Identifying key determinants of discharge capacity in ternary cathode materials of lithium-ion batteries

Citation: Xiangyue Li, Dexin Zhu, Kunmin Pan, Xiaoye Zhou, Jiaming Zhu, Yingxue Wang, Yongpeng Ren, Hong-Hui Wu. Identifying key determinants of discharge capacity in ternary cathode materials of lithium-ion batteries[J]. Chinese Chemical Letters, 2025, 36(5): 109870. doi: 10.1016/j.cclet.2024.109870 shu

Identifying key determinants of discharge capacity in ternary cathode materials of lithium-ion batteries

English

Identifying key determinants of discharge capacity in ternary cathode materials of lithium-ion batteries

a.
School of Materials Science and Engineering, Henan University of Science and Technology, Luoyang 471003, China
b.
Beijing Advanced Innovation Center for Materials Genome Engineering, Institute for Carbon Neutrality, University of Science and Technology Beijing, Beijing 100083, China
c.
Department of Materials Science and Engineering, Shenzhen MSU-BIT University, Shenzhen 518172, China
d.
Institute of Materials Intelligent Technology, Liaoning Academy of Materials, Shenyang 110004, China
e.
School of Civil Engineering, Shandong University, Ji'nan 250061, China
f.
National Engineering Laboratory for Risk Perception and Prevention, Beijing 100041, China
^* Corresponding authors.
E-mail addresses: pankunming2008@haust.edu.cn (K. Pan)
wangyingxue@cetc.com.cn (Y. Wang)
wuhonghui@ustb.edu.cn (H.-H. Wu).
¹ These authors contributed equally to this work.
Received Date: 15 March 2024
Accepted Date: 08 April 2024
Available Online: 15 May 2025

Abstract: Although lithium-ion batteries (LIBs) currently dominate a wide spectrum of energy storage applications, they face challenges such as fast cycle life decay and poor stability that hinder their further application. To address these limitations, element doping has emerged as a prevalent strategy to enhance the discharge capacity and extend the durability of Li-Ni-Co-Mn (LNCM) ternary compounds. This study utilized a machine learning-driven feature screening method to effectively pinpoint four key features crucially impacting the initial discharge capacity (IC) of Li-Ni-Co-Mn (LNCM) ternary cathode materials. These features were also proved highly predictive for the 50^th cycle discharge capacity (EC). Additionally, the application of SHAP value analysis yielded an in-depth understanding of the interplay between these features and discharge performance. This insight offers valuable direction for future advancements in the development of LNCM cathode materials, effectively promoting this field toward greater efficiency and sustainability.

Key words:

Lithium-ion batteries (LIBs) dominate the energy storage market due to their high energy density, long cycle life, and environmental compatibility [1-6]. Among cathode materials, Li-Ni-Co-Mn (LNCM) ternary materials, comprising Ni, Co, and Mn, are distinguished for their high energy density, durability, and environmental sustainability [7-10]. The initial discharge capacity (IC) of LNCM materials is a critical performance metric, affecting cycle life and indicative of overall performance [11,12]. Moreover, the discharge capacity at the 50^th cycle (EC) stands out as a critical parameter indicative of a battery's cycle endurance. Hence, the pursuit and discovery of strategies to enhance the IC and EC performances have become prominent subjects of inquiry in the realm of battery.

Discharge capacity, a pivotal metric for lithium-ion ternary battery performance, mirrors the available electric charge [13,14]. Doping is a prevalent approach to augment discharge capacity by optimizing LNCM material properties [15-18], and optimizing crystal-electronic structures to enhance capacity and cycle stability [19-21]. It increases specific surface area, boosting active material utilization and, consequently, battery discharge capacity [21,22]. Doping also optimizes the electronic structure, quickening lithium-ion diffusion, decreasing internal resistance, and uplifting overall battery efficacy [19]. The recent synthesis of AI and materials science has sparked keen interest due to their synergistic significance [23,24]. Machine learning, a key component of AI, offers pathways for optimizing and developing material properties [25,26]. Neural networks, for example, fulfill the need for high-efficiency solid-state electrolytes (SSE) materials [27] and accurately forecast polymer electrolyte ion conductivity [28]. Similarly, machine learning can predict the discharge capacity of LNCM ternary materials, guiding improvements in LIBs' performance.

This study aims to identify key material descriptors influencing IC and EC of LNCM ternary materials using machine learning. Data on IC and EC for LNCM ternary materials, doped with various single elements, were collected from academic publications. Subsequently, 229 material descriptors were constructed using domain knowledge and XenonPy [29]. An effective feature screening method was developed, integrating the Pearson correlation coefficient (PCC), importance coefficient, optimal subset method [30], and SHapley Additive exPlanations (SHAP) plots. This method pinpointed key features with high prediction accuracy for both IC and EC of LNCM materials [31], the workflow was shown in Fig. 1. SHAP plots elucidated potential relationships between these features and the target performance.

Figure 1

Figure 1. A workflow of screening the key factors in the discharge capacity of LNCM.

DownLoad: Full-Size Img PowerPoint

This work integrated dual feature sets originating from XenonPy calculations and domain-specific knowledge, summarized in Tables S1 and S2 (Supporting information). This study compiled a dataset of 237 single-element doped LNCM materials, as shown in Tables S3 and S4 (Supporting information), and normalized the feature set by Eq. S8 (Supporting information) [32-34]. After screening for feature independence, importance, and interpretation (Figs. S1 and S2 in Supporting information), we distilled down to 13 vital features. To derive the optimal subset, the optimal subset method is adopted for screening. As shown in Fig. 2a, observing a declining trend in PCC beyond four features, we concluded that the optimal subset comprises four attributes: var: ground_state_magnetic_moment (VMM), MTE, Vmax, and CD.

Figure 2

Figure 2. Selecting the optimal feature set and regression model. (a) PCC value was obtained by the optimal subset selection method. (b) The performance of 5 regression models with PCC, MAE, and RMSE values.

DownLoad: Full-Size Img PowerPoint

According to the "no free lunch" law, it is impractical to use a single algorithmic model to solve all machine learning problems [30,35-37]. Therefore, this study selects 5 typical models by ten-fold cross-validation [38-40], as shown in Fig. 2b. A series of indexes, PCC, MAE, and RMSE, were employed to gauge model performance [41-44]. Inspection of the results in Fig. 2b demonstrates that the XGB model consistently exhibits optimal performance across all three indexes, with PCC at 0.9034 and MAE and RMSE measured at 13.8033 and 18.9776 mAh/g, respectively. This indicates that through its exploration of the underlying relationship between the optimal set of features and the target performance, the XGB model achieves accurate and low-bias predictions. Consequently, the XGB model is selected as the preferred output model for predicting IC.

After feature engineering screening, four features demonstrating high predictive accuracy for target performance were identified, and Fig. 3a visually illustrates their predictive strength. By inputting the optimal subset of features into various regression models to select the model demonstrating the best performance, as illustrated in Fig. 2a, the XGB model was chosen as the optimal output model for predicting IC. Similarly, following the same rationale, the RF model was selected as the top-performing model for forecasting EC (Fig. 3b). After 100 random tests, the target performance yielded PCC = 0.9012, MAE = 13.9376 mAh/g, and RMSE = 19.1462 mAh/g. Beyond the notable predictive capability for IC, this screening approach also demonstrates heightened performance in predicting EC. Following 100 random tests, the performance indexes were PCC = 0.8461, MAE = 15.2549 mAh/g, and RMSE = 20.8239 mAh/g, highlighting the robustness of the model.

Figure 3

Figure 3. The scatterplot of the IC optimal model and EC optimal model: The scatterplot of (a) IC model and (b) EC model based on the same feature set.

DownLoad: Full-Size Img PowerPoint

The study further analyzed the impact of specific features on target performance using SHAP plots. Figs. 4a-d presented SHAP scatter plots, where Figs. 4a and c highlighted the positive correlation between Vmax and MTE with target performance, suggesting improved performance with higher Vmax and MTE values. Conversely, Figs. 4b and d associate CD and VMM with target performance, revealing a negative relationship as well.

Figure 4

Figure 4. The relationship between each feature and the IC based on the SHAP value diagram: The scatterplot of (a) Vmax, (b) CD, (c) MTE, and (d) VMM with their SHAP value, respectively.

DownLoad: Full-Size Img PowerPoint

In summary, this research successfully identified four key features significantly affecting the IC and EC of LNCM ternary materials. By examining the potential relationships between these features and the target performance using SHAP, we contribute to advancing battery technology and energy storage solutions. Furthermore, this methodology offers a basis for exploring performance prediction and optimization strategies in other battery materials.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CRediT authorship contribution statement

Xiangyue Li: Resources, Software, Writing – original draft, Writing – review & editing. Dexin Zhu: Software, Supervision, Writing – review & editing. Kunmin Pan: Funding acquisition, Resources, Supervision, Writing – review & editing. Xiaoye Zhou: Investigation, Supervision. Jiaming Zhu: Investigation, Validation. Yingxue Wang: Data curation, Software, Funding acquisition. Yongpeng Ren: Methodology, Visualization. Hong-Hui Wu: Funding acquisition, Resources, Supervision, Writing – review & editing.

Acknowledgments

This study was financially supported by the National Natural Science Foundation of China (Nos. 52122408, 52071023), the Program for Science & Technology Innovation Talents in the University of Henan Province (No. 22HASTIT1006), the Program for Central Plains Talents (No. ZYYCYU202012172), the Ministry of Education, Singapore (No. RG70/20) and the Opening Project of National Joint Engineering Research Center for Abrasion Control and Molding of Metal Materials, Henan University of Science and Technology (No. HKDNM201906).
1. [1]
  J. Hou, M. Yang, D. Wang, et al., Adv. Energy Mater. 10 (2020) 1904152. doi: 10.1002/aenm.201904152
2. [2]
  N. Zhang, T. Deng, S. Zhang, et al., Adv. Mater. 34 (2022) 2107899. doi: 10.1002/adma.202107899
3. [3]
  M. Armand, J.M. Tarascon, Nature 451 (2008) 652–657. doi: 10.1038/451652a
4. [4]
  M.R. Palacín, Chem. Soc. Rev. 38 (2009) 2565–2575. doi: 10.1039/b820555h
5. [5]
  B. Scrosati, J. Hassoun, Y.K. Sun, Energy Environ. Sci. 4 (2011) 3287–3295. doi: 10.1039/c1ee01388b
6. [6]
  G. Zhu, K. Wen, W. Lv, et al., J. Power Sources 300 (2015) 29–40. doi: 10.1016/j.jpowsour.2015.09.056
7. [7]
  Y. Liu, M. Liu, Int. J. Hydrogen Energy 42 (2017) 18189–18195. doi: 10.1016/j.ijhydene.2017.04.155
8. [8]
  Y.R. Kim, Y.W. Yoo, D.Y. Hwang, et al., Solid State Ionics 389 (2023) 116108. doi: 10.1016/j.ssi.2022.116108
9. [9]
  K. Xu, Energy Environ. Mater. 2 (2019) 229–233. doi: 10.1002/eem2.12057
10. [10]
  H. Jeong, S.J. Lim, S. Chakravarthy, et al., J. Power Sources 451 (2020) 227764. doi: 10.1016/j.jpowsour.2020.227764
11. [11]
  C. Wang, W. Zhang, T. He, et al., Int. J. Electrochem. 26 (2020) 12.
12. [12]
  G.G. Eshetu, H. Zhang, X. Judez, et al., Nat. Commun. 12 (2021) 5459. doi: 10.1038/s41467-021-25334-8
13. [13]
  L. Liu, M. Li, L. Chu, et al., Prog. Mater Sci. 111 (2020) 100655. doi: 10.1016/j.pmatsci.2020.100655
14. [14]
  J. Zhang, X. Li, G. Zhang, et al., Int. J. Energy Res. 44 (2020) 3134–3147. doi: 10.1002/er.5155
15. [15]
  A.S. Maiti, A.H. Sclar, D.R. A, et al., Energy Storage Mater. 45 (2022) 74–91. doi: 10.1016/j.ensm.2021.11.044
16. [16]
  W.W. Li, X.J. Zhang, J.J. Si, et al., Rare Met. 40 (2020) 1–8.
17. [17]
  S.B. Kim, S.Y. Ahn, J.H. Kim, et al., Electrochem. Commun. 146 (2023) 107426. doi: 10.1016/j.elecom.2022.107426
18. [18]
  H. Kim, M.G. Kim, H.Y. Jeong, et al., Nano Lett. 15 (2015) 2111–2119. doi: 10.1021/acs.nanolett.5b00045
19. [19]
  L. Bao, Z. Yang, L. Chen, et al., ChemSusChem 12 (2019) 2294–2301. doi: 10.1002/cssc.201900226
20. [20]
  L. Sun, X. Yi, X. Ren, et al., J. Electrochem. Soc. 163 (2016) A766–A772. doi: 10.1149/2.1071605jes
21. [21]
  G. Xue, Q. Xue, J. Li et al., Solid State Ionics 293 (2016) 7–12. doi: 10.1016/j.ssi.2016.04.025
22. [22]
  G. Wang, T. Fearn, T. Wang, et al., ACS Cent. Sci. 7 (2021) 1551–1560. doi: 10.1021/acscentsci.1c00611
23. [23]
  R. Ramprasad, R. Batra, G. Pilania, et al., npj Comput. Mater. 3 (2017) 54. doi: 10.1038/s41524-017-0056-5
24. [24]
  J. Liu, C.J. Tian, W. Chang, J. Chin. Ceramic Soc. 51 (2023) 6.
25. [25]
  W. Li, T. Yang, C. Liu, et al., Adv. Sci. 9 (2022) e2105550. doi: 10.1002/advs.202105550
26. [26]
  D. Yue, Y. Feng, X.X. Liu, et al., Adv. Sci. 9 (2022) e2105773. doi: 10.1002/advs.202105773
27. [27]
  Z. Lu, P. Adeli, C.H. Yim, et al., ACS Appl. Energy Mater. 5 (2022) 8042–8048. doi: 10.1021/acsaem.2c00493
28. [28]
  G. Bradford, J. Lopez, J. Ruza, et al., ACS Cent. Sci. 9 (2023) 206–216. doi: 10.1021/acscentsci.2c01123
29. [29]
  Y.R. Yoshida, Mol. Inf. 39 (2020) 1900107. doi: 10.1002/minf.201900107
30. [30]
  X. Li, D. Zhu, K. Pan, et al., Int. J. Refract. Met. Hard Mater. 117 (2023) 106386. doi: 10.1016/j.ijrmhm.2023.106386
31. [31]
  S. Wu, G. Lambard, C. Liu, et al., Mol. Inf. 39 (2020) 1900107. doi: 10.1002/minf.201900107
32. [32]
  Y. Chen, S. Wang, J. Xiong, et al., J. Mater. Sci. Technol. 132 (2023) 213–222. doi: 10.3390/math11010213
33. [33]
  B. Kumarsingh, K. Verma, A.S. Thoke, Int. J. Comput. Appl. Technol. 116 (2015) 11–15. doi: 10.5120/20443-2793
34. [34]
  J. Li, N. Wu, J. Zhang, et al., Nano-Micro Lett. 15 (2023) 227. doi: 10.1007/s40820-023-01192-5
35. [35]
  K. Sharma, M. Cerezo, Z. Holmes, et al., Phys. Rev. Lett. 128 (2020) 070501. doi: 10.1103/PhysRevLett.128.070501
36. [36]
  C. Shang, C. Wang, H. Wu, et al., Sci. China Technol. Sci. 66 (2023) 2069–2079. doi: 10.1007/s11431-023-2372-x
37. [37]
  F. Wang, H.H. Wu, L. Dong, et al., J. Mater. Sci. Technol. 165 (2023) 49–65. doi: 10.1117/12.2649022
38. [38]
  T.T. Wong, P.Y. Yeh, IEEE Trans. Knowl. Data Eng. 32 (2020) 1586–1594. doi: 10.1109/tkde.2019.2912815
39. [39]
  R. Malhotra, S. Meena, 2021 S International Conference on Electronics and Sustainable Communication Systems (ICESC), Coimbatore, India (2021) 431–438 https://ieeexplore.ieee.org/document/9533030.
40. [40]
  R. Kohavi, IEEE Trans. Knowl. Data Eng. 32 (2019) 1586–1594.
41. [41]
  T.O. Hodson, Geosci. Model Dev. 15 (2022) 5481–5487. doi: 10.5194/gmd-15-5481-2022
42. [42]
  D. Chicco, M.J. Warrens, G. Jurman, Peer J. Comput. Sci. 7 (2021) e623. doi: 10.7717/peerj-cs.623
43. [43]
  T. Chai, R.R. Draxler, Geosci. Model Dev. 7 (2014) 1247–1250. doi: 10.5194/gmd-7-1247-2014
44. [44]
  J. Adler, I. Parmryd, Cytometry Part A 77 (2010) 733–742. doi: 10.1002/cyto.a.20896
Figure 1 A workflow of screening the key factors in the discharge capacity of LNCM.

下载: 全尺寸图片幻灯片

Figure 2 Selecting the optimal feature set and regression model. (a) PCC value was obtained by the optimal subset selection method. (b) The performance of 5 regression models with PCC, MAE, and RMSE values.

下载: 全尺寸图片幻灯片

Figure 3 The scatterplot of the IC optimal model and EC optimal model: The scatterplot of (a) IC model and (b) EC model based on the same feature set.

下载: 全尺寸图片幻灯片

Figure 4 The relationship between each feature and the IC based on the SHAP value diagram: The scatterplot of (a) Vmax, (b) CD, (c) MTE, and (d) VMM with their SHAP value, respectively.

下载: 全尺寸图片幻灯片