Developing Multivariate Linear Regression Models to Predict the Electrochemical Performance of Lithium Ion Batteries Based on Material Property Parameters

Abstract

Predicting the electrochemical performance of active materials before their assembly in lithium ion batteries would be a path to cutting costs and time for assembling coin cells and running charging and discharging tests. Therefore, it is valuable to establish a statistical model to precisely predict the electrochemical performance of active materials in lithium ion batteries before cell assembly. In this study, we employed LiFePO 4 as the cathode active material. We measured the properties of LiFePO 4 powder, and then prepared cathode electrodes and ran electrochemical experiments. The acquired material property parameters and the electrochemical scores were correlated using multivariate linear regression models. Using these models, we can predict electrochemical performance prior to cell assembly. In this study, we collected 11 different commercial LiFePO 4 powders. We first used XRD, FTIR, and EA techniques to measure the crystal structure, vibration of PO 4 3− functional group, and the carbon content, respectively. Next we made the cathode electrodes using these 11 LiFePO 4 products and assembled them into coin cells, we then ran capacity tests at various current rates and cycleability tests at a 2 C current rate for 1000 cycles. Estimates of the regression coefficients in the regression models were calculated by the least squares method, and thus the regression models were established. We hope the models can act as references for effectively estimating capacity, rate and cycle performances based on the structure, functional groups, and carbon content properties of LiFePO 4 . The regression models can be expanded to consider the effects of additional material properties, and to predict other electrochemical properties. We expect to popularize this powerful material science statistical predictive strategy, to allow future researchers to predict the performance of products in a cost-effective and timely manner.

Modeling Macroscopic Material Behavior With Machine Learning Algorithms Trained by Micromechanical Simulations

“…Most of the processes that happen in nature are too complex to analyze, have too many independent parameters, and sometimes even the interrelation between parameters is unknown. In materials science, Machine Learning (ML) techniques such as Support Vector regression (SVR) (Swaddiwudhipong et al, 2005;Owolabi et al, 2014Owolabi et al, , 2015, linear regression models (Cheng et al, 2017) and Neural Networks (Ihom and Offiong, 2015) are becoming more and more important to describe complex phenomena for which the governing principle is not known or the proper implementation of which is too tedious and prone to errors. Among others, ML techniques have also been used in the field of material science to predict material properties (Swaddiwudhipong et al, 2005;Lin et al, 2008;Versino et al, 2017), characterize microstructure (Lubbers et al, 2017;Gola et al, 2018) and even to design better and efficient materials (Liu et al, 2015).…”