DikshantBattery RUL predictor
Two-stage model for the remaining useful life of Li-ion cells; a physics-informed network estimates degradation features from BMS signals, then a tree and neural-network ensemble predicts RUL.
Model details
- Architecture
- Physics-informed neural network, then gradient-boosted trees and neural-network ensemble
- Input
- BMS signals (voltage, current, temperature, resistance, capacities, dQ/dV and dV/dQ peaks), cycle, C-rate, depth of discharge
- Output
- RUL (cycles)
- Training data
- Not stated
- Uncertainty
- Not reported
- Physics constraints
- Monotonic degradation features, Arrhenius temperature gating and non-negative outputs, from SEI, lithium-loss, active-material and plating models
- License
- mit