The science

How we determine the lifetime
of a battery system

We publish our method because a forecast that cannot be audited is not much use for a decision. This page sets out what we observe, how it becomes a degradation trajectory, how we validate it against measured outcomes, and — just as important — what this approach cannot tell you.

1 · Behaviour is observable

A battery reveals how it is being used through how it operates in its market. That behaviour — not a datasheet — is what determines how it ages, and it can be characterised at fleet scale.

2 · The physics is known

Lithium-ion ageing follows mechanisms documented over two decades: calendar loss driven by time, temperature and resting state of charge, and cycling loss driven by the depth and frequency of each cycle.

3 · Scale is what makes it accurate

Generic literature parameters are a starting point. Every additional asset, chemistry, climate and vintage in the corpus sharpens the prior for all the others. Accuracy is a function of coverage.

The pipeline

From operating behaviour to an auditable forecast

Assemble

We continuously assemble what is knowable about each asset: how it charges and discharges, what services it provides, its configuration and chemistry, the climate it sits in, and — for clients — its own plant telemetry. Every input is versioned with its provenance and confidence.

Reconstruct

Power flows become an energy state trajectory: a state-of-charge profile, from which we derive equivalent full cycles, throughput, effective C-rate and time spent at high state of charge — the variables ageing actually responds to.

Count cycles

Rainflow counting — the fatigue-analysis standard adapted to batteries — decomposes the profile into individual cycles and their depth. Depth distribution matters enormously: one deep cycle damages far more than several shallow ones moving the same energy.

Project

The fingerprint drives a calendar + cycling ageing model parameterised by chemistry, with Monte Carlo propagation of parameter uncertainty. Output: a state-of-health trajectory with p05–p95 band, remaining life as a range, and augmentation timing.

Calibrate

Where measured outcomes exist — capacity demonstrations, accredited capacity, published fleet performance, client capacity tests — the model is calibrated against them. Bands narrow as evidence accumulates, per asset and across the fleet.

The degradation model

Calendar plus cycling, with uncertainty propagated

We use the semi-empirical family that constitutes the academic and industrial standard — the same foundation used by national laboratories and the leading vendors:

  • Calendar loss: a function of time (∝ t0.5–0.75 depending on chemistry), temperature (Arrhenius) and mean resting state of charge. Minimised near 50% SoC; penalised by sustained high SoC and heat.
  • Cycling loss: the sum of a stress function Φ(d) over the cycles identified by rainflow, with Φ strongly non-linear in depth d — a cell cycled at 100% DoD degrades on the order of ten times more than at 10% DoD for the same energy throughput.
  • Datasheet seeding: the cycling model is seeded from the manufacturer's cycle-life-versus-depth curve, then corrected by observed operation — so results are useful from day one, without waiting years for history.
  • Reproducible Monte Carlo: uncertain parameters are sampled and the trajectory is reported as a p05–p95 band with a fixed seed — the same report can be regenerated identically in an audit.

Methodological references: ageing models after Xu et al. (IEEE Transactions on Smart Grid) and Schmalstieg et al. (Journal of Power Sources, 2014); rainflow counting and depth-of-discharge stress functions after Shi et al. (University of Washington, 2017); chemistry parameters cross-checked against NREL's open-source BLAST-Lite. Full assumptions accompany every report.

Field evidence that anchors the model

Published evidenceValue
ERCOT fleet (Modo Energy, 2025)≈7% / 365 EFC
Great Britain fleet (Modo Energy, 2024)≈4.4% / 365 EFC
Utility BESS, Italy — 3 years (Energies, 2026)≈1.4% / year
Grid-scale LFP, literature (first decade)2–3% / year
Peak shaving vs frequency regulation≈1.8× degradation
BESS failure incidents 2018→2025 (EPRI)rate −99%

The ERCOT-versus-Britain gap is the point: two fleets of similar technology ageing at rates that differ by 1.6× because they are operated differently. Degradation is a behavioural outcome, and behaviour is measurable.

What we do not do

  • We do not publish accuracy figures without independent validation.
  • We do not issue a single lifetime number — always a range with its assumptions.
  • We do not promise guaranteed savings; economics come with sensitivities.
  • We do not let one client's confidential data appear in another's report.
The evidence base

What the model is built on

Six independent classes of evidence. No single one is sufficient; together they are what makes a forecast defensible.

Layer 1

Operating behaviour

How storage assets are actually charged and discharged across the world's power markets, at the finest resolution each market allows — the basis for every cycling fingerprint and the only way to know how an asset is truly being used.

time seriesper assetcontinuous
Layer 2

Asset configuration

Capacity, duration, chemistry, cell and container model, integrator and commissioning date. Without this layer a time series is uninterpretable — degradation modelling is cohort modelling, and the cohort is defined here.

registersfilingsdatasheets
Layer 3

Environment

Ambient and site thermal history at hourly resolution anywhere on earth. Temperature is the second driver of calendar ageing after time, and it is why an identical system ages differently in Atacama, Texas and Scotland.

reanalysishourlyglobal
Layer 4

Measured outcomes

Capacity demonstrations, accredited and de-rated capacity, published fleet performance, incident and failure records, and client capacity tests. This is the ground truth that turns a model into a validated model.

ground truthvalidation
Layer 5

Cell science

Laboratory cycling and ageing datasets, peer-reviewed degradation models and open-source reference implementations, spanning chemistries from LFP and NMC to sodium-ion and LMFP.

laboratorypeer-reviewed
Layer 6

Your plant

BMS exports, meter data, temperatures, capacity test results and configuration. The layer that converts a well-founded estimate into a measurement of your asset — and the one that narrows the band the most.

client dataisolatedoptional
Methodological honesty

What observation can — and cannot — tell you

From observed behaviour alone

  • Equivalent full cycles, depth-of-discharge distribution and throughput
  • Duty-cycle intensity benchmarked against comparable assets worldwide
  • A first health trajectory with an explicit uncertainty band
  • Changes in operating pattern that signal a problem or a strategy shift

Only with plant telemetry

  • Measured state of health — verified capacity and internal resistance
  • Cell and container temperatures, the dominant local ageing driver
  • Rack imbalance and module-level health
  • Bands tight enough for warranty and augmentation decisions

This is why the service is progressive. We tell you what we can see, and we label it as what it is. Your data turns an estimate into a measurement. We never present the first as the second.

Want your technical team to audit the method?

We will walk through the pipeline, the assumptions, the references and a reproducible worked example on an asset of your choosing.

Book a technical session