Title

(Semi-) Automated identification and Labelling of Defects in Images of Battery Cells

Description

Battery storage systems, and lithium-ion batteries in particular, are considered a key technology for the transport and energy transition. To meet the exponentially increasing demand for batteries and ramp up production capacities, new production facilities for battery cells need to be opened, which, in the best case scenario, produce both cost-effective and high-quality batteries. Until the battery cell is finished, production can be divided into electrode production, assembly, and finalisation. Even a faulty coating in electrode production can have a massive impact on the energy density and longevity of a battery cell. Imaging sensors are used for quality control, which in turn provide the data basis for machine learning applications. Such applications are usually part of supervised learning and require annotated data in order to train good models. Manual annotation of image data by humans is not only time-consuming and cost-intensive, but often also inaccurate. Automated methods are therefore a desirable solution.

The goal is to examine methods that simplify, and in the best case automate, labelling in the application of electrode coating. During annotation, relevant defects in the image should be identified and classified.

Requirements

Experience with image processing helpful

Person working on it

Lukas Quade

In cooperation with

Kevin Pouls, Fraunhofer Research Institution for Battery Cell Production

Category

Master thesis