Kampker, A.; Heimes, H.H.; Dorn, B.; Clever, H.; Drescher, M.; Ludwigs, R.: Synthesis of Artificial Coating Images and Parameter Data Sets in Electrode Manufacturing. In: Herberger, D.; Hübner, M.; Stich, V. (Eds.): Proceedings of the Conference on Production Systems and Logistics: CPSL 2023 - 1. Hannover : publish-Ing., 2023, S. 654-664. DOI:
https://doi.org/10.15488/13485
Zusammenfassung: |
Driven by continuous cost pressure and increasing market requirements, the optimisation of the lithium-ion battery production is focus of attention. In order to save time and costs, machine learning (ML) represent a promising tool. ML methods are able to analyse highly complex correlations and abstract data sets. But a considerable amount of training data is needed. Since data is not always available to the required extent, approaches for synthesising artificial data were investigated.
In this study, the quality and corresponding measurement parameters in electrode production were assessed and selected. Based on this selection, coating trials have been conducted and the corresponding data set collected. The data set forms the basis for synthesis of artificial coating images and parameters. The selection and design of the synthesis models was divided into two sub-steps. First, the synthesis of artificial coating images was investigated. This was followed by the consideration of a procedure for the synthesis of structured data sets.
A promising method for data synthesis of (coating) images are Generative Adversarial Networks (GAN). The basic idea of GANs is to oppose two models: a discriminator and a generator. The generator generates artificial data samples that match the input of the training dataset. Afterwards those data samples (both input and artificial data) are introduced to the discriminator. The discriminator's function is to identify whether the data presented originates from the training dataset or whether it is a counterfeit (artificial data) of the generator. The requirements for the synthesis of tabular data sets correspond in principle to those for a multivariate regression analysis.
The combination of the models resulted in a method that allows the prediction of the corresponding measured quality values for arbitrarily selected process parameters, as well as the visualisation of the associated coating result in the form of an artificial image.
|
Lizenzbestimmungen: |
CC BY 3.0 DE - http://creativecommons.org/licenses/by/3.0/de/
|
Publikationstyp: |
BookPart |
Publikationsstatus: |
publishedVersion |
Erstveröffentlichung: |
2023 |
Schlagwörter (deutsch): |
Konferenzschrift
|
Schlagwörter (englisch): |
Machine Learning, Generative Adversarial Networks, Data Synthesis, Electrode Manufacturing, Battery Cell Production
|
Fachliche Zuordnung (DDC): |
620 | Ingenieurwissenschaften und Maschinenbau
|