Gradient-based PIV using Neural Networks

Recently, the whole field measurement using PIV has been becoming common to analyze many flow fields. The conventional PIV, however, often fails to measure an entire velocity vector field from deficient visualized images or to obtain velocity vectors in near-wall boundary layers. We have proposed a novel method using an artificial neural network for acquiring the characteristics of a two-dimensional flow field to solve the problems. We apply the proposed PIV to experimentally visualized tracer images of a rectangular channel flow with reverse flow. The examination proves that even velocity vectors very close to walls are measurable by putting a non-moving pattern on the wall.