Determining Multi-Component Phase Diagrams with Desired Characteristics Using Active Learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Chemical Engineering

Background:

• Predicting phase diagrams for multi-component systems is crucial for materials discovery but computationally challenging.
• Existing experimental data offers limited insight into the vast space of possible material compositions.

Purpose of the Study:

• To develop a machine learning-driven framework for predicting and validating multi-component phase diagrams.
• To minimize experimental validation through intelligent experimental design.
• To discover novel materials with specific properties, such as high transition temperatures.

Main Methods:

• Utilizing machine learning (ML) for phase diagram prediction from high-dimensional virtual spaces.
• Employing Bayesian experimental design within an active learning loop for efficient validation.
• Applying the framework to ferroelectric ceramics and NiTi-based alloys.

Main Results:

• Successfully predicted and experimentally validated phase diagrams for complex ferroelectric and NiTi-based systems.
• Achieved high transition temperatures and triple points in synthesized materials.
• Required only three new experiments for validation and optimization of each phase diagram.

Conclusions:

• The ML-based approach effectively navigates complex compositional spaces to predict phase diagrams.
• This method significantly reduces the experimental burden for materials discovery.
• The framework enables the design and synthesis of advanced materials beyond current computational capabilities.