3D AI and Plenoptic PIV in Fluid Flow Tracking

This article provides a comprehensive overview of how plenoptic particle image velocimetry (PIV) combines with 3D tomographic artificial intelligence to track turbulent vortex rings. By using a single light-field camera to capture volumetric data and leveraging deep learning algorithms for fast, accurate reconstruction, this technology solves traditional fluid dynamics challenges. Below, we explore the mechanics of plenoptic imaging, the role of AI in processing complex 3D particle fields, and how this system is directly applied to improve aerodynamics in wind tunnels and cardiovascular diagnostics in artificial heart valves.

Understanding Plenoptic Particle Image Velocimetry

Traditional Particle Image Velocimetry (PIV) measures fluid velocity by seeding a flow with microscopic reflective particles, illuminating them with a laser, and capturing their movement with cameras. To see in three dimensions, conventional Tomographic PIV requires four or more bulky cameras positioned at different angles.

Plenoptic PIV replaces this multi-camera array with a single plenoptic (or light-field) camera. Inside a plenoptic camera, an array of microscopic lenses—called a microlens array—is placed right in front of the image sensor. This setup captures not only the intensity of light reflecting off the seeded particles, but also the exact direction the light rays are traveling in space. By recording this directional light data, a single snapshot contains all the information needed to reconstruct a 3D volume of the fluid flow.

The Role of 3D Tomographic Artificial Intelligence

While plenoptic cameras simplify the hardware needed for 3D fluid tracking, they create a massive computational bottleneck. Converting 2D light-field data into a 3D volume of particles—a process called tomographic reconstruction—traditionally takes hours of processing per frame and suffers from elongation artifacts along the camera's line of sight, as well as "ghost particles" (false particles created by overlapping light rays).

3D tomographic artificial intelligence eliminates these problems. Researchers train deep learning models, specifically Convolutional Neural Networks (CNNs) and Physics-Informed Neural Networks (PINNs), on synthetic and experimental fluid data. The AI steps in to perform two critical functions:

1. Instant Volumetric Reconstruction: Instead of using slow, iterative mathematical algorithms to guess where particles are located in 3D space, the trained AI instantly translates the raw microlens data into a sharp, accurate 3D particle field. It recognizes and deletes ghost particles while correcting depth-perception errors.
2. Motion Tracking of Vortex Rings: Once the 3D volume is built, the AI tracks how clusters of particles move from one millisecond to the next. Turbulent vortex rings—spinning, donut-shaped regions of fluid—are notoriously chaotic and fast. The AI applies fundamental physics equations (like the Navier-Stokes equations) to predict and trace the exact 3D swirl, velocity, and decay of these vortex rings without losing track of individual particles.

Tracking Turbulence in Wind Tunnels

In aerospace and automotive wind tunnels, engineers must understand how air separates from wings, fuselages, and spoilers. This separation creates turbulent vortex rings that cause drag, structural vibration, and loss of lift.

Conventional multi-camera setups are often impossible to mount around wind tunnels due to limited viewing windows and tight spaces. Because Plenoptic PIV requires only one camera and a single optical port, engineers can easily point it at hard-to-reach areas, such as the inside of a jet engine intake or the wake behind a landing gear. The 3D tomographic AI processes the high-speed turbulent airflow in near real-time, mapping the exact birth and breakdown of vortex rings. This allows engineers to quickly tweak aerodynamic designs to smooth out airflow and boost fuel efficiency.

Analyzing Flow in Heart Valves

The exact same physics that govern air in a wind tunnel apply to blood flowing through the human heart. When designing prosthetic heart valves or studying cardiovascular disease, researchers place transparent, anatomically correct silicone models of the heart in a rig that pumps a blood-mimicking fluid seeded with tracer particles.

As a heart valve snaps open and shut, it generates complex, three-dimensional vortex rings inside the aorta and ventricles. In a healthy heart, these rings help wash blood smoothly through the chambers. However, in a poorly designed artificial valve, turbulent vortex rings create high shear stress, which tears apart red blood cells (hemolysis) and triggers lethal blood clots.

Optical access inside a curved, pulsing artificial heart valve is severely restricted, making traditional multi-camera setups useless due to light refraction and blocked views. A single plenoptic camera, peered through a lone viewing window, captures the entire interior volume of the valve. The 3D tomographic AI corrects for the optical distortions caused by the curved silicone walls and maps the microscopic turbulent eddies. Medical engineers use these precise 3D velocity maps to reshape valve leaflets, ensuring blood flows smoothly and safely for the patient.