Neural-Network Particle Velocimetry for Pixel-Level Flow Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing particle image velocimetry methods can only calculate one velocity vector per large window, limiting the resolution of velocity measurements to 16×16 or 32×32 pixels, which does not capture the detailed velocity distribution in a flow field.
Innovation Solution
Employ a neural network model trained on previous and subsequent frames of particle images to predict the movement trajectory of individual particles, allowing for pixel-level velocity measurement by determining the velocity at each pixel position in the flow field.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cross-correlation algorithm is used on particle distributions in windows, then velocity vectors can be calculated, but the resolution is limited to one velocity vector per 16×16 or 32×32 pixel window
Solution Approach 1:
The patent segments the flow field measurement into multiple overlapping windows, and further segments each window into multiple sub-windows. This hierarchical segmentation allows velocity vectors to be calculated at multiple resolution levels, achieving pixel-level velocity measurement while managing computational complexity through progressive refinement from coarse to fine scales.
Solution Approach 2:
The patent transitions from traditional single-scale window-based measurement to a multi-scale hierarchical measurement approach. By introducing the dimension of scale levels (coarse to fine), the system achieves pixel-level resolution without proportionally increasing overall system complexity, as each level builds upon the previous level's results.
2Measurement precision
If window size is reduced to increase velocity vector density, then measurement resolution improves, but signal-to-noise ratio deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the measurement process into multiple hierarchical levels. At coarser levels, larger windows provide good signal-to-noise ratio. At finer levels, smaller sub-windows provide high resolution. The multi-level approach ensures that each segmentation level operates at its optimal scale, maintaining reliability while achieving high overall resolution.
Solution Approach 2:
The patent performs preliminary velocity measurements at coarser scales before proceeding to finer scales. This preliminary action at higher levels provides a foundation that guides and constrains the subsequent fine-scale measurements, ensuring that small window measurements are performed in context and maintaining signal-to-noise ratio while achieving pixel-level resolution.
3Measurement precision
If traditional PIV method is used, then velocity field can be measured, but detailed velocity distribution at pixel level cannot be captured
Solution Approach 1:
The patent segments the computational domain into a hierarchical structure of windows and sub-windows across multiple levels. This segmentation enables the system to achieve pixel-level velocity measurement by processing at appropriate scales, avoiding the need to process all pixels at full resolution simultaneously, thus maintaining computational efficiency while achieving high measurement precision.
Solution Approach 2:
The patent performs preliminary coarse-scale velocity measurements that provide initial velocity field information. This preliminary action guides subsequent fine-scale measurements, reducing the search space and computational burden at higher resolutions. The multi-level hierarchical approach ensures that detailed pixel-level velocity distribution is captured without requiring prohibitively expensive full-resolution processing from the start.
Data Source
AI summary
A velocity measurement method, system, device and apparatus and storage medium, as well as a velocity field measurement method and system are provided. The velocity measurement method includes: inputting a single-particle image to a trained neural network model to obtain a predicted result image; a pixel position of a particle in the predicted result image is a predicted position of a particle in the single-particle image; the particle in the single-particle image is configured to characterize one particle in a two-dimensional (2D) flow field measurement region; the trained neural network model is trained by taking a previous frame of particle image in a particle image pair as an input and a subsequent frame of particle image as an output true value; and determining a predicted velocity at the pixel position of the particle in the single-particle image.