This article proposes a deep-learning framework for estimating lithium-ion battery state of health (SOH) without requiring additional degradation experiments. The framework uses a swarm of deep neural networks with domain adaptation to achieve accurate estimations, even without target battery labels. Validation results on 71,588 samples from 65 commercial batteries showed absolute errors below 3% for 89.4% of samples and below 5% for 98.9% of samples, with a maximum error under 8.87%. This method significantly reduces the time and resource demands of traditional SOH estimation.

Jiahuan Lu, Rui Xiong, Jinpeng Tian, Chenxu Wang, Fengchun Sun