This paper presents a structured end-to-end design and development methodology for a 150 kW DC–DC boost converter intended for fuel-cell-propelled aircraft. The proposed workflow progresses from requirements capture and device specification through system modelling, scaled prototyping, and software design, enabling early validation of performance and control, specific to aerospace fuel-cell systems. Detailed electromechanical design, component selection, and model-based testing are integrated with custom hardware development, software integration, and equipment-level testing to ensure robustness and scalability. The methodology culminates in construction and Technology Readiness Level (TRL) testing, demonstrating a systematic pathway for delivering high-power, aviation-grade power converters suitable for electrified propulsion architectures.

Over voltage protection devices are essential for safeguarding Solid-State Circuit Breakers (SSCBs) against transient overvoltages in applications such as cryogenically - cooled all electric aircraft (AEA) power systems. Despite their significance, their performance under cryogenic conditions remains inadequately characterized. This study experimentally evaluates the behaviour of these devices at ambient (298 K) and cryogenic (77 K) temperatures, focusing on breakdown voltage (V-BR), nonlinearity coefficient (alpha), temperature coefficient beta s), and leakage current (I-leak). Using a Keithley 4200A-SCS parameter analyser with I-V sweep configuration, four devices were tested: MOV Metal, MOV Znr, a Schottky diode, and a TVS diode. Results indicate that MOVs exhibit a negative beta, with VBR increasing at 77 K, (e.g., MOV Metal: +9.7%), and a significant rise in leakage current, attributed to carrier freezeout and defect induced tunnelling. In contrast, Schottky and TVS diodes display a positive beta, with V-BR decreasing (Schottky: -17.5%, TVS: -13.2%) and reduced leakage currents, due to enhanced carrier mobility, resulting in improved reliability and device clamping performance. Among the devices investigated, the Schottky diode exhibits the highest sensitivity to temperature variations, yet its performance improves under cryogenic conditions. These findings highlight trade-offs in cryogenic performance, MOVs are faced reliability challenges due to increased leakage, while diodes risk sudden failures due to elevated nonlinearity coefficient. Future work will involve integrating these devices into SSCBs for low-voltage cryogenic applications, ensuring robust overvoltage protection in extreme low-temperature environments.

The technology for implementing stacked bus planes in half-bridge power modules is the most prominent solution for enabling high current capability and minimizing switching noise. A significant increase in stray inductance is observed in the commutation path with the use of two parallel coplanar busbars, leading to severe voltage overshoots and ringing during turn-off switching transients. This paper presents three bus plane configurations that characterize and optimize the design of a two-layer bus plane for SiC power modules. All prototypes are constructed by stacking the drain and ground plates with a Kapton sheet placed in between to provide insulation. An energy-based method is performed to mathematically estimate the potential stray inductance of a bus plane or busbar design. A comprehensive investigation into the common direct measurement methods for low impedance characteristic and connection setups incorporating a test fixture is undertaken to identify an appropriate empirical methodology for stray inductance measurements. The measured stray inductance of the proposed design is 22.2 nH, which is improved by 33 % compared to the minimal plane design. The proposed bus plane, with its heavily overlapping design and the shortest terminals, is validated to achieve a significant decrease in stray inductance and maximize magnetic flux cancellation.

The DC circuit breaker (DCCB) is a critical component in cryogenically cooled electrified aircraft propulsion (EAP), which needs to be operated under extremely low temperature condition for the reliable fault protection. As the device selection is the key factor to design a reliable DCCB, this study assesses the behaviour of semiconductor switches as DC interrupters to ensure the reliability and performance of solid-state circuit breakers (SSCBs) in steady-state and fault conditions. To achieve this, an experimental setup and a modular SSCB was designed to evaluate and compare commercial Si, SiC, and GaN field-effect transistors (FETs) for the temperature range of 300K to 77K. The study characterized the static and dynamic on-state resistance of the power devices and analyzed SSCB behaviour during fault condition. The experimental results show that GaN FET devices performance improved to maintain the fault current interruption process and energy absorption capabilities at 77K, enhancing their potential for improving SSCB design for cryogenic applications, while Si and SiC devices show degraded performance.

Online temporal action segmentation shows a strong potential to facilitate many HRI tasks where extended human action sequences must be tracked and understood in real time. Traditional action segmentation approaches, however, operate in an offline two stage approach, relying on computationally expensive video wide features for segmentation, rendering them unsuitable for online HRI applications. In order to facilitate online action segmentation on a stream of incoming video data, we introduce two methods for improved training and inference of backbone action recognition models, allowing them to be deployed directly for online frame level classification. Firstly, we introduce surround dense sampling whilst training to facilitate training vs. inference clip matching and improve segment boundary predictions. Secondly, we introduce an Online Temporally Aware Label Cleaning (O-TALC) strategy to explicitly reduce oversegmentation during online inference. As our methods are backbone invariant, they can be deployed with computationally efficient spatio-temporal action recognition models capable of operating in real time with a small segmentation latency. We show our method outperforms similar online action segmentation work as well as matches the performance of many offline models with access to full temporal resolution when operating on challenging fine-grained datasets.

Multi-sensor technology is now commonly used in many applications to monitor a wide range of complex systems. In many cases, the outputs from the individual sensors are multiplexed through relays or electronic switching systems into the detection and signal conditioning electronics so that only one set of measurement equipment is required. These switching systems are often very expensive and limited to relatively benign laboratory type environments. Within the work presented here, the performance of commercial off-the-shelf (COTS) optical relay ICs within a digital switching system are characterized to access their suitability in a cost-effective switching system. A low-cost (sub £30) optical-relay switch system (ORSS) is developed and characterized that achieves remote switching speeds of over 2800 switches/s. Combining the ORSS with a sourcemeter results in measurement speeds of up to 66 sensors/s with less than 1% error compared to measuring the sensors individually. Finally, in demonstrating the application of the ORSS, multiple temperature sensors were used to monitor the temperature profile of a microcontroller board where a clear temperature profile for the individual components and the printed circuit board (PCB) is measured. Hence demonstrating that this ultra-fast and cost-effective ORSS can be used for a wide area of applications including large-scale device characterization, monitoring multichannel sensors for workplace safety, or data acquisition in industry and academia.

Transformers have rapidly become the dominant architecture for analyzing sequential data, utilizing their self-attention mechanism to effectively capture long-term temporal patterns, outperforming recurrent-based methods across various applications. In this paper, we explore the application of transformers to wearable sensor data, focusing on the analysis of human gait, which is often complex and sensitive. We propose two novel frameworks: Data Efficient Sensor Transformer (DesT) for centralized learning and Federated Data Efficient Sensor Transformer (FeDesT) for federated learning (FL) in edge-computing environments. Both frameworks employ knowledge distillation to improve the generalization of transformers, which can be prone to over-fitting due to the limited labeled data available in wearable sensor applications. Experimental results using human gait data collected from uneven and irregular surfaces show that DesT improves the accuracy by 14.8% when compared to existing transformers. FeDesT reduces computational demands on edge devices while outperforming traditional FL methods for transformers. This work demonstrates the potential of transformers for wearable sensor data analysis in both centralized and federated contexts, particularly where privacy and computational efficiency is paramount.

An advantage of bio-inspired robots is the versatility of their locomotion on a wide range of terrains that conventional robots are not able to traverse. The snake-like robot, which is a mechanism designed to move in the manner of a biological snake, is an example of a bio-inspired platform. This paper aims to implement and validate, using real and simulation tests on flat terrain, a snake-like endoskeleton robot that permits serpentine movement with adaptive control capabilities. The prototype mechatronic robot was comprised of eleven segments that were 3D printed sequentially. A simulation using Matlab’s Multibody Mechanics tool determines the necessary torques for each motor. In this simulation, a parameterized virtual model of the robot is created, where rectilinear gaits are programmed. In accordance with simulation and experimental results, the robot undertakes different times to reach the goal trajectory or end points depending on the frequency, angular position, and wave length (duty cycle). As a result, there is a high degree of similarity between the simulation tests and those conducted with the prototype endoskeleton. Furthermore, the robot is equipped with machine vision capabilities that allow it to detect faults within oil and gas pipes for the purpose of inspection and maintenance. Furthermore, a Model Reference Adaptive Control (MRAC) method based on Lyapunov stability analysis and MIT rules is proposed. The theoretical analysis and numerical simulation show that the designed trajectory tracking control law can make the multi-joint snake-like robot track the trajectory of the front joint when the robot encounters disturbances and stabilize the angular position and velocity of the first two head joints when disturbances occur.

It has become evident that pneumatic muscle-based robots are the most compliant of all soft and semi-soft robotic testbeds under investigation, despite the fact that the artificial muscle-based robot is the most challenging due to its hyperelastic behavior resulting from the constitutive elastic material. The purpose of this paper is to examine the development of a physical pneumatic actuation testbed and the application of control strategies for precision small scale soft robotics control. Moreover, an adaptive PI-Fuzzy like control is proposed for reference signal tracking. It is imperative to note that in this application, the controller's performance is determined by its ability to track a reference signal as well as its ability to minimize the absolute tracking error. The Fuzzy Logic Controller (FLC) controller, including the nature of the membership functions and the rule base, is described in detail. Simulations and hardware testing have revealed that the adaptive PI-Fuzzy like controller outperforms the traditional FLC controller in terms of experimental results. Additionally, the actuation speed is controlled under different loading conditions. It is considered that the proposed testbed is capable of controlling and managing multi-configurations of medium to small scale soft-based robots efficiently and effectively.

Large submarine flows of sediment (sand and mud), known as turbidity currents, transfer and bury significant amounts of organic carbon and pollutants to the deep sea via submarine canyons. They are also significant geohazards, regularly breaking networks of seabed telecommunications cables that carry > 99% of global data that underpin the internet. Despite this, key parameters (notably their sediment concentration) in these flows are yet to be directly measured in real-time due to their inherently harsh environment that is unsuitable for commercial conductivity sensors. To address this issue, a subsea datalogger (SSDL) is developed with a planar conductivity sensor head that can measure the sediment concentration within dense turbidity currents. Unlike conventional sensors, the planar design of the SSDL’s sensor prevents clogging at high sediment concentrations, allowing for continuous measurements within turbidity currents. The conductivity sensor is developed with a temperature sensor which is measured using an external 16-Bit ADC which is controlled with a SAMD21 32-Bit ARM microcontroller. The SSDL measures the temperature and the conductivity of the seawater once every 4 seconds for over a year. In an initial device test, the SSDL can record a turbidity current within the Bute Inlet, Canada. It is found that the seawater’s conductivity increases with salinity concentration and decreases with sediment concentration. The SSDL developed here can thus be used for both conventional subsea datalogging applications and high turbidity current applications.

Due to the rapid temporal and fine-grained nature of complex human assembly atomic actions, traditional action segmentation approaches requiring the spatial (and often temporal) down sampling of video frames often loose vital fine-grained spatial and temporal information required for accurate classification within the manufacturing domain. In order to fully utilise higher resolution video data (often collected within the manufacturing domain) and facilitate real time accurate action segmentation - required for human robot collaboration - we present a novel hand location guided high resolution feature enhanced model. We also propose a simple yet effective method of deploying offline trained action recognition models for real time action segmentation on temporally short fine-grained actions, through the use of surround sampling while training and temporally aware label cleaning at inference. We evaluate our model on a novel action segmentation dataset containing 24 (+background) atomic actions from video data of a real world robotics assembly production line. Showing both high resolution hand features as well as traditional frame wide features improve fine-grained atomic action classification, and that though temporally aware label clearing our model is capable of surpassing similar encoder/decoder methods, while allowing for real time classification.

This study introduces the Data Efficient Separable Transformer (DeSepTr) architecture, a novel framework for Human Activity Recognition (HAR) that utilizes a light-weight computer vision model to train a Vision Transformer (ViT) on spectrograms generated from wearable sensor data. The proposed model achieves strong results on several HAR tasks, including surface condition recognition and activity recognition. Compared to the ResNet-18 model, DeSepTr outperforms by 5.9% on out-of-distribution test data accuracy for surface condition recognition. The framework enables ViTs to learn from limited labeled training data and generalize to data from participants outside of the training cohort, potentially leading to the development of activity recognition models that are robust to the wider population. The results suggest that the DeSepTr architecture can overcome limitations related to the heterogeneity of individuals’ behavior patterns and the weak inductive bias of transformer algorithms.

In nuclear inspection environments, a tether cable is used to transfer power and data between the underwater robotic system and the surface control unit. During underwater nuclear inspection, the tether cable can become entangled and loop with the environment such as nuclear waste boxes and objects. The risk of colliding with underwater objects is increased by the presence of more inspection robots underwater, especially if they are equipped with manipulator arms. As a result of the loops and knots around the cable, the inspection process may be affected and the ROV may not be able to perform its job. The present article is an extended development of the previous Collision Avoidance Robotic Tether (CART-I) model [1]. The CART-I system consists of micro thrusters that are attached to the base unit by a tether cable. The micro thrust unit is capable of generating a small amount of thrust that can move the tether away from obstacles in the water, particularly in restricted spaces. The use of light detection technologies such as IR or LiDAR for obstacle detection is not effective underwater due to the complex motion dynamics of the tether underwater and the size of obstacles, which makes it impossible to provide definite identification of the objects within a given time period. In order to provide the surroundings of the micro thrust units with obstacle detection capability, we have developed an autonomous force soft sensor. Additionally, the soft moulded sealed encase was developed for effective force detection underwater, and was experimentally tested in a water tank to validate our proposed design. Simulation and experimental results of the sensor is provided. The overall goal of the CART-II is to provide a smart autonomous vision of obstacle avoidance using soft force sensing capabilities. This paper presents the full kinematic model and the simulation with finite element analysis of the CART-II system with the hardware and physical implementation of the soft sensor in order to enhance the performance of traditional tether systems.

In recent years generative adversarial networks (GANs) have been used to supplement datasets within the field of marine bioacoustics. This is driven by factors such as the cost to collect data, data sparsity and aid preprocessing. One notable challenge with marine bioacoustic data is the low signal-to-noise ratio (SNR) posing difficulty when applying deep learning techniques such as GANs. This work investigates the effect SNR has on the audio-based GAN performance and examines three different evaluation methodologies for GAN performance, yielding interesting results on the effects of SNR on GANs, specifically WaveGAN.

Photo-identification (photo-id) is one of the main non-invasive capture-recapture methods utilised by marine researchers for monitoring cetacean (dolphin, whale, and porpoise) populations. This method has historically been performed manually resulting in high workload and cost due to the vast number of images collected. Recently automated aids have been developed to help speed-up photo-id, although they are often disjoint in their processing and do not utilise all available identifying information. Work presented in this paper aims to create a fully automatic photo-id aid capable of providing most likely matches based on all available information without the need for data pre-processing such as cropping. This is achieved through a pipeline of computer vision models and post-processing techniques aimed at detecting cetaceans in unedited field imagery before passing them downstream for individual level catalogue matching. The system is capable of handling previously uncatalogued individuals and flagging these for investigation thanks to catalogue similarity comparison. We evaluate the system against multiple real-life photo-id catalogues, achieving mAP@IOU[0.5] = 0.91, 0.96 for the task of dorsal fin detection on catalogues from Tanzania and the UK respectively and 83.1, 97.5% top-10 accuracy for the task of individual classification on catalogues from the UK and USA.

A quick, cheap, and portable system for measuring accurately basic electrical characteristics of electronic devices is presented-using common off-the-shelf components linked through a simple breadboard. The system is controlled using a MATLAB app that communicates to an Arduino Nano that, in turn, controls two AD5761 16-bit, 20-V, digital-to-analog converters and an INA219 high-side shunt sensor for driving the voltage and measuring the current through a device under test (DUT). The raw data from the system are plotted in real time in MATLAB and are automatically saved to a.csv file based on the current date and time. For higher current devices (above 10 mA), a simple direct-wired connection between the DUT and the measurement circuit effectively eliminates the breadboard parasitic resistance. With these techniques, the system introduced here shows a negligible difference to the data measured using a Keithley 2410 sourcemeter and a B1500 semiconductor parameter analyzer, thereby presenting a cheap and reproducible alternative to high-end instruments for measuring two- and three-terminal devices.

Gait data collected using wearable sensors offers non-intrusive, affordable, real-time monitoring of human motion. Recognizing surface conditions from wearable sensor data has the potential to help systems discriminate between ‘poor quality’ walking data. This research investigates the predictive capabilities of machine learning models, trained on both centralized and decentralized datasets, at categorizing uneven and irregular surface conditions. The results showed that machine learning classification algorithms, trained with data originating from a single sensor positioned on the left-shank, were able to accurately discriminate between different types of surface conditions. We found the Support Vector Machine, when trained with the data centralized, had a test-set accuracy of 94%. Federated Learning offers a way to increase privacy and security for healthcare applications by avoiding the centralization of data. Our simulated federated Deep Neural Network converged to a test-accuracy of 85%, which was 8% less than the centralized counterpart.

In Underwater exploration, umbilical cord or tether cables are used to provide data transfer, electrical power and control commands to underwater robots and surface vessels. Umbilicals are also useful for anchoring, mooring, and for the deployment of underwater sensors. In general, tethers are naturally buoyant and are designed to work under rough sea conditions and is managed via the Tether Management Systems(TMS). In a limited-size environment such as nuclear environment inspection, multiple ROVs may be deployed and due to the density of various objects underwater, tangles and loops tend to form in low-tension zones due to residual torsion and flexure. The formation of tangles could hinder the operation of the ROV, and attenuate signal transmission in fibre-optic cables due to formed knots and kinks on the cables. Moreover, the risk of collection and entanglements is even increased if the ROV is equipped with manipulators that may tangle with the tether. In this paper, a novel robotics tether system is developed that is capable of navigating in the water to prevent loops and tangles and avoid obstacles. The developments of the system is a bio-inspired design that mimics the motion of a snake in the water. The design of the system single unit consists of a double $\mu$thrust system that is designed to be fitted on the tether. The developed system is distributed in the form of sections on the tether cable. The $\mu$thruster implement a specific pulse thrust according to the distance from an object to prevent collection and entanglement. To effectively simulate the proposed system, a tether cable is simulated for three types of cases, traditional tether, constant thrust tether and constant thrust with wave pattern. Moreover, the initial experiments in the hydrodynamics lab tank showed efficient locomotion of the tether cable away from obstacles.

In many areas of science, Arduino based data loggers have become common enabling instruments because of their low cost and ease of use. However, battery life is commonly the limiting factor - particularly in respect of writing data to embedded SD cards. In this paper, various methods by which to optimise an SD card based data logger using an Arduino UNO, Atmega328P at 5 V at 16 MHz and the Atmega328P at 3.3 V at 8 MHz is explored. With the bare Atmega328P chip in sleep mode, the lifetime of a 2400 mAH battery can theoretically exceed 10 years, although this is reduced to only 3 months following the introduction of an SD card. The exact power consumption of an Arduino/SD card during saving events is analysed for the first time and is found to take up to 200 ms with current spikes up to 80 mA for every initialisation and saving event dramatically increasing the average current consumption of fast data loggers. Through the use of a power control MOSFET with proper initialisation and timing of SD saving events, it is found that the Atmega328P can be set up to measure data once every two seconds whilst also ensuring a battery lifetime of one year. With the novel techniques presented here, a new method for maximising the lifetime of Atmega328P microcontroller circuits for environmental data logging applications has been achieved; allowing researchers to record data using a cheap and reproducible system.

We introduce the Northumberland Dolphin Dataset 2020 (NDD20), a challenging image dataset annotated for both coarse and fine-grained instance segmentation and categorisation. This dataset, the first release of the NDD, was created in response to the rapid expansion of computer vision into conservation research and the production of field-deployable systems suited to extreme environmental conditions -- an area with few open source datasets. NDD20 contains a large collection of above and below water images of two different dolphin species for traditional coarse and fine-grained segmentation. All data contained in NDD20 was obtained via manual collection in the North Sea around the Northumberland coastline, UK. We present experimentation using standard deep learning network architecture trained using NDD20 and report baselines results.