Interpretable machine learning for battery capacities prediction and coating parameters analysis

Battery manufacturing plays a direct and pivotal role in determining battery performance, which, in turn, significantly affects the applications of battery-related energy storage systems. As a complicated process that involves chemical, mechanical and electrical operations, effective battery propert...

Full description

Saved in:
Bibliographic Details
Published in:Control engineering practice Vol. 124; p. 105202
Main Authors: Liu, Kailong, Niri, Mona Faraji, Apachitei, Geanina, Lain, Michael, Greenwood, David, Marco, James
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.07.2022
Subjects:
Battery capacity predictions
Battery manufacturing analysis
Control engineering
Data science
Energy storage system
Machine learning
Data science
Control engineering
Energy storage system
Battery capacity predictions
Battery manufacturing analysis
Machine learning
ISSN:0967-0661, 1873-6939
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Battery manufacturing plays a direct and pivotal role in determining battery performance, which, in turn, significantly affects the applications of battery-related energy storage systems. As a complicated process that involves chemical, mechanical and electrical operations, effective battery property predictions and reliable analysis of strongly-coupled battery manufacturing parameters or variables become the key but challenging issues for wider battery applications. In this paper, an interpretable machine learning framework that could effectively predict battery product properties and explain dynamic effects, as well as interactions of manufacturing parameters is proposed. Due to the data-driven nature, this framework can be easily adopted by engineers as no specific battery manufacturing mechanism knowledge is required. Reliable battery manufacturing dataset particularly for coating (one key stage) collected from a real battery manufacturing chain is adopted to evaluate the proposed framework. Illustrative results demonstrate that three types of battery capacities including cell capacity, gravimetric capacity, and volumetric capacity can be accurately predicted with R2 over 0.98 at the battery early-manufacturing stage. Besides, information regarding how the variations of coating mass, thickness, and porosity affect these battery capacities is effectively identified, while interactions of these coating parameters can be also quantified. The developed framework makes the data-driven model become more interpretable and opens a promising way to quantify the interactions of battery manufacturing parameters and explain how the variations of these parameters affect final battery properties. This could assist engineers to obtain critical insights to understand the underlying complicated battery material and manufacturing behavior, further benefiting smart control of battery manufacturing. •Interpretable machine learning is designed for battery smart manufacturing.•Designed method can effectively predict three types of battery capacities.•Designed method can quantify dynamic effects and interactions of coating parameters.•Designed framework is validated on a real battery manufacturing chain.
ISSN:0967-0661
1873-6939
DOI:10.1016/j.conengprac.2022.105202