The Measurement Problem Nobody Talks About
Ask most BESS operators what their State of Health is, and they'll pull a number from their BMS dashboard. That number comes from somewhere — typically a capacity look-up table embedded in firmware by the cell manufacturer, indexed against cycle count and temperature history. It's an estimate derived from population-averaged aging curves measured in a laboratory, not from your specific cells, your specific load profile, or your specific installation environment.
The result is a reading that tends to drift optimistically over time. Population-averaged curves assume cells age uniformly. Real installations don't. String imbalance, localized thermal gradients, uneven State-of-Charge (SoC) distribution across parallel branches — all of these accelerate fade in specific cells while neighbors remain healthier. The look-up table sees average behavior; your asset is experiencing worst-case cell behavior in some strings and best-case in others.
By the time the BMS SoH drops below a warranty trigger — typically 80% of rated capacity for NMC chemistries — the actual degraded cells may already be at 72–74% capacity. You've been dispatching against a nameplate you no longer have.
What First-Principles Measurement Actually Means
First-principles SoH measurement means inferring battery state from electrochemical observables rather than calendar-and-cycle counters. There are four primary methods used in field deployments, and they're not mutually exclusive:
Capacity Test (Full Charge / Discharge)
The most direct method: discharge the pack to the low-voltage cutoff at a controlled C-rate (typically C/5 to C/10), integrate current over time, compare to nameplate. This gives actual remaining capacity in Ah or MWh with minimal modeling assumptions. The catch: a full capacity test takes the asset offline for 4–10 hours depending on pack size and C-rate, and introduces a significant DoD cycle that itself accelerates aging. For utility-scale BESS in active dispatch, capacity tests are practical once per quarter at most.
Incremental Capacity Analysis (ICA)
ICA — sometimes called dQ/dV analysis — computes the derivative of capacity (Q) with respect to voltage (V) during a slow charge or discharge. The resulting curve shows characteristic peaks and troughs that correspond to electrochemical phase transitions in the electrode materials. In fresh NMC cells, you'll see a prominent peak near 3.7–3.9 V corresponding to the lithium ordering transition in the cathode. As the cell ages, this peak flattens and shifts: peak height reduction correlates with loss of active lithium inventory (LAM-LLI), and peak position shift indicates structural changes in the electrode lattice.
ICA doesn't require taking the asset offline for a deep discharge — you can compute it from partial charge cycles that occur naturally during operation, provided the C-rate is slow enough (typically <C/5) and the voltage data is sampled at high enough resolution (1-second intervals or finer). A fleet of 50 MWh NMC assets cycling daily will generate usable ICA data from normal operations within 2–3 weeks.
Electrochemical Impedance Spectroscopy (EIS)
EIS measures the complex impedance of a cell across a range of AC frequencies — typically from 10 mHz to 10 kHz. The resulting Nyquist plot decomposes into components corresponding to distinct electrochemical processes: ohmic resistance (R0), charge-transfer resistance (RCT), Warburg diffusion (representing lithium-ion transport through the solid electrolyte interphase, or SEI), and double-layer capacitance.
Aging manifests in EIS as a growing R0 (SEI thickening and contact degradation) and an enlarged RCT arc (reduced active surface area). EIS is the most information-rich diagnostic, but traditionally required benchtop equipment. Field-deployable EIS — injecting a small AC perturbation through the existing inverter — is now achievable with careful signal processing, though the technique is sensitive to temperature and SoC at the time of measurement.
dQ/dV Differential Voltage Analysis (DVA)
Related to ICA but computed as dV/dQ (voltage derivative with respect to capacity), DVA identifies the same electrode phase transitions from a different mathematical vantage. DVA is less sensitive to measurement noise than ICA in some cell chemistries, particularly LFP, where the voltage plateau is flat enough that the dQ/dV peak structure is compressed and harder to resolve.
Why Percentage Capacity Is the Wrong Primary KPI
Here's where the subtle issue appears. Most operator dashboards — and most warranty agreements — anchor SoH to a single capacity percentage: "battery is at 91% health." This collapses a rich degradation landscape into a single scalar that obscures the mechanism and location of fade.
Consider a hypothetical 50 MWh NMC installation, 24 months into operation, with BMS-reported SoH of 93%. ICA analysis of the same pack reveals two distinct string clusters: 18 strings show ICA peak heights consistent with ~96% capacity, while 6 strings show peak degradation consistent with ~79% capacity. Fleet average: 93%. But those 6 strings are already in warranty-gap territory, contributing disproportionately to capacity fade acceleration and creating thermal runaway risk at high SoC. The single-number SoH hides this entirely.
The more useful KPI structure pairs capacity percentage with a string-level dispersion metric. At Cellanchor we track SoH quartile spread — the difference between the 25th and 75th percentile string SoH within a pack. A spread of less than 3 percentage points is healthy; above 8 points, maintenance prioritization is warranted regardless of the fleet average. This is the kind of measurement that changes a dispatch decision: if your worst-quartile strings are below 80%, you shouldn't be running a full-depth RegD discharge cycle, even if the BMS says 91%.
The Physics-Based Model vs the Curve-Fit Lookup
A physics-based SoH model tracks the underlying electrochemical state variables: active lithium inventory (nLi), active material concentration in the cathode and anode, SEI layer thickness, and electrode porosity. These don't have to be solved in full from first principles at every timestamp — a semi-empirical approach uses electrochemical parameterization for the governing equations but allows key parameters to be fit to field measurement data in real time.
The operational advantage over a pure look-up table is drift correction. A look-up table has no feedback mechanism — it can't know that your cells are aging faster than the population average because your installation runs at an average ambient of 34°C instead of the standard 25°C. A physics-based model, fed real temperature and voltage telemetry, continuously updates its parameter estimates. After 6–8 weeks of operation, the model's SoH estimate for your specific strings diverges meaningfully from the manufacturer's generic curve — often by 3–7 percentage points, in the direction of faster-than-expected fade for high-temperature deployments.
Where This Matters for Operators: A Realistic Scenario
Consider a mid-size utility operating a 100 MWh NMC installation in a Texas climate — ambient temperatures regularly reaching 38–42°C in summer months. The manufacturer's warranty quotes 0.03% capacity fade per equivalent full cycle (EFC) at 25°C. At 250 EFCs per year, that's 7.5% fade per year, reaching the 80% warranty floor after roughly 2.7 years.
Arrhenius-corrected calendar aging at 38°C mean temperature accelerates this. Depending on the specific NMC cathode formulation, the Arrhenius activation energy for SEI growth (Ea) falls in the 50–75 kJ/mol range. A 13°C increase from the reference temperature corresponds to roughly 1.8–2.3× acceleration in calendar aging rate. If that installation hits 80% SoH in under 2 years instead of 2.7, the operator is looking at a warranty dispute based on cycle count alone, while the real culprit is thermal calendar aging that the cycle-count contract doesn't capture.
First-principles SoH monitoring doesn't prevent that aging — no monitoring system can. What it does is give the operator documented, timestamped evidence of what the cells were actually experiencing: temperature logs, ICA-derived active material loss curves, impedance growth records. That evidence matters when the warranty conversation arrives.
What We're Not Saying
We're not saying that BMS-reported SoH is useless. For basic monitoring and alarm thresholds, manufacturer firmware performs adequately. We're also not saying that EIS should be running continuously on every pack — that's technically demanding and not always justified by the asset value and dispatch intensity. The right measurement stack depends on asset size, chemistry, dispatch profile, and warranty exposure.
What we're arguing is that for assets above 10 MWh in active dispatch markets — FCAS, RegD, energy arbitrage — the cost of SoH measurement infrastructure is small relative to the value of the decisions it informs. A 3% SoH overestimate on a 50 MWh pack running 250 EFCs annually at $150/MWh FCAS revenue equals roughly $56K/year in overbid revenue at risk. The economics of measurement are not subtle.
First-principles characterization — ICA from operational charge cycles, impedance tracking from periodic EIS, capacity tests quarterly — provides the information layer that transforms SoH from a firmware readout into an operational decision variable.