Lithium-plating-type defects in lithium-ion batteries pose severe safety risks due to their potential to trigger thermal runaway.To prevent defective batteries from entering the market,developing in-line detection met...Lithium-plating-type defects in lithium-ion batteries pose severe safety risks due to their potential to trigger thermal runaway.To prevent defective batteries from entering the market,developing in-line detection methods during manufacturing is critical yet challenging.This study introduces a deep learning-based method for detecting lithium-plating-type defects using formation and capacity grading data,enabling accurate identification of defective batteries without additional equipment.First,lithiumplating-type defect batteries with various types and area ratios are fabricated.Formation and capacity grading data from 154 batteries(48 defective,106 normal)are collected to construct a dataset.Subsequently,a dual-task deep learning model is then developed,where the reconstruction task learns latent representations from the features,while the classification task identifies the defective batteries.Shapley value analysis further quantifies feature importance,revealing that defective batteries exhibit reduced coulombic efficiency(attributed to irreversible lithium loss)and elevated open-circuit voltage/K-values(linked to self-equalization effects).These findings align with the electrochemical mechanisms of lithium plating,enhancing the model's interpretability.Validated on statistically robust samples shows that the framework achieves a recall of 97.14%for defective batteries and an overall accuracy of 97.42%.This deep learning-driven solution provides a scalable and cost-effective quality control strategy for battery manufacturing.展开更多
基金supported by the National Natural Science Foundation of China(NSFC,52277223 and 51977131)the Shanghai Pujiang Programme(23PJD062)。
文摘Lithium-plating-type defects in lithium-ion batteries pose severe safety risks due to their potential to trigger thermal runaway.To prevent defective batteries from entering the market,developing in-line detection methods during manufacturing is critical yet challenging.This study introduces a deep learning-based method for detecting lithium-plating-type defects using formation and capacity grading data,enabling accurate identification of defective batteries without additional equipment.First,lithiumplating-type defect batteries with various types and area ratios are fabricated.Formation and capacity grading data from 154 batteries(48 defective,106 normal)are collected to construct a dataset.Subsequently,a dual-task deep learning model is then developed,where the reconstruction task learns latent representations from the features,while the classification task identifies the defective batteries.Shapley value analysis further quantifies feature importance,revealing that defective batteries exhibit reduced coulombic efficiency(attributed to irreversible lithium loss)and elevated open-circuit voltage/K-values(linked to self-equalization effects).These findings align with the electrochemical mechanisms of lithium plating,enhancing the model's interpretability.Validated on statistically robust samples shows that the framework achieves a recall of 97.14%for defective batteries and an overall accuracy of 97.42%.This deep learning-driven solution provides a scalable and cost-effective quality control strategy for battery manufacturing.
