27716 research outputs found
Sort by
Development of a Para-hydrogen Delivery System for in situ SABRE with Low-Field NMR Detection
Get PDFLow-field and benchtop Nuclear Magnetic Resonance (NMR) spectrometers offer significant advantages in terms of cost and accessibility, but their widespread application is limited by their inherently low sensitivity. Hyperpolarisation techniques like Signal Amplification By Reversible Exchange (SABRE) can overcome this limitation by boosting NMR signals by several orders of magnitude, creating new opportunities for chemical analysis and process monitoring outside of specialist laboratories. However, to realise this potential, methods for generating and measuring hyperpolarised signals must be robust and reproducible.
This work details the development of an automated para-hydrogen gas flow system to enable the probing of polarisation transfer and relaxation dynamics in SABRE. The system was developed to overcome the limitations of ex situ measurements, with applications for both in situ ultra-low-field studies and experiments on benchtop NMR spectrometers. The system provides reproducible para-hydrogen delivery with precise control over experimental parameters, enabling time-resolved studies of the hyperpolarisation process.
Application of this system to a solution of SABRE catalyst and pyridine enabled the indirect observation of para-hydrogen spin-order relaxation at the Earth's field. The decay of the hyperpolarised signal followed a biexponential model, dominated by a fast relaxation process t≌1-4 s. This relaxation rate appears to be governed by the competitive exchange of ligands at the iridium catalyst, providing insight into the solution-state dynamics of para-hydrogen.
The utility and modularity of the polariser unit were further established by its integration with a 1.4 T benchtop spectrometer with an integrated motor-driven sample shuttle and capillary bubbling system. Software was developed for the motor system, spectrometer and user interface to make all components compatible for synchronous motor-driven polarisation transfer, bubbling and detection. This integrated hardware implementation demonstrates a viable pathway for performing advanced, automated SABRE experiments on accessible low-field platforms, moving beyond the limitations of manual sample handling
The role of plant-parasitic nematodes in modulating plant symbioses with arbuscular mycorrhizal fungal networks
Get PDFArbuscular mycorrhizal (AM) fungi form widespread symbioses with plants, exchanging soil nutrients for carbon (C) fixed through photosynthesis. While this bidirectional exchange is thought to be regulated by both partners, natural interactions involve multiple co-occurring organisms and complex mycorrhizal networks (MNs) linking neighbouring plants, complicating resource distribution.
This thesis investigated the relationship between AM networks, potato plants, and potato cyst nematodes (PCN). First, using multi-plant systems likely connected to the same MN and isotope tracing, I quantified the movement of fungal-acquired phosphorus (P) and plant-derived C under herbivory by PCN. Fungal-acquired P was preferentially allocated to uninfected plants relative to their infected neighbours, while plant-fixed C was consistently redistributed within the MNs away from infected hosts.
To assess the influence of the wider microbial environment, I characterised root and soil communities using metabarcoding. PCN infection reduced fungal diversity, including AM fungal richness, and altered community composition, whereas bacterial communities remained largely unchanged. This suggests fungal communities are sensitive to PCN infection, with associated shifts potentially influencing C-for-P exchange. However, using simplified microcosms stripped of microbial complexity, I found that AM networks redirected C towards non-infected plants, confirming the MNs capacity to regulate allocation independent of other microbes.
Finally, I explored plant metabolic responses. Non-volatile metabolites in leaves were overall unaffected by PCN infection, with only infected plants showing elevated defence-related compounds relative to their uninfected neighbours, indicating limited below-ground signalling. In contrast, volatile metabolites emitted from PCN-free plants were subtly altered by neighbouring PCN infection, suggesting MN-mediated below-ground signalling can influence some above-ground plant responses.
Collectively, these results show that AM networks can, at least partly, regulate both resource exchange and plant volatile metabolite responses, and that these dynamics are modified by herbivory. This highlights the ecological importance of AM networks and their role in shaping plant–microbe–herbivore interactions
Active Form Error Control during Robotic Assisted Milling
Get PDFRobotic milling is becoming increasingly popular, as an alternative to the use of conventional CNC (Computer Numerical Control) machines, due to the added dexterity, expansive working envelope and multi-station capability of robotic arms. However, one of the main issues that arises is the positional error that comes with using them and their low stiffness, compared to conventional industrial machines. There is therefore a great need to compensate these errors. Various methods have been investigated in the past in order to improve robotic milling errors, among which: robot command modification, manipulator model modification, optimisation of the existing robotic machining cell and the augmentation of the robotic machining cell. The concept of robotic assisted machining which was first proposed by Ozturk et al in conventional CNC milling is now gaining popularity as a viable solution to reduce form errors in robotic milling.
This study focusses on the use of a collaborative robot to mitigate form errors in both conventional CNC and robotic milling. The first setup is similar to the one proposed by Ozturk et al in peripheral milling, and the second setup is comprised of a milling robot and a colinear collaborative robot supporting the backface of the workpiece while the milling robot performs face milling.
The study begins with an in-depth analysis of cutting conditions, workpiece materials, and the various factors contributing to form errors in robotic milling. Simulations were conducted to model the performance of the proposed control methods under varying conditions, including robotic support forces, static stiffness, and position errors. Results from the simulations show that both force minimisation and thickness control significantly reduce form errors compared to traditional robotic milling approaches, with thickness control being particularly effective in mitigating errors across a range of scenarios. Under load ratings from the robot’s ball caster, the thickness control method achieved a 62% reduction in form error across rectangular paths when compared to unsupported milling trials, while the force minimisation method achieved a 9% decrease.
Experimental validation was conducted using a collaborative robot system equipped with a force sensor to measure form errors during milling trials. The experimental setup was carefully designed to benchmark the force control method against conventional robotic milling without error compensation. There was a decrease of 69% and 50% in form error in peripheral and pocket milling operations. The findings from both the simulation and experimental work demonstrate that integrating active form error control enhances machining precision, especially in challenging robotic milling tasks involving complex geometries and varying material properties
The Satanic Cult Conspiracy: How online conspiracy theory discourses construct moral panic
Get PDFThe Satanic cult conspiracy theory alleges the existence of evil, secret, Satan-worshipping cults that seek to morally subvert society. From the Middle Ages to the late 20th century, its accusations have ebbed and flowed – peaking in the form of periodic ‘moral panics’ whereby Satanism becomes depicted as an urgent moral threat to society. These panics have consistently led to the identification and persecutions, including murders, of innocent individuals accused of Satanic cult activity. The last decade has seen a concerning resurgence of Satanic cult conspiracy theories online, however currently there is no research that analyses the overall breadth of themes found within this discourse today. This thesis evidences and presents a detailed and comprehensive analysis of the content of contemporary Satanic cult conspiracy theory discourse across Twitter/X, Instagram, and TikTok, with the aim of determining whether it indicates a new wave of Satanic moral panic. Highlighting the differences between interest-group and grassroots moral panics, it also pays attention to exploring how this notion of a ‘Satanic moral panic’ can even be identified, and why accurately identifying it matters in the first place. Research is currently limited in its understanding of the exact relationship between conspiracy theories and moral panics. To address this, I then also develop and present in this thesis a new research framework for identifying when conspiracy theory discourses are indicative of moral panics, and when they are not
Sensorless control of single and dual three phase interior permanent magnet synchronous machines with inductance asymmetries
No full textThis thesis presents a comprehensive study on the effect of inductance asymmetries, including self-inductance asymmetries and single open phase faults, and their compensation methods, in high-frequency (HF) signal injection (HFSI) sensorless control of single three-phase (STP) and dual three-phase (DTP) interior permanent magnet synchronous machines (IPMSMs).
Inductance asymmetries, caused by manufacturing tolerances and winding faults etc., can introduce harmonic distortions in the estimated rotor position, deteriorating the performance of HFSI sensorless control systems. It is found in this thesis that the self-inductance asymmetries in multiple phases can cause the 2nd- and 4th-order estimated position errors in HFSI sensorless controlled STP- and DTP-IPMSMs. When self-inductance asymmetries in multiple phases exist in both sets of three-phase windings of DTP-IPMSMs, apart from the 2nd- and 4th-order estimated position errors, there will also be a DC offset estimated position error. Besides, the single open phase fault can lead to a 2nd-order estimated position error in the HFSI sensorless controlled DTP-IPMSM system.
Compensation methods are proposed in this thesis to suppress the estimated position error caused by the inductance asymmetries. For the STP-IPMSM with self-inductance asymmetries in multiple phases, this thesis proposes a synchronous reference frame filter-based compensation method and an anti-rotating d-axis signal injection-based method to mitigate the 2nd-order harmonic estimated position errors. Then, for the DTP-IPMSM with one-set self-inductance asymmetries in multiple phases, this thesis proposes a current summation with a ratio-based method to compensate the 2nd-order harmonic estimated position error and an adaptive notch filter (ANF) -based method to compensate both the 2nd- and 4th-order harmonic estimated position errors. For the DTP-IPMSM with two-set self-inductance asymmetries in multiple phases, an Adaline filter-based compensation method is introduced to suppress both the 2nd- and 4th-order harmonic estimated position errors, and a current direct summation-based method is proposed to cancel the DC offset estimated position error. Finally, this thesis explores the effect of the single open phase fault on HFSI sensorless controlled DTP-IPMSMs, where the 2nd-order harmonic estimated position error can destabilise the HFSI sensorless control system. To address this, an ANF-based compensation method and a current summation with a ratio-based compensation method are proposed, offering a fault-tolerant HFSI sensorless control
Uncertainty-aware Deep Learning Methods for Image Classification, Object Detection and Segmentation
No full textDeep learning has garnered significant attention over the last decade, emerging as a transformative force in science and technology. As the foundation of modern artificial intelligence (AI), deep learning models have revolutionized computer vision, impacting numerous areas such as facial recognition, autonomous driving, medical image segmentation, and generative AI systems. These advancements have profoundly changed daily life, demonstrating the immense potential and versatility of AI technologies.
Despite the rapid progress and widespread adoption of deep learning technologies, critical challenges persist. Modern deep learning models often suffer from overconfidence, susceptibility to distributional shifts, and vulnerabilities to adversarial attacks. These limitations raise serious concerns about the reliability and trustworthiness of AI systems, particularly in safety-critical applications where erroneous predictions can have life-threatening consequences. As a result, building reliable, explainable, and uncertainty-aware AI systems has become a focal point of research.
This thesis addresses pressing challenges in large-scale computer vision by focusing on both Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). While these models have achieved significant advances in performance, accuracy, and scalability, ensuring their reliability under distribution shifts remains a fundamental barrier for deployment in real-world and safety-critical scenarios. To address this, the thesis emphasizes the quantification and management of uncertainty as a means to achieve robustness, safety, and trustworthiness in deep learning models. Specifically, this work develops practical and scalable uncertainty quantification methods, including a novel Bayesian object detection framework that leverages Gaussian weight sampling from pre-trained networks for effective out-of-distribution (OOD) detection without incurring high computational costs or requiring synthetic data. Experimental results demonstrate substantial improvements, including up to an 8.19\% reduction in FPR95 and a 13.94\% increase in AUROC, thus enhancing the reliability of object detection in open-world settings. For Vision Transformers, the Prior-augmented Vision Transformer (PViT) is introduced, which utilizes prior knowledge from pre-trained models to robustly separate in- and out-of-distribution samples by quantifying the divergence in predicted class logits, outperforming state-of-the-art OOD methods across the ImageNet benchmark and seven additional OOD datasets. Furthermore, this thesis contributes a new benchmark dataset for complex cell morphology segmentation and proposes an uncertainty-aware framework that incorporates virtual outlier sampling, leading to up to a 7.74\% increase in Dice Similarity Coefficient and significant reductions in boundary errors. Collectively, these contributions advance the interpretability, reliability, and deployment-readiness of computer vision models in complex, real-world scenarios. The public release of codebases and datasets further amplifies the scientific and societal impact, paving the way toward safer and more dependable AI systems
Design of Next-Generation Degradable Block Copolymer Nanoparticles by Polymerization-Induced Self-Assembly
No full textDevelopment of Deep-learning-based Technologies for Optical Coherence Tomography (OCT) and OCT-Angiography in Clinical Applications
Get PDFSkin and oral cancers demand early detection to improve outcomes, but conventional biopsy is invasive, while non-invasive methods like dermoscopy struggle with resolution-depth trade-offs. Optical coherence tomography (OCT) and angiography (OCTA) offer non-invasive, high-resolution imaging (1-3 mm depth) for visualizing tissue microstructure and microvasculature without contrast agents, showing promise in dermatology and oral medicine. This thesis advances software and algorithms to optimize a lab-built swept-source OCT system, enhancing diagnostic accuracy and improving imaging speed through deep learning and image processing.
Key contributions address five areas: OCT-based segmentation and denoising, and OCTA-based reconstruction, super-resolution, and vasculature extraction. For OCT, lightweight networks improve skin tissue boundary segmentation, while a Swin transformer-based pipeline reduces speckle noise, enabling faster scans without quality loss. The Efficient Segmentation-Denoising Model concurrently denoises and segments oral tissues. For OCTA, deep-learning-based protocols reduce scan repetitions, shortening imaging time by >80%: the Image Reconstruction U-Net and U-shaped Fusion Convolutional Transformer generate high-quality angiograms from minimal two-repetition scan. Super-resolution transformers (Intraoral Micro-Angiography Super-Resolution Transformer and Angiography Reconstruction Transformer) enhance spatial resolution for sub-second OCTA imaging. Besides, the Vasculature Extraction Transformer directly extracts vascular signals from a single OCT scan, cutting acquisition time by 75% while preserving diagnostic quality.
These innovations address critical limitations in OCT/OCTA imaging, such as slow acquisition, motion artifacts, noise, and time-saving by automatic segmentation. By integrating deep learning, this study improves image quality, accelerates workflows, and enables precise, non-invasive assessment of lesion depth and vascular morphology. The advancements improve diagnostic workflow in skin cancer and oral pathologies, facilitating early detection and reducing reliance on invasive biopsies. Collectively, this research strengthens the clinical translation of OCT/OCTA, offering a patient-friendly paradigm for improving outcomes in dermatology and dentistry
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇