Battery Lifetime Modelling — Empirical Arrhenius, Semi-Empirical Coupled Models, and Indian Operating Profile Validation

A battery lifetime model transforms the physical and chemical degradation mechanisms described at the molecular level into a numerical prediction: given a known temperature history, SOC profile, and cycling pattern, how much capacity will remain after N years? Answering this question with engineering precision — not just "it depends" — requires a model that links the Arrhenius temperature sensitivity of SEI growth to the actual temperature distribution experienced by cells in a specific operating environment, and that accounts for the power-law scaling of cycle aging with both depth and rate.

For Indian EVs, the most critical insight from any properly calibrated model is this: the temperature distribution experienced by a car parked on-street in a North Indian summer for 8 years is substantially different from the European cycling protocol used to parameterise most commercial battery lifetime models. Bridging this gap — with Indian-specific validation data — is the open research problem whose solution will determine whether Indian EV battery warranties remain commercially defensible.

Why Lifetime Models Are Necessary

Battery chemistry is deterministic in principle — given a complete specification of the thermal and electrochemical history of a cell, one could in principle calculate the resulting degradation state. But in practice, operating a battery under unknown future conditions requires probabilistic projection: what is the expected SOH distribution across a fleet of vehicles with varying operating patterns, climate zones, and user behaviours, after a given number of years?

This fleet-level question drives every commercial decision in the EV battery value chain:

• Warranty terms: what degradation threshold, what time period, at what claim rate?
• Second-life qualification: when a pack leaves its primary EV service, how much usable capacity remains for stationary storage?
• Residual value: how does a 4-year-old EV's battery compare to a 6-year-old's, and how should this reflect in used EV pricing?
• BMS real-time algorithms: what is this specific pack's current SOH, and how much range can it reliably promise?

All of these questions require a model.

Tier 1: Empirical Arrhenius Calendar Aging Model

The simplest useful model separates calendar aging from cycle aging and parameterises each with measured laboratory data.

Calendar aging component:

Q_cal(t, T, SOC) = A_cal × exp(-Ea / (R × T)) × f(SOC) × t^0.5

Where:

• Q_cal: fraction of original capacity lost to calendar aging
• A_cal: pre-exponential factor (fitted to lab data)
• Ea: activation energy (fitted from temperature-varying experiments — typically 50–70 kJ/mol for NMC)
• f(SOC): SOC scaling function (exponential or polynomial, fitted from constant-SOC hold experiments)
• t^0.5: square root of time — reflects diffusion-limited SEI growth (parabolic kinetics: rate ∝ 1/√t as the SEI thickens and slows its own growth)

Cycle aging component:

Q_cyc(N, T, DoD, C-rate) = A_cyc × exp(-Ea_cyc / (R × T)) × g(DoD) × h(C-rate) × N^z

Where:

• N: cycle count
• z: cycle count exponent (0.5–0.7, reflecting decreasing per-cycle damage as SEI stabilises)
• g(DoD): DoD scaling (higher DoD = more damage per cycle, often polynomial or exponential)
• h(C-rate): C-rate scaling (higher rate = more damage per cycle, particularly for rates > 1C)

Total aging:

Q_total = Q_cal + Q_cyc + ε

Where ε is a small cross-term interaction (calendar aging accelerates at high SOC, which also happens to be where cycles complete — they are correlated).

Model limitations: This purely additive empirical model assumes calendar and cycle aging are independent and simply sum. They are not fully independent (high C-rate cycling generates heat that accelerates calendar aging within the same event). The error from this assumption is 2–5% SOH over a 10-year simulation.

Tier 2: Semi-Empirical Coupled Model

The Schmalstieg et al. (2014) model for NMC 18650 cells is a widely referenced example of the coupled approach:

Q_loss = (a_cal(SOC) × exp(-b/T) × t^0.75) + (c_cyc × exp(-d/(RT)) × (|ΔQ|)^1.12)

Where:

• The first term models calendar aging with a 3/4 power time dependence (empirically better than t^0.5 for this cell)

• The second term models cycle aging as proportional to total charge throughput |ΔQ|^1.12 rather than discrete cycle count — this naturally handles variable depth cycles
• Cross-terms couple the two if significant

All constants (a, b, c, d) are fitted from a designed experiment matrix covering temperature, SOC, and C-rate ranges relevant to the application.

Tier 3: Physics-Based DFN Model

The Doyle-Fuller-Newman (DFN) model solves the coupled partial differential equations governing:

• Solid-phase diffusion: ∂c_s/∂t = D_s(1/r² × ∂/∂r(r² × ∂c_s/∂r))
• Liquid-phase transport: ε_e × ∂c_e/∂t = ∂/∂x(D_eff_e × ∂c_e/∂x) + a × j_n × (1 - t₊)
• Charge conservation: ∂/∂x(κ_eff × ∂φ_e/∂x) + j = 0
• Butler-Volmer kinetics: j = j₀ × (exp(αF/RT × η) - exp(-αF/RT × η))

For lifetime prediction, SEI growth is added as a parasitic reaction on the anode:

• SEI growth flux: j_SEI = -i₀_SEI × exp(-Ea_SEI/(RT)) × exp(-α_SEI × F × η_SEI / (RT))
• SEI layer thickening: dδ_SEI/dt = -M_SEI × j_SEI / (2F × ρ_SEI)
• Lithium loss rate: dQ_Li/dt = ∫ j_SEI dA (integrated over anode surface)

This model is accurate across the full range of temperatures and C-rates without extrapolation concerns — but it requires ~50 material parameters per chemistry, and a single 10-year simulation at 1-second time resolution takes hours on a workstation.

Reduced-Order DFN Models

For on-board BMS use, the full DFN is replaced by reduced-order models (ROM): single-particle models (SPM) that assume uniform current distribution in each electrode, or state-space linearisations of the DFN that can be solved in real time. The SPM has approximately 10 parameters and runs in milliseconds. It is less accurate than the full DFN but captures the first-order temperature and rate dependence needed for BMS estimates.

Applying Models to Indian Operating Profiles

The critical input to any lifetime model is the temperature distribution experienced by the cells — not ambient temperature, but cell temperature, including soak from sun exposure and heat from operation.

Indian Urban Operating Profile (representative, Tier 2 city):

The comparison reveals the most important insight for Indian EV buyers: liquid cooling + LFP chemistry achieves European-equivalent battery longevity despite Indian thermal conditions. Air cooling + NMC + 100% SOC is catastrophically more damaging — a 20+ percentage point gap in 8-year SOH.

RUL Estimation Architecture in Production BMS

Real-time Remaining Useful Life estimation requires three components:

The Validation Gap for India

All current production battery lifetime models were primarily calibrated against laboratory data from European cycling protocols (WLTP, NEDC) and climate chamber tests designed to represent European ambient conditions. The Indian-condition validation data — from real vehicles with known initial capacity, operated in Indian conditions for 5+ years — is only beginning to emerge as India's first generation of mainstream EVs (Nexon EV launched 2020, Tiago EV launched 2022) approach their 5-year milestone.

Two specific gaps in current validation:

Thermal soak characterisation: Laboratory aging tests hold cells at constant temperature. Real Indian parking soak is a daily temperature excursion — cells start at 25°C in the morning, rise to 45–55°C mid-afternoon, fall overnight. The area under the temperature-time curve determines total aging, but the path (gradual vs. rapid heating) and its interaction with SOC during the soak are not fully captured in constant-temperature models. Data from Indian fleet deployment will enable refinement of the soak model.

DC fast charging contribution under Indian ambient: European models characterise DC fast charging degradation at 25°C. In India, DC fast charging often occurs in 35–40°C ambient — the cell starts the charge already warmer, and the thermal management system has less margin. The interaction of elevated pre-charge temperature with DC fast charging degradation rate has only partially been characterised in the open literature.