William Manley

An overview of the project and its current state can be found here

Project Outline:
Lab based microscopy, at a non-specialist level involves the tedious task of focusing onto the prepared sample. Often this process is needed to be replicated multiple times and can become incredibly time consuming. Moreover, the human error involved in this task often results in a sub-optimal perception of absolute focus. This is therefore a process that can be optimised by the aid of machine input.

Under supervision of the Micron research group at the University of Oxford, I worked on early stage development of the MicroscoPi imaging platform. The aim of the project was to produce a fully automated, robust, 3D printed microscopy platform. Specifically, my task was to develop and implement an autonomous autofocus module, that the user could run on any pre-prepared sample slide, under any light source and readily obtain a focused sample perspective. The crux of the problem was the reformulation of focus into a computer vision context.

Main Challenges:
No Reference Image: One thing that made this project unique was having no ‘reference image’ to compare against, since we needed to apply our module to any given prepared sample, and we were starting from a completely unfocused image. Hence, a “no reference, pixel-based” method (NRP) as commonly referenced in relevant literature was employed as the logical approach.
Quantifying Focus: The main problem that we needed to address was how to devise a metric to quantify how in/out of focus a sample is at each stage of the automated process. This corresponded to characterising the image artefact of ‘blur’ at each stage of the focal refinement process. Two metrics were developed, one using edge feature analysis and the other using Fourier spectra analysis.

Project Outcomes:
Project Report: A report detailing the progress that I made can be found here.
Project Presentation: A presentation of my research (Optimised for viewing in Microsoft PowerPoint) can be found here.

Project Outline:
Turbulence is one of the most abundant, yet poorly understood phenomena in the physical world. Huge difficulty arises due to the inherent chaotic fluctations that can be readily observed in even relatively simple flow scenarios. The most widely ueed analyitcal turbulence models are Navier-Stoke type models and decompose the flow into a mean flow component and a stochastically fluctauting flow componenet. Averaging the resulting equations leads to a tractable model for the mean field flow. However, there is a problem of closure of the Reynold’s stress term. Most commonly, the Bousinessq hypothesis is used to close the mean flow equations. This hypothesis is an energy model that accounts for the energy dissipation in a cascade down the length scales of eddies. This allows us to relate the Reynold’s stress solely to the mean flow and obtain a fully closed set of equations. In most urban settings models for the mean flow are usually sufficient - the effect of stochastic fluctations is relatively mild. However, in higher turbulent scenarios, (e.g. oceanic models) these models become insufficient. In this project, I explore these techniques, and apply then in a computational fluid dynamics (CFD) framework.

The project proposal involved working alongside a structural engineering company and using OpenFOAM CFD turbulence modelling to conduct stress analysis on the boundaries of a given house structure, subject to a widely accepted model for the flow profile in the atmospheric boundary layer. I also decided to conduct some pedestrian analysis - looking at how the wind flow is distorted by the presence of the house structure.

Main Challenges:
Computational Demand: One of the biggest challenges in CFD is the large computational demand asscoiated with running a simulation on a sufficiently refined computational mesh. Whilst a sufficiently refined mesh is desireable to allow for adequate ‘blending’ to occur through the viscous sublayer that forms close to the solid boundaries, a trade off can be made in reducing the mesh resolution and using interpolating ‘wall functions’, that are widely studied and proposed in the literature, and based on idealised experiments. This was a particularly necessary approach, since I didn’t have access to a powerful machine.
Numerical Discretisation: It was important to develop an understanding of how to translate the theory of fluid dynamics into a numerical framework. This included making decisions on the most suitable finite volume schemes (FVS) to use for each operator and how to discretise the physical quantities: space, time, energy & angle. It was also necessary to develop an understanding of exactly how the OpenFOAM software operates, for example, how each field is mapped through the geometry of successive cells in the mesh.

Project Outcomes:
Project Presentation: A presentation of my project can be found here.

Project Outline:
Traditionally, classical theories of elasticity and magnetism are studied in isolation. However, a recent emergence of exciting engineering applications has prompted the need for a unified theory of the magneto-active elastica. An example application is to implement non-invasive surgery/drug delivery systems. In this context, microscale, hard-magnetic ‘robots’ that cannot faesibly be motorised due to the associated length scale are able to be locomoted deterministically via. an externally applied magnetic field.

The aim of this dissertation is two-fold. Firstly, to unify the theory of elastic rods with magnetism. This will be done by considering suitable conservation/balance laws and enrgy minimisation principles, which upon applying constitutive relationships specific to the material, will result in a closed, non-linear, differential system. After the governing equations have been set out, we solve for the steady state deflections under various classes of applied magnetic fields, which in principle can be mapped onto a target location in space when viewed as an inverse problem.