Inicio/Recursos/Conocimiento/Testing&Analysis/Battery Digital Twins: How They Work in Cell R&D

Battery Digital Twins: How They Work in Cell R&D

Canrud November 7, 2026 126

A battery digital twin is a virtual model of a specific cell, module or pack that is kept up to date with measured data from that real battery. In cell R&D it is used to test design and operating choices on the computer first, so fewer physical cells and test channels are needed to reach a decision.

The idea is simple; doing it well is not. This article explains what sits inside a twin, which lab tests feed it, how it is checked, and where it does and does not save lab time.

What is a battery digital twin?

A battery digital twin has three parts that work together:

Part

What it is

Physical twin

The real cell, module or pack being tested or used

Digital model

A model of that battery: electrical, electrochemical, thermal or data-driven

Data link

The flow of measured data (current, voltage, temperature, capacity) that keeps the model in step with the real battery

 

The data link is what separates a twin from an ordinary model. As the real cell is cycled, the model is updated with new measurements, so its estimates of capacity, resistance and health follow that cell rather than an average one.

Digital twin vs battery model vs BMS

A battery model describes how a type of cell behaves. It can be run for any conditions, but it is not tied to a specific cell.

A digital twin uses a model, but ties it to one real asset and keeps updating it with that asset's data.

A battery management system (BMS) measures and protects a pack in real time. It often runs a simple model to estimate state of charge. A twin can use BMS data, but it usually sits outside the BMS and answers broader questions, such as how fast the pack is ageing.

One practical gap shows why twins are useful. In modules with cells in parallel, the BMS usually does not measure each cell's current and temperature. A model can estimate them, which helps spot cells that age faster than their neighbours.

What models sit inside a battery digital twin?

There are four main model families. The right one depends on the question and the data you have.

Equivalent circuit models

An equivalent circuit model (ECM) represents the cell as a voltage source with resistors and capacitors. It is fast and needs only electrical test data, which is why it is common in BMS state estimation. It describes what the cell does, not why, so it cannot answer electrode design questions.

Physics-based electrochemical models

These models describe lithium transport and reactions inside the electrodes and electrolyte. The single particle model (SPM) is the simpler form; the Doyle-Fuller-Newman (DFN) model, also called the pseudo-two-dimensional or P2D model, is more detailed. Open-source tools such as PyBaMM implement both. They can answer design questions, such as what happens if the electrode is thicker, but they need many parameters.

Data-driven models

Machine learning models learn patterns from large amounts of cycling data, for example to predict remaining useful life. They can be accurate within the range of data they were trained on, but they are weak outside it, such as with a new chemistry or new operating conditions.

Hybrid models

A hybrid model combines a physics or circuit model with a data-driven correction. The physics keeps predictions sensible, and the learned part captures effects the physics leaves out. Many practical twins end up here.

What lab data builds a battery digital twin?

A twin is only as good as the test data behind it. These are the tests that typically supply its inputs:

Test

What it gives the model

Mainly used by

OCV vs state of charge

The open-circuit voltage curve, the backbone of most models

ECM, physics-based

Pulse / HPPC tests

Resistance at different states of charge and temperatures

ECM

Impedance (EIS)

How resistance splits into different processes over frequency

ECM, physics-based

GITT

Lithium diffusion in the active material

Physics-based

Cycle and calendar ageing

How capacity and resistance change over time and use

All, especially data-driven

Thermal tests

Heat generation and temperature response

Thermal sub-models

Material and electrode data

Electrode thickness, porosity, particle size, conductivity

Physics-based

 

Test conditions matter as much as the tests. A model fitted only at room temperature will not be reliable in the cold, and one fitted on new cells will drift as the cells age.

How is a digital twin calibrated and validated?

Building a twin is a loop, not a one-off:

  1. Calibrate. Fit the model's parameters to one set of test data.
  2. Validate. Run the model against data it has not seen, such as a different drive cycle or temperature, and compare.
  3. Update. As the real cell keeps cycling, feed in new data and adjust parameters that change with ageing, such as capacity and resistance.

Skipping validation is the most common shortcut. A model can match its fitting data closely and still fail on new conditions. Keep a validation set aside before fitting starts.

Common mistakes when building a battery digital twin

  • Fitting and validating on the same data. The model looks accurate but fails on anything new.
  • Testing at one temperature. Resistance and ageing depend strongly on temperature, so a single-temperature model breaks in real use.
  • Ignoring cell-to-cell variation. A model fitted to one cell may not describe the next cell from the same build.
  • Choosing the model before the question. A detailed physics model is wasted on a simple state-of-charge task, and an equivalent circuit model cannot answer an electrode design question.
  • Letting the twin go stale. If new data stops flowing in, the twin becomes an ordinary model that slowly drifts away from the real cell.

Where a twin saves lab time in cell R&D

  • Screening designs. Compare electrode thicknesses, loadings or formats on the computer and build only the most promising cells.
  • Projecting ageing. Use early cycling data to estimate how capacity will fade, and stop tests that are clearly heading the wrong way. To understand what drives that fade, see the main causes of capacity fade.
  • Narrowing the test plan. Run fewer temperature and current combinations physically, and use the model to fill the gaps between them.
  • Explaining results. A physics-based model can suggest why a cell behaved as it did, which points to the next experiment.

Where a digital twin does not replace the lab

A twin works best where you already have data. It is much weaker in these cases:

  • New chemistries and materials, where there is no data to fit.
  • Safety and abuse behaviour, such as thermal runaway, which still needs physical testing.
  • Manufacturing variation, unless the model is fed data from many real cells.
  • Untested conditions, where the model is extrapolating rather than interpolating.

In every case, the twin's output should be treated as a strong estimate that guides testing, not a replacement for it.

Where to start

If you are considering a digital twin for a cell programme, settle these five points first:

  1. The question: design choice, ageing prediction or state estimation.
  2. The model family that can answer it with the data you can get.
  3. The tests needed to fit it, and at which temperatures and states of charge.
  4. The validation data, set aside before fitting.
  5. Who owns the data and the model, and how often it is updated.

 

Frequently asked questions

Is a battery digital twin the same as a battery simulation?

No. A simulation can describe any cell of a given design. A battery digital twin is tied to a specific real cell, module or pack and is updated with that asset's measured data. The model inside a twin is often a standard simulation model; the live data link is what makes it a twin.

How much test data do you need to build a cell-level twin?

It depends on the model and the question you want answered. An equivalent circuit model can be fitted from a small set of electrical tests. A physics-based model needs many more parameters, including material and electrode data. Plan a separate validation data set as well, not just the data used for fitting.

Can a digital twin replace cycle life testing?

No. A twin can project ageing from shorter tests and help you choose which conditions to test, which can shrink a test plan. But its ageing predictions are only as good as the real ageing data used to calibrate and check it. New chemistries and safety behaviour still need physical tests.

 

Need test data to build or validate a model?

Canrud provides end-to-end battery R&D services, from cell building to electrochemical testing. Tell us your cell format and what you want the model to answer, then contact us. Work is quoted after a short scope call, all costs listed upfront.

 

Canrud provides battery R&D and testing services that produce the data models are built on. Learn more at canrud.com.