Autonomous adaptive sampling in PIV image analysis with extension to multi-dimensional PIV and application to vortex shedding behind a flapping wing

As an experimental measurement technique Particle Image Velocimetry (PIV) allows the
measurement of flow velocity of air or water by injecting small particles which reflect light
when illuminated, followed by analysis of the image recordings. Its non-intrusive nature
together with its intrinsic simplicity and capability of retrieving instantaneous planar velocity
measurements have made PIV a mature, standardized measurement technique in the field
of experimental fluid-related dynamics both in academic and industrial environments for a
wide range of applications. However, although the PIV image analysis procedure is
inherently simplistic, the user is required to carefully optimise the dominant processing
parameters. Consequently, the imposed parameters will rarely be optimal throughout the
image and no indication is available as to how accurate the obtained results are. Moreover,
the location of the interrogation areas is independent of underlying flow physics restricting
the spatial resolution in obtained velocity data. These drawbacks of the measurement
technique are problems which have gained concern and attention over the last few years.
Adaptivity on the basis of signal density and velocity gradients is able to maximise
measurement accuracy and resolution with minimal user input. However, curvature of the
flow field offers the potential to further improve the accuracy as it is known that this
parameter inherently modulates obtained displacement estimates. Moreover, no adaptation
has ever been performed utilising measurement uncertainty although it is logical that
regions of heightened data ambiguity demand increased sampling.
The objective of the work performed during Mr. Edwards' PhD at the University of Bristol will
therefore be to ameliorate the adaptive image interrogation in PIV analyses by automatically
coupling flow, signal and error adaptivity. Concepts will be inspired by numerical techniques
common in Computational Fluid Dynamics (CFD) and implemented in recursive PIV image
analysis routines to automatically adjust interrogation parameters accordingly.
The PhD program will be initiated with a background study to comprehend the concept and
implementation of existing PIV image processing algorithms. To this extent Mr. Edwards will
have access to state-of-the-art image processing codes developed at the University of
Bristol by his supervisors Dr Theunissen and Prof. Allen. The project will gradually evolve
into devising reliable curvature heuristics for PIV image analysis. To avoid erroneous
vectors driving the sampling, curvature estimates must be sufficiently reliable. This in itself
will imply some form of automated adaptivity based on data reliability. Reliability itself can
subsequently be used to ameliorate the sampling strategy. Once such parameters are
obtained, they can be combined in an iterative solver for PIV image analyses. This solver
will adaptively place and size a variable number of interrogation windows as to minimise
computational effort while maximising spatial resolution and accuracy. In other words,
interrogation parameters will be optimised adaptively and iteratively. Findings will be
juxtaposed using numerical flow fields and/or experimental two-component LDA
measurements and/or traditional and stereoscopic PIV measurements. The developed
concepts will then be extended to multi-dimensional PIV, whereby flow adaptivity criteria can
be refined. Of particular interest is the extension of the developed concepts to stereoscopic
PIV (3 velocity components in a single plane) and time-resolved PIV (temporal adaptivity).
Finally the developed algorithms can be applied to fluid dynamics problem of vortex
shedding of wings, which can now be better characterised due to the improved PIV
metrology.
Because of the widespread use of PIV and the novelty of the proposed techniques, the work
will be of high significance and impact both in academia and industry.