Unlocking new possibilities in ionic thermoelectric materials: a machine learning perspective.

Area of Science:

• Materials Science
• Thermoelectrics
• Machine Learning

Background:

• Ionic thermoelectric (i-TE) materials offer high thermopower for waste-heat recovery and thermal sensors.
• Current material discovery relies on inefficient trial-and-error methods lacking theoretical guidance.

Purpose of the Study:

• To develop a machine learning model for predicting the Seebeck coefficient of i-TE materials.
• To overcome the challenge of inconsistent i-TE material types using a simplified molecular-input system.
• To accelerate the discovery of high-performance i-TE materials.

Main Methods:

• Introduced a simplified molecular-input line-entry system.
• Developed and validated a machine learning model to evaluate the Seebeck coefficient (R² = 0.98).
• Conducted experimental identification of a novel ionogel material and utilized molecular dynamics simulations for analysis.

Main Results:

• Achieved a high prediction accuracy (R² = 0.98) for the Seebeck coefficient using the machine learning model.
• Experimentally identified a waterborne polyurethane/potassium iodide ionogel with a Seebeck coefficient of 41.39 mV/K.
• Identified key molecular descriptors (rotatable bonds, octanol-water partition coefficient) negatively impacting Seebeck coefficients.

Conclusions:

• The machine learning-assisted framework significantly accelerates the discovery of i-TE materials.
• The developed model provides theoretical underpinning for material design, reducing reliance on trial-and-error.
• This pioneering approach holds significant promise for advancing the field of ionic thermoelectrics.