The platform
ZAPPP EdgeAI: Vehicle intelligence
Battery degradation is invisible until it is expensive. EdgeAI is on-vehicle AI that reads what the OEM summary never shows you. The intelligence lives on the asset, with or without connectivity.
- CAN-FD telemetry at 1Hz from a ruggedised edge device
- Anomaly detection at cell and pack level
- SOC correction for usable range on today's route
- Remaining-useful-life signals and duty-cycle classification
01 · The gap
Your OEM's SOC tells you charge remaining. Not usable range on today's route in today's conditions.
OEM state-of-charge figures are route-naive and temperature-naive. They are calibrated to laboratory conditions, and on LFP packs the flat voltage plateau makes them coarser still. EdgeAI fuses coulomb counting with sequence models trained on real fleet data, corrects the figure against route gradient, HVAC load and ambient temperature, and shows the confidence interval with it. You get the answer to the question that matters: can this vehicle complete its next duty.
02 · Edge inference
The models run on the asset. Not in a dashboard.
Model 01
Anomaly detection
Unsupervised models watch per-cell voltages, temperature gradients and charge/discharge patterns against each vehicle's own baseline. Cell imbalance, thermal spikes and early degradation inflections are flagged as they happen: no cloud round-trip, no labelled dataset required.
Model 02
SOC correction
Sequence models correct the reported figure against recent history and operating context. Updated continuously, with a confidence band.
Model 03
RUL signals
Short-horizon capacity-fade signals at the edge; full probabilistic remaining-useful-life (P10, P50, P90) from the cloud twin. A distribution, not a point estimate, because finance and warranty decisions need the uncertainty stated.
Model 04
Duty-cycle classification
Every frame is tagged with how hard the vehicle is being worked: gradient, speed, HVAC load, ambient temperature. The same health score means different things on a flat urban loop and a hilly route in 40-degree heat. Context travels with the data.
72+ hours of full telemetry buffered on-device through connectivity dropouts. Models and firmware update over the air: no hardware replacement, no vehicle recall.
03 · The twin
Every vehicle has a model. The model is its own.
Each enrolled vehicle carries a continuously updated digital twin, calibrated to its own cycle history, temperature exposure and charge behaviour, not to a fleet average. Two identical buses diverge the day they enter service; their twins diverge with them. The twin drives degradation modelling, residual-value curves, warranty evidence and second-life assessment. It is the battery's logbook, written automatically.
04 · The segmentation
Every battery in the fleet, in one of four segments.
| Segment | Rule (illustrative) | What it means |
|---|---|---|
| A · Defer | Lower-bound margin of 24 months or more, module gap of 5 points or less | Defer up to 24 months, re-test every 6 |
| B · Intervene | Pack average healthy, one module group more than 5 points behind | Rebalance or replace the module, then re-score in 90 days |
| C · Replace | Margin under 24 months, pack ageing uniformly | Replace on schedule at the gate |
| D · Diagnose | Margin under 24 months with uneven ageing or thermal flags | Diagnose within 30 days; repair and re-segment, or replace |
Degradation is non-uniform. Route profile, charging infrastructure, climate and duty cycle produce a wide distribution of health across the same fleet. The segmentation makes that distribution visible, and makes blanket replacement decisions indefensible.
05 · Two buyers
The operator reads uptime. The financier reads collateral.
The operator
Fewer roadside failures. Warranty evidence with timestamps. Vehicles matched to routes by actual health, not age. Degraded vehicles moved to lighter duties before they strand.
The financier
Independent collateral assessment. Covenant evidence. Residual-value curves with confidence bounds. Portfolio benchmarking across OEMs and climates.