7103 research outputs found
Sort by
Modelling on-domestic buildings energy performance using machine learning methods, a case study of the UK
Get PDFThis thesis was previously held under moratorium from 29th July 2020 to 29th July 2022.In the UK, only 7% of non-domestic buildings are newly built, whilst this sector generates 20% of total gas emission. Consequently, the government has setregulations to decrease the amount of energy take-up by buildings. It is apparent from the seminal literature that deep energy retrofit is the primary solution to achieve that goal. Due to the size and complexity of non-domestic buildings, finding optimum plans is cumbersome. To that end, artificial intelligence has been employed to assist this decision-making procedure, yet limited to high time-complexity of energy simulations. Surrogate modelling seems a promising alternative for simulation software, developing accurate energy prediction models requires an understanding of the building physics and a vision on the use of data-driven models. This study evaluated the accuracy and time complexity of most popular Machine Learning (ML) methods in the buildings energy efficiency estimation. It established an approach based on evolutionary optimisation to reach the highest potential of MLs in predicting buildings energy performance. It then developed an energy performance prediction model for the UK non-domestic buildings with the aid of ML techniques. The ML model amid at supporting multi-objective optimisation of energy retrofit planning by accelerating energy performance computation. The study laid out the process of model development from the investigation of requirements and feature extraction to the application on a case study. It outlines a framework to represent the building records as a set of features in away that all alterations produced by applying retrofit technologies can be captured by the model to generate accurate energy ratings. The model provides a reliable tool to explore a large space of the available building materials and technologies for evaluating thousands of buildings going under retrofit to fulfil the energy policy targets and enables building analysts to explore the expanding solution space meaningfully.In the UK, only 7% of non-domestic buildings are newly built, whilst this sector generates 20% of total gas emission. Consequently, the government has setregulations to decrease the amount of energy take-up by buildings. It is apparent from the seminal literature that deep energy retrofit is the primary solution to achieve that goal. Due to the size and complexity of non-domestic buildings, finding optimum plans is cumbersome. To that end, artificial intelligence has been employed to assist this decision-making procedure, yet limited to high time-complexity of energy simulations. Surrogate modelling seems a promising alternative for simulation software, developing accurate energy prediction models requires an understanding of the building physics and a vision on the use of data-driven models. This study evaluated the accuracy and time complexity of most popular Machine Learning (ML) methods in the buildings energy efficiency estimation. It established an approach based on evolutionary optimisation to reach the highest potential of MLs in predicting buildings energy performance. It then developed an energy performance prediction model for the UK non-domestic buildings with the aid of ML techniques. The ML model amid at supporting multi-objective optimisation of energy retrofit planning by accelerating energy performance computation. The study laid out the process of model development from the investigation of requirements and feature extraction to the application on a case study. It outlines a framework to represent the building records as a set of features in away that all alterations produced by applying retrofit technologies can be captured by the model to generate accurate energy ratings. The model provides a reliable tool to explore a large space of the available building materials and technologies for evaluating thousands of buildings going under retrofit to fulfil the energy policy targets and enables building analysts to explore the expanding solution space meaningfully
An integrated machine learning framework for enhanced vessel operational efficiency
Get PDFThis thesis was previously held under moratorium from 28th October 2021 until 28th October 2022.Inadequate machinery maintenance and inefficient sailing performance comprise two major hindrances to vessel operational sustainability and profitability. To ensure that vessel operation remains competitive while its environmental impact is mitigated, the development of a systematic approach for vessel monitoring and operational enhancement is required. Currently, the maritime industry predominantly operates on a hybridisation of corrective and preventive maintenance, along with monitoring and decision making based on past experience. More intelligent, data-driven approaches are slowly permeating the industry; these offerings however remain largely rudimentary, retaining considerable assumptions and data requirements for their application. In this respect,this thesis aims to enhance operational efficiency in the maritime industry through the development of an integrated machine learning framework combining efficient and robust machinery anomaly detection, vessel performance degradation monitoring, and routing decision support. This is achieved through a number of key objectives, including: a) the identification of research gaps; b) the extraction of meaningful information for available data sources; c) the monitoring of machinery condition and detection of incipient anomalies; d) the identification of optimal data-driven Fuel Oil Consumption (FOC) modelling architectures; e) the monitoring of vessel performance based on FOC modelling; the facilitation of optimal routing through a suitable Decision Support System (DSS); and f) the demonstration and validation of the above through appropriate case studies. The proposed aim and objectives are accomplished through the combination of a robust pre-processing methodology with a number of data-driven modelling methods (e.g. One-Class Support Vector Classifiers (OCSVCs), Deep Neural Networks(DNNs)), and a novel modification of Dijkstra’s algorithm. A key novelty aspect of this proposed framework is derived by the development and combination of a number of data-driven methodologies for the operational efficiency enhancement of a vessel. Moreover, a novelty of the approach lies upon the minimisation of the inherent assumptions required, streamlining its use in a diverse set of applications. In the same vein, a novel aspect of the proposed framework concerns its flexibility to operate using datasets from different sources, exhibiting different levels of granularity and frequency. This framework is applied to a number of case studies, covering data pre-processing, engine condition monitoring, a FOC modelling comparison, FOC-based performance monitoring, and optimal routing. This helps verify the framework’s robustness in a range of realistic scenarios applicable to a variety of vessel types (e.g. reefer, containership, bulk carrier). These case studies, among others, demonstrated the robustness of the anomaly detection methodology when examining different parameters and systems, the accuracy deviation when predicting a vessel’s FOC using Automated Data Logging & Monitoring (ADLM) or noon-report data and the optimal models for each case, a successful evaluation of the performance monitoring methodology as a vessel’s fouling increases; and the identification of optimal routes as a vessel sails from the Gulf of Guinea to Marseille anchorage.Inadequate machinery maintenance and inefficient sailing performance comprise two major hindrances to vessel operational sustainability and profitability. To ensure that vessel operation remains competitive while its environmental impact is mitigated, the development of a systematic approach for vessel monitoring and operational enhancement is required. Currently, the maritime industry predominantly operates on a hybridisation of corrective and preventive maintenance, along with monitoring and decision making based on past experience. More intelligent, data-driven approaches are slowly permeating the industry; these offerings however remain largely rudimentary, retaining considerable assumptions and data requirements for their application. In this respect,this thesis aims to enhance operational efficiency in the maritime industry through the development of an integrated machine learning framework combining efficient and robust machinery anomaly detection, vessel performance degradation monitoring, and routing decision support. This is achieved through a number of key objectives, including: a) the identification of research gaps; b) the extraction of meaningful information for available data sources; c) the monitoring of machinery condition and detection of incipient anomalies; d) the identification of optimal data-driven Fuel Oil Consumption (FOC) modelling architectures; e) the monitoring of vessel performance based on FOC modelling; the facilitation of optimal routing through a suitable Decision Support System (DSS); and f) the demonstration and validation of the above through appropriate case studies. The proposed aim and objectives are accomplished through the combination of a robust pre-processing methodology with a number of data-driven modelling methods (e.g. One-Class Support Vector Classifiers (OCSVCs), Deep Neural Networks(DNNs)), and a novel modification of Dijkstra’s algorithm. A key novelty aspect of this proposed framework is derived by the development and combination of a number of data-driven methodologies for the operational efficiency enhancement of a vessel. Moreover, a novelty of the approach lies upon the minimisation of the inherent assumptions required, streamlining its use in a diverse set of applications. In the same vein, a novel aspect of the proposed framework concerns its flexibility to operate using datasets from different sources, exhibiting different levels of granularity and frequency. This framework is applied to a number of case studies, covering data pre-processing, engine condition monitoring, a FOC modelling comparison, FOC-based performance monitoring, and optimal routing. This helps verify the framework’s robustness in a range of realistic scenarios applicable to a variety of vessel types (e.g. reefer, containership, bulk carrier). These case studies, among others, demonstrated the robustness of the anomaly detection methodology when examining different parameters and systems, the accuracy deviation when predicting a vessel’s FOC using Automated Data Logging & Monitoring (ADLM) or noon-report data and the optimal models for each case, a successful evaluation of the performance monitoring methodology as a vessel’s fouling increases; and the identification of optimal routes as a vessel sails from the Gulf of Guinea to Marseille anchorage
Novel applications of advanced optical microscopy for microbiology
No full textThis thesis was previously held under moratorium from 30th October 2020 until 30th October 2022The study of bacteria often requires visualisation by optical microscopy, but the use of advanced optical microscopy methods is uncommon by many microbiologists. Therefore, there remains many areas of microbiology which require exploration using newly developed techniques. This thesis describes the application of advanced optical microscopy methods to three distinct microbiological questions centred on bacterial gliding motility, the spatial organisation of biofilms, and the growth of bacteria in a mimetic three-dimensional (3D) culture environment. The gliding motility of Myxococcus xanthus has been described as a lateral process, and it was unclear if single bacteria were capable of moving in three dimensions. This was due to three-dimensional imaging of bacteria often being unachievable by optical methods due to the height of a bacterial cell being on the order as the axial resolution of the conventional optical microscope. To overcome this a novel variant of the livecell label-free technique, interference reflection microscopy (IRM), was developed. This method relies on the interference of multiple wavelengths of incident and reflected light and results in a series of intensity maxima and minima which encode 3D information. A specimen of known geometry was used to characterise this method before application to gliding M. xanthus cells. Multi-wavelength confocal IRM revealed that M. xanthus exhibited aperiodic oscillations during gliding, which challenged the theory that gliding motility was a lateral phenomenon. By use of deleterious mutants, it was deduced that the oscillatory behaviours were not linked to the main driving force of gliding, proton motive force. A hypothesis was proposed which suggested that these behaviours were caused by recoil and force transmission along the cell body following firing of the Type IV pili. Bacterial biofilms have been studied by conventional microscopy methods for over 50 years; however due to a technology gap, the structure of large microbial aggregates remained unclear. The development of the Mesolens, an optical system which uniquely allows simultaneous imaging of individual bacteria over a 36 mm2 field of view, enabled the study of mature Escherichia coli macro-colony biofilm architecture like never before. The Mesolens enabled the discovery of intra-colony channels on the order of 10 µm in diameter that are integral to E. coli macro-colony biofilms and form as an emergent property of biofilm growth. These channels have a characteristic structure and reform after total mechanical disaggregation of the colony, facilitate transport of particles, and play a role in the acquisition of and distribution of nutrients through the biofilm. Furthermore, intra-colony channels potentially offer a previously unobserved route for the delivery of dispersal agents or antimicrobial drugs to biofilms, which would ultimately lower their impact on public health and industry. The practice of bacterial culture has remained unchanged for over a century. Therefore, almost all observations of bacterial behaviour have been made using synthetic laboratory condition which are not representative of the natural environment. To address this, a mimetic 3D transparent soil culture medium was fabricated and designed specifically for bacterial culture. This novel culture medium was optimised for two wide-ranging genera of soil bacteria, Streptomyces coelicolor and Bacillus subtilis. Following careful design of the transparent soil platform, each strain was imaged using the Mesolens to provide a better understanding of how they colonised their natural habitat. Each species was found to colonise the surface of soil independently of their growth behaviours on traditional two-dimensional culturemethods. Moreover, the viability of bacteria grown in transparent soil was found to be uncompromised. Therefore, transparent soil stands as a readily tailored platform for bacterial culture and is compatible with any optical microscope to study bacterial behaviours in a mimetic soil environment.The study of bacteria often requires visualisation by optical microscopy, but the use of advanced optical microscopy methods is uncommon by many microbiologists. Therefore, there remains many areas of microbiology which require exploration using newly developed techniques. This thesis describes the application of advanced optical microscopy methods to three distinct microbiological questions centred on bacterial gliding motility, the spatial organisation of biofilms, and the growth of bacteria in a mimetic three-dimensional (3D) culture environment. The gliding motility of Myxococcus xanthus has been described as a lateral process, and it was unclear if single bacteria were capable of moving in three dimensions. This was due to three-dimensional imaging of bacteria often being unachievable by optical methods due to the height of a bacterial cell being on the order as the axial resolution of the conventional optical microscope. To overcome this a novel variant of the livecell label-free technique, interference reflection microscopy (IRM), was developed. This method relies on the interference of multiple wavelengths of incident and reflected light and results in a series of intensity maxima and minima which encode 3D information. A specimen of known geometry was used to characterise this method before application to gliding M. xanthus cells. Multi-wavelength confocal IRM revealed that M. xanthus exhibited aperiodic oscillations during gliding, which challenged the theory that gliding motility was a lateral phenomenon. By use of deleterious mutants, it was deduced that the oscillatory behaviours were not linked to the main driving force of gliding, proton motive force. A hypothesis was proposed which suggested that these behaviours were caused by recoil and force transmission along the cell body following firing of the Type IV pili. Bacterial biofilms have been studied by conventional microscopy methods for over 50 years; however due to a technology gap, the structure of large microbial aggregates remained unclear. The development of the Mesolens, an optical system which uniquely allows simultaneous imaging of individual bacteria over a 36 mm2 field of view, enabled the study of mature Escherichia coli macro-colony biofilm architecture like never before. The Mesolens enabled the discovery of intra-colony channels on the order of 10 µm in diameter that are integral to E. coli macro-colony biofilms and form as an emergent property of biofilm growth. These channels have a characteristic structure and reform after total mechanical disaggregation of the colony, facilitate transport of particles, and play a role in the acquisition of and distribution of nutrients through the biofilm. Furthermore, intra-colony channels potentially offer a previously unobserved route for the delivery of dispersal agents or antimicrobial drugs to biofilms, which would ultimately lower their impact on public health and industry. The practice of bacterial culture has remained unchanged for over a century. Therefore, almost all observations of bacterial behaviour have been made using synthetic laboratory condition which are not representative of the natural environment. To address this, a mimetic 3D transparent soil culture medium was fabricated and designed specifically for bacterial culture. This novel culture medium was optimised for two wide-ranging genera of soil bacteria, Streptomyces coelicolor and Bacillus subtilis. Following careful design of the transparent soil platform, each strain was imaged using the Mesolens to provide a better understanding of how they colonised their natural habitat. Each species was found to colonise the surface of soil independently of their growth behaviours on traditional two-dimensional culturemethods. Moreover, the viability of bacteria grown in transparent soil was found to be uncompromised. Therefore, transparent soil stands as a readily tailored platform for bacterial culture and is compatible with any optical microscope to study bacterial behaviours in a mimetic soil environment
Process intensification through electric field enhanced crystallisation and particle separation
Get PDFStrathclyde theses - ask staff. Thesis no. : T15813Previously held under moratorium from 18th February 2020 until 18th February 2023Crystallisation is a fundamental unit operation in purification and separation process. It has been amply adopted in the pharmaceutical and fine chemical industry as a product of high purity can be obtained in a particulate solid form. This unit operation allows for tailoring product characteristics, providing extensive benefits to industries that operate in highly regulated environments. In particular, the pharmaceutical industry must abide with a complex set of regulations and laws imposed on drug products intended to protect the health of the public. Thus, in order to gain a competitive edge, pharmaceutical industries invest significant time and resources to optimize the manufacturing process. In this context, increasing research is focused on the optimization of crystallisation. One of the major issues that may occur during the purification process is the concomitant crystallisation of other substances together with the Active Pharmaceutical Ingredient (API), which may reduce the purity of the product below the compliance level. This thesis explores the effects of an externally applied electric field on suspended crystals and on crystallisation, aiming at the understanding of the fundamental principles of the interaction of the electric field with suspensions of particles and solutions of small organic molecules such as pharmaceutical compounds. Then, electric fields can be exploited as a process intensification tool to aid purification processes. For instance, rather than alleviating a concomitant crystallisation problem, the use of an externally applied field could make use of a concomitantly crystallising system to purify and separate two crystalline products in a single process step. The manipulation of suspended particles driven by electric field requires elevated electric fields which might result in the generation of electric current into the fluid system, inducing undesired chemical reactions that may affect the final product. To minimize electric current, non-polar solvents constitute of small molecules such as dioxane can be used. In Chapter 3, the solution behaviour of a number of APIs in non-polar solvents are studied in order to define suitable systems that resist electrolysis under strong electric fields, and therefore allow for the design of crystallisation processes under such conditions. From a number of thermodynamic models, the van ’t Hoff equations presented good correlation values to the experimental data, and therefore it can be used to extrapolate and interpolate solubility data at any given temperature. In addition, solution behaviour was studied from determined activity coefficients, which showed that all the studied systems positively or negative ly deviated from ideality. The Wilson activity coefficient model was used to predict these deviations. From well-defined systems, particle and solution properties can then be related to the electric field phenomena, leading to the identification of the scientific principles behind the interactions between the particles and the electric field. Suspensions of APIs of small organic molecules in different apolar solvent were studied under the presence of non-uniform and uniform electric fields in Chapter 4, which enabled to assess the electrokinetic phenomena associated to the motion of the suspended crystalline particles. Under an electric field, suspended particles travelled to and accumulated on a particular electrode. The greater dielectric properties of the organic compounds compared to that of the apolar solvents induce s a motion of particles towards an electrode by positive dielectrophoresis. The collection of particles on an electrode of a particular charge is due to electrophoretic forces acting on the surface charge of the particle. The collection on a specific electrode was anticipated by the sign of the zeta-potential of the solid phase. Other phenomena such as electro-osmosis and electrorheological fluids were observed in the studied systems in the presence of a strong electric field. The investigation of equilibrated suspensions in an electric field raises the question of the effects of the energy input in metastable systems. Chapter 5 explores the effect of strong electric fields in crystallisation processes of solutions of a small organic compound (isonicotinamide) in a non-polar solvent (dioxane). An electric field directly applied to a supersaturated solution of isonicotinamide in 1,4-dioxane enhances the nucleation kinetics of the small organic molecules. Thus, in the presence of an electric field, nucleation occurs at higher temperatures and shorter induction times for a given solution composition compared to crystallisation processes in the absence of an externally applied field. A plausible explanation is the accumulation of monomers of the crystalline compound in the vicinity of the electrodes due to electrokinetic forces. This phenomenon could increase the local supersaturation, resulting in an effective increase of the frequency of successful attachment of building units. Then, the combined action of the electric field effects on suspended particles and crystallisation from solution can be applied to developing means to exploit the electric field phenomena for the separation of heterogeneous suspensions. In Chapter 6, a number of multicomponent mixtures have been successfully separated in-situ by the use of inhomogeneous electric fields. The separation of two crystallising compounds that collect on opposite electrodes under the electric field was achieved with high purity values by cooling crystallisation in the presence of the field. In addition, the separation of compounds that collect on counter electrodes was carried out by a two-step seeding/cooling crystallisation method driven by electric fields. Thus, the separation method can potential be implemented as a tool to aid separation and purification processes. This thesis investigated the scientific principles behind the interaction of the electric field with both particle suspensions and crystallising solutions. By relating particle and solution properties to the observed phenomena, we were able to anticipate the behaviour of particles and solutions in an electric field, and then exploit the phenomena to separate complex mixtures of two crystallising compounds in situ. The gained knowledge can be applied as an intensification tool to aid purification and separation processes.Crystallisation is a fundamental unit operation in purification and separation process. It has been amply adopted in the pharmaceutical and fine chemical industry as a product of high purity can be obtained in a particulate solid form. This unit operation allows for tailoring product characteristics, providing extensive benefits to industries that operate in highly regulated environments. In particular, the pharmaceutical industry must abide with a complex set of regulations and laws imposed on drug products intended to protect the health of the public. Thus, in order to gain a competitive edge, pharmaceutical industries invest significant time and resources to optimize the manufacturing process. In this context, increasing research is focused on the optimization of crystallisation. One of the major issues that may occur during the purification process is the concomitant crystallisation of other substances together with the Active Pharmaceutical Ingredient (API), which may reduce the purity of the product below the compliance level. This thesis explores the effects of an externally applied electric field on suspended crystals and on crystallisation, aiming at the understanding of the fundamental principles of the interaction of the electric field with suspensions of particles and solutions of small organic molecules such as pharmaceutical compounds. Then, electric fields can be exploited as a process intensification tool to aid purification processes. For instance, rather than alleviating a concomitant crystallisation problem, the use of an externally applied field could make use of a concomitantly crystallising system to purify and separate two crystalline products in a single process step. The manipulation of suspended particles driven by electric field requires elevated electric fields which might result in the generation of electric current into the fluid system, inducing undesired chemical reactions that may affect the final product. To minimize electric current, non-polar solvents constitute of small molecules such as dioxane can be used. In Chapter 3, the solution behaviour of a number of APIs in non-polar solvents are studied in order to define suitable systems that resist electrolysis under strong electric fields, and therefore allow for the design of crystallisation processes under such conditions. From a number of thermodynamic models, the van ’t Hoff equations presented good correlation values to the experimental data, and therefore it can be used to extrapolate and interpolate solubility data at any given temperature. In addition, solution behaviour was studied from determined activity coefficients, which showed that all the studied systems positively or negative ly deviated from ideality. The Wilson activity coefficient model was used to predict these deviations. From well-defined systems, particle and solution properties can then be related to the electric field phenomena, leading to the identification of the scientific principles behind the interactions between the particles and the electric field. Suspensions of APIs of small organic molecules in different apolar solvent were studied under the presence of non-uniform and uniform electric fields in Chapter 4, which enabled to assess the electrokinetic phenomena associated to the motion of the suspended crystalline particles. Under an electric field, suspended particles travelled to and accumulated on a particular electrode. The greater dielectric properties of the organic compounds compared to that of the apolar solvents induce s a motion of particles towards an electrode by positive dielectrophoresis. The collection of particles on an electrode of a particular charge is due to electrophoretic forces acting on the surface charge of the particle. The collection on a specific electrode was anticipated by the sign of the zeta-potential of the solid phase. Other phenomena such as electro-osmosis and electrorheological fluids were observed in the studied systems in the presence of a strong electric field. The investigation of equilibrated suspensions in an electric field raises the question of the effects of the energy input in metastable systems. Chapter 5 explores the effect of strong electric fields in crystallisation processes of solutions of a small organic compound (isonicotinamide) in a non-polar solvent (dioxane). An electric field directly applied to a supersaturated solution of isonicotinamide in 1,4-dioxane enhances the nucleation kinetics of the small organic molecules. Thus, in the presence of an electric field, nucleation occurs at higher temperatures and shorter induction times for a given solution composition compared to crystallisation processes in the absence of an externally applied field. A plausible explanation is the accumulation of monomers of the crystalline compound in the vicinity of the electrodes due to electrokinetic forces. This phenomenon could increase the local supersaturation, resulting in an effective increase of the frequency of successful attachment of building units. Then, the combined action of the electric field effects on suspended particles and crystallisation from solution can be applied to developing means to exploit the electric field phenomena for the separation of heterogeneous suspensions. In Chapter 6, a number of multicomponent mixtures have been successfully separated in-situ by the use of inhomogeneous electric fields. The separation of two crystallising compounds that collect on opposite electrodes under the electric field was achieved with high purity values by cooling crystallisation in the presence of the field. In addition, the separation of compounds that collect on counter electrodes was carried out by a two-step seeding/cooling crystallisation method driven by electric fields. Thus, the separation method can potential be implemented as a tool to aid separation and purification processes. This thesis investigated the scientific principles behind the interaction of the electric field with both particle suspensions and crystallising solutions. By relating particle and solution properties to the observed phenomena, we were able to anticipate the behaviour of particles and solutions in an electric field, and then exploit the phenomena to separate complex mixtures of two crystallising compounds in situ. The gained knowledge can be applied as an intensification tool to aid purification and separation processes
Acoustic-based machine learning diagnostic tool for voice disorders
Get PDFPreviously held under moratorium from 15th April 2021 until 2nd May 2023.The research presented in this thesis addresses the application of deep neural networks and digital signal processing algorithms in the pathological voice detection. In this thesis, the novel methods are presented, including deep acoustic recurrent model that combines frame-based cepstral and spectral features and Bi-directional Long short-term memory (Bi-LSTM) network, a 10-layer convolutional neural network (CNN) model with spectrogram of the speech as input, transfer learning from image recognition applications to pathological voice detection field using timefrequency representation as input, and a novel CNN model using data augmentation idea with scalogram of the speech as input. The deep acoustic recurrent model explores the relationship of frame-based cepstral features with RNN model. Two novel cepstral features based on cepstrum are proposed: Second Peak Perturbation (SPP) and standard deviation of cepstrum (CepStd). These novels cepstral features are validated to improve the classification performance on three databases. In addition, traditional acoustic analysis is compared with the proposed deep acoustic recurrent model. It is shown that framebased cepstral features shows overall better performance on deep recurrent model than traditional classifiers. A 10-layer convolutional neural network is proposed with spectrogram of the speech as input. This is the first model that applies time-frequency representation in deep learning for pathological voice detection. The experimental results have shown that it is an effective and efficient model for detecting pathological speech data. However, it shows overfitting problem to some extent. This is a commonly seen problem due to the small data size. In order to address this issue, transfer learning with state-of-the-art CNN networks from image recognition field is applied in the pathological voice detection field. The results shows that transfer learning improves the testing data accuracy. However, the overfitting problem is still severe. Finally, the concept of data augmentation is explored and a novel CNN model called the R-Net is proposed. This method uses continuous wavelet transform to obtain the scalograms of the speech onset, and data augmentation within a CNN environment. This model significantly reduces the overfitting problems, and improves the testing performance between 15% to 20% on the most challenging SVD database. It validates the efficiency of data augmentation on small-data-size problems.The research presented in this thesis addresses the application of deep neural networks and digital signal processing algorithms in the pathological voice detection. In this thesis, the novel methods are presented, including deep acoustic recurrent model that combines frame-based cepstral and spectral features and Bi-directional Long short-term memory (Bi-LSTM) network, a 10-layer convolutional neural network (CNN) model with spectrogram of the speech as input, transfer learning from image recognition applications to pathological voice detection field using timefrequency representation as input, and a novel CNN model using data augmentation idea with scalogram of the speech as input. The deep acoustic recurrent model explores the relationship of frame-based cepstral features with RNN model. Two novel cepstral features based on cepstrum are proposed: Second Peak Perturbation (SPP) and standard deviation of cepstrum (CepStd). These novels cepstral features are validated to improve the classification performance on three databases. In addition, traditional acoustic analysis is compared with the proposed deep acoustic recurrent model. It is shown that framebased cepstral features shows overall better performance on deep recurrent model than traditional classifiers. A 10-layer convolutional neural network is proposed with spectrogram of the speech as input. This is the first model that applies time-frequency representation in deep learning for pathological voice detection. The experimental results have shown that it is an effective and efficient model for detecting pathological speech data. However, it shows overfitting problem to some extent. This is a commonly seen problem due to the small data size. In order to address this issue, transfer learning with state-of-the-art CNN networks from image recognition field is applied in the pathological voice detection field. The results shows that transfer learning improves the testing data accuracy. However, the overfitting problem is still severe. Finally, the concept of data augmentation is explored and a novel CNN model called the R-Net is proposed. This method uses continuous wavelet transform to obtain the scalograms of the speech onset, and data augmentation within a CNN environment. This model significantly reduces the overfitting problems, and improves the testing performance between 15% to 20% on the most challenging SVD database. It validates the efficiency of data augmentation on small-data-size problems
Investigating design solutions for high-rise social housing in Kuala Lumpur with reference to thermal comfort and indoor air quality
No full textAn extensive programme of construction of high-rise social housing is being carried out in Kuala Lumpur which does not address in full the issues of thermal comfort and indoor air quality. This situation impacts human's health and comfort, and it becomes even more critical considering the climate change. As a hot-humid country, Malaysia experiences uniformly high temperature and humidity as well as low wind speeds. Approximately 75% of the time in the year air temperature and humidity lie outside the thermal comfort zone established by ASHRAE and CIBSE. As household incomes rise, residents resort to retro-fitting wall mounted split, air conditioning units to provide indoor comfort, a strategy that is neither cost nor carbon effective. The indoor and outdoor air quality conditions also surpass the World Health Organization (WHO) limits and there are insufficient local regulations on indoor comfort. Therefore, this research addresses the four main issues identified during the fieldwork: high temperature, high humidity, air pollution and low air movement with a proper and possible solution. Following a previous outline proposal of an 'Airhouse' Concept, several systems have been tested. The combination of 'Dynamic Hybrid Air Permeable Ceiling' (DHAPC) and 'Dynamic Hybrid Chilled Beam Ceiling' (DHCBC) could produce better indoor thermal comfort and air quality in the housing units, reducing the air temperature, humidity, airborne particle and gases as well as constantly providing an adequate airflow rate. This integrated system has been tested through physical and computer models and is based on a combination of dynamic insulation, hybrid ventilation and chilled beam techniques which reduces the ambient air temperature and humidity by up to 20%. The DHAPC alone could efficiently filter particulate matters (PM10 and PM2.5) circa 90% from the incoming air intake. If this outcome can be delivered in practice, it would represent an overall saving of circa 66% in power consumption and carbon emission for cooling purposes. The system could be incorporated in the 'Airhouse' Concept, for efficiently providing thermal comfort and healthy indoor air quality in high-rise residential buildings in Kuala Lumpur and perhaps in other tropical countries. However, this is only one of the possible systems and the research should encourage further studies.An extensive programme of construction of high-rise social housing is being carried out in Kuala Lumpur which does not address in full the issues of thermal comfort and indoor air quality. This situation impacts human's health and comfort, and it becomes even more critical considering the climate change. As a hot-humid country, Malaysia experiences uniformly high temperature and humidity as well as low wind speeds. Approximately 75% of the time in the year air temperature and humidity lie outside the thermal comfort zone established by ASHRAE and CIBSE. As household incomes rise, residents resort to retro-fitting wall mounted split, air conditioning units to provide indoor comfort, a strategy that is neither cost nor carbon effective. The indoor and outdoor air quality conditions also surpass the World Health Organization (WHO) limits and there are insufficient local regulations on indoor comfort. Therefore, this research addresses the four main issues identified during the fieldwork: high temperature, high humidity, air pollution and low air movement with a proper and possible solution. Following a previous outline proposal of an 'Airhouse' Concept, several systems have been tested. The combination of 'Dynamic Hybrid Air Permeable Ceiling' (DHAPC) and 'Dynamic Hybrid Chilled Beam Ceiling' (DHCBC) could produce better indoor thermal comfort and air quality in the housing units, reducing the air temperature, humidity, airborne particle and gases as well as constantly providing an adequate airflow rate. This integrated system has been tested through physical and computer models and is based on a combination of dynamic insulation, hybrid ventilation and chilled beam techniques which reduces the ambient air temperature and humidity by up to 20%. The DHAPC alone could efficiently filter particulate matters (PM10 and PM2.5) circa 90% from the incoming air intake. If this outcome can be delivered in practice, it would represent an overall saving of circa 66% in power consumption and carbon emission for cooling purposes. The system could be incorporated in the 'Airhouse' Concept, for efficiently providing thermal comfort and healthy indoor air quality in high-rise residential buildings in Kuala Lumpur and perhaps in other tropical countries. However, this is only one of the possible systems and the research should encourage further studies
Artificial intelligence in business analytics, capturing value with machine learning applications in financial services
Get PDFThis Ph.D. thesis explores the strength and applicability of machine learning-based classifiers within the context of business analytics for data-driven decision making. The focus is on supervised binary classification on structured datasets, which are vastly present in relational databases across all enterprises. Advanced analytics has become indispensable for today's corporate world and it is demonstrated that predictive analytics is one of the major contributors to capture business value across the financial services value chain. To test this hypothesis different models as Generalized Linear Models, Random Forest, Gradient Boosting, and Artificial Neural Networks were tested, compared, and combined to test their predictive strength and robustness in different scenarios and use cases. The results indicate the superiority of Gradient Boosting when it comes to structured datasets compared to all other classifiers. This is a major reason why the diffusion of Deep Learning within business analytics is lacking behind. Also, the ensemble learning method stacking - which uses several base learners to create a more powerful super learner - proved to be a viable tool to consistently improve upon the accuracy of even the most powerful candidate models - including Gradient Boosting. Automated Machine Learning (AutoML) was benchmarked against manually tuned models and proved to be a valuable tool to democratize predictive analytics for small to medium-sized corporations and to tackle the skill shortage for ML experts. AutoML has the potential to completely automate the predictive modeling process, but it is mainly concerned with model tuning and selection while ignoring steps at the beginning and end of the pipeline. Also, an ML pipeline setup is suggested that would - once it is automated - be able to reach human expert-level prediction accuracy for binary classification on structured datasets. All those models were tested and applied in the context of different business analytics use cases- with a focus on financial services - to solve problems in credit risk management, insurance claims prediction, and marketing and sales. All use cases demonstrate improvements in prediction accuracy and hence offer direct value gains. Throughout the thesis, there is a consideration of the advantages and constraints when it comes to the use of ML models in the industry including a translation into managerial implications. Also, general economic and business implications are discussed to understand how the field will evolve in the future.This Ph.D. thesis explores the strength and applicability of machine learning-based classifiers within the context of business analytics for data-driven decision making. The focus is on supervised binary classification on structured datasets, which are vastly present in relational databases across all enterprises. Advanced analytics has become indispensable for today's corporate world and it is demonstrated that predictive analytics is one of the major contributors to capture business value across the financial services value chain. To test this hypothesis different models as Generalized Linear Models, Random Forest, Gradient Boosting, and Artificial Neural Networks were tested, compared, and combined to test their predictive strength and robustness in different scenarios and use cases. The results indicate the superiority of Gradient Boosting when it comes to structured datasets compared to all other classifiers. This is a major reason why the diffusion of Deep Learning within business analytics is lacking behind. Also, the ensemble learning method stacking - which uses several base learners to create a more powerful super learner - proved to be a viable tool to consistently improve upon the accuracy of even the most powerful candidate models - including Gradient Boosting. Automated Machine Learning (AutoML) was benchmarked against manually tuned models and proved to be a valuable tool to democratize predictive analytics for small to medium-sized corporations and to tackle the skill shortage for ML experts. AutoML has the potential to completely automate the predictive modeling process, but it is mainly concerned with model tuning and selection while ignoring steps at the beginning and end of the pipeline. Also, an ML pipeline setup is suggested that would - once it is automated - be able to reach human expert-level prediction accuracy for binary classification on structured datasets. All those models were tested and applied in the context of different business analytics use cases- with a focus on financial services - to solve problems in credit risk management, insurance claims prediction, and marketing and sales. All use cases demonstrate improvements in prediction accuracy and hence offer direct value gains. Throughout the thesis, there is a consideration of the advantages and constraints when it comes to the use of ML models in the industry including a translation into managerial implications. Also, general economic and business implications are discussed to understand how the field will evolve in the future
Self-paced treadmill as a rehabilitation tool for recovering functional gait in people with stroke
Get PDFBackground. The underlying mechanism operating during the rehabilitation of walking after a stroke is not fully understood. Treadmill training is a rehabilitation tool used to improve the walking capacity of people affected by stroke with evidence of improvement to fitness and walking speed. These changes are not, however, translated to improved ability or participation in community walking which is an important goal of rehabilitation. When combined with an immersive virtual reality environment, self-paced treadmill (in which the belt speed automatically adapts to the user's intended speed) training can be used to simulate an overground community walking experience with the potential to train more complex walking skills such as speed adaptation, obstacle avoidance and dual tasking which are essential components of successful independent community walking. The electrical activity of muscles (electromyograms) can reveal the underlying motor control strategy employed when walking in different contexts, for example indoors or community walking. Mathematical tools such as the variance ratio and the occurrence frequency provide a means of quantifying this muscle activity including variability and phasic activity. These tools may help to understand the potential of walking simulators (self-pacing treadmills synchronised with virtual reality) as tools in the rehabilitation of community mobility following stroke.;Methods. Two studies were conducted; an initial study to explore the motor control and biomechanical differences across overground, standard (fixed pace) treadmill and self-paced treadmill walking in able-bodied adults and a second study further exploring these differences in a post stroke population and including outdoor and gradient walking. Data collection for both studies included EMG and kinematic data during treadmill (standard and self-pacing), indoor and outdoor gait of varying speeds and gradients and required the development of bespoke software and novel algorithms to identify the underlying differences in motor control and muscle activity variability in particular. Results. Using the variance ratio, self-paced (SP) treadmill walking and overground outdoor walking presented similar value during level walking for both the able-bodied and stroke participants. Variance ratio values during self-paced treadmill walking ranged from 0.36 to 0.51, and from 0.38 to 0.77 during outdoor walking. A variance ratio value close to 1 represents low repeatability of the muscle pattern. The results suggest that SP treadmill walking, which allows natural speed variability, is a closer analogue to outdoor walking, in terms of muscle activation consistency, than fixed pace treadmill walking. For the able-bodied participants, fixed pace treadmill walking and indoor presented similar low VR values (0.26 and 0.22 respectively), which indicated a highly repeatable (cycle to cycle) EMG patterns.;Discussion. It was found that it is possible to use quantitative measures of EMG variability to characterise the differences of muscle recruitment strategies between different walking situations such as treadmill walking at a fixed pace, treadmill walking in self-pace and overground walking indoors and outdoors. The self-paced treadmill walking and overground outdoor situations presented the most similarities in muscle activity variability. The number of participants, especially stroke survivors (n=2) was limited and cannot be generalised. Nonetheless, the use of self-paced treadmills coupled with an immersive virtual environment and, targeting community walking training present a promising platform for gait adaptability training. Conclusion. The use of the self-paced treadmills within an immersive virtual environment, present similarities to outdoor walking. Using these treadmills as a complementary rehabilitation tool, have potential for the training gait adaptability for community walking after a stroke.Background. The underlying mechanism operating during the rehabilitation of walking after a stroke is not fully understood. Treadmill training is a rehabilitation tool used to improve the walking capacity of people affected by stroke with evidence of improvement to fitness and walking speed. These changes are not, however, translated to improved ability or participation in community walking which is an important goal of rehabilitation. When combined with an immersive virtual reality environment, self-paced treadmill (in which the belt speed automatically adapts to the user's intended speed) training can be used to simulate an overground community walking experience with the potential to train more complex walking skills such as speed adaptation, obstacle avoidance and dual tasking which are essential components of successful independent community walking. The electrical activity of muscles (electromyograms) can reveal the underlying motor control strategy employed when walking in different contexts, for example indoors or community walking. Mathematical tools such as the variance ratio and the occurrence frequency provide a means of quantifying this muscle activity including variability and phasic activity. These tools may help to understand the potential of walking simulators (self-pacing treadmills synchronised with virtual reality) as tools in the rehabilitation of community mobility following stroke.;Methods. Two studies were conducted; an initial study to explore the motor control and biomechanical differences across overground, standard (fixed pace) treadmill and self-paced treadmill walking in able-bodied adults and a second study further exploring these differences in a post stroke population and including outdoor and gradient walking. Data collection for both studies included EMG and kinematic data during treadmill (standard and self-pacing), indoor and outdoor gait of varying speeds and gradients and required the development of bespoke software and novel algorithms to identify the underlying differences in motor control and muscle activity variability in particular. Results. Using the variance ratio, self-paced (SP) treadmill walking and overground outdoor walking presented similar value during level walking for both the able-bodied and stroke participants. Variance ratio values during self-paced treadmill walking ranged from 0.36 to 0.51, and from 0.38 to 0.77 during outdoor walking. A variance ratio value close to 1 represents low repeatability of the muscle pattern. The results suggest that SP treadmill walking, which allows natural speed variability, is a closer analogue to outdoor walking, in terms of muscle activation consistency, than fixed pace treadmill walking. For the able-bodied participants, fixed pace treadmill walking and indoor presented similar low VR values (0.26 and 0.22 respectively), which indicated a highly repeatable (cycle to cycle) EMG patterns.;Discussion. It was found that it is possible to use quantitative measures of EMG variability to characterise the differences of muscle recruitment strategies between different walking situations such as treadmill walking at a fixed pace, treadmill walking in self-pace and overground walking indoors and outdoors. The self-paced treadmill walking and overground outdoor situations presented the most similarities in muscle activity variability. The number of participants, especially stroke survivors (n=2) was limited and cannot be generalised. Nonetheless, the use of self-paced treadmills coupled with an immersive virtual environment and, targeting community walking training present a promising platform for gait adaptability training. Conclusion. The use of the self-paced treadmills within an immersive virtual environment, present similarities to outdoor walking. Using these treadmills as a complementary rehabilitation tool, have potential for the training gait adaptability for community walking after a stroke
Mechanistic characterisation of the Escherichia coli ammonium transporter AmtB
Get PDFThe exchange of ammonium across cellular membranes is a fundamental process in all domains of life. In plants, bacteria, and fungi, ammonium represents a vital nitrogen source, which they seek to scavenge from the external environment. In contrast, ammonium is a cytotoxic metabolic waste product in animal cells and must be excreted to prevent cell death. Transport of ammonium is facilitated by the ubiquitous Amt/Mep/Rh transporter superfamily. In addition to their function as transporters, Amt/Mep/Rh proteins play roles in a diverse array of biological processes. For example, Mep proteins signal the onset of pseudohyphal growth, a transition associated with virulence in pathogenic fungi. The human Rh proteins are also essential in maintaining acid-base homeostasis, and their malfunction can lead to various pathologies, including hereditary anaemias, overhydrated stomatocytosis, and early-onset depressive disorders. Despite this clear physiological importance, the mechanism of Amt/Mep/Rh proteins has remained elusive. Crystal structures of AmtB from Escherichia coli, the most intensely studied member of the family, suggest electroneutral transport, whilst functional evidence supports an electrogenic mechanism. The overall goal of this project was to combine electrophysiology, yeast functional complementation, and extended molecular dynamics simulations (MDS) to characterise the mechanism of ammonium transport in AmtB.;An in vitro assay based on Solid Supported Membrane Electrophysiology (SSME) was developed to confirm electrogenic activity in AmtB and characterise activity, selectivity, and kinetics of WT AmtB (Chapter 3). MDS revealed two ordered water chains embedded within the pore of AmtB, representing a potential polar transfer network. Subsequent SSME and in vivo yeast complementation characterisation of AmtB variants demonstrated that these wires were vital for AmtB-mediated NH4+ transport. This led to the proposal of a novel mechanism wherein NH4+ is deprotonated and H+ and NH3 are carried separately across the membrane (Chapter 4).;Disruption of the twin-His motif, a highly conserved histidine dyad within the pore of AmtB, had a significant impact on the kinetics of ammonium transport and a deleterious impact on selectivity, resulting in passage of potassium ions through AmtB. It is imperative that transporters maintain substrate selectivity, as uncontrolled entry of charged molecules can prove fatal to the cell, thus this explains the conservation of the twin-His motif within the Amt/Mep/Rh family (Chapter 5).;This work provides a novel model of AmtB-mediated electrogenic transport and offers valuable insight into the mechanism of the highly important Amt/Mep/Rh family. Given their physiological relevance, this work may form the foundation for future medical interventions and treatments.The exchange of ammonium across cellular membranes is a fundamental process in all domains of life. In plants, bacteria, and fungi, ammonium represents a vital nitrogen source, which they seek to scavenge from the external environment. In contrast, ammonium is a cytotoxic metabolic waste product in animal cells and must be excreted to prevent cell death. Transport of ammonium is facilitated by the ubiquitous Amt/Mep/Rh transporter superfamily. In addition to their function as transporters, Amt/Mep/Rh proteins play roles in a diverse array of biological processes. For example, Mep proteins signal the onset of pseudohyphal growth, a transition associated with virulence in pathogenic fungi. The human Rh proteins are also essential in maintaining acid-base homeostasis, and their malfunction can lead to various pathologies, including hereditary anaemias, overhydrated stomatocytosis, and early-onset depressive disorders. Despite this clear physiological importance, the mechanism of Amt/Mep/Rh proteins has remained elusive. Crystal structures of AmtB from Escherichia coli, the most intensely studied member of the family, suggest electroneutral transport, whilst functional evidence supports an electrogenic mechanism. The overall goal of this project was to combine electrophysiology, yeast functional complementation, and extended molecular dynamics simulations (MDS) to characterise the mechanism of ammonium transport in AmtB.;An in vitro assay based on Solid Supported Membrane Electrophysiology (SSME) was developed to confirm electrogenic activity in AmtB and characterise activity, selectivity, and kinetics of WT AmtB (Chapter 3). MDS revealed two ordered water chains embedded within the pore of AmtB, representing a potential polar transfer network. Subsequent SSME and in vivo yeast complementation characterisation of AmtB variants demonstrated that these wires were vital for AmtB-mediated NH4+ transport. This led to the proposal of a novel mechanism wherein NH4+ is deprotonated and H+ and NH3 are carried separately across the membrane (Chapter 4).;Disruption of the twin-His motif, a highly conserved histidine dyad within the pore of AmtB, had a significant impact on the kinetics of ammonium transport and a deleterious impact on selectivity, resulting in passage of potassium ions through AmtB. It is imperative that transporters maintain substrate selectivity, as uncontrolled entry of charged molecules can prove fatal to the cell, thus this explains the conservation of the twin-His motif within the Amt/Mep/Rh family (Chapter 5).;This work provides a novel model of AmtB-mediated electrogenic transport and offers valuable insight into the mechanism of the highly important Amt/Mep/Rh family. Given their physiological relevance, this work may form the foundation for future medical interventions and treatments
Lysozyme encapsulated gold nanocluster for studying protein denaturation
Get PDFProtein denaturation is a change in a protein's structure from its native folded state to a non-native misfolded state. Protein denaturation is the cause of many diseases. Current methods used for protein denaturation studies have provided useful information regarding protein structures, but have limitations, such as their inability to detect early aggregation of protein; hence, a new method for detecting early aggregation is needed. Lysozyme-encapsulated gold nanoclusters (Lyz-AuNCs) have interesting fluorescence properties that can be used in a variety of fluorescence measurement techniques and maybe a promising tool for studying protein denaturation. So far, studies of the fluorescence characteristics of Lyz-AuNCs under protein denaturation conditions, and their correlation to protein unfolding, have been limited. The goal of this research was, therefore, to discover the influence of environmental factors and protein denaturation on the fluorescence properties of Lyz-AuNCs and further explore the potential of Lyz-AuNCs to inhibit human beta-amyloid (1-40) Aβ40 aggregation.;In our study, changes in pH were observed to alter the fluorescence properties of Lyz-AuNCs. At an excitation wavelength 470 nm, the fluorescence intensity of AuNCs increased and redshifted when the pH is increased from pH 7 to 12. This increase correlated to a decrease in its fluorescence lifetime, suggesting a possible mechanism of the enhanced radiative process. Moreover, hen egg-white lysozyme (HEWL) was added to Lyz-AuNCs solutions to study the effect of HEWL concentrations on their fluorescence properties. Especially, the fluorescence lifetime was found to be sensitive to the concentration of HEWL at pH 3, possibly due to the aggregation that changed the local environment. Furthermore, unfolding of Lyz-AuNCs was induced by urea, sodium dodecyl sulphate and elevated temperature. It was found that the fluorescence intensity of Lyz-AuNCs decreased due to increased collisional quenching. Finally, the interaction between Aβ40 and Lyz-AuNCs was studied. The observed decrease in fluorescence intensity was believed to be due to static quenching. Significantly, Lyz-AuNCs was found to inhibit Aβ40 fibre formation. This result suggested Lyz-AuNCs as a promising candidate for Alzheimer's disease (AD) treatment as well as a probe to study Aβ40 accumulation in AD pathology.Protein denaturation is a change in a protein's structure from its native folded state to a non-native misfolded state. Protein denaturation is the cause of many diseases. Current methods used for protein denaturation studies have provided useful information regarding protein structures, but have limitations, such as their inability to detect early aggregation of protein; hence, a new method for detecting early aggregation is needed. Lysozyme-encapsulated gold nanoclusters (Lyz-AuNCs) have interesting fluorescence properties that can be used in a variety of fluorescence measurement techniques and maybe a promising tool for studying protein denaturation. So far, studies of the fluorescence characteristics of Lyz-AuNCs under protein denaturation conditions, and their correlation to protein unfolding, have been limited. The goal of this research was, therefore, to discover the influence of environmental factors and protein denaturation on the fluorescence properties of Lyz-AuNCs and further explore the potential of Lyz-AuNCs to inhibit human beta-amyloid (1-40) Aβ40 aggregation.;In our study, changes in pH were observed to alter the fluorescence properties of Lyz-AuNCs. At an excitation wavelength 470 nm, the fluorescence intensity of AuNCs increased and redshifted when the pH is increased from pH 7 to 12. This increase correlated to a decrease in its fluorescence lifetime, suggesting a possible mechanism of the enhanced radiative process. Moreover, hen egg-white lysozyme (HEWL) was added to Lyz-AuNCs solutions to study the effect of HEWL concentrations on their fluorescence properties. Especially, the fluorescence lifetime was found to be sensitive to the concentration of HEWL at pH 3, possibly due to the aggregation that changed the local environment. Furthermore, unfolding of Lyz-AuNCs was induced by urea, sodium dodecyl sulphate and elevated temperature. It was found that the fluorescence intensity of Lyz-AuNCs decreased due to increased collisional quenching. Finally, the interaction between Aβ40 and Lyz-AuNCs was studied. The observed decrease in fluorescence intensity was believed to be due to static quenching. Significantly, Lyz-AuNCs was found to inhibit Aβ40 fibre formation. This result suggested Lyz-AuNCs as a promising candidate for Alzheimer's disease (AD) treatment as well as a probe to study Aβ40 accumulation in AD pathology
Access Repository Dashboard
Do you manage STAX (Strathclyde Repository)? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!