SecondLife Grader
Retired EV batteries often still hold 70–90 % of their capacity, but proving that takes a 4-hour capacity test per module, so most go straight to the shredder. This grader predicts state-of-health, remaining life and knee-point risk from a 3-minute pulse test plus BMS history. It then sends each module to the second-life market where it's worth the most, and flags the few modules whose uncertainty could actually change that decision.
Random forests trained on 5,600 simulated modules · semi-empirical ageing modelPulse-test a module
Markets, ranked by value
Triage a truck-load
Every module gets the 3-minute test. A module only goes to the full capacity test when its uncertainty band crosses a market boundary. Market capacity is limited, so the best modules are placed first.
Predicted vs. true state of health
What drives the prediction
Why it matters
Europe will retire hundreds of GWh of EV batteries over the next decade. A capacity test ties up equipment for hours per module, so in practice grading is the bottleneck for reuse. A fast, uncertainty-aware grader turns "test everything or recycle everything" into a per-module decision with known risk.
Method
- Semi-empirical ageing: √t calendar fade, power-law cycle fade, chemistry-specific knee point, temperature, fast-charge and DoD stressors.
- Three random forests (SoH, remaining life, knee risk) on 9 features. The tree spread is calibrated to 90 % intervals on held-out modules.
- Decisions use the lower bound. A full test is requested only when the upper bound would unlock a different market (value of information).
Limits & next steps
- Synthetic data: the real test is validating against lab datasets of aged cells and pulse tests.
- Market values (€/kWh) are illustrative placeholders.
- Planned: EIS spectra as extra features; active learning to pick which modules to fully test.