Hybrid machine learning framework for predictive maintenance and anomaly detection in lithium-ion batteries using...

Area of Science:

• Materials Science and Engineering
• Electrical Engineering
• Computer Science

Background:

• Lithium-ion batteries are crucial for electric vehicles and energy storage, necessitating advanced predictive maintenance.
• Current battery management systems face limitations in real-time health diagnostics and state-of-charge estimation.

Purpose of the Study:

• To develop a comprehensive predictive maintenance framework for lithium-ion batteries.
• To improve battery health assessment and state-of-charge estimation using an Improved Random Forest algorithm.

Main Methods:

• Integration of physics-informed methodologies with data-driven machine learning models.
• Utilizing an Improved Random Forest (IRF) algorithm for dynamic battery health assessment.
• Analysis of features including state-of-charge, energy efficiency, and capacity decline.

Main Results:

• The IRF algorithm achieved superior performance over Gradient Boosting and standard Random Forest, with a Root Mean Square Error of 1.575 and R² of 0.9995.
• Achieved 99.99% classification accuracy for anomaly detection with no false negatives.
• Demonstrated real-time adaptability and robust anomaly detection capabilities.

Conclusions:

• The developed framework offers a transformative solution for sustainable energy systems by enhancing lithium-ion battery reliability, safety, and longevity.
• The IRF model provides accurate predictions and facilitates proactive interventions, reducing operational risks.
• The framework addresses scalability and computational efficiency challenges, positioning it as a critical tool for modern battery applications.