In-situ particle analysis with heterogeneous background: a machine learning approach.

Area of Science:

• Manufacturing
• Materials Science
• Computer Vision

Background:

• Traditional particle detection methods struggle with heterogeneous particle-substrate (HPS) interfaces in manufacturing.
• Existing techniques like size analyzers and basic deep learning models are often inadequate for complex backgrounds.

Purpose of the Study:

• To develop a flexible, AI-guided framework for robust particle detection in complex manufacturing environments.
• To enhance the accuracy and reliability of particle identification on heterogeneous particle-substrate interfaces.

Main Methods:

• A four-step modular framework combining preprocessing, AI model selection, AI-guided detection, and postprocessing.
• Image enhancement and sharpening for preprocessing; Transfer Learning with MobileNet as a model selector for heterogeneity classification.
• Utilizing distinct YOLO models for particle identification based on classified heterogeneity, with domain knowledge for postprocessing to reduce false positives.

Main Results:

• The AI-guided framework demonstrates consistent precision and recall across various HPS conditions.
• The harmonic mean of precision and recall is comparable to individual AI model outcomes.
• The system effectively handles diverse particle and substrate properties and varying ambient lighting.

Conclusions:

• The developed framework offers a versatile solution for particle detection in complex manufacturing settings.
• This tool has significant potential for advancing in-situ process monitoring in diverse manufacturing operations.
• Applications include 3D printing, powder metallurgy, coatings, particle categorization, and semiconductor manufacturing.