[SY-D2] Designing mesoscale structures of Li-ion battery electrode using FIB-SEM image via machine learning
Optimizations of the mesoscale structure of Li-ion battery electrode have been demonstrated by using the advanced simulation method by single 2D slice image (Quasi-3D modeling) and machine learning. The mathematical model is based on the electrochemistry and physics model1), and developed in order to calculate Li/Li+ concentration on the 2D plane, in consideration of virtual 3D structure. In this study, we firstly confirm the validity of the Quasi-3D model, and secondly optimize the electrode structure in mesoscale using Bayesian optimization, a method of machine learning.
In order to confirm the validity of our proposed model, full 3D discharge simulations with random packed active material particles have been performed and compared. By use of an appropriate value of “connection factor”, quasi-3D model reproduce well a sliced Li/Li+ concentration calculated by the full 3D model in charge/discharge process, in addition that this model makes it possible to reduce computation time dramatically. Next, we have carried out optimizations of the mesoscale structure of the positive electrode Li(Ni1/3Mn1/3Co1/3)O2 based on the actual FIB-SEM image via Bayesian optimization. As a result, statistical parameters of the optimized meso-scale structures, including the dispersion of active material size and location, remarkably differ depending on the objective functions for high rate charge/discharge performance or for long cycle performance.
1) M.Doyle et al. J. Electrochem.Soc. 1996, Vol.143, No.6, p.1890
In order to confirm the validity of our proposed model, full 3D discharge simulations with random packed active material particles have been performed and compared. By use of an appropriate value of “connection factor”, quasi-3D model reproduce well a sliced Li/Li+ concentration calculated by the full 3D model in charge/discharge process, in addition that this model makes it possible to reduce computation time dramatically. Next, we have carried out optimizations of the mesoscale structure of the positive electrode Li(Ni1/3Mn1/3Co1/3)O2 based on the actual FIB-SEM image via Bayesian optimization. As a result, statistical parameters of the optimized meso-scale structures, including the dispersion of active material size and location, remarkably differ depending on the objective functions for high rate charge/discharge performance or for long cycle performance.
1) M.Doyle et al. J. Electrochem.Soc. 1996, Vol.143, No.6, p.1890