Editorial - Materials Science and Nanotechnology (2017) Volume 1, Issue 2

Applying machine learning to the design of materials for lithium ion battery

Qiang Sun^1,2* and Wei Wu¹

¹Department of Materials Science and Engineering, College of Engineering, Peking University, Beijing 100871, China

²Center for Applied Physics and Technology, Peking University, Beijing 100871, China

*Corresponding Author:

Qiang Sun
Department of Materials Science and Engineering
Peking University, China
E-mail: sunqiang@pku.edu.cn

Accepted date: September 26, 2017

Abstract

Lithium ion batteries (LIBs) are widely used in electrical vehicle, portable electronic products and energy storage devices. It is in urgent need to develop novel battery materials with higher energy density, preeminent rate performance and better safety performance. Traditional experimental methods for developing new materials are time-consuming, while for conventional simulation methods a huge gap remains between computations and practical materials. In this regard, machine learning becomes promising due to its flexibility in dealing with complex problems. Since machine learning is a data driven method, no physical or chemical models are needed in the process. The results will be reliable as long as the data are consistent with the practical situation. An encouraging example is AlphaGo [1] that has exhibited the charm of machine learning.

Lithium ion batteries (LIBs) are widely used in electrical vehicle, portable electronic products and energy storage devices. It is in urgent need to develop novel battery materials with higher energy density, preeminent rate and better safety performance. Traditional experimental methods for developing new materials are time-consuming, while for conventional simulation methods a huge gap remains between computations and practical materials. In this regard, machine learning becomes promising due to its flexibility in dealing with complex problems. Since machine learning is a data driven method, no physical or chemical models are needed in the process. The results will be reliable as long as the data are consistent with the practical situation. An encouraging example is AlphaGo [1,2] that has exhibited the charm of machine learning.

Machine learning provides effective ways of searching for the candidate program space through data training and optimizing the program algorithm. Data and algorithm are the two key factors which determine the accuracy of the machine learning. In fact, data are the basis of machine learning; it provides the knowledge for computers to learn, as well as the criteria of evaluating algorithms. Choosing proper learning algorithms is the other key, which can improve the efficiency and accuracy of the system.

For LIBs, most cathode materials are transition metal oxides and anode materials are carbon based materials. As for electrolyte, liquid electrolyte dominates the market while solid state batteries are still under development. When exploring the structure-property relationships in LIBs, many factors must be considered. For example, large amounts of data will be generated during high-throughput screening and since the relationships between structures and properties in LIBs don’t have a one- to-one correlation, it’s hard to find potential links between the simulation results. In such as case, machine learning can play an important role by using data classification and regression. Besides, reverse design of materials may be feasibly implemented in machine learning. Different from the conventional methods from structure to properties, reverse design of materials begins with targeting the property of the materials and tries to search for satisfactory structures and compositions. Machine learning has the ability to find out the property-to-structure relationships due to its flexibility in mapping complex relations [3].

It is very encouraging to notice that machine learning has already been put into predicting some properties of LIBs. For instance, Shandiz et al. applied different algorithms to classify silicate-based cathode [2]. Tanaka et al.’s group used support vector machine algorithm to predict the ionic conductivity for superior lithium ion conducting solids [4]. Furthermore, a combination of DFT computations with machine learning methods was applied to the screening of garnet-type [5], olivine- type [6,7] and tavorite-type [8] materials. Besides, machine learning has also been used to study the cycling performance of LIBs. Since lithium ion battery is a dynamic system, the internal physical or chemical mechanisms are very complicated which makes it difficult to construct a simple model for estimating the remaining useful life. Different machine learning algorithms including relevance vector machine [9], particle filter [10], genetic algorithm and support vector machine [11] have been used to estimate the remaining useful life of LIBs [12-15].

In short, machine learning is an emerging branch of artificial intelligence with great potential in developing novel lithium ion battery materials, and it deserves full attention from the researchers due to its flexibility in dealing with complex problems in practical conditions. However, some limitations exist for machine learning: Firstly, machine learning is a data driven approach, so the prediction is based absolutely on mathematical statistics, the physical or chemical significance need to be further studied; Secondly, a dataset with suitable size and accuracy is needed in machine learning, and it is of vital importance to choose proper data and do the pre-processing, while currently there only exist several commercial lithium ion battery systems, the available data are limited; Thirdly, optimum algorithms need to be improved for accuracy and efficiency. Since machine learning is still in the early stage, there is still a lot of work to be done in the future that include: (1) Reliable data selection and effective pre-processing; (2) Developing effective algorithms; (3) Integrating machine learning with other physics methods such as density functional theory and molecular dynamics for effective design of new battery materials.

Acknowledgement

This work is partially supported by grants from the National Key Research and Development Program of China (2016YFB0100200).

References

1. Silver D, Huang A, Maddison CJ, et al. Mastering the game of go with deep neural networks and tree search. Nature. 2016;529:484.
2. Shandiz MA, Gauvin R. Application of machine learning methods for the prediction of crystal system of cathode materials in lithium-ion batteries. Comput Mater Sci. 2016;117: 270-8.
3. Sheppard D, Xiao PH, Chemelewski W, et al. A generalized solid-state nudged elastic band method. J Chem Phys. 2012;136(7): 074103.
4. Tanaka I, Fujimura K, Seko A, et al. Accelerated materials design of lithium super ionic conductors based on first-principles calculations and machine learning algorithms. Adv Energy Mater. 2013;3: 980-5.
5. Kireeva N, Pervov VS. Materials space of solid-state electrolytes: Unraveling chemical composition–structure–ionic conductivity relationships in garnet-type metal oxides using cheminformatics virtual screening approaches. Phys Chem Chem Phys. 2017;19:20904-18.
6. Jalem R, Aoyama T, Nakayama M, et al. Multivariate ethod-assisted ab initio study of olivine-type limxo4 (main group m2+–x5+ and m3+–x4+) compositions as potential solid electrolytes. Chem Mater. 2012;24:1357-64.
7. Jalem R, Nakayama M, Kasuga T. An efficient rule-based screening approach for discovering fast lithium ion conductors using density functional theory and artificial neural networks. Chem Mater. 2014;2:720-34.
8. Jalem R, Kimura M, Nakayama M, et al. Informatics-aided density functional theory study on the Li ion transport of tavorite-type limto4f (m3+–t5+, m2+–t6+). Chem Inf Model. 2015;55:1158-68.
9. Datong L, Zhou J, Pan D, et al. Lithium-ion battery remaining useful life estimation with an optimized relevance vector machine algorithm with incremental learning. Measurement. 2015;63:143-5.
10. Datong L, Yang C, Peng Y, et al. Data-driven hybrid remaining useful life estimation approach for spacecraft lithium-ion battery. Microelectron Reliab. 2017;73:142-53.
11. Sun F, Li X, Liaio H, et al. A Bayesian least-squares support vector machine method for predicting the remaining useful life of a microwave component. Adv Mech Eng. 2017;9(1):1-9.
12. Li F, Xu J. A new prognostics method for state of health estimation of lithium-ion batteries based on a mixture of Gaussian process models and particle filter. Microelectron Reliab. 2015;55:1035-45.
13. Davoudi A, Zou Y, Camacho IC, et al. Guest editorial modeling and control of electrified vehicles and transportation systems. IEEE Transactions on Transportation Electrification. 2016;2:2.
14. Hu C, Jain G, Schmidt C, et al. Online estimation of lithium-ion battery capacity using sparse Bayesian learning. J Power Sources. 2015;289:105-13.
15. Du C, Zhu Y, Yan F, et al. State of health estimation based on OS-ELM for lithium-ion batteries. Int J Electrochem Sci. 2017;12(7):6895-907.