Detecting faulty lithium-ion cells in large-scale parallel battery packs using current distributions

Area of Science:

• Battery technology
• Machine learning applications
• Electrochemical systems

Background:

• Battery pack safety is a major concern hindering adoption, primarily due to fire risks from cell faults.
• Understanding large pack dynamics and improving battery management systems (BMS) are crucial for mitigating these risks.

Purpose of the Study:

• To develop a machine learning classifier for detecting cell faults in large battery packs.
• To utilize a limited number of current sensors for efficient fault detection.

Main Methods:

• A machine learning classifier based on a support vector machine (SVM) was developed.
• A modeling framework for parallel-connected battery packs was introduced and validated against Doyle-Fuller-Newman electrochemical models.
• The classifier was trained and tested using current sensor data.

Main Results:

• The developed classifier achieved a satisfactory accuracy of 83% in detecting cell faults.
• Effective fault classification was accomplished using current information from only 27% of the cells.
• The model's performance was further validated using experimental pack data.

Conclusions:

• Combining mathematical modeling and machine learning offers a promising approach to enhance battery management systems.
• This method can effectively address the complexities associated with large battery packs and improve overall safety.
• The study demonstrates the potential for accurate cell fault detection with reduced sensor deployment.