Theses - Engineering

Permanent URI for this collection

https://www.repository.cam.ac.uk/handle/1810/219485

Browse

Now showing 1 - 20 of 1154

Embargo
Unlocking SiC MOSFET Switching Performance Through Packaging and Instrumentation
Shillaber, Luke
Electrical energy conversion is vital to the distribution and use of electricity. With the increased use of renewable energy sources and the move to cleaner electrified transport methods, the demands of power electronic converters and inverters continue to increase. A major component in any power electronics converter or inverter is the transistor. Advances in wide-bandgap semiconductors such as Gallium Nitride (GaN) and Silicon Carbide (SiC) are leading to improved transistor technologies. These materials allow for smaller transistors with reduced on state resistances and input capacitances, this reduces losses and supports higher switching frequencies. In contrast, packaging and measurement technologies have not significantly evolved to support these emerging high-speed devices. Traditional wire-bonding packaging technologies that were originally developed in the 1950's are still being used today, with the majority of power modules on the market using bond-wires for transistor connections. Wire-bonding introduces significant stray inductances into the transistor connections. When combined with high-speed switching, such inductance can cause device damaging voltage overshoots, unwanted resonances and parasitic turn-on of the transistor. Due to such effects, the switching speed of the semiconductor device has to be restricted in order to ensure reliability and prevent damage. This negates many of the advantages that wide-bandgap devices can offer. Existing switching measurement technologies are also unable to support the high speed potential of wide-bandgap semiconductor devices. Currently available systems either lack bandwidth, have significant bandwidth distortion or introduce a large stray inductance into the circuit under test. Insufficient bandwidth can slew the measured rise-time of the switching edge leading to incorrect switching loss measurement and can attenuate high-frequency resonances reducing their apparent severity. A distorted bandwidth can significantly distort the measured time-domain signal, causing overshoots and tail-currents that are not present in the “true” waveform. The stray inductance of the current measurement system will introduce inductance into the power module, exacerbating the inductance problems previously mentioned. These factors result in many proposed high speed power modules omitting current measurement completely. This leaves the true high-speed switching nature of wide-bandgap semiconductors unknown. In this Thesis, a hybrid PCB-Ceramic half-bridge packaging structure is proposed, developed and electrically optimised. This novel structure combines single layer ceramic with conventional multi layer circuit boards. The structure allows for very low power loop inductances of <0.8 nH whilst integrating auxiliary circuits in order to unlock extremely high device switching speeds in excess of 50 A/nS and 70 V/nS. To characterise the proposed power module, a zero-compensated shunt resistor based current measurement system is proposed, achieving 3 dB bandwidths in excess of 1.6 GHz. Through the integration of this current measurement system within the power module, insertion inductances as low as 20 pH have been achieved. An insertable oscilloscope-probe version of this measurement system is also developed using a novel new connection method known as the multi-layer interconnect (MLI). The MLI structure retains the connection convenience offered by existing current measurement probes whilst utilising multi-layer mutual inductance cancellation to achieve insertion inductances as low as 67 pH. This is considerably lower than a traditional coaxial current shunt which typically has an inductance in the range of 2-7 nH depending on the model.
Open Access
High Efficiency Topologies for Dual Voltage Inputs
Ezra, Noam
Growing concerns about climate change have led governments to update their policies to require better efficiency in appliance power supplies. Additionally, consumer demand and economic drives in the last few decades have reduced the power supply cost and size. Another emerging trend in recent years is creating a single power supply design that can operate at the full Universal Mains range. The main reasons for this trend are logistics and travel reasons. The above reasons drive the requisite for low-cost, small, high-performance, improved efficiency and broad input range power supplies. The main limitation of the existing solutions is poor device utilization and difficulties optimizing the design in Universal Mains (90 – 265 Vac). In this thesis, a novel new solution, named the “Dual Range” (or Dual Voltage), is introduced and explored. The solution suggested is mainly attractive for high-efficiency Universal Mains power supplies by providing a smaller dynamic operating range in Universal Mains. The result is a reduction of device stress, better design optimization and increased efficiency. The Dual Range converter comprises reconfigurable primary power loops enabled by additional State switches. This combination allows the converter to run in parallel or series modes, optimizing the performance at the high line, 230 Vac, and low line, 115 Vac Mains and enhancing the performance over the 90 – 265 Vac Universal Mains. The parallel/series configuration provides device size and cost reduction, two optimized working points instead of one and better device utilization. This thesis examines the theoretical effectiveness of this configuration. Later, analysis and simulations were performed to evaluate the design of 75 W prototype Dual Range Flyback and Dual Range Forward converters switching at 100 kHz. Last, the thesis presents experimental results of efficiency, operation and limitation compared to the conventional solutions.
Open Access
Underlying Mechanisms of Halide Perovskites Properties and Metal Insulator Transitions in Vanadium Dioxide
Zhang, Xuewei
Halide perovskites, as potential substitutes for silicon, stand out as one of the most promising photovoltaic materials for next-generation solar cells, owing to their superior properties such as high light-absorption ability, high carrier mobilities, easy synthesis, thin-film fabrication, and low production costs. The efficiency of perovskite solar cells has rapidly increased, reaching 32.5% in 2023 from a modest 3.8% in 2009. Despite this rapid growth, the underlying mechanisms driving the high light-absorption and small effective masses in halide perovskites are not fully understood. Here, we show that these properties mainly arise from the multi-centred X-B-X bonding formed by the linearly aligned B-p and X-p orbitals, resulting in large matrix elements, especially in the visible spectrum. The normalized matrix elements are twice that of typical two-centre bonds, and the transition probabilities are four times larger, leading to high light absorption. Additionally, strong coupling between B-p and X-p orbitals in the X-B-X bonds leads to broad valence and conduction bands, along with the small band gap, resulting in small effective masses for both carriers. Another challenge is the long-term instability of perovskite solar cells due to the unstable cubic phase of halide perovskites and poor interfacial quality with the hole transport layer. We propose that Co₃O₄ offers greater benefits as a hole transport material with perovskites compared to NiO, due to smaller valence band offsets, higher interfacial adhesion energies, and greater formation energies of interfacial metal vacancies. Additionally, FA_0.4MA_0.6PbI₃ exhibits superiority over MAPbI₃ as a light-absorbing layer with Co₃O₄ or NiO due to higher adhesion energy and better band alignment without defects or dopant facilitation. Thus, the Co₃O₄/FA_0.4MA_0.6PbI₃ combination holds promise for efficient and stable perovskite solar cells. Vanadium dioxide (VO₂) undergoes a metal-insulator transition (MIT) at 340K from its semiconducting monoclinic phase to the metallic rutile phase. In polycrystalline VO₂, the MIT is less sharp due to grain boundaries (GBs). Despite the crucial role of GBs in the MIT, the mechanisms driving them to be semiconducting or metallic are not fully understood. Here, we demonstrate that V-V pairing, which renders monoclinic VO₂ semiconducting in bulk, also causes band gap opening at GBs. Polycrystalline monoclinic VO₂ thin films are expected to have more metallic GBs without V-V dimerization due to energetic favour, leading to decreased resistivity and thus less sharp MIT. Additionally, the energy of twin GBs is primarily determined by re-binding across the GB, rather than by surface energy. Although introducing oxygen vacancies is a potential method to reduce the transition temperature, semiconducting GBs become metallic upon induction, resulting in less sharp MIT.
Open Access
Advances in Optimisation of Model Parameters and Hyperparameters for Neural Networks
Clarke, Ross MacKinnon; Clarke, Ross [0000-0001-9884-046X]
Machine Learning has exploded in popularity in recent years, and now sees use in a huge variety of applications, from advertisement targeting and image generation to the recent proliferation of Large Language Models. At the heart of each setting is the problem of optimising model parameters to minimise some loss metric, so the chosen optimisation algorithm plays a fundamental role in the training process — both through the optimisation logic itself, and the auxiliary *hyperparameters* which configure the optimiser’s behaviour. Moreover, Machine Learning tasks often demand unique properties from optimisers, distinct from those explored in classical optimisation research. In this Thesis, we explore three novel techniques for optimising these parameters and hyperparameters, each sharing the theme of an awareness of the underlying curvature of the optimisation space. We begin with a study of Hyperparameter Optimisation. Many existing algorithms require complete training runs to evaluate each proposed hyperparameter configuration, which carry considerable computational cost. Methods based on hypergradients use only one training pass, but these either cannot be applied to arbitrary optimiser hyperparameters (such as learning rates and momenta) or suffer considerable additional training time. In an extension to these methods, we develop an approximate hypergradient-based hyperparameter optimiser, which is applicable to any continuous hyperparameter appearing in a differentiable model weight update. Our algorithm requires only one training episode (with no restarts), has a motivating argument for convergence to the true hypergradient, and scales to optimising independent learning rates for each model parameter, which we demonstrate in a wide-ranging empirical study. This contribution advances our ability to effectively configure a parameter optimiser for Neural Network training. Having developed a technique for selecting the hyperparameters controlling the behaviour of an optimisation algorithm, we turn to the model parameter optimisation task itself to explore which algorithmic properties might be useful in this application. The intractably-large Hessian matrices of many Machine Learning models make it challenging to apply the second-order quasi-Newton methods which are popular in the broader continuous optimisation literature. Attempts to address non-convexity in the problem, for instance by modifying eigenvalues as in Saddle-Free Newton methods, only exacerbate the issue of intractability. Addressing both these concerns, we propose the first (to our knowledge) efficiently-scalable optimisation algorithm to asymptotically use the exact, eigenvalue-modified inverse Hessian. Our method uses a power series to principally square-root and invert the squared Hessian, then precondition a gradient vector, all without explicitly computing or storing the Hessian. A truncation of this infinite series yields an optimisation algorithm which is competitive with common first- and second-order approaches, a claim which our experiments verify. However, while performing this work, we noted unexpected differences between the computational efficiency of first-order, gradient-based methods and the theoretical efficiency of second-order, curvature-based methods. In particular, we were intrigued by the relative strength and popularity of the Adam algorithm and the performant but fragile nature of the K-FAC algorithm. Inspired by this behaviour, we conclude this Thesis by seeking to unify the benefits of both approaches, combining the stabilising heuristics of second-order methods (such as Levenberg-Marquardt damping) with the efficient update direction selection of first-order methods. Our resulting algorithm recasts the Adam optimiser from a second-order optimisation perspective by adopting features from the K-FAC optimiser, and raises interesting questions about the dynamics of both popular methods; we explore these in an empirical validation over a range of settings. To conclude, we summarise our contributions and explore some possible future research directions they raise. We briefly consider the imminent challenges faced by the Machine Learning research community, and make a case for an ethical, human-centred emphasis on potential developments.
Embargo
PEDOT:PSS Fibre Building Blocks for Adaptive and Sustainable Bioelectronic Structures
Pan, Yifei
Bioelectronics facilitate information exchange by signal transfer and transductions between living biological systems and artificial electronic systems, as a critical infrastructure in various bio-related research and industries. Fibre is a bioelectronic device format that may embody the building block functionality, by hierarchically assembling functional fibre structures with vast design possibilities across a wide range of sizes, achieving enhanced bio-integration, functional performance, and sustainability. Modern bioelectronic fibres were reviewed by fabrication methods, lifecycle sustainability, adaptabilities, and system hierarchies; the author found gaps and avenues for further development: reversible fibre assembly for repair and upgrade; spider web-like biomimetic construction; down-scale fibre coating process for reduced wastage and improved control. They were arranged into experimental research projects based on a primary material poly(3,4-ethylenedioxythiophene) polystyrene sulfonate. The first study was based on the “orbital printing” that deploys imperceptible bioelectronic fibres on living surfaces in a substrate-free direct-write manner at sub-mN deposition force and at miniscule material and power costs, resembling spider webs. The author studied functional circuit construction and multi-modal sensing capabilities of these fibres, then developed a fibre structure framework to biomorphically augment living surfaces, which supports in-situ repair and upgrades by wipe-and-rewrite for adapting to continuous changes in the biological surfaces and/or application demands. The second study was developing a liquid drop coating technique for functionalising thermoplastic microfibre core, which achieved μL-level downscaled coating solution batch sizes and sub-mm multi-layer coating length control for improved customisability at low power. The produced fibres are individually handleable, embodying multimodal sensing and electrical stimulation functionalities tested in artificial biological tissue models and ex vivo tissues. They were assembled into a novel non-bonded and non-interlocked fibre structure that enables spatiotemporal sensing while allowing individual fibre-level repair and upgrade. These fibre technologies demonstrated a new paradigm of bio-interface electronics based on versatile assemblies of fibre elements that are sustainable and functionally adaptive, which could inspire the development of novel bioelectronics following the same building block principles with expanded ranges of functionalities and applications.
Open Access
Biomechanics of Eustachian Tube Dysfunction: From Mechanics to Assessment
Nava, Tobia; Nava, Tobia [0000-0003-0622-9298]
The Eustachian tube (ET) is a narrow canal connecting the throat with the middle ear. In the UK 0.9 % of the population suffers from obstructive ET dysfunction (ETD), a condition where the ET does not open. Recent breakthroughs in the treatment of ETD highlighted the need for more accurate assessment methods for the ET. Sonotubometry is the only method to measure the ET function under physiological conditions. However, sonotubometry research has been stagnant due to inconsistent results and a lack of understanding of the acoustical processes. In addition, the opening mechanics of the ET is also poorly understood, which limited progress in ET research. This work investigates the acoustics of sonotubometry with the aim of optimising the methodology. In addition, the mechanics of the ET is studied in both human and sheep cadaver specimens to gain a better understanding of the ET opening process. First, this research investigated the acoustics of the ET in a controlled environment using three models of the ET transmission line. Based on these results, a healthy volunteer study was designed where the impact of the sound amplitude and sound type (8.3 kHz sinus tone, white noise and sweep) was evaluated and optimised. Subsequently, a patient study was performed to assess the sensitivity and specificity, as well as the clinical usability of sonotubometry. Based on the volunteer study, the best results were achieved at high sound amplitudes (i.e., above 115 dB) using the sinus tone. Most studies in the past used a significantly lower sound level, which is expected to have caused the inconsistency in the reported results. Nevertheless, certain gaps persist in understanding the ET opening process. In a second project, the mechanics of the ET was studied using an in situ approach. Five sheep cadaver heads were mechanically loaded and imaged using CT scanning. The data was also compared to a CT scan of a human cadaver specimen. The *in situ* study allowed for the first time to gain a detailed insight into the ET mechanics. However, when comparing the human and sheep anatomy significant differences were observed, which questions the validity of the commonly used sheep model to understand the human ET. The results provided novel insight into the challenges of studying the ET. However, they also demonstrated the potential of the developed *in situ* approach, which could also be applied to a human cadaver study to gain more insight into the mechanics of ET.
Open Access
Torque Actuated Rear Steering for Urban Electric Commercial Vehicles
Isted, Carl; Isted, Carl [0000-0001-6941-1324]
This thesis describes research in rear steering, using a torque differential generated by in-wheel motors acting on steerable axles ('torque actuated steering') to enhance the performance of rigid (non-articulated) commercial vehicles. The main objectives are to: 1. Increase vehicle length while maintaining manoeuvrability at low speeds, 2. At high speed, improve roll-over stability and emergency braking and cornering performance while maintaining accurate path following. Chapter 1 reviews the current state of greenhouse gas emissions within the freight industry and methods for decarbonisation, predominantly the use of electric vehicles and higher capacity vehicles. Rear steering using in-wheel motors is presented as an attractive option to both reduce emissions by switching to an electric powertrain, and also by enabling the vehicle to be made larger without degradation to manoeuvrability. In chapter 2, a set of case studies of urban commercial vehicles are analysed to determine the potential emissions reductions that can be achieved using higher capacity electric vehicles with rear steering. Results showed that a 4.25 tonne light goods van would benefit most from the technology. In chapter 3, a low speed path-following controller using two nested PID controllers is developed for the 4.25 tonne light goods van, and axle geometry and dynamics are parameterised. In chapter 4, realistic non-linear models of the 4.25 tonne van and a 32 tonne, 4 axle refuse truck are developed using Simscape Multi-body for the purpose of controller testing. Chapter 5 investigated torque actuated rear steering at high speeds using a Linear Quadratic Regulator (LQR) to enhance roll stability and reduce off-tracking. Results showed that the rear steering could provide significant improvements to both. In chapter 6, the use of rear steering in braking scenarios is investigated by using adaptive Model Predictive Control (MPC) to manage the longitudinal wheel slips and the rear steering. The results showed that enhanced directional stability could be achieved in both a vehicle with rear steering provided by an independent actuator, or by in-wheel motors, however the use of in-wheel motors resulted in longer stopping distances. In chapter 7, the path following controller developed in chapter 3 is implemented on commercial controller hardware and validated through Hardware-in-the-loop (HIL) tests. It is shown that the use of torque actuated rear steering would allow this vehicle to carry a 54% greater payload volume and maintain comparable manoeuvrability to the original-length vehicle without rear steering.
Open Access
Investigation of Severe Abrasive Truck Tyre Wear
Liu, Chen
In the design of tyres, rolling resistance and wear resistance cannot be optimized at the same time. Tyres with high rolling resistance have good wear performance and good adhesion, whereas tyres with low resistance have poor wear performance and poor adhesion. It is therefore necessary to understand wear in order to design better tyres so as to reduce rolling resistance. Tyre wear is more serious for trucks as the tyres of multi-axle suspensions are subjected to severe abrasive wear caused by large slip angles resulted from the turning geometry. In Chapter 1, studies about tyre wear are reviewed. Most of the studies focused on car tyre wear. No systematic study of severe abrasive truck tyre wear has been found in the open literature. A model needs to be developed and validated to accurately predict severe abrasive truck tyre wear. In this project, a conventional trailer tyre, Goodyear KMax T Gen 2, and a low-rolling-resistance trailer tyre, Goodyear FuelMax T, are studied and compared. In Chapter 2, a tyre wear model is developed by combining physical models of tyre rolling contact, including the ‘brush’ model and the ‘string’ model, with a local wear law. The physical models simulate the contact conditions of tyres subjected to large slip angles. Surface tractions and sliding speeds are calculated. Combining these outputs with a local wear law - which describes the wear rate as a function of surface tractions, sliding speed, contact temperature, surface roughness, etc, - wear of a single tyre is calculated. Contact pressure distributions used in the models are measured in Chapter 3 and the necessary tyre tread-band stiffness parameters are simulated in Chapter 5. In Chapter 4, a laboratory test rig to measure the local wear law on different surfaces is designed. Rubber samples are pressed against a rough moving platen at various vertical pressures and sliding speeds. Contact temperatures are monitored using an infrared camera and controlled within a range by intermittent loading. The mass loss of each sample is measured to determine the wear rate. It is found that wear rate is proportional to vertical pressure, and increases linearly with sliding speed with a positive intercept. Wear rate is found not to be sensitive to temperature up to 80⁰C. Two microscopic models are developed to explain different aspects of the observed local wear behaviour. In Chapter 6, a test track test using a special trailer with steerable axles is presented for the verification of the tyre wear model. Two axles of the trailer are steered at the same angle but opposite directions and the trailer is towed along to form two trails of rubber wear particles. The rubber particles deposited over a fixed area are collected and weighted. The wear model simulation results are found to fit well with measurements with a scale factor accounting for the difference between the abrasion characteristics of the asphalt surface and the laboratory test rig. An empirical model is also summarized using the measurement data. In Chapter 7, an in-service vehicle test is presented. Two identical tractors and trailers are equipped with new tyres of each type and are instrumented to measure their motion, driver inputs, tyre pressures and temperatures, engine performance, and many other quantities. All journeys of the two vehicles are recorded for a period of about 18 months and regular measurements of tread depths of each tyre are carried out. A vehicle wear model is developed to simulate the tyre-related variables and tyre wear during the journeys. The simulated wear is converted into tread depths and compared with the tread depth measurements. Both the tyre wear model and the empirical law are used with the vehicle model for wear calculations, and both results fit the measured tread depths well, with scale factors accounting for differences in the abrasive properties of the road surfaces. The scale factors for the tyre wear model are closer than the empirical law, indicating the model predicts more consistent results. Finally, conclusions are drawn in Chapter 8 and recommendations are made for future work.
Open Access
Vision-Based Solar Forecasting with Deep Learning
Paletta, Quentin
Solar power is expected to play a leading role in the current electrification of our economy and its shift towards a low-carbon energy supply. This source of energy has numerous advantages including a wide availability and low costs, but also some limitations such as space and material usage. In addition, the inherent spatio-temporal variability induced by the cloud cover dynamics causes a large uncertainty in solar power supply, limiting its contribution to the energy mix. The diverse solutions developed to increase the reliability of solar energy, including improved storage systems or demand flexibility, require accurate estimations of the future solar energy yield. At a short-term scale, the main source of variability can be modelled by observing the cloud cover dynamics from sky cameras or satellites. Besides traditional physics-based algorithms, neural networks have recently shown considerable potential in tackling this computer vision task. Benefiting from a wider availability of resources and rapid progress in the field of machine learning, this interest is expected to grow together with the value of solar power forecasting and the addition of new solar capacities. In that respect, this thesis aims to evaluate, interpret and, thus, advance the applicability of machine learning to vision-based solar forecasting. Observed limitations of this data-driven approach are first quantified and illustrated. To address the difficulty of deep learning models to predict critical events on time, a novel spatio-temporal deep learning architecture named ECLIPSE is introduced and compared with existing models. Further, several scene representation and data augmentation strategies applied to cloud coverage modelling with neural networks are shown to improve predictions. Following this, a hybrid approach combining sky and satellite observations in a single machine learning framework is tested for intra-hour solar forecasting. This novel approach shows improved forecasts from a 25-min forecast horizon and realistic probabilistic predictions in various weather conditions. Finally, the possibility of combining diverse sky image datasets collected at different locations to improve training via transfer learning and dataset integration is assessed. Despite several ongoing challenges such as the generalisation of algorithms to new solar sites and the development of foundation models based on multilocation data, this thesis shows that the application of deep learning to cloud cover observations has a considerable potential to improve solar power forecasting. This, in turn, would facilitate the integration of solar energy into power systems by increasing its reliability and thus contribute to the ongoing energy transition.
Open Access
Numerical and physical modelling of piled rafts on soft clay under combined loading
Katsanevaki, Zacharoula
Piled raft foundations are typically used in cases where a shallow foundation is inadequate to sustain the loads of the superstructure either because of lack of capacity or because of excessive displacement. They consist of two components, namely the raft and the pile group, contributing to the overall response of the foundation. Once the piles are introduced under a raft, the conventional design practice requires that the pile group should be designed to be able to sustain the entire load, ignoring the contribution of the raft, with significant associated financial and environmental cost. In the last four decades, a number of researchers have investigated the behaviour of piled rafts under vertical load, with the objective of defining more rational and economic methods for their design. However, piled rafts are often used for structures that are subject to complex load, involving combinations of vertical, lateral, and moment loading, under cyclic and seismic conditions. Examples of structures commonly founded on piled rafts include, e.g., high-rise buildings, bridge piers and wind turbines on weak soils, where the possibility of high differential settlements or the lack of space make the use of a shallow foundation insufficient or impossible and the use of piles mandatory. At present, research on the behaviour of piled rafts under combined loading is still limited and knowledge on the developed soil-structure interaction mechanisms insufficient to affect practically design codes. This study examined three key aspects of the behaviour of piled rafts in soft clay soil, namely overall capacity, load sharing between the piles and the raft, and failure mechanisms. The problem was tackled by both numerical and physical modelling and different configurations of the system were examined, including the isolated components of the piled raft. An extended numerical parametric study was carried out using the general-purpose finite element code Abaqus to examine systematically the effect of different factors, such as the number of piles, the raft width, and the raft-soil contact, under various load combinations. The principal challenges of the numerical work were to develop a representative model for the reinforced concrete pile section and to model the pile-soil and raft-soil interfaces. These should allow relative displacements between the soil and the structural elements, such as sliding and gapping, ensuring that the correct overall response is predicted. To this end, contact interfaces with damage were introduced with properties varying with depth in accordance with the soil profile. Experimentally, the study examined specific piled raft configurations under different load combinations using centrifuge modelling. In this case, the main challenge was to model miniature reinforced concrete piles, with sections exhibiting realistic moment-axial load failure loci. Traditionally, reduced-scale model of piles for centrifuge testing are made with aluminium hollow sections. In this study, miniature reinforced concrete piles to use in the centrifuge tests were constructed following an original procedure. The miniature piles were initially tested in four-point bending and then inserted in the reduced-scale model foundation for centrifuge testing, successfully predicting the expected failure mechanisms. The findings of this study are organised according to the three key theme areas. As regards the overall capacity, load-displacement curves show the effect of the raft width and the number of piles on the piled raft capacity and on the level of displacement required to fully mobilise the capacity. For the different piled raft configurations, the ultimate response of the system can be described by failure envelopes in the vertical-horizontal and vertical-moment domain with important conclusions about the coupling in different directions of loading. Moreover, the share between the raft and the piles for different load combinations at working or ultimate loading conditions is expressed by establishing load sharing coefficients. The coefficients can be used to predict the overall capacity of the piled raft based on a function of the raft width, the pile spacing and the number of piles. Finally, critical points of the failure mechanism of the piled raft in soft clay under different loading types are highlighted, to provide directions for design.
Open Access
Enhancing the Resolution and Predicting the Accuracy of Point Clouds Generated by Handheld 3D Scanners in Construction
Trzeciak, Maciej Piotr; Trzeciak, Maciej [0000-0001-8188-487X]
3D scanning serves as a fundamental element in a range of applications within the architecture, engineering, and construction industry. It provides point clouds that are utilised for construction progress monitoring, scan-to-BIM workflows and construction surveys. Nevertheless, data acquisition relying on terrestrial laser scanners or photogrammetry methods is labor-intensive during both the scanning and post-processing phases. Handheld scanners theoretically present a solution to this challenge due to their potential to significantly reduce on-site scanning efforts and obviate the need for post-processing tasks. However, existing mobile mapping devices are limited in their ability to produce accurate and high-resolution point clouds as compared to terrestrial laser scanners. The primary objective of this thesis is to formulate, implement, and evaluate two methods designed to enhance the resolution and estimate the accuracy of point clouds produced by handheld 3D scanners. The first method is novel in two key ways: (1) it boosts the resolution of a series of sequential sparse lidar scans by fusing them with high-resolution colour images, and (2) it employs these higher-resolution scans to progressively reconstruct a scene. The second method is designed to make real-time predictions on point cloud accuracy, which are based on the estimations of uncertainty levels in SLAM algorithms running in handheld scanners. This method also compares the estimates to the accuracy levels established by surveying standards, displaying the results to the user through colour overlays on the progressively built point cloud, hence enabling user-friendly and real-time assurance of point cloud accuracy. To assess these methods in real-world on-site scenarios, the author of this thesis assembled a unique dataset, ConSLAM, facilitating the evaluation and comparison of SLAM algorithms used by handheld 3D scanners and autonomous robots in a construction setting. The significant contributions of this thesis are primarily threefold: (1) the proposed camera-lidar fusion method increases the point cloud density approximately sixfold and reduces noise by around 11\%. This results in the improvement in point cloud resolution, which facilitates superior recognition of building elements in point clouds; (2) the introduction of ConSLAM, the world's first dataset which enables the accuracy measurements of SLAM algorithms on construction sites. The research community can also utilise this dataset to measure how the performance of their algorithms changes along with on-site progress; (3) the point cloud accuracy estimation method demonstrates a statistically relevant correlation between accuracy estimations and the actual error in point clouds. This means that the method can flag sections of point clouds with potentially higher spatial error, thereby safeguarding users from making incorrect measurements. To the best of the author's knowledge, this is the first such method of its kind.
Open Access
Sample-Efficient Reinforcement Learning for Spoken Dialogue Systems
Wu, Yen-Chen
Conversational Artificial Intelligence (Conversational AI) platforms such as Siri or Alexa have become deeply integrated into human lives in recent years. Despite this widespread adoption, the development of a general open-domain dialogue system capable of engaging in natural conversations with humans remains a challenging task. Developing effective dialogue systems poses a challenge in the face of non-deterministic environments. Users exhibit diverse behaviors, making it difficult to provide an optimal demonstrative response based on user inputs. Additionally, achieving the desired outcome often requires multiple turns of responses, and the consequences of an action may not manifest immediately. Consequently, dialogue management, which involves determining how to respond to users, is commonly approached as a reinforcement learning (RL) problem. RL algorithms are specifically designed to adapt and learn in the presence of non-deterministic probabilities associated with different states and actions. Moreover, RL algorithms excel at handling delayed rewards, enabling agents to understand the long-term consequences of their actions. The second challenge pertains to the sample efficiency of reinforcement learning algorithms, particularly in the context of enabling online learning with human interaction. Dialogue managers are expected to achieve effective training using minimal sample data. Model-based reinforcement learning (MBRL) offers a promising approach to enhance sample efficiency by constructing an environment model capable of predicting user behaviour. However, the problem of noisy environment model poses an inevitable hurdle in MBRL. Incorrect predictions from environment model could hinder rather than aid the training process. In this thesis, we address this noisy environment model problem in three distinct stages: The first concept revolves around extracting valuable information from noisy future predictions. Building upon the classical actor-critic architecture, we introduce the Actor-Double-Critic (ADC) algorithm, which enhances the available information through the inclusion of a model-based critic. The model-based critic utilizes the noisy future predictions as input and ensembles the outputs of both critics to optimize the policy. Through experiments, we demonstrate the robustness of ADC in noisy environments where accurate modeling of user behavior is challenging. Furthermore, ADC exhibits superior sample efficiency and stability compared to its model-free baseline. This work represents one of the pioneering applications of model-based reinforcement learning in the realm of dialogue systems. Instead of learning how to select useful information like ADC, the second approach involves leveraging domain knowledge to select useful information. Specifically, actions that result in repetitive or undesirable termination of dialogues are excluded from the action space. Drawing inspiration from the concept of action masks, a manually crafted component traditionally employed in dialogue managers to eliminate unfavorable actions, we introduce the Trainable-Action-Mask (TAM) algorithm. TAM automates the construction of action masks by utilizing an environment model. The simplified environment model in TAM solely focuses on predicting repetitive patterns and unfavorable dialogue terminations. Experimental results indicate that the simplified environment model facilitates faster learning, and evaluating the accuracy of the environment model becomes a more manageable task. The final solution entails leveraging domain knowledge to optimise dialogue policy directly, eliminating the need for constructing an environment model. We present Loop-Clipping Policy Optimisation (LCPO), a methodology that directly re-estimates the advantages of taking unfavorable actions in dialogue management policy. LCPO stands out as it does not necessitate environment model training, has no additional hyper-parameters to tune, and is straightforward to implement. LCPO realises online learning in the context of the Cambridge Restaurant Booking task, achieving 80% success rate within only 260 training dialogues. This efficiency is more than eight times greater than that of its baseline model, PPO (Proximal Policy Optimization). In comparison, the state-of-the-art online learning algorithm, GP-SARSA, requires 680 dialogues to achieve a similar level of performance (Gašic ́ et al., 2011). It is important to highlight that GP-SARSA, is a complex algorithm with a time complexity of O(N³), whereas LCPO is a lightweight and versatile algorithm with a time complexity of O(N²).
Embargo
High-throughput characterisation and device integration of nanomaterials
Potocnik, Teja
This thesis introduces high-throughput methods for the characterisation and fabrication of nanomaterial devices using InAs nanowires, bilayer graphene, MoS2, and MoSe2/WSe2 lateral heterostructures. First, a novel method for nanofabrication is introduced using a lithography-optimised fiducial marker system LithoTag, which is used for the identification of individual InAs nanowires and for automatically designing >200 devices with high alignment accuracy. The presented method provides the foundation for subsequent research presented in this thesis. The focus then moves to other nanomaterial characterisation methods and investigates spectroscopic ellipsometric contrast microscopy (SECM) for mapping twist angle disorder in optically resonant twisted bilayer graphene. Ellipsometric angles were optimised to enhance the image contrast based on measured and calculated reflection coefficients of incident light, allowing the mapping of individual twisted bilayer domains to <1° accuracy. The high-throughput methodology is then applied to ALD encapsulation. Hundreds of MoS2 field-effect transistors (FETs) were encapsulated with different Al2O3 recipes to evaluate their electrical performance. MoS2 FETs encapsulated with O3 pre-treated Al2O3 exhibit the highest mobility and lowest hysteresis. It was also found that in-situ oxidising pre-treatments counteract the n-doping of conventional alumina ALD processes. The last chapter applies the results of previous chapters to MoSe2/WSe2 heterostructures. SECM was used to optimise the contrast between MoSe2 and WSe2 and used to identify >1000 heterostructure domains. This was followed by the automatic fabrication of 48 p-n junction devices aligned to the MoSe2/WSe2 interface using the LithoTag system and encapsulation with in-situ O3 pre-treated Al2O3. The results illustrate the advantages of combined high-throughput methodology for characterisation and device integration of nanomaterials.
Embargo
Aerothermal Sentencing for Manufacturing Variations on Turbine Blade Shrouds
Hulhoven, Bram
The tolerances used to sentence HPT blades need to be well-matched to the corresponding influence on aerothermal performance caused by the geometric deviation. If this condition is not fulfilled, this can lead to well-performing parts costing around $8000 being scrapped and suboptimal parts being used in service, which can trigger an early shop visit with an associated cost of up to $2M. The current tolerances used on shroud platform radial displacements, which are required to limit shroud platform steps, are at risk of being unmatched to the corresponding influence on aerothermal performance and introduce uncertainty in the sentencing. Therefore, aerothermal tolerances for shroud platform variations are developed in this research project. Steady RANS simulations are run on a three passage HPT rotor model with engine-representative shroud platform manufacturing variations applied to the middle blade passage. Engine representative shroud platform manufacturing variations, consisting of platform steps and inter-platform gap width variations, are obtained based on a statistical analysis of step heights and gap widths occurring in a sample of 100 casting scans and 26 finished part scans. Both a 4σ backward-facing step and the equivalent height forward-facing step, as perceived by the flow in the aftchord region, are studied. The gap widths is varied from the nominal value by both an increase and reduction with 1σ. This study shows that the shroud endwall flow is aligned with the wedge face until midchord and crosses the step in the aftchord region, where the flow field resembles the corresponding canonical 2D step flow. Since both the heat transfer enhancement and the step-normal component of velocity is largest in this quasi 2D (Q2D) aftchord region, a step heat transfer and loss correlation is developed using a parametric study on a Q2D model of the shroud endwall. The heat transfer correlation calculates the Nusselt number at the reattachment point as a function of the included parameters. The change in total pressure loss caused by a step is derived using a control volume analysis. Both correlations are tested on 3D platform steps. The heat transfer correlation predicts the reattachment Nusselt number within 20% for all but one test case, where the prediction error is increased to 30% due to a 3D effect not included in the model. Both the predicted loss and the value obtained from CFD are below the numerical uncertainty for platform steps on a shroud endwall. More notably, the total blade passage loss scales linearly with inter-platform gap width, which is the main shroud platform manufacturing variations altering the aerodynamic loss. When the loss correlation is tested on steps in the Harrison cascade, prediction errors below 20% are obtained.
Open Access
An Experimental Investigation into the Dimensional Quality of the Material Extrusion Additive Manufacturing Process
Golab, Mark; Golab, Mark [0000-0002-7729-1841]
Material Extrusion Additive Manufacturing (ME AM) also known as fused deposition modelling is a popular manufacturing process. It is seen as an attractive alternative to many conventional manufacturing techniques due to its cost effectiveness, ease of adoption, and the ability to produce complex geometries. However, it is commonly characterised by producing components that are of poor dimensional quality, particularly when it comes to part accuracy and geometry. Prior work has indicated that poor accuracy is primarily a function of printing parameter optimisation and flow behaviour during deposition. The majority of prior work in this domain has focused on complete artefacts. However, more recent landmark studies have demonstrated that poor dimensional accuracy occurs at a local strand level, which can significantly influence the macro dimensions and geometries of fabricated artefacts. These initial studies have presented some insights into the morphology of deposited strands, and how they are affected by printing parameter modulation. However, it is not yet fully known how sensitive the material extrusion process is, what magnitude of errors can occur at a local strand level, and how precise strand deposition is. Thus, this Thesis experimentally investigated the dimensional qualities of the material extrusion additive manufacturing process at a local strand level. The first study examined the deposition of single and multiple strands using extreme printing parameter values. The influence of extreme parameters was significant, illustrating that the process is highly sensitive to parameter changes particularly at a local level. The central portion of deposited strands exhibited far less morphological variation than the starts and ends. Deposition at a local level demonstrated how complex parameter interactions are for influencing the morphologies of single and multiple strands. The second study experimentally investigated single and multiple strand deposition using default printing parameter values. It was established that slicer cross-sectional dimensional approximations were not representative of what was actually deposited. Multiple strand deposition highlighted the misalignment between strand layers in multiple strand stacks. The third study employed a novel etching method to characterise single strand deposition precision. Stand misalignments and displacements relative to the nozzle position were significant and recurrent. High speed imaging established that lateral displacements were due to the intrinsic and unpredictable filament flow behaviour which are presently not correctable. Displacements in strands’ starts and ends raised questions about the prevalence of system hysteresis. The fourth and final study developed and employed a novel analysis technique to investigate if system hysteresis causes strands’ starts and ends displacements. Although hysteresis was detected in one of the machines, displacements were significantly affected by printing parameter values and poor adhesion. The cause of the hysteresis was not entirely certain, and it is difficult to determine if it was a mechatronic phenomenon or due to filament flow. It was concluded that material extrusion additive manufacturing accuracy and precision is a function of strand geometry, lateral and longitudinal position, which are influenced by machine design, processes parameters, and material. However, the patterns of influence are not easily predicted or consistent between the errors. As a result, optimisation is exceptionally challenging. Future work should focus on in-situ observation of flow behaviour in the nozzle.
Embargo
Inventions in climate technologies: A patent analysis
Elsen, Maximilian
Climate technologies are urgently needed to mitigate the effects of climate change and adapt to the adverse effects of global warming. Governments and organisations worldwide have pledged to becoming carbon neutral by the second half of the century, necessitating the need to monitor the development of climate technologies. Patents, in this context, offer valuable insights to research the development and diffusion of (climate) technologies. This research addresses the complexities involved in identifying and classifying climate technologies from patent data, while acknowledging the inherent limitations and specific characteristics associated with patent data. By reviewing available means for identifying climate technology patents, this research examines advantageous features of existing classification schemes and discusses present challenges. It identifies the Cooperative Patent Classification Y02 class as the most comprehensive patent classification scheme for climate technologies, owing to its cross-sectional and hierarchical structure. Based on the Y02 class, this research constructs a dataset of global high-value patented inventions in climate technologies and conducts three studies: The first study (i) investigates patent applications in climate technologies between 1980 and 2019 and critically evaluates them compared to their non-climate related counterparts. The second study (ii) examines contributions from inventors in low- and middle-income countries (LMIC) to the development of climate change adaptation technologies. The third study (iii) develops metrics to assess the contribution of organisations towards the development of climate technologies based on their patent portfolios. The first study reveals that despite the urgent need for the development and diffusion of climate technologies, we find that patented inventions in climate technologies have seen limited growth in recent years. While patent filings in climate technologies have increased particularly between 1999 and 2012, they stagnated since then and experienced a decrease relative to the general patenting trend. Moreover, the study compares patent applications in climate technologies against patenting in non-climate related technologies and evaluates climate inventions based on value metrics from the patent literature. Climate technology patents are perceived of higher technological impact than non-climate patents, a phenomenon that is particularly prevalent during the period 2005-2012. The second study examines the contribution of LMIC towards the development of climate change adaptation technologies. This is particularly relevant, considering that although most climate inventions originate in high-income countries (HIC), low- and middle-income countries are exposed to great risks from global warming. The study reveals that while the contribution of LMIC is generally low in comparison to HIC, patenting activity varies widely across LMIC. The study further highlights considerable growth in patenting activity of Chinese inventors. The third study develops metrics that quantify the contribution of organisations towards the development of climate inventions based on their patent portfolios. Companies are crucial in global innovation processes as most climate technologies are developed by corporates. The current literature lacks climate-related metrics, which are pivotal for the development and diffusion of applications and technologies for mitigation and adaptation against climate change. The study identifies metrics regarding the quantity, quality, and specialisation of climate technology patents in patent portfolios. Based on the conducted studies, this dissertation critically discusses current patent classification schemes for climate technologies and identifies avenues for the field of machine learning and natural language processing.
Embargo
Scalable and Reconfigurable Optical Switches: From Network Design to System Application
Xia, Junfei; Xia, Junfei [0000-0002-1649-838X]
The exponential growth in internet traffic causes increasing demands on network switching technologies, both in the communication networks and datacentre networks. The optical switch has been the primary research topic in recent years since it shows great potential for meeting the requirements of large scalability, flexibility, energy efficiency and high bandwidth. The current optical switch fabric using Microelectromechanical systems (MEMs) technology can achieve up to 1000 ports but is limited by the low tolerance to external disturbance and complex control panels. With the development of InP and silicon integrated platforms, mature optical components with thermo-optics, electro-optics and semiconductor optical amplifiers (SOAs) become promising candidates for the large scalable switch fabric with a more compact design. However, the SEs with binary status require balanced non-blocking switching architectures to form the planar switching fabrics or networks. Mach-Zehnder interferometers (MZIs), Micro-ring Resonators (MRRs) and SOAs can provide the on-off switching states and thus are widely applied in the switch design. However, challenges on the MZI SEs still exist to realise the switches with the high port count, low insertion loss and crosstalk ratio, whereas switches built by MRR SEs have limitations on extinction ratio, insertion loss and narrow bandwidth. SOAs give optical amplification for loss compensation; however, they also induce noise and distortion. In the cascaded switching architecture, noise and distortion accumulate and degrade the signal performances. The study focuses on the design factors that affect the characterization of MZI, MRR, and SOA SEs for the purpose of designing large port count optical switching networks. To meet the design requirements, the MZI SEs need coupling coefficients between 0.45-0.55 and an imbalanced loss of less than 2dB caused by phase arms, while MRR SEs need to meet the critical coupling condition for perfect add-drop configuration. Therefore, this thesis studies the feasibility of the 1024-port optical switching network design used in data centre networks. InP-based dilated hybrid MZI and SOA switches are applied to achieve fast switching on nanoseceond scales. Over 15dB input power dynamic range (IPDR) with a power penalty below 2dB is assessed. The switching scheme of 4096-port is for the first time proposed with the performance assessment, indicating the IPDR of 12dB with a power penalty of less than 3dB. It has been recognised that the InP-based switch has a large footprint and thus is limited for scaling up the port count further. Silicon photonics, with its low-cost, compatible and compact platform, becomes attractive for scalable switch design. However, it can only support the passive components that induce excess losses. Hybrid integration technology subsequently gives the possible solutions by providing switching functionalities and the amplification for on-chip losses simultaneously. We, therefore, propose the 8×8 gain-integrated silicon switch fabric using adiabatic couplers, in which the low-loss MZI SEs on silicon platforms provide crosstalk isolation, whereas SOAs are implemented to give the additional gain. Detailed physical simulations are executed to verify and assess the performance of the switches with the formation of 64×64 switch fabric. To further reduce the excess loss induced by adiabatic couplers between InP and silicon platforms, the coupling strategies are also studied in terms of fabrication deviation and coupling misalignment. Another alternative for building switch fabric is MRR SEs because they have more compact volumes and fewer control signals. However, it is challenging to design the MRR SEs in the O-band because of high waveguide bending loss and more compact volumes. We introduce the extra degree of freedom by applying racetrack resonators and bent waveguides. The switch-and-select architecture is used to construct the monolithic switch fabric on the chip, and the design space is explored to obtain the applicable configurations. Experiment demonstrations illustrate the designed racetrack-based MRR chip cannot provide switching functionality, whereas the bent waveguide-based MRR switch has a low loss ranging from 5-20dB and an average crosstalk ratio of below -40dB. A 3dB passband of 43.6GHz gives promising solutions for high-performance switching applications in optical networks. In order to demonstrate the system implementation, the switch-and-select MRR switch fabric is applied in the optical fronthaul network as switching application scenarios. An envisaged converged optical network with the combination of digital radio-over-fibre (DRoF) and data service is proposed, where the silicon-based switch-and-select MRR switch fabric provides multiple functionalities such as wavelength-selective switching, multicast switching and space switching. The proposed switching network consolidates fixed and wireless services between C-RAN and PON architectures, experimentally demonstrating power penalties below 1.5dB and RF dynamic ranges of over 40dB with a bandwidth of 8Gb/s. The 14dB optical link budget enables scaling up the port count to more than 100. The studies in this thesis, therefore, illustrate the great potential of optical switches for future scalable and reconfigurable switching network design and diverse optical network applications.
Open Access
A high-current and low-loss high-temperature superconducting cable for nuclear fusion applications
Hao, Luning
The primary challenge in applying high-temperature superconducting (HTS) coated conductors (CCs) to AC energy applications, including motors, power transmission, and fusion magnets, is their relatively high AC loss. In this dissertation, a novel high-current and low-loss HTS cable, Patterned superconductors for AC loss minimization and current maximization (PSALM), is proposed and investigated to tackle the AC loss problem. Compared with the existing HTS cables, PSALMs are more compact, more capable of carrying large currents with minimized losses, and easier to manufacture. A 3D finite element method (FEM) model of the PSALM based on the T-A formulation is built to optimize the structure and verify the performance. Samples of PSALMs with a single superconducting layer are manufactured, tested, modeled, and analyzed. In the experiments, the critical current degradation and the transport AC losses are measured. In the simulation, the transport AC losses are calculated and compared with the experimental data, and the current distribution is analyzed in detail. The results show that compared with normal REBCO CCs, PSALMs with optimum patterns can reduce AC transport losses by 60\% through field-canceling and can reduce magnetization losses by the factor of the filament number through patterning. This model is then used to simulate long PSALMs which could be used in practice and calculate the transport AC losses. The losses are found over 30\% lower than that of the normal HTS REBCO tapes, which verifies the loss-reduction performance of the PSALMs. To explore the underlying physics of the PSALM, we then focus on the field-canceling effect of the PSALM introduced by the specially designed patterns on the buffer layers and superconducting layers. From the results obtained by the 2D and 3D models using the T-A formulation, we find this effect can make the current distributed more uniformly and can significantly reduce AC losses. The reasons and the basic principles behind are revealed and analyzed. To sufficiently exploit it to reduce losses, the structure parameters of the PSALM are optimized and an optimum solution is obtained. Then we extend the study to the PSALM with a multiple-layer structure. Current distribution and AC losses of the PSALMs with up to 10 REBCO layers are calculated. The results show that the tapes with even numbers of layers generate lower losses than those with odd numbers, which indicates the superconducting layers should always appear in pairs. Increasing layers can increase the engineering current density of the tape, but the average loss per layer also rises. Therefore, we suggest increasing the distance between the layer pairs or using double-layer PSALM stacks instead. This work provides both experimental and simulation proof of the advantage of the PSALM as a high-current and low-loss HTS cable. It could potentially become the key to compact fusion magnets, which require high power density and suppressed losses.
Open Access
Providing Personalised Experience in Text-based Customer Service Conversations
Blümel, Jan Hendrik; Blümel, Jan Hendrik [0000-0001-9578-2495]
It is commonly acknowledged that a positive customer experience is essential to maintaining a competitive advantage. Delivering relevance to each customer at the right time through personalising interactions, information, and the customer experience is a fundamental com- ponent of an excellent customer experience. However, it is becoming more and more difficult to personalise interactions and enable a personal touch due to increased digitalisation and a decline in human contact in customer service. When companies automate their customer service through the use of technologies such as conversational artificial intelligence (AI), the lack of a human touch and conversational care often gets exacerbated. This research aims to examine how a personal touch can be facilitated in text-based com- munication by enabling interpersonal communication and addressing a customer’s affective experience. Therefore, the research draws upon the advancements of conversational AI and Natural Language Processing (NLP). First, the research provides a thorough background of the core concepts of customer experience, personalisation and conversational AI in customer service to reveal the research gap and derive the research questions. To answer the research questions, the exploratory sequential design of the mixed-methods approach is applied, utilising both qualitative and quantitative studies. In a first qualitative study, a conceptual framework is developed to better understand the relationship between personalised affective communication and the affective customer experience in digital customer service. Therefore, the theory of relational personalisation is adapted with the social information processing theory for the context of text-based commu- nication. Building on a systematic literature review, the conceptual framework delineates propositions which suggest that conversation styles such as empathy, small talk and lexical diversity need to be personalised based on psychological and individual customer knowledge such as emotions and relational history. The developed propositions are subsequently tested with two quantitative studies. The first study uses a large-scale dataset of real-life customer service interactions from social media and tests how relational conversation styles impact the affective customer experience in human-to-human interactions. The second quantitative study builds upon the findings of the previous studies and carries out an experiment with a generative AI chatbot to investigate how relational conversation styles impact the affective customer experience in AI-performed interactions. The results show that cognitive empathy considerably improves the affective customer experience in both chatbot and human interactions, across different industries. However, affective empathy has been found to have a negative effect on the affective customer experi- ence. Similarly, regardless of the interaction type (human or AI), expressed small talk and a lack of lexical diversity have a negative effect. The findings were further delineated as it was found that the effect of the expressed relational conversation styles depends on the stage of the conversation they are expressed in as well as on the initial customer emotion and relational history. The research holds significant theoretical and practical implications. By integrating social information processing theory to extend the application of relational personalisation to text-based communication the research provides a new perspective on conversational AI ap- plication design in customer service. The novel conceptual framework further shifts the focus of personalisation towards language and linguistic styles, filling a gap in existing research. The quantitative studies challenge and refine existing theories, such as the role of small talk and lexical diversity in customer service, adding new dimensions to the understanding of digital interactions. The findings offer actionable insights for practitioners to enhance digital customer service by incorporating personalisation strategies into service scripts and chatbots. It advocates for the training of human agents in recognising and adapting to customers’ emotional states and conversation contexts, promoting the use of diverse lexicons and less scripted, more empathetic responses. For AI, it highlights the benefits of using advanced Natural Language Processing and generative AI technologies to analyse customer data and autonomously adapt conversation styles, enhancing the efficiency and quality of customer service.
Open Access
The microstructure, mechanical and superconducting properties of (RE)BCO bulk single grains
Baumann, Josef; Baumann, Josef [0000-0003-3996-749X]
High-temperature superconducting RE-Ba-Cu-O bulk single grains [(RE)BCO, where RE = Y, Gd, Eu, Sm or Nd] show a wide variability in their mechanical and magnetic flux trapping properties. These variations attribute primarily to variations in the microstructure of (RE)BCO bulk materials. While the influence of secondary (RE)₂ BaCuO₅ (RE-211) particles has been investigated extensively, less is known about the effect of cracks, pores and solidified liquid phase on the mechanical and superconducting properties of (RE)BCO bulk single grains. This thesis initially investigates the correlation between porosity and the critical current density *J_c*. The porosity was determined from optical tracking microscope images, and *J_c* was calculated from the maximum measured trapped field in 22 different sized YBCO bulk superconductors. A correlation between porosity and *J_c* was established by evaluating the experimental results statistically using Spearman’s correlation coefficient. The direct correlation between porosity and *J_c* was investigated in one representative YBCO bulk material using 3D X-ray computer tomography (XCT) and a superconducting quantum interference device (SQUID). It was concluded that on the micrometre scale, *J_c* is impacted by both Y-211 particles and porosity. Subsequently, decreasing porosity can significantly enhance *J_c* and trapped field. The microstructure also significantly impacts the mechanical properties of (RE)BCO bulk materials. Mechanical failure limits the maximum magnetic field to which these single grains can be exposed at low temperature. Subsequently, their mechanical properties limit indirectly the ability of (RE)BCO bulk superconductors to trap high magnetic fields. The second part of this thesis investigates the impact of the microstructure on the mechanical and superconducting properties of 11 partially oxygenated YBCO, 11 oxygenated YBCO and 10 oxygenated YBCO(Ag) bulk single grains using Brazilian testing. The mechanical failure in YBCO bulk samples is influenced mainly by pores and pre-oxygenation cracks but not by cracks introduced due to the tetragonal to orthorhombic phase transition during the post-melt processing oxygenation process. The increased mechanical properties of YBCO(Ag) bulk materials come with a significant decrease in trapped field, originating from secondary grain growth and the accumulation of liquid phase in these materials. Finally, the mechanical and flux trapping properties of 19 EuBCO(Ag) and 20 thin-wall EuBCO(Ag) bulk materials were investigated and compared to their YBCO equivalents. On average, the trapped field and the tensile strength of the EuBCO(Ag) bulk, single grain samples were superior to YBCO bulk samples. Despite the superior properties of EuBCO(Ag) bulk materials, they do not enhance performance compared to YBCO bulk samples when exposed to high magnetic fields at low temperature.