Nested State and Degradation Estimation of a Satellite Battery with In-flight Data

TL;DR

This paper addresses the challenging estimation of hidden battery states, specifically $SOC$ and $SOH$, for satellite Li-ion cells under space conditions. It introduces a nested multi-timescale EKF that couples a physics-based P2D battery model with an SEI-driven degradation law to track fast $SOC$ dynamics and slow aging (through $L_{SEI}$). The approach is validated on synthetic data and REIMEI flight data, achieving accurate $SOC$ and $SOH$ estimates and using the Kalman gain correction as a reliability indicator of model fidelity. The results demonstrate robust state estimation and provide practical guidance for model selection and degradation monitoring in satellite missions.

Abstract

Li-ion batteries are essential for the energy supply of satellites. The accurate estimation of their states is important for the reliable and safe operation in space. This paper introduces a new algorithm for the estimation of SOC and SOH. The multi-timescale algorithm combines Kalman filters and physics-based models for batteries. We use a P2D model combined with a degradation model that describes capacity fading due to SEI growth. The state estimation algorithm combines two extended Kalman filters for the two states evolving on different timescales, with one filter nested within the other one. We test the algorithm with synthetic data as well as with in-flight data from Japanese satellite REIMEI. The algorithm adequately estimates the SOC and SOH in both cases. Furthermore it gives insight into the reliability of the chosen model.

Figures (10)

• Figure 1: P2D cell model with incorporated degradation model. a) Li-ion transport in electrolyte. b) Reaction at anode particle surface. c) Growth of SEI during storage (A) and during charging (A+B) caused by transport of Li-ions and electrons through SEI. Adapted from Ref. Bolay2022.
• Figure 2: Flowchart of the multi-timescale filter algorithm. From Ref. Bolay2024.
• Figure 3: Five cycles of cell current in-flight data with averaged constant current profiles.
• Figure 4: Synthetic data of five typical LEO cycles. Noise is added to the simulated output data to emulate measured data. a) Cell current. b) Cell voltage. c) Corresponding SOC and SEI thickness. From Ref. Bolay2024.
• Figure 5: SOC filtering with in-flight data. Comparison of anode and cathode SOC trends for different initial conditions. a) Trend of several hundred cycles, where red curves abort after about 300 cycles. b) Zoom-in to first ten cycles. Adapted from Ref. Bolay2024.