Prediction of Remaining Useful Life and Cell Temperature for Li-ion Batteries Using TinyML

• Published in: Conference proceedings : Midwest Symposium on Circuits and Systems pp. 562 - 566
• Main Authors: Weng, Yuqin, Guan, Wenkai, Ababei, Cristinel
• Format: Conference Proceeding
• Language: English
• Published: IEEE 11.08.2024
• Subjects: battery pack; deep neural network; electric vehicle; Hardware; Lithium-ion batteries; Microcontrollers; Predictive models; remaining useful life; Temperature measurement; Temperature sensors; thermal management; Tiny machine learning; tinyML
• ISSN: 1558-3899
• DOI: 10.1109/MWSCAS60917.2024.10658694

In this paper, we develop new tiny machine learning (tiny ML) temporal convolutional network (TCN) models for prediction of remaining useful life (RUL) and of cell temperature for lithium-ion batteries. The proposed models are developed, trained, optimized and verified in Python using TensorFlow. Ex-tensive simulation experiments, using datasets from the Battery Archive website and from Sandia National Lab (SNL), show that the proposed models provide better results compared to previous models. Furthermore, the proposed models are converted to Ten-sorFlow lite for microcontroller models, which are deployed on IoT hardware devices, specifically the popular Arduino Nano 33 BLE Sense board. We conduct hardware experiments that show that the tinyML models are very efficient and provide satisfactory prediction accuracy. Therefore, the proposed optimized tinyML models could be easily deployed in real practical scenarios, such as electric vehicles (EVs), to continuously monitor in real-time the health and temperature of batteries.