Performance Predictors for Organic Cathodes of Lithium-Ion Battery

Abstract

Organic cathodes for lithium-ion batteries are one of the most promising and significant materials toward a sustainable society. The molecular design is a key to achieve superior performances beyond inorganic cathodes. The present work shows predictors of the reaction potential, specific capacity, and ideal energy density for organic cathodes. Straightforward prediction models of the performance were constructed by a combination of machine learning and chemical insight, namely, sparse modeling for small data (SpM-S), on a small data set as training data found in the literature. The prediction accuracy was validated using different literature data. The predictors can be applied to explore high-performance organic cathodes in a wide search space efficiently. Moreover, SpM-S afforded straightforward, interpretable, and generalizable prediction models compared to other machine-learning algorithms. The small-data-driven methodology can be applied for further exploration of materials, enhancement of performances, and optimization of processes.

A Capacity-Prediction Model for Exploration of Organic Anodes: Discovery of 5-Formylsalicylic Acid as a High-Performance Anode Active Material

“…The descriptors were extracted by ES-LiR, a machine learning method. ,− , The multiple regression models were exhaustively constructed to prepare the weight diagram by Python.…”

“…Our group has proposed with combination of machine learning and chemical insight on experimental small data for efficient exploration of materials and optimization of processes, i.e. , sparse modeling for small data (SpM-S). − Sparse modeling is a data-scientific method for explanation of whole high-dimensional data using a limited number of strongly correlated factors, namely, descriptors. , In our SpM-S, the balanced small dataset containing explanatory and objective variables is prepared based on the experimental data. The descriptors are extracted by machine learning.…”

“…The accuracy of the prediction model is not so high because the scale of the training data and number of the descriptors are small. Nevertheless, the model contributes to finding new materials with a reduced number of experiments. − In the present work, a linear regression model for the capacity prediction of organic anode active materials was constructed by SpM-S (Figure ). The specific capacity as an anode active material was experimentally measured for 25 compounds (Figure a–c).…”

Challenges and advances of organic electrode materials for sustainable secondary batteries

“…[197,198] Oaki's research group constructed straightforward prediction models with combination of machine learning and chemical insight (Figure 19B), and the prediction models were efficient to explore high-performance OEMs in a certain search space by the sparse modeling for small data (SpM-S). [199] Araujo et al developed an artificial intelligence (AI)assisted framework in combination with DFT calculations and machine learning for efficient discovery of novel OEMs (Figure 19C), and the AI-based workflow achieved highthroughput screening of 20 million molecules to efficiently search for new organic cathodes candidates. [200] The AIbased workflow is a useful tool for accurately identifying the functionalities of organic molecules without time-confusing experiments, which will pave the way for the high-efficient data-driven exploration and structure design of future electroactive OEMs.…”

“…Reproduced with permission. [ 199 ] Copyright 2022, The American Chemical Society. (C) The workflow of the artificial intelligence‐based method enabled high‐throughput screening and the efficient searching for novel organic materials.…”