Parts | Days | Selection | Search | Updates | Downloads | Help

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 1: Energy Storage and Batteries I

CPP 1.3: Talk

Monday, March 18, 2024, 10:15–10:30, H 0106

Introducing the LECA package for machine-learning guided optimization of the ionic conductivity — •Mirko Fischer¹, Harrison Martin¹, Peng Yan², Christian Wölke², Anand Narayanan Krishnamoorthy², Isidora Cekic-Laskovic², Diddo Diddens², and Andreas Heuer¹ — ¹Institute of Physical Chemistry, University of Münster, Corrensstraße 28/30, 48149 Münster — ²Helmholtz-Institute Münster (IEK-12), Forschungszentrum Jülich GmbH, Corrensstraße 46, 48149 Münster

We present the Liquid Electrolyte Composition Analysis (LECA) package as a versatile tool, which implements a simplified and semi-automatic workflow for data-driven and machine-learning guided analysis of large data sets, particularly designed for but not limited to High-Throughput-Experiments (HTE). The LECA package combines popular python-based libraries like scikit-learn, Mapie and GPyOpt to enable fast parallel training, hyperparameter-optimization, model comparison, and uncertainty calculation for various regression models. An active learning approach to reduce the amount of data needed to fit a model with high accuracy is under current development and testing.

We demonstrate the performance of the LECA package on a large HTE dataset for the ionic conductivity as an important bulk property of liquid electrolytes with over 200 individual compositions measured, including the organic electrolytes EC, EMC, and PC, and the lithium salts LiPF6 and LiFSI. Furthermore, we show how the LECA package can be used to optimize the ionic conductivity and discover new compositions.

Keywords: Battery; Ionic Conductivity; Electrolyte; Machine learning