Data-Efficient Characterization of Battery Slurry Viscosity using Uncertainty-Based Active Learning with Neural Networks

Abstract

Reducing experimental effort is a major challenge in materials science research, especially in battery technology, where high material and production costs and complex dependencies between process parameters require targeted experimental design. In this work, we present a data-efficient approach that combines neural networks with an uncertainty-based active learning strategy to efficiently predict the viscosity of battery slurries. Since there are currently no reliably validated simulation models available for real multi-component systems, we rely entirely on experimental data. A neural network is first trained on a small initial Design of Experiments data set. Using Monte Carlo dropout and an uncertainty-based loss function, we quantify epistemic and aleatory uncertainties. Based on these estimates, we specifically select additional measurement points that are particularly informative for the model. Through iterative re-measurement and retraining, the prediction quality can be significantly improved while maintaining the same number of measurements. The approach enables targeted experimental design with minimal effort and is flexible in terms of model architecture and experimental conditions. It is particularly suitable for data-poor, resource-intensive applications and thus contributes to the accelerated development of high-performance batteries.

Authors 7

1. Rui Yan Li Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Production Engineering of E-Mobility Components (PEM),Aachen,Germany

2. Christian Wittke

   Helmut Schmidt University

   Affiliation as printed

   Helmut Schmidt University,Professorship Computer Science in Mechanical Engineering,Hamburg,Germany

3. Robert Ludwigs Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Production Engineering of E-Mobility Components (PEM),Aachen,Germany

4. Paul Lingohr Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Production Engineering of E-Mobility Components (PEM),Aachen,Germany

5. Heiner Heimes Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Production Engineering of E-Mobility Components (PEM),Aachen,Germany

6. Oliver Niggemann

   Helmut Schmidt University

   Affiliation as printed

   Helmut Schmidt University,Professorship Computer Science in Mechanical Engineering,Hamburg,Germany

7. Achim Kampker Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Production Engineering of E-Mobility Components (PEM),Aachen,Germany