Machine-learning-based computationally efficient particle size distribution retrieval from bulk optical properties.

Area of Science:

• Optical Physics
• Computational Physics
• Materials Science

Background:

• Determining particle size distribution from bulk optical properties is crucial in various scientific fields.
• Traditional methods relying on evolutionary algorithms for inverse scattering problems are computationally intensive.
• There is a need for efficient numerical approaches to accelerate these calculations.

Purpose of the Study:

• To develop and validate an efficient numerical method for solving the inverse scattering problem.
• To accelerate the calculation of bulk optical properties using machine learning.
• To enable real-time particle size measurement.

Main Methods:

• Approximation of forward scattering calculations (Mie scattering theory) using machine learning for spherical particles.
• Application of the particle swarm optimization algorithm to optimize particle size distribution parameters.
• Minimization of deviation between target and simulated bulk optical properties.

Main Results:

• Machine learning accurately predicts scattering efficiency for a given particle size distribution in approximately 0.5 µs.
• The proposed method demonstrates high accuracy for both monomodal and bimodal size distributions, even with random noise.
• The combined approach significantly reduces computational expense compared to traditional evolutionary algorithms.

Conclusions:

• The machine learning-accelerated approach offers a computationally efficient solution for inverse scattering problems.
• This method shows significant potential for real-time particle size measurement applications.
• The simplicity and high efficiency make it a powerful tool for scientific analysis.