Battery Intelligence from EV Charging

What a session exposes

A charging session is a controlled experiment that happens to also deliver energy. The charger knows what it offered, the vehicle decides what it took, and the difference between those over time is informative.

• How long the session ran, and how that compares to the same vehicle's previous sessions
• How much energy went in, against what the vehicle reported it needed
• Charge acceptance across the session, which is where most of the signal lives
• Where in the state of charge range the session started and ended
• How the vehicle's acceptance changed as it filled
• Ambient conditions, which affect all of the above and have to be accounted for rather than ignored

AC and DC see different things

This is why battery intelligence is a DC proposition first. On AC the onboard charger sits between the observer and the thing being observed, and it is the onboard charger's behaviour that dominates what you can see.

Why repetition matters more than any single session

One session tells you very little. The same vehicle across many sessions tells you a great deal, and this is the property that makes charging infrastructure a good place to do this at all.

1. 1

   Establish a baseline

   The first sessions with a given vehicle characterise how it behaves now, not how healthy it is. There is nothing to compare against yet.

2. 2

   Build the pattern

   As the vehicle returns, its behaviour under comparable conditions becomes a distribution rather than a point. Conditions are rarely identical, so comparability has to be established rather than assumed.

3. 3

   Detect the trend

   Change over months is the signal. A battery declining slowly looks normal in any one session and obvious across thirty.

4. 4

   Flag the anomaly

   A session that departs from that vehicle's own established pattern is more informative than one that differs from the fleet average.

The practical consequence is that this works best where vehicles return: fleets, depots, and networks with regular users. A vehicle seen once is a vehicle you cannot say much about.

Where inference stops

This is the section that keeps the whole proposition honest, so it is deliberately blunt.

• Energy delivered is measured. State of health is inferred, and the two should never be presented in the same register
• The charger cannot see individual cells, so cell-level problems appear only as their aggregate effect, if at all
• It cannot see how the vehicle is driven or how it is parked, both of which affect degradation more than charging does
• It cannot distinguish a battery problem from a thermal management problem without more information than a session provides
• It is not a diagnostic and it is not a warranty judgement. It is a signal that something may be worth looking at

Presenting an inference with the confidence of a measurement is the failure mode of this entire category. A number without an honest accuracy statement attached is worse than no number.

Where this is worth doing

1. 1

   Fleets

   The strongest case. Known vehicles, regular return, a real decision attached: which vehicles to rotate, retire or investigate. The economic value is concrete.

2. 2

   Resale and residual value

   Battery condition dominates a used EV's value and is currently opaque to buyers. Independent evidence has obvious worth, and equally obvious integrity requirements.

3. 3

   Warranty and claims

   Repeated field observation supports or contradicts a claim in a way a single workshop test cannot.

4. 4

   Second life assessment

   Deciding whether a pack has a use after the vehicle needs condition data collected all along.

5. 5

   Driver-facing reporting

   A service the network offers rather than a decision it makes. Covered on its own page.

What has to be known before a vehicle can be assessed

The model is calibrated per vehicle model and per battery type, which is the reason it can say anything useful at all. A single population curve across every car on the road would describe none of them.

So the vehicle model and its battery chemistry and configuration have to be loaded before that vehicle's sessions mean anything. A vehicle the system has no profile for produces energy delivered and nothing else.

Where this runs is a deployment choice rather than a fixed answer. It runs at the edge, on the site, and it runs in the cloud, and which of those suits depends on how many sites there are and where the fleet history needs to live.

Where this currently stands

This is in development rather than shipping, and it is worth being plain about that.

What exists is the method and the model. What does not yet exist is a deployed installation with years of accumulated sessions behind it. If this is relevant to something you are building, we can show the capability against your requirement rather than against a brochure.