.png){kind=icon}
.png){kind=icon}
Return
Data-driven battery electrode production process modeling enabled by machine learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jmatprotec.2023.117967.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The modeling of electrode production process remains a crucial challenge due to the complexity of physics under the process. In this work, a data-driven method enabled by machine learning is proposed to model the rela-tionship between intermediate product properties and process parameters for individual electrode production sub-processes. This method circumvents the need of thoroughly understanding physics by leveraging the exceptional ability of machine learning in recognizing hidden patterns from high-dimensional and highly nonlinear data. The proposed method includes four steps: data generation, data clustering, machine learning modeling, and post-modeling analyses. First, a design of experiments method is used to generate data for modeling. Then, the data correlations among product properties are analyzed, and the correlated product properties are grouped by a graph-based clustering algorithm. Based on the clustered data, a generic machine learning framework that includes adaptable model templates was proposed to formulate the relationship be-tween intermediate product properties and process parameters. Afterward, the process models are evaluated in terms of their prediction accuracy and robustness. Finally, post-modeling analyses are conducted to make the process models more understandable. The proposed method was demonstrated by modeling mixing and other electrode production processes. It was found that considering data correlation in machine learning modeling could improve the prediction accuracy of process models and the efficiency of model development was greatly expedited by reusing the machine learning model templates. The result shows the great potential of using ma-chine learning techniques to model battery cell manufacturing process.
Keywords:
Electrode
Machine learning
Process model
Data correlation
Design of experiments
Journal
J
IF:
7.5
Papers:
1.6W
Citations:
4.5W