution budgets. Dynamic memory allocation is prohibited in safety-critical automotive software stacks, requiring all state-space matrices, covariance vectors, and lookup arrays to reside in static, pre-allocated memory spaces.
Microcontroller architectures in cell supervision circuits vary widely, ranging from simple eight-bit peripheral processors handling voltage monitoring up to powerful thirty-two-bit dual-core lockstep microcontrollers running primary pack estimation algorithms. Fixed-point microcontrollers demand rigorous numerical scaling of state variables to prevent integer overflow while maintaining resolution across tiny concentration differentials. Modern architectures featuring dedicated hardware floating-point units eliminate fixed-point scaling hazards, executing double-precision arithmetic without execution penalties.
- Spatial model reduction truncates continuous diffusion systems into sparse tri-diagonal matrix structures, minimizing floating-point operations during state transition updates.
- Lookup table interpolation accelerates the evaluation of open-circuit voltage gradients and non-linear kinetic overpotentials, avoiding trigonometric and transcendental runtime library calls.
- Asynchronous scheduling executes fast ohmic current corrections in ten-millisecond task loops while delegating slow solid diffusion particle updates to one-hundred-millisecond foreground threads.
- Static memory partitioning reserves fixed flash memory segments for electrochemical parameter matrices, isolating model calculations from low-level communication drivers.
A typical software failure mode involves numerical instability inside recursive observer covariance matrices. Repeated subtraction operations in Kalman filter gain updates can cause positive semi-definite covariance matrices to lose mathematical symmetry, producing negative variance states and causing the state estimator to crash. Implementing square-root filtering algorithms, such as the Potter or Bierman-Thornton UD factorizations, guarantees numerical symmetry and ensures stable long-term observer execution under noisy sensor inputs.

Commercial Stakes of Physics-Based Algorithms
The transition from empirical equivalent circuit modeling to physics-informed solid-phase diffusion estimation transforms pack commercial viability. Traditional battery management strategies utilize broad operating safety margins, derating usable capacity bands between ten and ninety percent state of charge to prevent inadvertent overpotential excursions and lithium plating. This empirical buffering leaves twenty percent of purchased cell energy unutilized over field operating lifespans.
Accurate solid-state concentration tracking reduces required safety margins by calculating true internal interfacial states dynamically. Physics-based software architectures safely expand usable state of charge windows from five to ninety-five percent, reclaiming ten percent of available pack energy density without accelerating cell degradation rates. In a high-volume manufacturing run of passenger electric vehicles with eighty-kilowatt-hour packs, a ten percent increase in usable energy equates to eight kilowatt-hours of reclaimed capacity per pack.
At an average automotive pack procurement cost of one hundred dollars per kilowatt-hour, reclaiming eight kilowatt-hours represents eight hundred dollars in unspent cell procurement costs per vehicle platform. Across an annual production volume of one hundred thousand vehicles, this software optimization yields eighty million dollars in avoided capital expenditure. Concurrently, physics-informed charge tracking eliminates plating overpotentials, reducing premature pack replacement warranty liabilities over multi-year operational warranties.
Procurement teams negotiating long-term cell supply agreements increasingly encounter warranty disputes regarding early capacity fade versus transport-induced power degradation. Establishing whether field performance losses stem from irreversible chemical degradation or unmodeled solid-phase diffusion impedance determines whether warranty claims fall on the cell manufacturer or the pack integrator’s software implementation.
How battery pack integrators will legally allocate warranty liabilities when proprietary BMS diffusion observers disagree with cell manufacturers’ physical degradation thresholds remains contested across major international energy storage contracts.





