Artificial Intelligence Investigation of NMC Cathode Manufacturing Parametersinterdependencies

Abstract

Nowadays, a great effort in terms of resources, time and money is dedicated to optimize Li-ion battery (LIB) manufacturing in order to improve batteries performances beyond the ones of current LIBs. This is particular true for next generation of active material chemistries formulated at the lab scale, which require significant trial-error to demonstrate their large scale manufacturability. Indeed, the number of parameters involved and the complexity of the physicochemical interactions throughout the associated manufacturing processes make highly difficult to find interdependencies between the final electrode characteristics and the fabrication parameters. The aim of the ERC-funded ARTISTIC project[1] is to develop computational tools able to capture the effect of the manufacturing process on the final electrode electrochemical performance thanks to the combination of multiscale modelling[2–5] and Artificial Intelligence (AI). AI algorithms offer powerful capabilities to automatically identify correlations in very large datasets (big data) containing many variables[6–8]. In this presentation[9], different machine-learning algorithms are analysed in order to find the best one in order to uncover the interdependencies between the slurry manufacturing parameters and crucial properties of NMC-based cathodes, as their mass loading and porosity. The results revealed that the Support Vector Machine combines high accuracy and the possibility to predict the influence of manufacturing parameters on the electrode mass loading and porosity in a straightforward graphical way. Thanks to this approach, several trends linking the electrode mass loading and porosity to the slurry characteristics were disclosed and their validity demonstrated by in house experimental results. These trends are explained in terms of the viscoelastic behavior of the slurries. Furthermore, the continuation of the latter work in terms of both methodology and fabrication process investigate (electrodes calendering) will be presented s well.It should be underlined that the proposed approach is chemistry neutral, and then it can be applied to next generations of active materials to rapidly assess and optimize their manufacture at pre-industrial scale. Moreover, it does not only considerably help to find interdependencies between LIB electrode manufacturing parameters, but it is designed in such a way to make it easily readable to both modelling and experimental researchers, simplifying the dialog between these two communities.Finally, a short perspective of the potential of AI methods to boost the discovery of ultra-high performance batteries (with particular reference to the European “Battery 2030+” initiative[10]) will conclude the talk.