Loading...
A physics informed machine learning framework for state-of health assessment of lithium-ion batteries in resilient infrastructure applications
Pereira, Eric L. Ogun, Damilola
Pereira, Eric L.
Ogun, Damilola
Citations
Altmetric:
Files
Loading...
Pereira_2025.pdf
Adobe PDF, 57.26 KB
Authors
Other Names
Location
Time Period
Advisors
Original Date
Digitization Date
Issue Date
2025-03-25
Type
Abstract
Genre
Keywords
Lithium-ion batteries,Degradation,State of health assessment,Differential voltage analysis,Machine learning
Subjects (LCSH)
Citation
Abstract
Lithium-ion batteries (LIBs) are widely used as energy sources in transportation and grid energy storage due to their high energy density, efficiency, and long lifespan. For instance, in regions like Kansas, which are prone to severe weather events such as tornadoes, LIBs offer a feasible solution for backup power during outages, playing an essential role in maintaining essential services and supporting resilient infrastructure. However, LIBs degrade over time, and if their degradation mechanisms are not properly monitored and managed, they can lead to operational failures or thermal runaway. State of health (SOH) assessment is crucial for monitoring battery performance. However, traditional methods that rely solely on capacity changes are insufficient due to the complexity of LIBs degradation. This work presents a framework for SOH assessment using differential voltage analysis (DVA) and machine learning. Commercial LIBs were tested under various charge/discharge rates, depths of discharge (DOD), and temperatures. Reference performance tests (RPTs) were conducted until end-of-life (EOL), and DVA extracted significant parameters, including internal resistance, active mass loss, and electrode stoichiometries. Additionally, to predict EOL, a Random Forest machine learning model was implemented on a Raspberry Pi computer, to enable real-time monitoring and remote data transmission to a cloud service for a centralized disaster control agency. Cells cycled at 100% DOD experienced greater cathode material loss. The Random Forest model processed data in about 2.9 seconds and achieved an accuracy of 86.67%. This accuracy demonstrates the potential of the low-cost model to facilitate remote monitoring of SOH and for secure power supply in extreme conditions.
Table of Contents
Description
Poster project completed at Wichita State University, Department of Electrical and Computer Engineering
Presented at the 22nd Annual Capitol Graduate Research Summit, Topeka, KS, March 25, 2025.
Presented at the 22nd Annual Capitol Graduate Research Summit, Topeka, KS, March 25, 2025.
Publisher
Wichita State University