A Data-Driven Quality Control Method for Li-Ion Cells

Abstract

The urgency to develop innovative electrochemical energy storage systems has increased exponentially as many governments across the world adopt net-zero strategies that rely heavily on transport electrification. This means that ensuring the repeatability and reproducibility of electrochemical data produced in battery research is one of the most important issues in the race to a carbon neutral economy. Li-ion batteries, the most mature and competitive battery technology currently available, has had some of its most significant advances in the discovery of new electrode materials, hence there is large research interest in this.Academic institutions have particular strengths and long experience in materials discovery, however, deployment of these materials in commercial devices lags behind for various reasons. One problem is the lack of quantitative parameters for quality control in electrochemical measurement as well as the difficulty of precisely evaluating cells prior to cycling. Cell assembly is a complex process with several steps, all of which could contribute to a faulty cell. Lack of reproducibility, especially when comparing from laboratory to laboratory, often ensue as a result. To optimize space and resource utilization many researchers do their testing in coin cells; the proliferation of best practice guides for coin-cell fabrication and measurement highlights the advantages of this cell format.(1-4) However, a simple quantitative method of quality control is still lacking.Figure 1: Graphical identification of outlier cells prone to poor cycling. OCV of pristine coin cells, NMC 811 half-cells. Outlier cells (black trace, bottom, red trace top) can be easily identified prior to cycling.In this presentation we demonstrate a proof-of-concept development of a quality control process based on the time evolution of the open circuit potential (OCV) of pristine Li-ion coin cells, and how that can be used to discard faulty cells before cycling. A pass/fail criterion is developed based on the OCV transients by fitting the behaviour of more than 20 different cells to a sigmoidal function, the goodness of fit is then compared. We found that cells that fall below a certain R2 value have shorter cycle life under the same cycling conditions. The threshold values showed a dependence on cell chemistry. Additionally, the graphical representation of the transients allows for identifying outliers that had significantly worse electrochemical performance (figure 1), therefore the goodness of fit can be used to quantitatively gauge the quality of the cells prior to cycling.The establishment of a quantitative pass/fail criterion makes the quality control process amenable to automation. To demonstrate this, we developed a Python program that takes experimental OCV transients as inputs to automate the data analysis. The program can process information from virtually hundreds of data files in a short period of time, easily identifying outlier cells. A spreadsheet file with the raw and transformed data, as well as with the generated plots is created for each transient OCV, allowing users to explore the data manually, if necessary.In summary, we have used data from a large number of coin cells to empirically fit a quantitative quality control parameter, establish a pass/fail criterion based on OCV transients, and develop code to greatly save time and resources and establish a statistical analysis of the effectiveness of the assembly process. This innovative data-driven approach showed great promise to improve repeatability and reproducibility in measuring the performance of new materials and cell-chemistries in a systematic and quantitative way.