Logo image
Second Life Prognostic for Li-Ion Battery: Cases of Missing Data with Polynomial Chaos and Time VAE
Conference proceeding

Second Life Prognostic for Li-Ion Battery: Cases of Missing Data with Polynomial Chaos and Time VAE

Pham Luu Trung Duong, Nagarajan Raghavan and IEEE
Prognostics and System Health Management Conference, pp.1-7
09/06/2025

Abstract

Battery charge measurement Chaos Discharges (electric) Li-Ion Batteries Lithium-ion batteries Missing Data Monitoring NASA Aging Battery Data Set Polynomial Chaos Polynomials Prognostics and health management Safety Second Life Second-Life State of Health Variational Auto-Encoder Voltage measurement
The global push for electrification and zero carbon emissions is introducing new challenges, particularly the growing volume of end-of-life (EoL) lithium-ion batteries (LiBs) from electric vehicles (EVs). Exposure to extreme temperatures, frequent partial cycling, and fluctuating discharge rates significantly accelerate battery degradation during the initial years of use. End-Of-life (EoL) batteries may no longer meet the power and energy requirements of most EV applications, but they remain viable for low power applications such as, electric golf cart, grid storage, and similar applications. Second-Life batteries (SLBs) generally operate without extensive measurement and monitoring systems, though maintaining some degree of monitoring is crucial to supports proactive maintenance planning, operational efficiency, and enhance safety. In the study presented in [1], the authors proposed a method for estimating the State of Health (SoH) of second-life batteries using a polynomial surrogate modeling method. This approach leverages data collected during the batteries' first-life phase and relies solely on terminal voltage measurements in their second-life phase. A generalized polynomial chaos (gPC) series is used to model the relationship between key features and discharge capacity during the first life. The trained model is then applied in the second-life phase to estimate the SoH. However, this method assumes the availability of complete data, which is often not the case for SLBs, as they typically operate without extensive measurement and monitoring systems. To address this limitation, a variational autoencoder (VAE) is employed to generate missing data, enabling the trained GPC model to effectively predict the SoH.

Metrics

1 Record Views

Details

Logo image