IEEE Transactions on Instrumentation and Measurement · 2022 · 89 citations · 37 references
ComputingComputer ScienceMaterials ScienceComputer VisionFeature EngineeringMachine LearningBattery DiagnosticsBattery MaterialsBattery SystemsBattery TechnologyDeep LearningPattern RecognitionBattery AgingBattery DegradationConvolutional Neural Network DetectionConvolutional Neural NetworksDegradationKnee Point DetectionLithium Iron PhosphateRemaining Useful Life Prediction
Estimating the capacity degradation curve and the remaining useful life (RUL) of lithium-ion batteries is of great importance for battery manufacturers and customers. Lithium Iron Phosphate (LiFePO4) and Lithium Nickel Manganese Cobalt (NMC) batteries exhibit a slow degradation of the capacity up to the knee point, after which the degradation accelerates rapidly until the end of life. In the existing literature, data-driven methods require higher percentages of training data for predicting the lithium-ion batteries’ RUL with reasonable accuracy. This study first presents a novel online and offline knee detection algorithm to detect the knee in the capacity degradation curve. Compared to the existing knee detection algorithms, the proposed algorithm has better algorithmic efficiency and superior performance. Using the knee point, we present a novel method to estimate the complete degradation curve using the data of the 1st cycle with the help of the Convolutional Neural Network (CNN). This study also presents the aging precursors of lithium-ion batteries, which are used as features for the CNN degradation curve estimation model. The proposed method successfully predicts the degradation curve using data of the 1st cycle with root mean squared error (RMSE) and mean absolute percentage error (MAPE) as low as 0.005 and 0.416, respectively.
37
Data-driven prediction of battery cycle life before capacity degradation
Kristen Severson, Peter M. Attia, Norman Jin et al. · Nature Energy · 2019 · 2.9K citations · Full text