The testing of battery cells is a long and expensive process, and hence understanding how large a test set needsto be is very useful. This work proposes an automated methodology to estimate the smallest sample size of...The testing of battery cells is a long and expensive process, and hence understanding how large a test set needsto be is very useful. This work proposes an automated methodology to estimate the smallest sample size ofcells required to capture the cell-to-cell variability seen in a larger population. We define cell-to-cell variationbased on the slopes of a linear regression model applied to capacity fade curves. Our methodology determinesa sample size which estimates this variability within user specified requirements on precision and confidence.The sample size is found using the distributional properties of the slopes under a normality assumption, andan implementation of the approach is available on GitHub.For the five datasets in the study, we find that a sample size of 8–10 cells (at a prespecified precision andconfidence) captures the cell-to-cell variability of the larger datasets. We show that prior testing knowledge canbe leveraged with machine learning models to operationally optimise the design of new cell-testing, leadingup to a 75% reduction in experimental costs.展开更多
基金funded by an industry-academia collaborative grant EPSRC EP/R511687/1 awarded by Engineering and Physical Sciences Research Council(EPSRC)&University of Edinburgh United Kingdom program Impact Acceleration Account(IAA).G.dos Reis acknowledges support from the Fundaçao para a Ciencia e a Tecnologia(Portuguese Foundation for Science and Technology)Portugal through the project UIDB/00297/2020 and UIDP/00297/2020(Center for Mathematics and Applications,CMA/FCT/UNL Portugal)P.Dechent was supported by Bundesministerium für Bildung und Forschung Germany(BMBF 03XP0302C).
文摘The testing of battery cells is a long and expensive process, and hence understanding how large a test set needsto be is very useful. This work proposes an automated methodology to estimate the smallest sample size ofcells required to capture the cell-to-cell variability seen in a larger population. We define cell-to-cell variationbased on the slopes of a linear regression model applied to capacity fade curves. Our methodology determinesa sample size which estimates this variability within user specified requirements on precision and confidence.The sample size is found using the distributional properties of the slopes under a normality assumption, andan implementation of the approach is available on GitHub.For the five datasets in the study, we find that a sample size of 8–10 cells (at a prespecified precision andconfidence) captures the cell-to-cell variability of the larger datasets. We show that prior testing knowledge canbe leveraged with machine learning models to operationally optimise the design of new cell-testing, leadingup to a 75% reduction in experimental costs.
