A Robust Battery Grouping Method Based on a Characteristic Distribution Model

Yuxiang Yang, Mingyu Gao, Zhiwei He, Caisheng WangEnergies 2017 · DOI

TL;DR

This paper introduces a novel battery grouping method that leverages a characteristic distribution model to overcome the limitations of traditional manual methods and existing clustering algorithms. By analyzing complete battery discharge curves, the proposed approach significantly improves grouping accuracy and efficiency, leading to better battery pack performance and reduced manufacturing costs.

What Problem Does This Paper Address?

The performance and lifespan of battery packs, especially in applications like electric vehicles, are severely hampered by the inherent inconsistencies between individual battery cells. These variations can lead to overcharging or undercharging of certain cells, reducing the overall pack's efficiency and longevity. While battery grouping is a known solution, current industrial practices rely on labor-intensive, time-consuming, and error-prone manual methods that often use only a few limited voltage measurements. Existing automated clustering algorithms, while better, struggle to directly address the practical requirement of forming fixed-size battery packs with highly similar characteristics.

Methods

The core of this paper is a two-pronged approach:

  1. Characteristic Distribution Model: A novel model is proposed to quantify the "density" of a battery's characteristics within the overall population of batteries. Batteries in low-density regions (i.e., those with unique characteristics) are prioritized for grouping first. This helps to isolate outliers and ensure that more similar batteries are grouped together.
  2. Improved k-Nearest Neighbor (k-NN) Algorithm: After identifying potential groups based on the characteristic distribution, an enhanced k-NN algorithm is used to select batteries for a pack. This algorithm considers the distances between batteries and a predefined threshold to ensure that selected batteries are not only close to each other but also sufficiently similar to form a cohesive group.

The method extracts features from the complete discharge characteristic curve of lead-acid batteries, going beyond the traditional charge static and discharge termination voltages. This richer feature set allows for a more nuanced understanding of battery behavior.

Datasets

The study uses data from 12V lead-acid batteries. Specifically, discharge curves from 200 batteries per shelf were automatically measured and used for evaluation. No new benchmark datasets were introduced; rather, the authors evaluated their method on a practical production line scenario.

Key Insights and Ideas

  • Beyond Simple Voltages: Relying solely on charge static and discharge termination voltages is insufficient for effective battery grouping, as demonstrated by examples where batteries with similar voltages exhibit vastly different discharge curves.
  • Density Matters: Prioritizing batteries in low-density characteristic regions for initial grouping is a clever way to handle outliers and ensure that the remaining batteries are more amenable to forming tight clusters.
  • Complete Discharge Curve is Key: Utilizing the entire discharge curve provides a much more comprehensive picture of a battery's performance than just a few discrete voltage points, leading to superior grouping results.
  • Efficiency Gains: The proposed automated method drastically reduces the time required for grouping compared to manual methods, offering significant cost savings in mass production.

Key Results

  • The proposed method, using the complete discharge curves, achieved significantly smaller intra-group variances (average of 0.1389) compared to using only charge static and discharge termination voltages (average of 0.7948).
  • The grouping failure rate ("orphan" batteries) was substantially lower with the proposed method (e.g., 4-12 "orphan" batteries per shelf) compared to the original k-nearest-neighbor algorithm (e.g., 8-24 "orphan" batteries per shelf).
  • The grouping time was reduced from 30 minutes per shelf for manual methods to just 2 seconds per shelf for the proposed automated system, a massive efficiency improvement.

Limitations and Future Work

  • Adaptability: While the paper demonstrates adaptability for different pack sizes (4 and 5 batteries), further exploration of its performance with a wider range of pack configurations and battery types would be beneficial.
  • Feature Engineering: The authors suggest that incorporating other battery charge/discharge characteristics could further improve the system.
  • Broader Applicability: The method is currently demonstrated on lead-acid batteries. Extending it to other battery chemistries, such as lithium-ion batteries, is identified as a key area for future research.

Summary updated

Abstract

The inconsistent characteristics of individual power batteries in a battery pack can seriously affect the performance and service life of the whole pack. Battery grouping is an effective approach for dealing with the inconsistency problem by grouping batteries with similar characteristics in the same battery pack. In actual production, the battery grouping process still relies on the traditional manual method, which results in high labor and time costs. In this paper, a robust and effective battery grouping method based on the characteristic distribution model is developed. Specifically, a novel characteristic distribution model is proposed to determine the grouping priority of different batteries. Then, an improved k-nearest-neighbor algorithm is used to decide which batteries should be group into the same battery pack. Experimental results demonstrate the effectiveness of the proposed method.
BibTeX
@inproceedings{yang2017robust,
  author = {Yuxiang Yang and Mingyu Gao and Zhiwei He and Caisheng Wang},
  title = {{A Robust Battery Grouping Method Based on a Characteristic Distribution Model}},
  booktitle = {Energies},
  year = {2017},
  doi = {10.3390/en10071035},
}

Loading related content...

Figure