Intelligent Battery Health Monitoring System for Sustainable Energy Storage

Problem Statement

Lithium-ion batteries power critical systems across transportation, defense, renewable energy, and microgrids, yet their performance degrades over time due to cycling, temperature effects, storage, and internal chemical changes. Traditional Battery Management Systems (BMS) rely on static, proprietary models that often fail to capture the nonlinear and dynamic nature of battery aging, leading to inaccurate State of Health (SoH) and Remaining Useful Life (RUL) estimates. This creates operational, safety, and cost challenges—especially for mission-critical applications such as those in the U.S. Navy, where reliability is paramount. There is a pressing need for transparent, adaptive, and data-driven battery health forecasting techniques that continuously learn from real-world conditions and enable smarter charging and asset-management decisions.

Approach

We develop AI-driven, data-based models to estimate SoH and predict RUL of Li-ion batteries using voltage and current time-series measurements captured during charge–discharge cycles. Our framework integrates machine learning, transfer learning, and validation using both public datasets (NASA, Oxford) and experimental data collected from a custom battery testbed built for this project. The models include a 1D-CNN for SoH estimation and a CNN-LSTM hybrid for capturing long-term degradation patterns, supported by a web-based platform for real-time visualization of diagnostic and prognostic outputs. This data-centric approach bypasses the need for proprietary battery-specific parameters, making it adaptable to diverse battery chemistries and applications.

Figure 1 – Schematic view of the AI-driven model for Li-ion battery health monitoring.

Ongoing Work

Current research is focused on leveraging the BatteryLife Benchmark to develop and validate unsupervised learning models. This approach aims to identify complex degradation patterns and failure precursors directly from raw time-series data.

Students and Scholars

Sunwoong Choi

Pranav Agrawal

Yannik Hahn (VGR)