Abstract:In this work, we use machine learning to discover new polymer electrolytes for lithium-ion batteries. Polymer electrolytes are solid materials that can conduct ions and are safer than liquid electrolytes. However, they have lower ionic conductivity, which means they cannot transport ions as fast or as far. This limits their performance in batteries. Developing high ionic conductivity polymer electrolytes will result in high energy density and high-power density energy storage devices with rapid charging ability. Furthermore, polymer electrolytes need to have high mechanical stability to make safer solid-state batteries. However, developing new polymer electrolytes using multiple experiments is a tedious process. As such, machine learning can help in identifying the important parameters responsible for the discovery of high ionic conductivity polymer electrolytes. As part of this project, we study different copolymer electrolytes in context to their ionic conductivities and build a framework to identify the parameters governing the ionic conductivity of these electrolytes. We use different machine learning models including random forest, XGboost, KNN, linear regression, and chemprop model to predict the ionic conductivity of polymer electrolytes based on their chemical composition. The chemprop model uses a message-passing neural network, which is a type of deep learning that can learn from graph data, such as molecular structures. The model was trained on data from experimental publications, particularly, from Bradford et al. [1] that measured the ionic conductivity of different polymer electrolytes. The results showed that XG boost outperformed other models in predicting the ionic conductivity of polymer electrolytes. The significance of discovering new polymer electrolytes lies in addressing the current limitations of ionic conductivity. By identifying materials that exhibit improved conductivity, this work contributes to the development of high-performance lithium-ion batteries. Enhanced ionic conductivity translates to batteries with faster and more efficient ion transport, leading to improved battery performance and durability. This advancement is crucial for meeting the growing energy demands and ensuring the continued evolution of safe and reliable energy storage solutions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to predict the conductivity of solid polymer electrolytes (SPEs) by using machine - learning methods. Specifically, polymer electrolytes are a type of solid - state material that can conduct ions and are safer compared to liquid electrolytes. However, their ionic conductivity is low, which means their performance in batteries is limited because they cannot transport ions quickly or over long distances. Developing polymer electrolytes with high ionic conductivity can increase energy density and power density, enable fast - charging capabilities, and requires high mechanical stability to manufacture safer solid - state batteries. To overcome these problems, researchers utilized different machine - learning models (such as Random Forest, XGBoost, KNN, Linear Regression, and Chemprop model) to predict the ionic conductivity of polymer electrolytes based on chemical composition. These models aim to accelerate the discovery process of new polymer electrolytes by reducing the number of experiments, thereby identifying important parameters that affect ionic conductivity. In particular, the Chemprop model uses a message - passing neural network, which is a deep - learning technique that can learn from graph data (such as molecular structures). By training these models, researchers hope to find new materials with improved conductivity, which is crucial for enhancing the performance of lithium - ion batteries, helping to meet the growing energy demand, and ensuring the continuous development of safe and reliable energy - storage solutions.

Analyzing the efficacy of different machine learning models for property prediction of solid polymer electrolytes

Chemometric and Machine Learning Analysis of Lithium Concentration and Solvation Behavior in Li-Ion Battery Electrolytes

Chemistry-Informed Machine Learning for Polymer Electrolyte Discovery

Exploring the Possibility of Machine Learning for Predicting Ionic Conductivity of Solid-State Electrolytes

Predicting the anion conductivities and alkaline stabilities of anion conducting membrane polymeric materials: development of explainable machine learning models

The Electrical Conductivity of Ionic Liquids: Numerical and Analytical Machine Learning Approaches

Navigating Materials Chemical Space to Discover New Battery Electrodes Using Machine Learning

Machine Learning Prediction Models for Solid Electrolytes based on Lattice Dynamics Properties

Early Prediction of Ion Transport Properties in Solid Polymer Electrolytes Using Machine Learning and System Behavior-Based Descriptors of Molecular Dynamics Simulations

Accelerating amorphous polymer electrolyte screening by learning to reduce errors in molecular dynamics simulated properties

Prediction of Electrical Conductivity of Ionic Liquids: From COSMO-RS Derived QSPR Evaluation to Boosting Machine Learning

A data science approach for advanced solid polymer electrolyte design

De novo Design of Polymer Electrolytes with High Conductivity using GPT-based and Diffusion-based Generative Models

End-To-End Learning of Classical Interatomic Potentials for Benchmarking Anion Polarization Effects in Lithium Polymer Electrolytes

Machine Learning Assisted Prediction of Donor Numbers: Guiding Rational Fabrication of Polymer Electrolytes for Lithium‐ion Batteries

Explainable machine learning to enable high-throughput electrical conductivity optimization and discovery of doped conjugated polymers

Frequency-dependent dielectric constant prediction of polymers using machine learning

Size Dependent Lithium-Ion Conductivity of Solid Electrolytes in Machine Learning Molcular Dynamics Simulations

Automatically Capturing Key Features for Predicting Superionic Conductivity of Solid-State Electrolytes Using a Neural Network