Evaluation of 1D CNN Autoencoders for Lithium-ion Battery Condition Assessment Using Synthetic Data

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Sep 22, 2019
Christopher J. Valant Jay D. Wheaton Michael G. Thurston Sean P. McConky Nenad G. Nenadic

Abstract

To access ground truth degradation information, we simulated
charge and discharge cycles of automotive lithium ion batteries
in their healthy and degrading states and used this information
to determine performance of an autoencoder-based
anomaly detector. The simulated degradation mechanism was
an abrupt increase in the battery’s rate of time-dependent capacity
fade. The neural network topology was based on onedimensional
convolutional layers. The decision-support system,
based on the sequential probability ratio test, interpreted
the anomaly generated by the autoencoder. Detection time
and time to failure were the metrics used for performance
evaluation. Anomaly detection was evaluated on five different
simulated progressions of damage to examine the effects
of driving profile randomness on performance of the anomaly
detector.

How to Cite

Valant, C. J., Wheaton, J. D., Thurston, M. G., McConky, S. P., & Nenadic, N. G. (2019). Evaluation of 1D CNN Autoencoders for Lithium-ion Battery Condition Assessment Using Synthetic Data. Annual Conference of the PHM Society, 11(1). https://doi.org/10.36001/phmconf.2019.v11i1.876
Abstract 335 | PDF Downloads 398

##plugins.themes.bootstrap3.article.details##

Keywords

Battery, Deep learning, Simulation, 1D Convolutional Neural Networks

Section
Technical Papers