Quantum computing - machine learning, Fudan team achieves new breakthrough in lithium metal batteries
As an important component of ionic polymer electrolytes (IPE), ionic liquids (ILs) with high ionic conductivity and wide electrochemical windows are promising candidates for achieving safe, high energy density lithium metal batteries (LMBs).
The research results, titled "Machine-learning-guided discovery of ionic polymer electrolytes for lithium metal batteries," were recently published in Nature Communications. Nature Communications.
In the May 15 issue of Nature Communications, Kai Li, Jifeng Wang, Yuanyuan Song, and Ying Wang's team at Fudan University describe a machine learning workflow embedded with quantum computing and graphical convolutional neural networks to discover potential ILs for IPE. by selecting a recommended subset of ILs, combining rigid rod-like polyelectrolytes and lithium salts, the experimental team developed a series of thin (~50 μm) and robust (>200 MPa) IPE membranes.
Ionic polymer electrolytes (IPE) containing non-flammable ions embedded in mechanically supported polymers with predetermined ion pathways have received considerable attention in revitalizing clean energy storage and conversion devices: e.g., batteries, fuel cells, supercapacitors, mechanical actuators, and reverse osmosis membranes. As promising candidates for safe and environmentally friendly electrolyte materials, ionic liquids (ILs) are room temperature (RT) molten salts with low vapor pressure, high thermal stability, wide electrochemical window, and high ionic conductivity.
In recent years, liquid crystalline polymers have shown the ability to effectively reduce interfacial resistance while proposing unique ionic conduction mechanisms in lithium metal batteries (LMBs). Lithium (Li) metal anodes coupled with high energy density cathodes, such as Li-air and Li-sulfur batteries, often require electrolytes with high conductivity, thermal and electrochemical stability to suppress inhomogeneous Li dendrites, overcome side reactions and break the balance between conductivity and modulus in the composite electrolyte.
As a key component of IPEs, it is desirable to develop a method to filter suitable ILs from a large number of IL candidates to develop successful LMB IPEs. machine learning (ML) has been widely discussed to predict the properties and learn the basic rules of the dataset, thus effectively simplifying the material discovery process.
This time, the Fudan University team describes an ML workflow embedded with quantum chemical calculations and graph convolutional neural networks (GCNN) to discover potential ILs with high ionic conductivity and sufficient electrochemical windows.This ML workflow requires two main steps: (1) unsupervised learning, followed by (2) supervised learning to target promising ILs.
Machine learning workflow for discovering ILs with high conductivity (σ) and wide electrochemical windows (ECW).
Unsupervised learning is performed on the dataset.
Supervised learning on the dataset.
Nowadays, predicting the accurate properties of new ILs without sufficient labeled data points is still a challenge. To overcome the problem of data scarcity, this team comprehensively combines object-oriented unsupervised learning and supervised learning, emphasizing the design of statistical regression and classification workflows (rather than independently predicting the absolute physical properties of IL pairs). In addition, this work demonstrates the efficiency of using GCNN for IL-based graph-attribute relationship classification tasks.
A series of thin (~50 μm) and robust (>200 MPa) IPE membranes were developed by selecting the recommended subset of ILs in combination with rigid rod-like polyelectrolytes and Li salts.
Li|IPEs|Li cells exhibit ultra-high critical current density (6 mA cm^-2) at 80°C. Li|IPEs|LiFePO4 (10.3 mg cm^-2) cells exhibit excellent capacity retention (>96% at 0.5C; >80% at 2C) over 350 cycles with fast charge/discharge. 80%), fast charge/discharge capability (146 mAh g^-1 at 3C) and excellent efficiency (>99.92%). This performance has rarely been reported in other monolayer polymer electrolytes that do not contain combustible organics.
Comparison of the IPE-based LMBs in this work with recent literature results.
In summary, the experimental team described an ML-guided screening scheme to screen promising ILs with high ionic conductivity and wide electrochemical windows for the preparation of IPE in LMBs.
In terms of ML modeling, unsupervised learning and multi-step supervised learning through unique object-oriented This comprehensive approach is essential to improve the efficiency of targeting promising ILs for practical applications.
The research team says the work focuses on unique, commercially available cations and anions from IoLiTec, rather than the widely used, decentralized NIST ILThermo database. "This helps the research effort better align with commercially available products, which we believe will also have implications for future practical research and new material design."
More importantly, the work provides new insights into overcoming data scarcity issues and enabling effective use of ML in materials design and optimization. "By investigating the golden rule, we can fabricate IPEs with tunable mechanical, structural and transport properties and apply them in large numbers to multifunctional devices, including batteries, fuel cells, supercapacitors, mechanical actuators and more."
Reference links:
[1] https://www.nature.com/articles/s41467-023-38493-7
[2] https://mp.weixin.qq.com/s/PRtJKzGf67TC3ab4pOTGDQ
