More than 13 times faster than manual work, "robot + AI" discovers the best electrolyte for batteries and accelerates materials research

Editor | Ziluo

The traditional material research and development model mainly relies on "trial and error" experimental methods or accidental discoveries, and its research and development process generally lasts 10-20 years.

Data-driven methods based on machine learning (ML) can accelerate the design of new materials for clean energy technologies. However, its practical application in materials research is still limited due to the lack of large-scale high-fidelity experimental databases.

Recently, research teams from the Pacific Northwest National Laboratory and Argonne National Laboratory in the United States designed a highly automated workflow that combines a high-throughput experimental platform with the most advanced active learning algorithms to Efficient screening of binary organic solvents with optimal solubility for the anolyte. The goal of this research is to improve the performance and stability of energy storage systems to promote the widespread application of renewable energy. Traditionally, research involving anolytes usually requires a lot of trial and error experiments, which is time-consuming and labor-intensive. Using this automated workflow, researchers can more quickly screen out suitable binary compounds. In addition to an efficient workflow designed to develop high-performance redox flow batteries, this machine learning-guided High-throughput robotic platforms provide a powerful and versatile approach to accelerate the discovery of functional materials.

The reviewer commented: "This study shows that an AI-guided robotic platform can effectively find non-intuitive combinations of solvents and electrolytes in energy applications. This work has important implications for the battery community."

The research is titled "

An integrated high-throughput robotic platform and active learning approach for accelerated discovery of optimal electrolyte formulations

" and was published in "Nature Communications" on March 29, 2024 "superior.

Paper link:

https://www.nature.com/articles/s41467-024-47070-5

for It is crucial to ensure the development of clean energy technology applications and achieve deep decarbonization of electricity, so designing materials with targeted functional properties of tools is critical to developing clean energy technology applications and achieving deep decarbonization of electricity. Traditional trial-and-error methods are costly and time-consuming, so design tools are inherently expensive and time-saving.

The solubility of redox active molecules is an important factor in determining the energy density of redox flow batteries (RFB). However, electrolyte materials discovery is limited by the lack of experimental solubility data sets that are critical to exploit data-driven approaches.

Nonetheless, the development of highly soluble redox-active organic molecules (ROMs) for non-aqueous RFBs (NRFBs) remains a daunting task due to the lack of standardization of organic solvent systems and application-relevant experimental solubility data. task.

By utilizing the automated high-throughput experiment (HTE) platform, the reliability and efficiency of the "excess solute" solubility measurement method can be improved and the NRFB's solubility database can be constructed. However, even with HTE systems, the diversity of potential solvent mixtures makes the screening process more time-consuming and expensive.

Active learning (AL), and specifically Bayesian optimization (BO), has proven to be a reliable method to accelerate the search for electrolytes needed for energy storage applications. Therefore, a closed-loop experimental workflow guided by BO can be used to minimize HTE execution.

ML-guided high-throughput experimental robotic platform

Here, researchers used 2,1,3-benzothiadiazole (2,1,3 -benzothiadiazole (BTZ), a high-performance anolyte with a high degree of delocalized charge density and good chemical stability, serves as a model ROM. The focus is on studying its solubility in various organic solvents, demonstrating the potential of a machine learning-guided high-throughput experiment (HTE) robotic platform to accelerate NRFB electrolyte discovery.

Illustration: Schematic diagram of the closed-loop electrolyte screening process based on a high-throughput experimental platform guided by machine learning (ML). (Source: paper)

Specifically, the researchers designed a closed-loop solvent screening workflow consisting of two connected modules, HTE and BO. The HTE module performs sample preparation and solubility measurements via a high-throughput robotic platform. The BO component consists of a surrogate model and an acquisition function, which together act as an oracle, making solubility predictions and suggesting new solvents for evaluation.

The workflow is shown in the figure below, the specific steps are:

• First, prepare a saturated solution of ROM and analyze the sample via the HTE platform. Next, nuclear magnetic resonance (NMR) spectra of these samples were acquired and the spectral data were used to calculate the solubility of ROM.
• This data set is then used to train a surrogate model that is used to predict the solubility of untested samples within the search space as part of the BO process.
• Subsequently, the acquisition function is applied within the BO framework to guide the selection of new samples, guiding the evaluation based on the balance of predicted solubility values ​​and associated uncertainties (i.e., fitness scores), thereby simplifying the discovery and analysis of potential solvents .

More than 13 times faster than manual sample processing

The automated platform can prepare saturated solutions with solute excess and quantitative nuclear magnetic resonance (qNMR) with minimal manual intervention ) sample.

With the automated HTE workflow, the total experimental time to complete the solubility measurements of 42 samples was approximately 27 hours (~39 minutes/sample, less time per sample when running more samples). This is more than 13 times faster than manually processing samples using the "excess solute" method (approximately 525 minutes per sample).

In addition to the speed increase provided by the HTE system, research also placed great emphasis on controlling experimental conditions, such as temperature (20°C) and stabilization time (8 hours), to ensure accurate measurement of BTZ solubility in various organic solvents .

Illustration: Overview of the automated high-throughput experiment (HTE) platform. (Source: Paper)

Based on a literature review and consideration of solvent properties, the researchers listed 22 potential candidate solvents for BTZ. Then, an additional 2079 binary solvents were further enumerated by combining these 22 single solvents in pairs, each with 9 different volume fractions.

Table: List of 22 candidate organic solvents and their physical and chemical properties. (Source: paper)

#The platform identifies multiple solvents from a comprehensive library of more than 2000 potential solvents, with prototype redox active molecules2,1,3 -The solubility threshold of benzothiadiazole exceeds 6.20 M. Notably, the comprehensive strategy required solubility assessment for less than 10% of drug candidates, highlighting the efficiency of the new approach.

Illustration: Identification of required electrolytes via Bayesian Optimization (BO). (Source: paper)

The research results also show that binary solvent mixtures, especially those incorporating 1,4-dioxane (1,4-dioxane), help improve the BTZ solubility.

In conclusion, the study demonstrates an ML-guided HTE platform for electrolyte screening, where ML predictions and automated experiments work together to effectively screen binary organic solvents with optimal solubility for BTZ.

This research not only helps connect the fields of data science and traditional experimental science, but also lays the foundation for the future development of an autonomous platform dedicated to battery electrolyte screening.

The above is the detailed content of More than 13 times faster than manual work, "robot + AI" discovers the best electrolyte for batteries and accelerates materials research. For more information, please follow other related articles on the PHP Chinese website!