State of charge and temperature-dependent impedance spectra regeneration of lithium-ion battery by duplex learning modeling. (1st August 2023)

• Record Type: Journal Article
• Authors: Temiz, Selcuk; Erol, Salim; Kurban, Hasan; Dalkilic, Mehmet M.
• Is Part Of: Journal of energy storage. Volume 64(2023)
• Publication Date: 2023-08-01
• Subjects: Electrochemical impedance spectroscopy -- Lithium-ion battery characterization -- Supervised learning -- Regression -- Machine learning; Energy storage -- Periodicals; Energy storage -- Research -- Periodicals; 621.3126
• DOI: 10.1016/j.est.2023.107085
• Languages: English
• ISSNs: 2352-152X

Abstract

Abstract: Impedance spectroscopy is a powerful technique and broadly used for battery characterization. In this study, we introduce a novel machine framework we call the duplex (for paired outputs) that constructs a linear ensemble of the best k models. Several impedance spectra of commercial lithium-ion battery coin cells at various states of charge and ambient temperatures are measured sensitively and compared with the spectra predicted by duplex learning modeling. The average difference between overall experimental and duplex estimated impedance data is approximately 3.374 %. The reliable predictions of battery impedance corresponding to a wide range of frequency are made over several operating conditions. The duplex can also be applied to different types of batteries at diverse operating conditions. Critical advantages of this approach are the speed of predicting the experimental space (the training/testing time for duplex is minimal), the small amount of data required to build the duplex, its performance compared to conventional machine learning techniques, as a new, integral component to battery management systems. All code/data are made freely available. Graphical abstract: Unlabelled Image Highlights: Machine Learning approach comprehends a good potential for Li-ion battery characterization and battery management system. The Duplex Learning Modeling is capable of regenerating the electrochemical impedance spectra by processing empirical data. Machine Learning-Estimated Impedance Spectra are revealed to be in perfect agreement with the experimental results.