Even normally shipped batteries may contain disturbing factors such as very small amounts of foreign matter or structural inhomogeneity, which can cause serious accidents such as thermal runaway or explosions. Since dismantling and identifying the cause is time-consuming and costly, nondestructive techniques for identifying the cause are required. A method for detecting abnormal current distribution by magnetic field inverse analysis has been proposed as one such method, but it is difficult to directly elucidate the factors that cause the abnormal current distribution itself. In this study, we first focused on the effect of different separator structures and analyzed the in-plane directional current distribution using machine learning in order to construct a non-destructive model for estimating the causes.

A one-dimensional model based on porous electrode theory ^[1] was used. Constant current charging, homogeneous electrode layer structure, uniform Li concentration in active material particles, and isothermal conditions were assumed. The electrode layer was assumed to be a porous body consisting of active material, electrolyte, and auxiliary materials (conductivity assistant and binder). Using the previously reported structural information and physical properties ^{[2], [3]}, Case 1 was a coin cell using LiNi_1/3Co_1/3Mn_1/3O₂ as the cathode active material, and Case 2 was a laminated cell using LiCoO₂ as the cathode active material. The anode active material was graphite and the electrolyte was 1.0 M LiPF₆ solution in EC・DEC solvent for both Case 1 and 2. The degree of flexure of the cathode and anode was determined using the formulas from previous studies ^[4]. Cross-sectional images of the actual biaxially oriented separator were obtained using a focused ion beam scanning electron microscope, and the three-dimensional simulated structure was reproduced based on these images (Sep. A). Similarly, a nonwoven fabric structure (Sep. B) and a foam structure (Sep. C) were also reproduced.

Since the collector foils are connected in-plane, the potentials at both ends of the cell must be uniform. Under this constraint, the ion current distribution can be estimated depending on the in-plane structure distribution of the separator. In this study, the correlation between the local porosity (ε) of the separator and the local ion current (denoted as C-rate) was determined using a nonlinear regression method called Support Vector Regression (SVR) ^[5]. 100 points were taken randomly from 0.4~1.0 and 5~10 for ε and C-rate, respectively, and used as explanatory variables. Each charging curve was compared with the charging curve at the in-plane mean value of (ε, C) = (0.545, 7.5), and the RMSE (root mean square error) of the voltage was calculated as the objective variable.

The in-plane reaction distribution was obtained from the voltage error predicted from ε and C-rate by SVR and the porosity distribution of the separator. The mesh size was 40 nm/pixel. For each separator, the difference between high and low reaction profiles was more clearly observed in Case 2. The reaction overvoltage distribution at a certain point in Case 2 was acquired, and it was confirmed that the reaction was concentrated near the separator of the negative electrode. This also made it possible to estimate the Li deposition point, which is the starting point of degradation.

Using machine learning to analyze the in-plane current distribution, it was found that the current distribution differs depending on the electrode conditions even for the same separator. We will use this technology to develop a technique for identifying factors based on the current distribution obtained by nondestructive measurement.

References

[1] G. M. Goldin et al., Electrochim. Acta., 64 118 (2012).

[2] G. Inoue et al., J. Chem. Eng. Japan., 54 (5) 207-212 (2021).

[3] K. Ikeshita et al., ECS Trans., 75 (20) 165-172 (2017).

[4] G. Inoue et al., J. Pow. Sour., 342, 476 (2017).

[5] C. Cortes et al., Mach. Learn., 20 (3) 273-297 (1995).