Enhancing corrosion-resistant alloy design through natural language processing and deep learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Chemical Engineering

Background:

• Machine learning models for materials design are often limited by their inability to process textual data from experimental descriptions.
• Manual extraction of data from text reduces information density and limits model accuracy.

Purpose of the Study:

• To enhance machine learning capabilities for corrosion-resistant alloy design.
• To develop automated methods for incorporating textual data into deep learning models.
• To identify key elemental descriptors influencing alloy pitting potential.

Main Methods:

• Implemented a natural language processing (NLP) approach to automatically convert textual data into a format suitable for deep neural networks.
• Developed a deep learning model utilizing a transformed input feature space.
• Replaced traditional alloy composition inputs with numerical descriptors based on elemental physical and chemical properties.

Main Results:

• Achieved a significant improvement in pitting potential prediction accuracy, surpassing state-of-the-art methods.
• Successfully identified critical elemental descriptors influencing pitting potential, including configurational entropy, atomic packing efficiency, local electronegativity differences, and atomic radii differences.

Conclusions:

• Automated NLP integration with deep learning offers a powerful strategy for enhancing materials design.
• Elemental physical/chemical properties are crucial predictors of alloy pitting potential, providing new insights for designing superior corrosion-resistant alloys.