High performance computing for a 3-D optical diffraction tomographic application in fluid velocimetry

Area of Science:

• Fluid dynamics
• Optical physics
• Computational imaging

Background:

• Optical Diffraction Tomography (ODT) is emerging in fluid velocimetry for 3D particle localization.
• Traditional methods often use linear approximations, insufficient for multiple scattering scenarios.
• Accurate 3D reconstruction requires addressing diffraction effects, especially at visible wavelengths.

Purpose of the Study:

• To develop and evaluate a non-linear, iterative image reconstruction method for ODT.
• To optimize the method for precise spherical particle location in fluid flows.
• To leverage GPU computing for efficient handling of computationally intensive tasks.

Main Methods:

• Implementation of a non-linear iterative reconstruction algorithm.
• Customization for spherical particle detection within ODT.
• Utilizing GPU computing and a specialized data storage format for performance enhancement.

Main Results:

• Demonstrated the feasibility of a non-linear ODT approach for 3D particle tracking.
• Achieved improved accuracy in spherical particle localization compared to linear methods.
• Showcased computational efficiency gains through GPU acceleration.

Conclusions:

• Non-linear ODT methods are essential for accurate fluid velocimetry when multiple scattering occurs.
• The presented GPU-accelerated, customized non-linear method enhances spherical particle location.
• This approach offers a computationally viable solution for complex fluid dynamics analysis.