Highly Available Li-Ion Batteries Sensors Readings Prediction Framework

Abstract

Lithium-ion batteries can experience internal failures due to issues like overcharging and excessive discharging. They can also encounter external failures, primarily involving sensors that monitor temperature, voltage, and current. External failures are often seen as the primary causes of internal failures [1]. Numerous battery health monitoring systems (BHMS) have been developed to ensure the proper functioning of batteries. These proposed BHMS systems could report sensor failures but are unable to recover the sensor readings. In this article, the main contribution is presenting a highly available sensor readings prediction framework to address situations where sensor readings are unavailable. The proposed prediction model compensates for faulty sensors by using readings from other healthy sensors. The generated reading sequences achieve high <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i><sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> accuracy, ranging from 94% to 99%. The proposed highly available sensor readings prediction framework is implemented using a lightweight long short-term memory (LSTM) network. The prediction model is trained and tested on batteries B6 and B7 from the NASA dataset and validated using the Toyota dataset, one of the largest available datasets.