Search

Data-Driven Design is an emerging area with the advent of big-data tools. Massive information stored in electronic and digital forms on the internet provides potential opportunities for knowledge discovery in the fields of design and engineering. The aim of the research reported in this thesis is to facilitate the design information retrieval process based on large-scale electronic data through the use of text mining and semantic network techniques. We have proposed a data-driven pipeline for design information retrieval including four elements, from data acquisition, text mining, semantic network analysis, to data visualisation and user interaction. Web crawling techniques are applied to fetch massive online textual data in data acquisition process. The use of text mining enables the transformation of data from unstructured raw texts into a structured semantic network. A retrieval analysis framework is proposed based on the constructed semantic network to retrieve relevant design information and provoke design innovation. Finally, a web-based platform B-Link has been developed to enable user to visualise the semantic network and interact with it through the proposed retrieval analysis framework. Seven case studies were conducted throughout the thesis to investigate the effectiveness and gain insights for each element of the pipeline. Thousands of design post news items and millions of engineering and design peer reviewed papers can be efficiently captured by web crawling techniques. Through the use of itemset mining and noun phrase chunking, a semantic network constructed based on these textual data is shown to capture more inherent design- and engineering-oriented concepts and relations, compared to the benchmarking approaches: WordNet, ConceptNet, NeLL and Wikipedia. A retrieval analysis framework has been developed with different retrieval behaviours to retrieve either common general or domain-specific concepts, explicit or implicit knowledge relations, which are found to satisfy various knowledge demands in our real design projects at the conceptual stage. Finally, the result of a user test is shown to be consistent with these findings.

As one of the most distinctive characteristics of English insurance law, the duty of utmost good faith is essentially stated in sections 17-20 of the Marine Insurance Act 1906. According to the statutory rules, both of the insurance parties must observe utmost good faith before the conclusion of an insurance contract. After one century of its application, both the judiciary and academics expressed their concerns in terms of its legislative defects and complexity in practice. Some developments have been made in recent judicial decisions and in statutory reform, e.g. the English Consumer Insurance (Disclosure and Representations) Act 2012, and Recommendations, Statutes and Explanations on the Amendments of Chinese Maritime Code of the People’s Republic of China. Therefore, debatable issues and law reform programs in both English and Chinese law are considered in the main body of this thesis. The examination is essentially based upon, (1) the materiality test of the concealed/misrepresented circumstances which can empower the injured party to rescind the insurance abinitio; (2) the duration of utmost good faith and specific issues; (3) the protective measures related to innocent misconduct; (4) the legal status of good faith and its application to fraudulent behaviour; and (5) whether the classic English utmost good faith doctrine can be extended to Chinese law. Therefore, the main objective of this thesis is to provide a comprehensive study of the current status and developments of the duty of utmost good faith in both English and Chinese law, which is of fundamental importance, not only at the negotiation stage, but also throughout the performance and at the claiming stage of an insurance contract. After identifying and analysing these crucial issues, this thesis concludes with some possible solutions.

We are now in an era of the Big Bang of artificial intelligence (AI). In this wave of revolution, both industry and academia have cast numerous funds and resources. Machine learning, especially Deep Learning, has been widely deployed to replace the traditional algorithms in many domains, from the euclidean data domain to the non-euclidean domain. As the complexity and scale of the AI algorithms increase, the system host these algorithms requires more computational power and resources than before. Using the design of the modules of the video analytic platform as the use cases, we analyze the workload cost for computational resource and memory allocation during the execution of the system. The video analytic platform is a complex system that comprises various computer vision and decision-making tasks. Every module accomplishing a specific task is a stage in the pipeline of the video analytic platform. With the analyses mentioned above, we synthesize the adaptive AI algorithms from availability and variability perspectives, such as optimization with tensorization or matricization. We conceive the sparse Transformer and segmented linear Transformer as the critical components for the human action recognition task. The Constraint Satisfaction Problem is employed to assist the decision-making in the scene parsing stage. To facilitate this fulfillment of this task, we designed a hybrid model for graph learning-based SAT solver. Graph matching is employed at the final stage for the scene understanding task. We implemented a hybrid model of GNN and Transformer architecture. Finally, we design the unified hardware acceleration architecture for both dense and sparse data based on the optimizations of algorithms. Our design of the architecture targets the arithmetic operation kernels, such as matrix multiplications, with the help of data transformation and rearrangement. We first transform the inputs and weights with Winograd transform for dense convolution operations, then we feed the transformed data to the matrix multiplication accelerator. While for sparse data, we need to utilize the index to nonzero to fetch data; therefore, the indexation, scattering, and gathering are crucial components, effective implementation will dramatically improve the system's overall performance. To improve the matrix multiplication accelerator's efficiency and reduce the number of heavy arithmetic operations and the number of memory accesses, we also conduct the hardware-based recursive algorithm, i.e., Strassen's algorithm for matrix multiplication.

Mode conversion at the magnetopause has been suggested to lead to the generation of kinetic Alfven waves (KAWs) and effective mass transport from the solar wind into the magnetosphere. To investigate the mode conversion process in the dynamics dayside system, a systematic 3-D global hybrid simulation that includes the solar wind, bow shock and the geomagnetic field is carried out for the dayside magnetosphere under various interplanetary magnetic field (IMF) conditions. In the quasi-parallel (Q-||) shock, the foreshock compressional pulses are self-consistently generated by the interaction between the solar wind and the dipolar geomagnetic field. The simulation results show that as the compressional pulses propagate from the magnetosheath to the magnetopause, short wavelength structures of k⊥ ρi ~ 1 with enhanced parallel electric field E|| are excited in the magnetopause boundary layer (MPBL), where k⊥ ~ kx is the perpendicular wave number nearly along the GSE x direction, and ρi is the ion Larmor radius. The wave phase relationship between the magnetic field and the density changes from in-phase in the magnetosheath to anti-phase in the short-wavelength MPBL perturbations. The wave polarization is predominantly compressive in the magnetosheath, whereas strong transverse wave powers appear abruptly around the MPBL. The mode conversion from the compressional pulses to KAWs is identified around the predicted Alfven resonance points in the MPBL. As these KAWs evolve, KAW modes dominated by azimuthal wave number with ky ρi ~ 1 are also generated. The short-wavelength KAWs in the MPBL are identified by the sharp increases in E|| and the transverse electromagnetic field polarization relation of Alfven mode. These KAW perturbations propagate poleward into the cusps along the MPBL. Due to the differential flow convection speeds at various latitudes, the KAW packets expand along the north-south direction, while they may also merge with newly formed KAWs generated by newly-arrived compressional waves. The mode conversion takes place downstream of the Q-|| shock, where the driver compressional waves are the strongest in their global distribution. In the cases with a radial IMF, the mode conversion takes place in the subsolar region. In the cases with a more general IMF direction, the Q-|| shock regions and thus the strongest mode conversion shift off the subsolar, although the KAWs may still be identified near the subsolar region around the MPBL. In the dipole field region of the magnetosphere, field line resonance associated with the standing shear Alfven waves is found due to the mode conversion. In the presence of the IMF By component, the field line oscillations in the inner magnetosphere are more intense downstream of the Q-|| shock than on the quasi-perpendicular (Q-⊥) side, indicating again that the KAWs from the mode conversion is in favor of the Q-|| side. Over several periods of incoming magnetic pulses, the KAWs approximately show a symmetric pattern from north to south. This symmetry, however, is not observed in the existence of the northward component (Bz in GSM) in the IMF, because the Q-|| is now located in the southern hemisphere where more KAWs could be perceived. While observational signatures are provided by the global simulation, the simulation results are also compared with satellite observations for the physical evidence of mode conversion.

A severe harmful algal bloom (HAB) devastated marine life in the Wellington Harbour in the late summer 1998. A novel species of Karenia dinoflagellates was isolated from this HAB and named Karenia brevisulcata (Chang). Toxins produced by this alga were categorised into i) brevisulcatic acids (BSXs), a group of six polycyclic ethers with molecular weights 800-950; ii) brevisulcenals (acronym KBTs from a previous name), a complex of 6-8 lipophilic compounds with molecular weights 1900-2200. This PhD thesis reports the isolation, purification and chemical investigation of BSXs, analytical and toxicological studies of BSXs and KBTs, and the toxicity evaluation of three Karenia species. LC-MS provides a powerful micro-analytical technology to detect novel toxins, determine structural characteristics, guide chemical isolation and provide quantitative data on levels of toxins in cultures. An LC-MS (SIR method) with six channels for BSXs and one channel for brevetoxin-2 (calibrant) was developed to guide BSX isolation. A more sensitive method based on solid phase extraction (SPE) and LC-MS (MRM) was validated for determination of 6 BSXs and 4 KBTs in cultures with LOQs for each toxin of ca 0.15 µg L⁻¹ in seawater. Barrels, carboys and a photobioreactor were used to optimise cell growth and toxin production of a K. brevisulcata isolate collected by Cawthron Institute in 1998. Carboys were the most efficient in batch culture mode with the highest growth rate (μ=0.086 day⁻¹ versus 0.050-0.057 day⁻¹ for barrels or photobioreactor) and the highest cell yield for total BSXs (0.77 pg cell⁻¹ versus 0.29-0.38 pg cell⁻¹ for barrel or photobioreactor). However, the large volume automated photobioreactor operated in continuous culture mode was more suited to large-scale toxin production from dinoflagellates. K. brevisulcata cultures (total 936 litres unlabelled; 492 litres ¹³C-labelled using added NaH¹³CO₃) were grown at Cawthron Institute. The toxins were extracted from the bulk cultures by cell disruption and absorption using HP20 polymeric resin followed by acetone extraction. Solvent partitioning of crude acetone extracts was used to prepare neutral fractions containing KBTs and acidic fractions containing BSXs. Semi-purification of BSXs from the acidic fractions used desalting by reverse phase SPE and defatting by normal phase SPE. BSX-1 (C₄₉H₇₂O₁₆, [M+H]⁺ 917.5) and BSX-2 (C₄₇H₆₈O₁₅, [M+H]⁺ 873.5) were separated by normal phase column chromatography and were purified in mg quantities using preparative C18 high performance liquid chromatography (HPLC). BSX-4 (C₄₉H₇₀O₁₅, [M+H]⁺ 899.5) and BSX-5 (C₄₇H₆₆O₁₄, [M+H]⁺ 855.5) are the lactones which were produced from BSX-1 and BSX-2 respectively by acid-catalysed ring closure for analytical, structural and toxicological studies. The neutral fractions of crude extracts were sent to the University of Tokyo (Prof. Masayuki Satake) where four KBTs (-F, -G, -H & -I) were isolated. The chemical structures of BSXs and KBTs were characterised by spectroscopic (UV, MS-MS) and micro-derivatisation techniques (methylation, acetylation, hydrolysis). BSXs are ladder-frame polycyclic ethers with close similarities to brevetoxins. A probable structure for BSX-4 has been determined based on partial NMR spectra obtained from Massey University. BSX-4 has structural similarities to brevetoxin-1 but with some differences in the polyether rings and a longer, carboxylated side chain. However, the full structures of BSXs have not been completely elucidated because of missing NMR peaks due to conformation changes in the time frame of the NMR pulse sequences. Missing signals, even at low temperatures, probably involve three flexible 7-, 8- and 9-membered ether rings in the central part of the molecule with one double bond and a tertiary methyl group. NMR studies are continuing at the University of Tokyo. The complex structure of KBT-F (C₁₀₇H₁₆₀O₃₈, [M+H]⁺ 2054.1) has been elucidated at the University of Tokyo using NMR and TOF-TOF MS. KBT-F is a novel ladder frame polycyclic ether compound with a 2-methyl-2-butenal side chain that is the same as that in gymnocin-A. It has the most contiguous ethers rings amongst the known polycyclic ethers. Acute toxicity, haemolytic activity and cytotoxicity of BSXs and KBTs were studied by mouse bioassay, Bignami assay, and in vitro mammalian cell bioassays respectively. BSXs are weakly toxic to mice with intraperitoneal (i.p.) LD₅₀ of > 1500 μg kg⁻¹ body weight (bw), and weakly haemolytic. Neuro2a cell assays revealed that BSXs are brevetoxin-like voltage gated Na channel (VGSC) agonists but with lower activity. KBT-F and KBT-G are strongly toxic to mice (i.p. LD₅₀ of 32-40 μg kg⁻¹ bw), strongly haemolytic, and highly cytotoxic to P388, Vero and Neuro2a cells, but are unlikely to be VGSC agonists. Further cytotoxicity assays using multiple end-points e.g. LDH/MTS and LDH/caspase-3/7 revealed that damage to P388 cells by KBT-F involved effects on membranes but apoptotic mechanisms were not involved. In vivo bioassays using marine biota (two fish species and larvae of six marine invertebrate species) were used to investigate the toxicity and potential modes-of-action of three toxic Karenia species. The bioassays confirmed field observations on the very high ecotoxic risks posed by blooms of K. brevisulcata. A provisional trigger of 1 x 10⁴ cells L⁻¹ is recommended to provide an early warning of a K. brevisulcata bloom. The mechanism for fish and invertebrate kills by K. brevisulcata is likely to depend on release of toxins from algal cells into seawater with KBTs being the main toxic agents.

A variety of nanocrystals, nanoparticles or quantum dots are fabricated using nucleation and growth processes. Therefore, a fundamental understanding of nucleation and growth is crucial to materials science and engineering on the nanoscale. In this dissertation, we explore the fundamental characteristics of nucleation and growth in multiple dimensional systems using several different methods. One method which has been found to be particularly useful is the Monte Carlo (MC) method. In particular, the kinetic Monte Carlo (KMC) method has made MC simulations of complicated many body systems very efficient. In this dissertation, we use KMC simulations to study nucleation, growth, and coarsening in a variety of different systems. In addition, we have carried out a theoretical analysis using rate equations. We have also carried out investigations of the fundamental characteristics of the coarsening process using parallel methods based on a newly developed parallel KMC algorithm.This dissertation is organized in two parts, the first part is about fundamental characteristics of multiple dimensional systems, the second part is about parallel KMC calculation of coarsening process. In Part I, we first study the fundamental characteristics of nucleation and growth in 3 dimensional (3D) systems using a simplified model of nucleation and growth. One of the main goals of this work is to compare with previous work on 2D nucleation and growth in order to understand the effects of dimensionality. The scaling of the average island-size, island density, monomer density, island-size distribution (ISD), capture number distribution (CND), and capture zone distribution (CZD) are studied as a function of the fraction of occupied sites (coverage) and the ratio D/F of the monomer hopping rate D to the (per site) monomer creation rate F. Our model may be viewed as a simple model of the early-stages of vacancy cluster nucleation and growth under irradiation. Good agreement is found between our mean-field (MF) rate-equation results for the average island and monomer densities and our simulation results. In addition, we find that due to the decreased influence of correlations and fluctuations in 3D as compared to 2D, the scaled CND depends only weakly on the island-size. As a result the scaled ISD is significantly sharper than obtained in 2D and diverges with increasing D/F. However, the scaled ISD obtained in kinetic Monte Carlo (KMC) simulations appears to diverge more slowly with increasing D/F than the MF prediction while the divergence occurs at a value of the scaled island-size which is somewhat beyond the MF prediction. These results are supported by an analysis of the asymptotic CND. The final goal for understanding the mechanism of nucleation and growth is to develop a theory to concisely and precisely disclose the law underlying the nucleation and growth process. From the theoretical point view, dimension can be taken as a variable to develop theory. In order to obtain the upper critical dimension corresponding to MF theories, we compare the results of kinetic Monte Carlo (KMC) simulations of a point-island model of irreversible nucleation and growth in four-dimensions with the corresponding mean-field (MF) rate equation predictions for the monomer density, island density, island-size distribution (ISD), capture number distribution (CND), and capture zone distribution (CZD), in order to determine the critical dimension d_c for mean-field behavior. The asymptotic behavior is studied as a function of the fraction of occupied sites (coverage) and the ratio D/F of the monomer hopping rate D to the (per site) monomer creation rate F. Excellent agreement is found between our KMC simulation results and the MF rate equation results for the average island and monomer densities. For large D/F, the scaled CND and CZD do not depend on island-size in good agreement with the MF prediction, while the scaled ISD also agrees well with the MF prediction except for a slight difference at the peak values. Coupled with previous results obtained in d = 3, these results indicate that the upper critical dimension for irreversible cluster nucleation and growth is equal to 4. While most of this work focusses on the correlations between various quantities and the island-size, it is also interesting to study the dependence of global quantities, such as the global CZD on dimensionality. In order to compare with recent analytic theories for the global CZD we have also studied the dependence of the global CZD using KMC simulations as a function of dimensionality in, one, two, three, and four dimensions. In general, we find that while the CZD depends on the short-range interaction for finite D/F (corresponding to the ratio of monomer hopping rate D to deposition rate F) in the asymptotic limit of infinite D/F there is no significant dependence. However, poor agreement is found between the asymptotic CZD and the predicted Wigner distributions. Our results also indicate that for d = 2 and 3 the asymptotic CZD is independent of model details and dimension. However, for d = 1 and d = 4 the resulting distribution is significantly more sharply peaked. We also find that in contrast to the island-size distribution, for which mean-field-like behavior is observed in d = 3 and above, the asymptotic CZD is significantly broadened by fluctuations even in d = 4. In addition to nucleation and growth, coarsening also plays a significant role in nano-particle fabrication and thin film growth. Therefore, in Part II, we explore the coarsening process using KMC simulations. Exploring the fundamental characteristics of coarsening process requires large-scale simulations which cover a large time scale. In this study, the results of parallel kinetic Monte Carlo (KMC) simulations of island coarsening based on a bond-counting model are obtained. Our simulations were carried out both as a test of and as an application of the recently developed semi-rigorous synchronous sublattice (SL) algorithm. By carrying out simulations over long times and for large system sizes the asymptotic coarsening behavior and scaled island-size distribution (ISD) were determined. Our results indicate that while cluster diffusion and coalescence play a role at early and intermediate times, at late times the coarsening proceeds via Ostwald ripening. In addition, we find that due to fluctuations which become important in 2D, the asymptotic scaled ISD is significantly narrower and more sharply peaked than the mean-field theory prediction. The dependence of the scaled ISD on coverage is also studied. Our results demonstrate that parallel KMC simulations can be used to effectively extend the time-scale over which realistic coarsening simulations can be carried out. These results also suggest that the SL algorithm is likely to be useful in the future in parallel KMC simulations of more complicated models of coarsening.