In chemical literature much information is given in the form of diagrams depicting chemical structures. In order to access this information electronically, diagrams have to be recognized and translated into a processable format. Although a number of approaches have been proposed for the recognition of molecule diagrams in the literature, they traditionally employ procedural methods with limited flexibility and extensibility. This thesis presents a novel approach that models the principal recognition steps for molecule diagrams in a strictly rule based system. We develop a framework that enables the definition of a set of rules for the recognition of different bond types and arrangements as well as for resolving possible ambiguities. This allows us to view the diagram recognition problem as a process of rewriting an initial set of geometric artefacts into a graph representation of a chemical diagram without the need to adhere to a rigid procedure. We demonstrate the flexibility of the approach by extending it to capture new bond types and compositions. In experimental evaluation we can show that an implementation of our approach outperforms the currently available leading open source system. Finally, we discuss how our framework could be applied to other automatic diagram recognition tasks.

Currently used rocket propellants such as hydrazine, monomethylhydrazine, unsymmetrical dimethylhydrazine and nitrogen tetroxide are carcinogenic and toxic to the environment and therefore special protective measures are required when producing, transporting, storing and handling them. Employing alternatives could possibly save costs and this has revived the research interest in so called green propellants. Hydrogen peroxide is such a possible alternative. It requires a catalyst bed to decompose the liquid peroxide into steam and oxygen. The purpose of this work is to design numerical tools that describe the processes in the catalyst bed and subsequently employ these tools to predict the performance of the catalyst bed and investigate the influence of design choices on the performance. In contrast to the models described in the literature, the tools developed in this thesis are two fluid models. In order to test the reliability of the tools results are compared with experimental data. A single control volume two-fluid model has been developed to investigate the pressure drop over the catalyst bed and the influence of the shape and size of catalyst pellets on the pressure drop. Parametric studies with this model revealed that the Tallmadge equation gives a better prediction of the pressure gradient than the more traditionally employed Ergun equation. It was also found that for a given bed length cylindrical pellets with a diameter to length ratio of 2 or more give a lower pressure drop than cylindrical pellets, while achieving the same level of decomposition. A one-dimensional two-fluid model has been developed to obtain longitudinal variations of fluid properties. This model revealed that the catalyst bed can be divided into 3 sections: a pre-boiling region, rapid conversion region and a dry-out region. It was shown that most of the mass transfer takes place due to evaporation. A sensitivity analysis showed that the gas-liquid interfacial area hardly influences the results.

In order to study computationally increasingly complex systems using theoretical methods model Hamiltonians are required to accurately describe the potential energy surface they represent. Also ab-initio methods improve the calculation of the excited states of these complex systems becomes increasingly feasible. One such model Hamiltonian described herein, the Vibronic Coupling Hamiltonian, has previously shown its versitility and ability to describe a variety of non-adiabatic problems. This thesis describes a new method, a genetic algorithm, for the parameterisation of the Vibronic Coupling Hamiltonian to describe both previously calculated potential energy surfaces (allene and pentatetraene) and newly calculated (cyclo-butadiene and toluene) potential energy surfaces. In order to test this genetic algorithm quantum nuclear dynamics calculations were performed using the multi-configurational time dependent hartree method and the results compared to experiment.

Hydrocarbon selective catalytic reduction (HC-SCR) is emerging as one of the most practical methods for the removal of nitrogen oxides (NOx) from light-duty-diesel engine exhaust gas. In order to further promote the chemical reactions of NOx-SCR by hydrocarbons, an understanding of the HC-SCR process at the molecular level is necessary. In the present work, a novel surface-reaction mechanism for HC-SCR is set up with emphasis on microkinetic analysis aiming to investigate the chemical behaviour during the process at a molecular level via detailed elementary reaction steps. Propane (C3H8) is chosen as the reductant of HC-SCR. The simulation is designed for a single channel of a monolith, typical for automotive catalytic converters, coated with a silver alumina catalyst (Ag/Al2O3). The complicated physical and chemical details occurring in the catalytic converter are investigated by using the numerical method of computational fluid dynamics (CFD) coupled with the mechanism. The C3H8-SCR reaction mechanism consists of 94 elementary reactions, 24 gas-phase species and 24 adsorbed surface species. The mechanism is optimised by tuning some important reaction parameters against some measurable data from experiments. The optimised mechanism then is validated with another set of experimental data. The numerical simulation shows good agreements between the modelling and the experimental data. Finally, the numerical modelling also provides information that is difficult to measure for example, gas-phase concentration distribution, temperature profiles, wall temperatures and the occupation of adsorbed species on catalyst surface. Consequently, computational modelling can be used as an effective tool to design and/or optimise the catalytic exhaust aftertreatment system.

Data collected in life sciences studies mostly include a genotype description of the organism, a phenotype characterisation of the organism, and experiment-specific covariates including a description of experimental procedures and laboratory (environmental) conditions. Here, phenotype measurements are taken for Neurospora crassa (wild type) growing on agar in the standard laboratory conditions. I define a phenotype as a set of traits including apical extension velocity, branching angle, and branching distance. I use the above measures (traits) to model (estimate) biologically complex filamentous fungi network as a simplified 'In Silico Fungus' consisting of series of straight lines. Phenotype data, under the central limit theorem, is often characterized by means and standard deviations. Subsequently, P values are used to show statistical validity. Here, I question whether making normality assumption based on the popularity of such approach is always justified. Therefore, I test three different scenarios by making different assumptions about the data collected. (1) Firstly, I use the most popular approach: I assume the phenotype data comes from the continuous, normal (Gauss) distribution. Thus, I predict the future measurement outcomes by using normal (Gauss) parametric approximation. (2) Secondly, I use the most intuitive approach: I do not make any assumptions about the data collected and use it to predict the future measurement outcomes by withdrawing values pseudo randomly from the actual, raw, and discrete dataset. (3) Finally, I use the strategy balanced between the previous two: I construct a customised, continuous, and non-parametric distribution based on the data collected. Thus, I predict the future measurement outcomes by using kernel density estimation method. Subsequently, I implement all of the strategies above: (1), (2), and (3) in the in silico fungus programme to compare the computer simulation outcomes. More specifically, I compare the surface coverage, expressed as the proportion of the surface occupied by the fungus. Obtained results show that the differences between different data regimes (1), (2), and (3) are significant. Therefore, I conclude that the correct assessment of the data normality is crucial for the correct interpretation and implementation of scientific observations. I suspect the described data classification process determines successful implementation of biological findings especially in the fields such as medicine and engineering.