Lithium-ion battery is the key component of spacecraft. Remaining useful life (RUL) prediction is the core issue of power management and facility maintenance. Due to changes in working conditions, the trained model will shift inevitably. This paper proposes an unsupervised transfer learning (TL) method utilizing monitor data without label to predict cross-device RUL. The Health Assessment Deep Subdomain Adaptation RUL Regression Network (HA-DSARN) contains general feature extractor, cycle feature extractor, RUL regressor and health subdomain evaluator. HA-DSARN learns invariant features from source and target domain data through exactor. It takes local maximum mean discrepancy (LMMD) as domain differences measurement, and health subdomain evaluator is introduced to guide invariant subdomains features learning. HA-DSARN is verified through case studies transferred from simulation data (generated by finite element simulation) to monitoring data (actual collected data). The experiment results and comparison with state-of-the-art methods demonstrate the accuracy and transfer effect.

remaining useful life prediction,transfer learning,domain adaptation,state of health estimation,lithium-ion battery