Interpretable noninvasive framework for lithium-ion battery degradation diagnosis: Joint prediction of capacity and power fade and mechanism-aware health index classification

Capacity and power fade in lithium-ion batteries arise from distinct degradation mechanisms, especially under fast-charging conditions. Simultaneous diagnosis of both fades from early-cycle data is therefore essential for effective battery management. This study presents a noninvasive feature evaluation framework that jointly predicts capacity and power fade while providing mechanism-informed interpretation of degradation pathways using early-cycle health indices (HIs). An interpretable elastic net model is employed to predict both fades, and two complementary metrics, Contribution and Relative Count, are introduced to quantify each HI’s overall importance across varying sparsity levels, and to classify HIs as capacity-specific, power-specific, or dual-impact. The framework is applied to two public fast-charging datasets with distinct chemistries, achieves competitive predictive accuracy, and categorizes critical HIs consistently in both datasets. It further reveals chemistry- and operating condition-dependent shifts in influential HI groups which are consistent with established electrochemical understanding. By organizing early-cycle HIs according to their output-specific roles, the proposed framework provides an interpretable basis for inferring which degradation pathways are likely to dominate, even in datasets where prior mechanistic knowledge is limited. These results highlight its potential as a generalizable tool for noninvasive battery degradation diagnostics across datasets with different chemistries and operating conditions. Overall, the proposed framework can support the design of adaptive fast-charging protocols that balance future capacity and power fade, and it can be further extended to energy fade prediction.