A novel clustering algorithm for grouping and cascade utilization of retired Li-ion batteries

https://doi.org/10.1016/j.est.2020.101303Get rights and content

Highlights

  • A complete process for grouping retired batteries is proposed including safety checking, performance evaluation, data processing, and clustering of batteries.

  • A novel clustering algorithm of retired batteries based on traversal optimization is proposed.

  • The proposed algorithm shows that the greatest differences are found between clusters, but the least differences between the samples within a single cluster, which indicates the effectiveness of the algorithm.

Abstract

The rapid deployment of lithium-ion batteries in clean energy and electric vehicle applications will also increase the volume of retired batteries in the coming years. Retired Li-ion batteries could have residual capacities up to 70–80% of the nominal capacity of a new battery, which could be lucrative for a second-life battery market, also creating environmental and economic benefits. Presently, retired batteries are first screened to select usable batteries and then a proper secondary application is choosen according to the battery performance. Here, a complete process for grouping used batteries is proposed including safety checking, performance evaluation, data processing, and clustering of batteries. Also, a novel clustering algorithm of retired batteries based on traversal optimization is proposed. The new method does not require defining the cluster numbers and centers in beforehand, but possesses immunity to outliers. It can be used both for small and large sample sizes, as the optimization parameters used do not require iteration. The Davies-Bouldin Index of the proposed algorithm shows that the greatest differences are found between clusters, but the least differences between the samples within a single cluster, which indicates the effectiveness of the algorithm.

Introduction

Due to environmental reasons, more clean energy and transport means are increasingly introduced. For example, electric vehicles (EVs) are emerging as an alternative to traditional vehicles [1]. Lithium-ion batteries are the most commonly used battery type in EVs due to their high storage capacity [2]. It is estimated that the lithium-ion battery market will grow up to tens of thousands of MWhs and exceed an annual market value of $30 billion by 2020 [3]. Usually, a Li-ion battery is employed in electric vehicles until the capacity has decayed below 80% to ensure operational safety and adequate mileage [4], [5], [6], [7]. Consequently, retired batteries could still have 70–80% of the nominal capacity and would be potential for re-use in other secondary applications such as energy storage in smart grids with renewable electricity, or, powering electric bicycles, telecommunication stations, and other small devices [2, 8]. The second-life of retired batteries could thus be a highly competitive market due to the large number of retired batteries expected in the future [9], [10], [11], [12], [13]. The global annual quantity and weight of retired Li-ion batteries could surpass 25 billion units and 500,000 tons in 2020 [14]. The number of post-vehicle application battery packs available for second-use would rise up to 6.8 millions by year 2035. The global second-life use of retired battery business may grow from $16 million in 2014 to $3 billion by 2035. Reusing recycling retired batteries would also be important for environmental reasons.

Retired batteries have very different operating histories meaning that the residual capacities and battery efficiencies may vary very much and there may be even major inconsistencies between the cells within a single battery pack [15,16]. If retired batteries with different states are reused together, this could induce over-charging and discharging resulting in lower efficiency, accelerated aging, heat runaway, and other safety loopholes [17]. Therefore, it is of great importance to assess and screen retired Li-ion batteries so that batteries with similar properties are regrouped.

Different clustering algorithms can be used to regroup retired batteries. Available clustering algorithms for grouping battery include the k-means method, affinity propagation (AP), support vector machines, and neural network clustering algorithms [18], [19], [20]. Jiang et al. [21] employed k-means clustering model to group weeded-out lithium batteries using three characteristic indicies namely capacity, ohmic resistance, and polarization resistance. Huang et al. [22] extracted discharging features from used batteries using first curve fitting and then pretreated feature vectors accomplishing battery grouping with the k-means algorithm.

Though the k-means method is widely used in grouping large-scale data because of its simplicity and high efficiency, it has some disadvantages, e.g. the cluster numbers and centers need to be defined in advance. Moreover, the clustering result largely depends on the shortcoming of the initial center and is also vulnerable to outliers. Affinity propagation is a relatively new clustering algorithm proposed in 2007 [23]. Compared with the k-means method, there is no need to define the clustering number in advance, just determining the clustering centers and the members with information passed between data points [24]. He et al. [18] studied the neural network algorithm, finding that its clustering efficiency was closely related to the method of defining the neurons and the topology of network. Lai et al. [4] proposed two approaches to accomplish the screening and regrouping of batteries, namely a neural network model and a piecewise linear fitting model. They concluded that the former one was suitable for large samples with higher estimation accuracy, while the latter one was applicable for small samples. Liu et al. [25] put forward an approach for battery screening using convolutional neural networks based on two-step time-series clustering and hybrid resampling for imbalance data, which was able to reduce the inconsistency rate of screened cells by over 90% in an industrial application. One drawback of these algorithms were the high computing time needed to reach good accuracy. Therefore, the aim of this paper is to propose an improved algorithm, which combines simplicity and effectiveness. The new algorithm does not require defining the cluster numbers and initializing the clusters centers in advance, but can ensure convergence and possess immunity to outliers with accelerated calculation speed.

The structure of this paper is as follows. In Section 2, the performance assessment of retired batteries and preprocessing of experimental data are described with detail. In Section 3, the proposed algorithm is introduced. Section 4 presents the clustering results and a comparison to the k-means method and AP algorithm. Moreover, the effect of outliers on these algorithms is expatiated. The paper ends with conclusions in Section 5.

Section snippets

Performance assessment of retired batteries

The assessment process mainly focuses on extracting characteristic parameters from the battery charging-discharging curves to quantify the battery state. Here 18 retired Li-ion battery packs from electric vehicles produced by Shanghai Electric Guoxuan New Energy Technology Co. Ltd, China was used in the model testing. The tested battery pack (Fig. 1) comprised of eight cells, which are connected in parallel to form two modules, and then in series to form a pack. Detailed information of the

Proposed traversal optimization algorithm

A novel clustering algorithm is proposed here to save in the calculation effort and to improve the clustering accuracy. This algorithm is based on traversal optimization, bringing each target point into an existing cluster for calculation, and then determining the ownership of the target point. The algorithm has many advantages: First, unlike the k-means method this algorithm does direct cluster without the need to determine the cluster number and to initialize clustering centers in advance;

Results and discussion

The main results of the clustering process as well as a comparison to k-means and AP algorithms are shown in Fig. 7. After optimizing the parameters, the new algorithm clusters the 18 batteries into 6 groups. The AP algorithm forms 4 groups. Referring to the cluster number of traversal optimization, the cluster number of k-means is artificially set to be 6 in advance and the initial centers are selected as 1–6 #, the clustering result is shown in Table 6.

In Table 6, the number in bold is the

Conclusion

Increasing deployment of lithium-ion batteries will also require more attention on how to deal with retired batteries. In this paper, a new effective method is proposed for reusing retired batteries. The method includes checking safety aspects and evaluating battery performance, and incorporating an effective clustering algorithm of retired batteries for their reuse.

The new algorithm is based on traversal optimization. Compared e.g. to the k-means method, it does not need defining the cluster

Declaration of Competing Interest

None.

Acknowledgments

The work was supported by the National Science Foundation of China (Grant number 51736006). This work was partially supported by Aalto University.

References (32)

Cited by (39)

View all citing articles on Scopus
View full text