About the research
This project was intended to create an intelligent prognostics platform for lithium-ion (Li-ion) batteries, which would equip existing battery management systems with the capability to perform predictive maintenance/control for failure prevention. The platform developed in this project consisted of two modules:
- Deep feature learning, which automatically learns the features of (capacity) fade from large volumes of voltage and current measurement data during partial charge cycles and estimates the real-time state of health (SOH) of a battery cell in operation
- Ensemble prognostics, which leverage the current and past SOH estimates in Module 1 to achieve robust prediction of the cell’s remaining useful life
Robust prediction of remaining useful life was achieved by ensemble learning-based prognostics, which synthesized the generalization strengths of multiple prognostic algorithms to ensure high prediction accuracy for an expanded range of battery applications and their operating conditions. The two modules aimed to learn features of fade from partial charge data, assess real-time health of individual battery cells, and predict when and how the cells are likely to fail. A case study involving implantable-grade Li-ion cells was conducted to demonstrate a deep learning approach to online capacity estimation, developed for Module 1.
Iowa State University ($50,258.00)
Midwest Transportation Center
Contract Number: DTRT13-G-UTC37