Battery reliability · KIT historical-data replay

Which cells are drifting—and does EIS improve the warning?

Move through the check-ups and watch 228 measured KIT cells age along their historical timelines. At each point, the page compares two feature views: B uses capacity history and experimental context; D_CORE adds current EIS to the same model. Three check-ups later, the forecast is checked against the SoH that was actually measured.

Development OOF · not a final model · not a production alarm

Source attribution: Matthias Luh and Thomas Blank (2024). This page uses transformed, derived replay data. KIT v2 result data ↗ CC BY 4.0 ↗

What you can do

Press play to move all 228 cells through their recorded check-ups.
Switch among calendar, cyclic, and driving-profile aging.
Switch HGB / ElasticNet to see how model choice changes EIS's incremental value.
Select any cell to inspect capacity history, future-SoH forecasts, and the measured answer three check-ups later.

What the demo is actually asking

It is not merely estimating current SoH—the measured capacity already gives that. The useful question is whether data available now can forecast the SoH loss three check-ups ahead, and whether adding EIS improves that future forecast consistently.

The most honest answer so far

There is no universal gain. For cyclic aging with HGB, adding EIS lowers MAE by 0.625 SoH percentage points. But cyclic ElasticNet gets clearly worse, calendar is essentially unchanged, and profile results are mixed. EIS value depends on aging mode and model family.

Open the interactive demo full screen ↗