Predicting the Health States of Second-Life Lithium Iron Phosphate Battery Using A Long Short-Term Memory Degradation Model

Abstract

LiFePO4 batteries are widely implemented in the maritime industry due to high energy density, low self-discharge rate, good thermal stability, and easy compliance with maritime safety regulations. These batteries reach the end of the first life at 80% of nominal capacity. However, they retain residual second-life capacity that can be repurposed for onboard auxiliary systems, backup during grid-ship integration, etc. Varying C-rates and depth of discharge in the first and second life lead to different degradation mechanisms and aging rates, causing second-life batteries to start from an undefined state. This shift makes it challenging for physics-based and empirical models to predict degradation in the second-life stages, necessitating reliable degradation models. Our research addresses the aforementioned challenge by investigating how early in the second-life cycle the State of Health can be predicted. A Long Short-Term Memory model is applied to answer how prediction varies with the amount of training data. Two experimental datasets acquired from LiFePO4/graphite pouch cells are used, subjected to a 1C- 1C and 1C-2C charge and discharge rate, respectively, at ambient temperature, and 100% Depth of Discharge. The results show the feasibility and potential of Long Short-Term Memory models for predicting battery health states for second-life applications, which can be implemented to enhance the reliability and efficiency of systems in the maritime industry