| The trend of new energy is unstoppable,and with the popularity of electric vehicles and energy storage devices,the market demands higher energy density and ultra-high safety for lithium-ion batteries,and hopes to increase the output of lithium-ion batteries while improving battery consistency and reducing production costs.Due to the limitations of traditional manual analysis and verification methods in terms of time and cost,it is difficult to speed up the development of lithium-ion batteries.Based on the above problems,an intelligent research architecture for lithium-ion battery production that integrates big data technology,computer science and battery-related mechanisms to promote the development of lithium-ion battery materials,devices and systems is designed in thesis.And the specific applications of machine learning in lithium-ion battery cathode material production and cell manufacturing are explored.Firstly,the design of intelligent lithium-ion battery production research architecture is described.The key modules such as data acquisition,high-throughput experimental platform and special algorithm library are introduced to describe the key bridging role of machine learning.To address the data processing problem,a special automation software is developed to assist in data screening,repair and organization to improve research efficiency.Secondly,the effectiveness of machine learning application in quality control of lithium-ion cathode material production line is explored.Machine learning modeling experiments were carried out in each link of the production line to achieve accurate prediction of product quality and feature screening of 103 production line inspection parameters.The performance of the model is generally improved after feature screening in each link,which proves that the method of screening the production line parameters with large contribution based on model analysis is correct.The results of the analysis of each link and the whole process analysis showed that the Zr content among the many detection parameters of the production line always has an important influence on the prediction of charging capacity,and the content of Ti plays a key role in the prediction of discharging capacity.Finally,the lithium-ion battery capacity grading technology in the device module is investigated in conjunction with machine learning.Using more than 20,000 data originating from the same lithium-ion battery manufacturing line,four nonlinear machine learning algorithms were applied to explore the correlation between the line monitoring data and its capacity of lithium-ion battery before capacity grading.The Cat Boost algorithm model exhibited the optimal capacity prediction accuracy in the two-stage study for the whole process and some links,with the root-mean-square error of the best prediction results in the test set accounting for 0.32% and 4% of the standard capacity,respectively.In addition,the results of the statistical analysis of the contribution of the model’s characteristic parameters are consistent with the traditional mechanistic understanding,and the parameters such as core weight are found to be the key influencing parameters for the capacity prediction.The research shows that the utilization of production line data and machine learning modeling to directly predict the product performance of lithium batteries is expected to optimize the production process of lithium batteries,support the improvement of production capacity and explore the production mechanism.In addition,the characteristic parameter analysis of the model can reveal the key production links that affect the battery capacity to a certain extent,and support the guarantee ability to improve product quality. |
