Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Arda Genc, Justin Marlowe, Anika Jalil, Libor Kovarik, Phillip Christopher

TL;DR

This study tackles the challenge of high-throughput, quantitative analysis of nanoparticle size distributions in TEM and STEM images of heterogeneous catalysts. It introduces a two-stage AI workflow that combines YOLOv8x-based NP detection with SAM-based, prompt-guided segmentation (using a lightweight $vit{-}b$ backbone) to achieve accurate, pixel-level NP segmentation without retraining. The approach demonstrates robust generalization across PtCo on carbon black, Cu on silica, and Ru on γ-alumina, achieving $mAP@0.5=0.78$, $mAP@0.5:0.95=0.41$, and $F1=0.91$, while effectively resolving overlapping particles and providing detailed size-distribution insights from large datasets. The method significantly enhances throughput and reliability for NP analysis in TEM/STEM, with public code and data enabling broad adoption and extension to additional catalyst systems.

Abstract

Accurate and efficient characterization of nanoparticles (NPs), particularly regarding particle size distribution, is essential for advancing our understanding of their structure-property relationships and facilitating their design for various applications. In this study, we introduce a novel two-stage artificial intelligence (AI)-driven workflow for NP analysis that leverages prompt engineering techniques from state-of-the-art single-stage object detection and large-scale vision transformer (ViT) architectures. This methodology was applied to transmission electron microscopy (TEM) and scanning TEM (STEM) images of heterogeneous catalysts, enabling high-resolution, high-throughput analysis of particle size distributions for supported metal catalysts. The model's performance in detecting and segmenting NPs was validated across diverse heterogeneous catalyst systems, including various metals (Cu, Ru, Pt, and PtCo), supports (silica ($\text{SiO}_2$), $γ$-alumina ($γ$-$\text{Al}_2\text{O}_3$), and carbon black), and particle diameter size distributions with means and standard deviations of 2.9 $\pm$ 1.1 nm, 1.6 $\pm$ 0.2 nm, 9.7 $\pm$ 4.6 nm, and 4 $\pm$ 1.0 nm. Additionally, the proposed machine learning (ML) approach successfully detects and segments overlapping NPs anchored on non-uniform catalytic support materials, providing critical insights into their spatial arrangements and interactions. Our AI-assisted NP analysis workflow demonstrates robust generalization across diverse datasets and can be readily applied to similar NP segmentation tasks without requiring costly model retraining.

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

TL;DR

This study tackles the challenge of high-throughput, quantitative analysis of nanoparticle size distributions in TEM and STEM images of heterogeneous catalysts. It introduces a two-stage AI workflow that combines YOLOv8x-based NP detection with SAM-based, prompt-guided segmentation (using a lightweight backbone) to achieve accurate, pixel-level NP segmentation without retraining. The approach demonstrates robust generalization across PtCo on carbon black, Cu on silica, and Ru on γ-alumina, achieving , , and , while effectively resolving overlapping particles and providing detailed size-distribution insights from large datasets. The method significantly enhances throughput and reliability for NP analysis in TEM/STEM, with public code and data enabling broad adoption and extension to additional catalyst systems.

Abstract

Accurate and efficient characterization of nanoparticles (NPs), particularly regarding particle size distribution, is essential for advancing our understanding of their structure-property relationships and facilitating their design for various applications. In this study, we introduce a novel two-stage artificial intelligence (AI)-driven workflow for NP analysis that leverages prompt engineering techniques from state-of-the-art single-stage object detection and large-scale vision transformer (ViT) architectures. This methodology was applied to transmission electron microscopy (TEM) and scanning TEM (STEM) images of heterogeneous catalysts, enabling high-resolution, high-throughput analysis of particle size distributions for supported metal catalysts. The model's performance in detecting and segmenting NPs was validated across diverse heterogeneous catalyst systems, including various metals (Cu, Ru, Pt, and PtCo), supports (silica (), -alumina (-), and carbon black), and particle diameter size distributions with means and standard deviations of 2.9 1.1 nm, 1.6 0.2 nm, 9.7 4.6 nm, and 4 1.0 nm. Additionally, the proposed machine learning (ML) approach successfully detects and segments overlapping NPs anchored on non-uniform catalytic support materials, providing critical insights into their spatial arrangements and interactions. Our AI-assisted NP analysis workflow demonstrates robust generalization across diverse datasets and can be readily applied to similar NP segmentation tasks without requiring costly model retraining.
Paper Structure (9 sections, 8 equations, 4 figures, 2 tables)

This paper contains 9 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Results and discussion
  3. Detection and segmentation
  4. Nanoparticle analysis
  5. Methods
  6. Materials and data acquisition
  7. Training and optimization
  8. Evaluation metrics
  9. Conclusion

Figures (4)

  • Figure 1: Schematic illustrating the main components of YOLO architecture (top) and the proposed two-stage AI-assisted TEM particle analysis workflow, consisting of detection and segmentation stages (bottom).
  • Figure 2: Sets of images showing object detection and segmentation results from BF-TEM images of PtCo NPs on carbon black (top), HAADF STEM images of Cu NPs on silica (middle), and Ru NPs on $\gamma$-alumina (bottom).
  • Figure 3: Object detection (top-left) and segmentation (top-right) results of Pt NPs on $\gamma$-alumina for 70 NPs. Side-by-side comparison of histograms generated using AI-assisted NP size analysis for the image containing 70 NPs (bottom-left) versus batch processing of 7 images containing 1,080 NPs (bottom-right). The total analysis times were 12 s and 212.7 s, respectively.
  • Figure 4: Object detection (left) and segmentation (right) of overlapped PtCo NPs. The AI-assisted particle analysis workflow disentangles four overlapping NPs into nine distinct NPs.