User-Friendly Freeware for Determining the Concentration of Electrolyte Components in Lithium-Ion Cells Using Fourier Transform Infrared Spectroscopy, Beer's Law, and Machine Learning. Issue 14 (1st January 2019)

• Record Type: Journal Article
• Authors: Buteau, Sam; Lee, Elizabeth; Young, Scott; Hames, Sam; Dahn, J. R.
• Is Part Of: Journal of the Electrochemical Society. Volume 166:Issue 14(2019)
• Page Start: A3102
• Page End: A3108
• Publication Date: 2019-01-01
• Subjects: Batteries - Lithium; Electrochemistry -- Periodicals; 541.3705
• DOI: 10.1149/2.0151914jes
• Languages: English
• ISSNs: 0013-4651

Abstract

Abstract : Understanding the changes in the electrolyte during lithium-ion cell aging is valuable in order to improve longevity. Studying this in hundreds or thousands of cells requires a fast and widely available measurement such as Fourier Transform Infrared Spectroscopy (FTIR) of electrolyte samples. This article expands on a previous work to use a new model more grounded in the physics of the measurement and machine learning to determine electrolyte composition from FTIR measurements. A carefully prepared dataset of mixtures of 5 electrolyte components (i.e. LiPF6, EC, EMC, DMC, and DEC), and the code to replicate and extend the model to different electrolyte mixtures are made available. With this new model, the mass ratio of salt to total is predicted within an error of 0.4%, and each solvent's mass ratio to total is predicted within an error of 2%. Furthermore, a calculated spectrum based on the predicted components can be compared to the measurement which allows one to detect if unexpected species are present in the electrolyte in significant quantity. A model for mixtures of 5 components can be calibrated well with between 25 and 50 carefully prepared samples so this work can be extended to other systems by simply adding more data and retraining.