Extracting particle size distribution from laser speckle with a physics-enhanced autocorrelation-based estimator...

Area of Science:

• Optical Physics
• Machine Learning
• Materials Science

Background:

• Quantifying particle size distribution (PSD) in highly scattering materials like powders is difficult due to complex light scattering patterns.
• Real-time, non-invasive monitoring of processes such as powder drying is crucial in industries like pharmaceuticals.
• Existing methods lack the necessary non-invasive and real-time capabilities for monitoring powder drying.

Purpose of the Study:

• To develop a theoretical framework linking particle size distribution (PSD) to speckle image patterns.
• To introduce a novel machine learning algorithm, the physics-enhanced autocorrelation-based estimator (PEACE), for speckle analysis.
• To enable real-time, non-invasive measurement of powder PSD during drying processes.

Main Methods:

• Developed a theoretical relationship between PSD and speckle image formation.
• Implemented a physics-enhanced autocorrelation-based estimator (PEACE) machine learning algorithm.
• Integrated physical laws to regularize the machine learning model for enhanced interpretability.

Main Results:

• Successfully established a method to measure powder PSD from speckle images.
• Demonstrated the capability of the PEACE algorithm to solve both forward and inverse problems simultaneously.
• Achieved increased interpretability in the machine learning model due to physical law regularization.

Conclusions:

• The developed PEACE algorithm provides a viable solution for non-invasive, real-time PSD measurement of powder surfaces.
• This approach offers significant advantages for monitoring dynamic processes like pharmaceutical powder drying.
• The physics-regularized machine learning model enhances understanding and reliability in speckle-based analysis.