Engineering: Theses and Dissertationshttp://hdl.handle.net/10092/8402019-10-21T14:47:54Z2019-10-21T14:47:54ZHigh throughput breeding for wood quality improvement.Davies, Nicholas T.http://hdl.handle.net/10092/174692019-10-17T14:02:27Z2019-01-01T00:00:00ZHigh throughput breeding for wood quality improvement.
Davies, Nicholas T.
Eucalypt species are fast-growing and can produce high quality timber for appearance and structural products including Laminated Veneer Lumber (LVL). Their use for solid wood products is hindered by the fact that they can contain large growth-strains, which impose substantial processing costs. Growth-strains are associated with log splitting, warp, collapse and brittle-heart. The body of work presented here focused on the possibility of very-early selection at two years of age, of Eucalyptus bosistoana trees for growth-strain and other wood properties, including an in-depth assessment of the accuracy of the methodology used.
Chapter 1 gives an introduction on growth-stresses in trees and how this knowledge developed over the last century. Chapter 2 describes a pilot study assessing wood properties at a young age. Growth-strain was assessed by measuring stem openings after splitting along the pith, which resulted in a left-censored dataset. A Bayesian approach to the analysis was used to increase the accuracy of genetic parameter prediction from the left- censored data. Chapter 3 tested the hypothesis that the reason for the left-censored data was tension wood formed early in growth resulting in a reversed stress profile. The testing showed this was not the case, at least under the given experimental conditions. Chapter 4 describes a very-early selection trial (age 2) of 81 Eucalyptus bosistoana families with seven measured traits (growth-strain, under-bark diameter, density, stiffness, volumetric shrinkage, height and acoustic velocity), which yielded heritability estimates of 0.23, 0.57, 0.70, 0.77, 0.39, 0.71, and 0.80 respectively. Following this the precision of the splitting test was investigated. Chapter 5 describes an experimental approach which found that the splitting test could predict surface growth-strains with a precision of ±1003 micro- strain (Chapter 5). The accuracy of the splitting test was further investigated in Chapter 6 using a classical mechanics model. The effect of differing surface strain fields on the results of both the splitting test and point measurements such as strain-gauges, indicated that the theoretically obtainable maximum accuracy of the splitting test is approximately ±281 micro-strains. This is similar to four evenly spaced strain-gauges. Finally, Chapter 7 reviews very-early selection and provides guidelines for future breeding projects where reduced cycle times are desired.
2019-01-01T00:00:00ZTree structure in phylogenetic networksSimpson, J. R.http://hdl.handle.net/10092/173992019-10-11T14:01:20Z2019-01-01T00:00:00ZTree structure in phylogenetic networks
Simpson, J. R.
Phylogenetic trees are widely used to express and explore evolutionary relationships.
In recent times, the observation of evolutionary processes that
cannot be expressed by individual phylogenetic trees has prompted interest
in the study of phylogenetic networks. Phylogenetic networks generalise
phylogenetic trees by allowing non-treelike events to be represented. A particular
consequence of this is that a phylogenetic network may be understood
to simultaneously express the relationships of a number of different phylogenetic
trees. These phylogenetic trees are then said to be embedded in the
network.
In this thesis, the connections between various classes of phylogenetic networks
and their corresponding sets of embedded phylogenetic trees are explored.
Among others, the following questions are expanded on and answered.
1. For a given set of trees does there exist a network that embeds each
tree? In the case of level-1 networks a polynomial time algorithm
is given that outputs, up-to a particular topological ambiguity, the
unique level-1 network with minimum reticulations that displays a
given set of trees or identities that no such network exists.
2. From a given set of trees embedded in a network can the network
be reconstructed? It is proven that a normal network can be reconstructed
from a subset of the trees it displays that grows linearly with
respect to the number of leaves in the network.
3. For a given network how many embedded trees are required to use
every vertex and every edge of the network? It is proven that the
class of stack-free network is precisely the class of networks for which
only two embedded trees are required to use every vertex and every
edge of the network.
4. For a given network and tree does the network embed the tree? In
the case of sibling-free networks a polynomial time algorithm is given
that outputs, for a given network and tree, whether or not the network
embeds the tree.
2019-01-01T00:00:00ZEnhancing packed bed geometry using computational fluid dynamic simulationsHoulton, Benhttp://hdl.handle.net/10092/173922019-10-08T14:01:13Z2019-01-01T00:00:00ZEnhancing packed bed geometry using computational fluid dynamic simulations
Houlton, Ben
This work comprises an investigation into how the efficiency of packed beds can be enhanced by manipulating the geometry of the channel network, without altering the process chemistry. A particular focus is placed on investigating how the morphology of the structures influences the mass transfer properties of a passively transported solute species. This includes the degree of axial dispersion and the breakthrough performance of an adsorption column. Using computational fluid dynamic modelling techniques (the Lattice Boltzmann Method) periodic representations of ordered sphere packing and Triply Periodic Minimal Surface (TPMS) structures were subject to laminar flow conditions to compare the mass transport properties. The model parameters were derived from a proposed novel sorption filter for removal of trace heavy metal contaminants from water sources, although the results are expected to be relevant to many other applications.
A detailed outline of the modelling techniques is given, before addressing the performance of each geometry tested. Initially, traditional packing media (arrangements of spherical particles) were compared with TPMS monolithic structures. After confirming that TPMS structures provide enhanced packed column properties when compared with particle arrangements, additional TPMS and manipulated TPMS structures were considered. Parameters describing the geometry and flow fields within each structure are shown to correlate with chromatographic performance and column breakthrough. The relationships between solid phase distribution and enhanced packed bed performance provide a good basis for further investigations into refining packed bed geometry. The Schwarz Diamond and Double Schoen Gyroid TPMS structures were identified as the best performing of the structures that were modelled. Additionally, several practical suggestions are made to assist with the implementation of these findings, such as the expected dimensions of a real system and the sensitivity of results to operating parameters.
2019-01-01T00:00:00Z3D pose estimation in videos using convolutional neural network.Shangguan, Huyuanhttp://hdl.handle.net/10092/173042019-10-17T23:29:04Z2019-01-01T00:00:00Z3D pose estimation in videos using convolutional neural network.
Shangguan, Huyuan
This thesis proposes, develops and evaluates different convolutional neural network based methods for 3D single-person pose estimation in RGB video. The research goals are achieved by studying image processing methods that use machine learning algorithms and applying them to different aspects of the task of pose estimation. The theoretical framework for fulfilling these goals is based on the design of the convolutional neural network for pose estimation task, which has been explored and extended in this work.
Different object detection, object tracking, and activity recognition methods have been compared and evaluated in this thesis. State-of-the-art pose estimation methods which can regress pose from images are extensively reviewed and used as the starting point of this thesis. The thesis also introduces pose-guided image synthesis methods which can be used to create images that contain a person in a given human pose.
This thesis proposes a three-stage CNN-based framework for 3D pose estimation for a single person in RGB video. The task of 3D pose estimation in RGB video is divided into three sub-tasks: human object detection, 2D pose estimation, and 3D pose regression. A state-of-the-art object detection method called Faster RCNN, a state-of-the-art 2D pose estimation method known as Stacked Hourglass, and a greedy-style 3D pose reconstruction method called Projection Matching Pursuit are applied to complete the three sub-tasks respectively. Then the proposed 3D pose estimation framework is evaluated on Human3.6M dataset and an Olympic figure-skating video. The results prove that the proposed framework produces a visually satisfactory 3D pose estimation for many of the poses but not for unusual poses such as those often seen in figure-skating.
One of the reasons that convolutional neural network performs poorly on images with unusual poses is a lack of training data. This thesis proposes a method to augment human pose dataset using generative adversarial network (GAN). The task of human pose dataset augmentation is to generate a large number of labeled pose-image data pairs from a small training dataset. Generative adversarial network shows potential on the area of conditional image synthesis. The dataset augmentation task can be divided into three sub- tasks: pose data augmentation, mask image generation, and RGB image generation. One Variational Autoencoder (VAE) network and two generative adversarial (GAN) networks are designed to complete the three sub-tasks respectively. This method is then evaluated on Human3.6M datasets. The experimental results show that this data augmentation method can help the training of the pose estimation neural network.
2019-01-01T00:00:00Z