Lithium-ion batteries (LIBs) are currently the primary energy storage devices for modern electric vehicles (EVs). Early-cycle lifetime/quality classification of LIBs is a promising technology for many EV-related applications, such as fast-charging optimization design, production evaluation, battery pack design, second-life recycling, etc. The key c. ••A deep learning method for the early classification of battery qualities is studied.••A deep network model deriving latent features indicating battery qualities is developed.••The developed method is effective and robust to different battery types.••The battery quality classification accuracy can reach 96.6% based on data of first 20 cycles.Lithium-ion batteriesRapid lifetime classificationEarly-cycle dataConvolutional sparse autoencoderUnder the global pursuit of the green and low-carbon future, lithium-ion batteries (LIBs) have played significant roles in the energy storage and supply for modern electrical transportation systems, such as new energy electric vehicles (EVs), electric trains, etc. [1,2]. However, there still exist quite a few key issues which need to be addressed in the further development of LIBs, such as the high cost in research and development (R&D) as well as the safety concern. Studying methods for rapidly predicting battery lifetime and classifying the battery quality via early degradation cycles is of a high importance since they are beneficial to accelerate the battery R&D cycle and ensure the battery product quality.To meet the fast-charging demand of modern EVs, one critical research direction in the battery R&D is the multi-step fast-charging design and optimization, which aims to identify the optimal fast-charge profile for minimizing the battery charging time while maximizing the battery lifetime. However, the battery lifecycle test is time-consuming which forms a significant obstacle for the optimization process. For example, a battery with a lifetime of 2000 cycles may require several months to reach its failure. Rapid battery lifetime prediction and quality classification in early cycles are designed to accelerate the battery design and optimization. For example, techniques requiring only first-5-cycle data as inputs can rapidly classify the test battery into long-lived good ones or. This study considers three types of commercial LIBs widely applied in electric vehicles and grid-scale energy storage systems in terms of materials, i.e., the lithium-iron phosphate (LFP) battery, lithium cobalt oxide (LCO) battery, and Li(NiMnCo)O2 (NMC) battery. Datasets of three types of LIBs aged under different conditions are studied, which include a total of 156 battery samples. Each cell sample has its corresponding detailed specification. Specifications of a few selected cell samples are listed in Table 1 as an illustrative example. Meanwhile, the battery lifetime is defined as the number of cycles that the battery state-of-health degrades to and touches 80% of its nominal capacity.Table 1. Specifications of studied battery samples.Dataset I. Dataset I is produced via test experiments conducted by research groups in the Massachusetts Institute of Technology and Stanford University, which contains cyclic data of 123 commercial LFP/graphite cells with a nominal capacity of 1.1Ah. The lower and upper cutoff voltages are 2.0 V and 3.6 V, respectively. In a temperature chamber of 30 °C, all cells were charged with 72 different fast-charging profiles denoted as “C1(S1%)-C2”. These cells first underwent a constant-current (CC) charging process with a current rate C1 until their state of charge (SOC) reached S1% and next underwent a CC charging.
