AILD: An Interpretable AI Framework for Designing Logical Descriptors in Predicting CO Adsorption on Catalysts

Area of Science:

• Catalysis Science and Engineering
• Materials Science
• Computational Chemistry

Background:

• Rational catalyst design requires predictive and interpretable structure-property relationships.
• Empirical metrics and opaque machine learning models limit catalyst development.
• A need exists for a science-based paradigm beyond trial-and-error.

Purpose of the Study:

• To develop a novel framework for generating interpretable and predictive descriptors for catalyst design.
• To bridge the gap between fundamental understanding and practical catalyst engineering.
• To accelerate the rational design of high-performance catalysts.

Main Methods:

• A 4-stage Artificial Intelligence-guided Logical Descriptors (AILD) framework was proposed.
• Methods include knowledge-driven feature generation, computational data processing, interpretable machine learning, and mechanistic interpretation.
• The framework integrates domain knowledge, first-principles simulations, and interpretable machine learning.

Main Results:

• The AILD framework successfully generates logical descriptors that are both predictive and physically interpretable.
• Implementation in open-source AILD software enables reproducible research.
• The framework facilitates a shift towards knowledge-driven catalyst design.

Conclusions:

• The AILD framework advances catalyst development by linking mechanism to design.
• It offers a powerful tool for accelerated rational catalyst design.
• This approach moves beyond empirical methods towards a science-based paradigm.