Machine learning enabled cross-stage prediction of cathode performance using manufacturing parameters and measured electrode descriptors
File(s)
Published version
Author(s)
Zhao, Zhen
Leung, Chu Lun Alex
Huang, Chun
Type
Journal Article
Abstract
Optimisation of lithium-ion battery manufacturing has traditionally relied on time-consuming and resource-intensive experimental trial-and-error. Here, we fabricated LiNi0.8Mn0.1Co0.1O2 (NMC811) cathodes using varied manufacturing parameters, generating an experimentally measured laboratory dataset comprising 160 electrodes and 99 coin cells across varied manufacturing conditions. Group A and Group B cathodes were matched within manufacturing batches. Group A was uncalendered, whereas Group B was calendered. We built gradient boosting regressor (GBR) models to predict the cathode properties (active mass loading, thickness) in the intermediate steps and final battery performance (capacity of each cycle) using directly controllable manufacturing parameters together with measured electrode descriptors (electrode mass and thickness) as inputs. In a cross-stage analysis, the uncalendered-electrode thickness and early-cycle capacity, together with the corresponding manufacturing inputs, were used to predict post-calendering thickness and capacity within the present laboratory dataset. This study provides a preliminary laboratory-scale proof-of-concept for cross-stage prediction and indicates a potential route to reducing selected experimental trial-and-error steps in electrode processing optimisation.
Date Issued
2026-12-30
Date Acceptance
2026-09-18
Citation
Journal of Energy Storage, 2026, 182, Part C
ISSN
2352-152X
Publisher
Elsevier BV
Journal / Book Title
Journal of Energy Storage
Volume
182, Part C
Copyright Statement
© 2026 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)
License URL
Publication Status
Published
Article Number
124825
Date Publish Online
2026-09-23