Efficient Closed-loop Maximization of Carbon Nanotube Growth Rate using Bayesian Optimization.

Area of Science:

• Materials Science
• Chemical Engineering
• Machine Learning

Background:

• Materials research faces challenges due to large, complex parameter spaces hindering experimental efficiency.
• Limited understanding of input-output correlations in single-walled carbon nanotube (CNT) synthesis leads to poor catalyst yields.

Purpose of the Study:

• To demonstrate the application of Bayesian optimization (BO) for efficient and robust CNT synthesis.
• To leverage machine learning for high-throughput materials research and accelerate development.

Main Methods:

• Employed Bayesian optimization (BO), an adaptive sequential design algorithm, for exploring and optimizing high-dimensional parameter spaces.
• Utilized autonomous closed-loop experimentation integrated with machine learning.

Main Results:

• Achieved up to an 8-fold improvement in CNT growth rate compared to seed experiments.
• Demonstrated rapid improvement in predictive power (learning) of the BO model.
• Consistently good performance irrespective of seed experiment variations.
• Successfully exploited a high-dimensional parameter space.

Conclusions:

• Bayesian optimization is a highly effective algorithm for optimizing CNT synthesis, significantly increasing growth rates.
• The BO approach accelerates materials discovery by achieving results in approximately 100 experiments (~8 hours), a 5x improvement over previous methods.