Predicting battery health for an electric bus fleet
Turning noisy telemetry into a measure of how fast each battery is ageing.
The situation
An electric bus fleet produced large volumes of raw, noisy telematics data, but no reliable way to tell how fast each battery was ageing.
The process
Pipeline
- Raw telemetry
- Cleaning and denoising
- Usable driving windows
- Statistical models
- Health per bus
What I did
- Extracted and cleaned the telemetry, and kept only the usage windows that say something about the battery.
- Studied the literature on lithium-ion degradation to choose the right indicators.
- Built probability and regression models adapted to each battery technology and bus model.
The result
A state-of-health model built on real data, so maintenance can be planned and battery replacements budgeted before they become urgent.