A versatile machine learning workflow for high-throughput analysis of supported metal catalyst particles.

Area of Science:

• Materials Science and Engineering
• Nanotechnology
• Computational Science

Background:

• Accurate nanoparticle (NP) characterization, especially size distribution, is crucial for understanding structure-property relationships and designing advanced materials.
• Existing methods for NP analysis can be time-consuming and may struggle with complex, heterogeneous samples common in catalysis.

Purpose of the Study:

• To introduce a novel, two-stage artificial intelligence (AI)-driven workflow for high-throughput nanoparticle analysis.
• To enable accurate characterization of particle size distributions in heterogeneous catalysts using transmission electron microscopy (TEM) and scanning TEM (STEM) images.

Main Methods:

• Leveraged state-of-the-art single-stage object detection and large-scale vision transformer (ViT) architectures with prompt engineering.
• Applied the AI workflow to TEM/STEM images of supported metal catalyst NPs across various metals (Ru, Cu, PtCo, Pt) and supports (SiO2, γ-Al2O3, carbon black).
• Validated NP detection and segmentation using F1 overlap score, Hausdorff distance, and robust Hausdorff distance metrics.

Main Results:

• Achieved an average F1 overlap score of 0.91 ± 0.01, demonstrating high accuracy in NP detection and segmentation.
• Successfully characterized particle size distributions ranging from 1.6 ± 0.2 nm to 9.7 ± 4.6 nm, even for overlapping NPs.
• Segmentation accuracy validated with robust Hausdorff distance errors between 0.4 ± 0.1 nm and 1.4 ± 0.6 nm.

Conclusions:

• The AI-driven workflow provides accurate, high-resolution, and high-throughput nanoparticle size distribution analysis.
• The methodology shows robust generalization across diverse catalyst systems and NP types without retraining.
• This approach facilitates efficient NP characterization, accelerating research and development in catalysis and nanotechnology.