Discovering Electrochemistry with an Electrochemistry-Informed Neural Network (ECINN)

Area of Science:

• Electrochemistry
• Computational Chemistry
• Machine Learning

Background:

• Machine learning is increasingly used in chemistry for data analysis and insight generation.
• Current methods often overlook valuable domain-specific knowledge held by chemists.
• Integrating prior chemical knowledge into machine learning can improve model performance and interpretability.

Purpose of the Study:

• To introduce an Electrochemistry-Informed Neural Network (ECINN) that embeds electrochemistry principles into neural networks.
• To develop a multi-task learning framework for discovering electrochemical parameters.
• To demonstrate the application of ECINN in analyzing voltammetry experiments.

Main Methods:

• Developed an ECINN by incorporating Butler-Volmer, Nernst, and diffusion equations into a neural network architecture.
• Applied ECINN to analyze voltammetry data for redox couples to determine electrode kinetics and mass transport parameters.
• Integrated mass transport and Butler-Volmer equations within the ECINN framework for comprehensive voltammogram analysis.

Main Results:

• ECINN successfully discovered electrode kinetics and mass transport parameters from voltammetry data.
• The model seamlessly integrated mass transport with the Butler-Volmer equation to infer transfer coefficients directly.
• ECINN provided a novel approach to Tafel analysis, potentially replacing traditional mass transport correction methods.
• The model demonstrated the ability to uncover the nature of electron transfer and identify incorrect physical assumptions.

Conclusions:

• ECINN represents a new paradigm in chemistry-informed machine learning by embedding domain knowledge.
• This approach enhances model accountability, interpretability, and generalization capabilities.
• The study encourages the integration of chemical prior knowledge into machine learning for more robust scientific discovery.