The existing practice of leveraging extensive life-cycle data combined with specialized post-test characterization for lithium-ion battery (LiB) aging mode classification and quantification is both time and resource intensive. To facilitate non-incremental advances in LiB-technology development and validation, innovative, efficient, and inexpensive methods that provide insight into underlying failure modes are needed. Such high-throughput methods would accomplish two primary objectives: (i) early design- and/or use-modification guidelines to battery developers and (ii) accurate prediction of battery state of health. However, developing these high-throughput methods is extremely challenging due to the inherently complex and convoluted nature of battery aging, particularly early in the life cycle, and the sparse data sets associated with limited experimental designs.Conventionally, incremental capacity (IC) analysis has been widely used for investigating electrochemical aging phenomena in batteries by capturing signatures such as peak position, height, width, and area.1, 2 However, classic forward-looking IC analysis involves manual, time-consuming, trial-and-error-based processes that often require supplemental test data. Computationally assisted approaches, such as deep learning (DL), have great potential to make the process automatic and efficient. However, like any other machine-learning (ML) method, DL requires extensive training data for different combinations of aging modes, and these data are difficult and resource intensive to capture experimentally. This hinders the widespread application of ML in battery research.In this presentation, we will discuss how to leverage synthetic IC data associate with different aging modes to establish an automatic DL-model framework that pinpoints aging modes and compositions. We have generated ~26,000 IC-based synthetic data sets for Gr/NMC532 cells and used a DL-model-based framework to automatically classify and quantify the failure modes for 22 single layer pouch cells aged at different charging rates, from 1C to 9C. The framework classifies the dominant aging-mode combination—namely, pure loss of lithium inventory (LLI), LLI plus loss of active material (LAM) from positive electrode, and LLI plus LAM from negative electrode within 100 cycles with more than 99% confidence. Upon classification, evolution of aging modes is quantified as cycling progresses up to 600 cycles. The automatic classification and quantification provide significant benefit over conventional aging-mode identification and quantification analysis which are manual and require substantial time, effort, and experimental data. Our work marks the first efforts utilizing synthetic data to generate training datasets needed for DL and creating models for aging-mode identification and quantification. References C.R. Birkl et al., J. Power Sources, 341, 373–386 (2017)M. Dubarry et al., J. Power Sources, 219 , 204–216 (2012).