Research poster competition

Design of a hybrid environmental sensor for water quality monitoring

Water quality monitoring is a critical aspect of environmental sustainability and public health. Ensuring water quality and safety requires the effective detection of emerging contaminants, which present significant risks to both human health and the environment. The hybrid environmental sensor integrates multiple sensor technologies to monitor different aspects of water quality.

Firstly, optical sensors are utilized to measure parameters such as turbidity and organic matter content, providing insights into water clarity and contamination levels. Additionally, electrochemical sensors are employed to detect the presence of specific pollutants such as heavy metals and dissolved oxygen levels, crucial for assessing water quality and ecosystem health. However, this research work will establish a knowledgeable technique to the novel applications of machine learning (ML – Python) algorithm in water environment. One of the key features of the hybrid sensor is its ability to operate autonomously and transmit real-time data wirelessly to a central monitoring system.

This capability enables timely detection of water quality issues and facilitates prompt interventions to mitigate potential risks. Moreover, the sensor is designed to be robust and weather-resistant, making it suitable for deployment in various environmental conditions.

Affordable and portable IoT-based smart weather observation device for urban microclimate and thermal comfort studies

According to the United Nations, the world’s urban population will increase by 68% by 2050. The smart city is seen as a concept of addressing urbanisation challenges, such as air pollution, water quality, energy efficiency, water scarcity etc., including the Urban Heat Island (UHI) phenomenon, where cities are warmer than the surrounding rural area, that worsens the thermal discomfort in cities. UHI is due to the replacement of the natural surface with infrastructures for the growing population needs.

In any thermal comfort studies, Air Temperature (AT), Relative Humidity (RH), Wind Speed (WS) and Mean Radiant Temperature (MRT) are the four variables that need to be measured. Among them, MRT needs to be calculated using the Globe Temperature (GT) values. However, none of the existing sensor devices were found to have a standardisation for measuring GT for outdoor thermal comfort studies. On the other hand, such sensor devices are expensive.

To overcome this limitation, we developed an affordable IoT-based smart sensor device capable of measuring AT, RH, WS, and GT. It is portable and equipped with GPS, and enables real-time monitoring in any location. This cost-effective solution supports urban microclimate studies helping cities mitigate UHI and improve thermal comfort.

An autonomous distributed timing signal in-space as alternative to GNSS time synchronisation

Accurate synchronised time is an important requirement for many ground- and space-based services. Currently, the primary source of this time data comes from Global Navigation Satellite Systems (GNSS). However, the past years have shown the vulnerability of GNSS’ centralised system to not only space weather events, but also attacks, spoofing, jamming, and DoS (Denial of Service), creating a single point of failure.

Thus, as the number of satellites launched keeps on increasing, eyes are now turning to Internet of Things (IoT) technologies and their associated research into the field of distributed, energy-efficient, and robust time synchronisation. My research has focused on the development of a framework, AUDITS (AUtonomous DIstributed Timing-signal in Space), meeting these requirements.

Inspired by the opinion-dynamics of social systems, AUDITS presents a dynamic protocol enabling efficient time synchronisation to be reached in satellite systems without the need for an external terrestrial reference source. This approach, which is transferrable to IOT systems and other types of dynamic distributed networks, also allows satellite systems to cope with unknown number of disruptors without knowledge of the network topology.

RFA-PVTInvoNet: enhancing image quality for building datasets

High-resolution building datasets derived from remote sensing imagery are essential for urban planning and emergency response. However, such imagery is not always readily accessible, necessitating advancements in spatial resolution enhancement, which is known as Super-Resolution (SR).

In SR methods, Transformer-based networks have emerged as state-of-the-art approaches. However, limitations in feature representation and self-attention create significant opportunities for improving performance and reducing computational costs. To address the issue, this research introduces an innovative method called Residual Feature Aggregation with Pyramid Vision Transformer and Involution Network (RFA-PVTInvoNet).

This approach can enhance performance while mitigating high computational costs in super-resolution tasks. With extensive comparative experiments against established methods, the proposed RFA-PVTInvoNet demonstrates superior performance across three building datasets, including the Wuhan University (WHU) building dataset, the Massachusetts (MAS) building dataset and the Waterloo (WAT) building dataset, compared to established methods. The most significant improvement is observed on the WAT dataset, where it surpasses MSCA-RFANet with PSNR increasing from 33.10 dB to 33.66 dB, SSIM from 0.85 to 0.87, and testing speed from 3.52 to 3.66 FPS. These results confirm RFA-PVTInvoNet’s efficiency in super-resolution tasks, especially for building datasets.

Real-time monitoring of childhood asthma using microphone acoustic sensors

Children with asthma often have difficulty accurately communicating signs of physical discomfort to their caregivers. The aim of this project is to use microphone sensors to monitor children and capture real-time signals of their heart and lung sounds in order to determine whether an asthma attack is occurring. The development of the device seeks to allow children to wear it daily without impacting their everyday lives. The use of a silicone Ecoflex layer as a medium between the microphone and the body enhances the wearing experience while reducing the transmission loss of heart and lung sounds.

Decentralised autonomous community in space

The number of spacecraft in low-Earth orbit, along with the data and services they provide, is growing rapidly and will continue to expand over the coming decades. However, space operations and governance remain largely unchanged, limiting the potential of space-based services and posing risks to the sustainability of the space environment.

As these spacecraft become increasingly interconnected, they will evolve into an Internet of Things (IoT) network, creating a vast, distributed system of sensors in orbit. My research envisions transforming this network into a Decentralised Autonomous Community in Space, governed by its members via a distributed ledger, independent of centralised government or corporate control. Decisions on spacecraft separation and orbital manoeuvres would be made autonomously, enhancing safety and sustainability in space. This would be achieved through a system of coordinated movements, machine-to-machine tasking, and smart contracts that minimise collision risks and reduce the creation of space debris.

My current focus is assessing different distributed ledger technologies to determine their suitability for the unique challenges of the space environment. It involves evaluating various characteristics, such as consensus mechanisms and architecture, to identify how these technologies can be optimised for space and secure, decentralised control and communication between spacecraft.

Implementation of 5G NR PHY Layer Algorithms on RFSoC

5G New Radio (NR) aims to provide a technological solution to the rising demand for faster data rates and lower latency in mobile communications through its improved features and flexibility compared to previous generations, however, this comes at the cost of increased complexity. The research involves the design and implementation of a hardware compatible 5G NR uplink transmitter model focusing on the Physical Uplink Shared Channel (PUSCH), using Simulink with the HDL Coder Support Package to create a hardware design targeted at an AMD Zynq UltraScale+ Radio Frequency System-on-Chip (RFSoC) device. The design contains the main processing stages of the PUSCH such as scrambling, symbol modulation, layer mapping, codebook precoding, and resource block mapping with OFDM, and makes use of the ZCU216 development board’s multiple DACs and ADCs for MIMO transmission. The design is compared at various key points against a software reference developed in MATLAB using the 5G Toolbox to confirm the expected operation and ensure standard compliancy. The hardware resource utilisation and timing are essential evaluation metrics to ensure the design can produce data at the required rate for 5G NR PUSCH

A novel deep learning method for estimating cortical thickness trajectories in Alzheimer’s patients and healthy population 

Alzheimer’s disease (AD) is a neurodegenerative disease that presents critical challenges in diagnosis and treatment. Emerging research indicates that AD-related cortical changes, such as cortical thickness (CTh), can appear up to a decade before cognitive symptoms. Accurately measuring cortical thickness can therefore offer a significant avenue for early AD diagnosis and monitoring of clinical progression. Automatic techniques, such as FreeSurfer and CAT12 Toolbox, offer out-of-the-box cortical thickness estimates, but with an excessively long computational time (up to 10 hours per volume), systematic differences between approaches (Seiger et al., 2018) and significant errors when applied to clinical data (Ozzoude et al., 2020). We propose DeepThickness; the first Deep Learning-based approach for estimating cortical thickness from structural MRI in just a few seconds. The method relies on recent advances in deep learning, leveraging two artificial neural networks and a graph neural network to generate topologically correct white matter and pial surface meshes with cortical thickness estimates as an overlay. We report promising preliminary findings, highlighting our method’s similarity to FreeSurfer in mesh generation and cortical thickness estimations. Leveraging comprehensive clinical datasets, we also showcase our method’s use for mapping cortical thickness and clinical trajectories over time for healthy, MCI and AD populations

Aeroelastic Validated Lightweight Blade Design for FOWT

This study integrates our previous proposed fluid structure interaction (FSI) analysis framework with advanced Non Sorting Genetic Algorithm II (NSGA-II) optimization strategy to find possible solutions for a blade design with lighter weight and higher strength, which performs better on the floating offshore wind turbines (FOWTs) in aeroelastic terms. In this work, we demonstrated the effective optimization and aeroelastic validation and verification (V&V) workflow for the NREL 5 MW reference FOWT blade under a platform surge load condition. To accelerate the searching during optimization process, a Radial Basis Function (RBF) neural network surrogate model is trained to approximate the complicated blade finite element (FE) model, with mathematical relation established between the design variables of stiffness and thickness of the blade shear web structures and the objectives of blade mass and max. Von Mises stress. With this approach, the efficiency of the optimization is significantly increased, also, comprehensive blade aeroelastic inspections can be achieved using our previously proposed FSI framework.

A novel foundation model for estimating brain MRI health

We present a novel foundation model for brain health assessment using MRI, representing a significant advance in the integration of neuroscience with deep learning. The developed model quantifies the brain age gap—a comparison of chronological age to predicted brain age—serving as a biomarker of healthy aging. To tackle the scarcity of clinical data, we utilize an extensive synthetic MRI dataset consisting of 100,000 volumes created by generative AI. Our model has been further refined using real-world data from approximately 56,000 MRI volumes from the UK Biobank, enhancing its versatility and application range.

The fine-tuned model excels in several key areas: it can estimate overall brain health, track patient trajectories over time, and correlate changes in brain anatomy with cognitive developments. We demonstrate state-of-the-art accuracy with a mean absolute error of 3.2 years in brain age prediction on the synthetic dataset. Preliminary findings highlight significant disparities in brain age gaps among individuals with neurological conditions such as dementia, head injuries, and Parkinson’s disease compared to healthy controls.

Our foundation model is the first to employ synthetic neuroimaging data for brain health assessment, promising a new direction in personalized medicine.

Digital twin model for IoT based smart cities

The robust and adaptive modeling and monitoring of Internet of Things (IoT) networks with Digital Twins (DTs) provide qualified application services for smart cities. However, the increase in the number of smart services and their diversity pose extra challenges for integrating DTs into IoT networks. If these problems are not handled, they cause crucial Quality of Service (QoS) degradations for smart cities. In addition, by considering the constraints of IoT devices in terms of computational capability and energy resources, it gets harder to work on a unified framework to realize and manage all IoT services.

Considering these modelling and communication challenges, my research focuses on creating a DT model of an IoT-based smart city network by optimizing both DT and IoT network performances in terms of modeling accuracy, communication interoperability and qualified DT services regarding timeliness.

Eddy Current Inspection of uncured Carbon Fibre Reinforced Polymer Composites with radii geometries

Carbon Fibre Reinforced Polymer (CFRP) composites play a crucial role in sectors like aerospace, automotive, and energy. Eddy-current sensors, which operate without contact, have shown promise in detecting defects and damage in these materials. However, due to the complex electromagnetic interactions within their multi-layered, anisotropic structure, accurately detecting and characterising defects, such as in and out of plane waviness, fibre breakage, missing and misalignment of fibres remains a challenge. This study seeks to develop finite element models using COMSOL and leveraging detailed experimental characterizations for detecting and characterising defects in uncured CFRP with radii geometries. The research will also facilitate the extraction of essential material and structural properties, optimize sensor designs, and create a comprehensive database of material properties to support the training of machine-learning inversion algorithms.

Hardware-software co-design of FPGA-based neural network accelerators for edge inference

Developing a smart venturi wet-gas meters to foster efficient gas production and digital transformation

Wet gas metering is essential to enable optimized and cost-effective production of natural gas. There is demonstrably significant industrial demand for a smart metering device with diagnostic capabilities. The project addresses major technological challenges:
• measurement of the liquid loading at low cost.
• asset health monitoring with diagnostic capabilities.
• models/correlations proven at field conditions.
The project partners, GCU, TUV-SUD-NEL and McMenon, will contribute significant resources to design, manufacture, test and train an AI enabled instrument and bring a device to market which can play a vital role in bringing marginal fields into operation and in optimizing production of natural gas.

Cost-effective soft strain sensors using laser-induced graphene for breathing monitoring in healthcare applications

This research introduces the development of cost-effective, flexible strain sensors using laser-induced graphene (LIG) for real-time breathing monitoring, with potential applications in healthcare. The sensors are fabricated by directly patterning graphene onto a polymer substrate using a CO2 laser, a method that offers both scalability and affordability. These LIG sensors are embedded in a soft elastomer matrix, making them highly adaptable to the subtle movements of the human body, particularly in respiratory monitoring. The sensors exhibit excellent electromechanical performance, with high sensitivity to the minute strains associated with breathing. This makes them particularly suited for integration into wearable healthcare devices aimed at continuous, non-invasive monitoring of respiratory conditions, such as asthma, sleep apnea, or chronic obstructive pulmonary disease (COPD). In addition to their promising application potential, the study also focuses on optimizing the laser parameters and material composition to enhance sensor performance, including sensitivity, durability, and response time. By refining these aspects, this research aims to contribute to the advancement of low-cost, reliable healthcare monitoring systems, potentially improving patient outcomes through early detection and management of respiratory issues.

A mechanism for preventing access to adult content

Preventing children from accessing online adult content has become a challenging issue for parents and governments. This is due to the digital revolution and the fact that people use Internet services from a very young age. Conventional methods of age verification have proven to be ineffective and have raised considerable security and privacy concerns. Based on previous research, the proposed PhD suggests a fast, secure and effective identification mechanism for classifying a digital user as an adult or a child using a number of sensors, including keystroke information, webcam assessment of saccadic eye movement and sensors to effect voice recognition.

Using additive manufacturing to develop a soft, wearable sensor mesh for medical applications

This project aims to use additive manufacturing to develop a controllable, wearable mesh which uses thermoelectric stimulation to actuate and act as a compressive sleeve while simultaneously measuring vital signs such as body temperature or blood pressure, to be used in biomedical applications. The focus is to create pressure and temperature sensors which can be easily integrated into a wearable sleeve and potentially explore multimodality, such that a single device measures changes in pressure and temperature simultaneously. The use of additive manufacturing and optimising printing parameters allows for rapid production of personalised devices for improved patient care while tailoring the sensors’ sensitivity, pressure range, and creating complex multi-material structures in a single process.

These sensors consist of a flexible Polydimethylsiloxane (PDMS) substrate with Laser-Induced Graphene (LIG), for measuring changes in resistance across the graphene structure, and are placed onto printed conductive electrodes to detect these electrical changes. Varying the surface geometry of these electrodes changes the contact area between the LIG and electrode with increasing pressure, allowing for improved linearity of the sensor across a greater pressure range which is desirable in applications beyond the biomedical industry.

A look into the reproducibility of Bluetooth attacks

With the prevalence of Bluetooth in the IoT industry, it is essential that vendors are able to create products that are robust against critical security vulnerabilities. In our research project, we explore the feasibility of two major Bluetooth attacks, BIAS and BLUFFS, in modern hardware. We demonstrate those attacks are still possible even after the standardization of the mitigation techniques, and we discuss the importance of reproducibility in cyber security as a whole.

Stable-sketch: a versatile sketch for accurate, fast, web-scale data stream processing

Data stream processing plays a pivotal role in various web-related applications, including click fraud detection, anomaly identification, and recommendation systems. Accurate and fast detection of items relevant to such tasks within data streams, e.g., heavy hitters, heavy changers, and persistent items, is however non-trivial. This is due to growing streaming speeds, limited fast memory (L1 cache) available in current systems, and highly skewed item distributions encountered in practice. In effect, items of interest that are tracked only based on their features (e.g., item frequency or persistence value) are susceptible to replacement by non-relevant ones, leading to modest detection accuracy, as we reveal. In this work, we introduce the notion of bucket stability, which quantifies the degree of recorded item variation, and show that this is a powerful metric for identifying distinct item types. We propose Stable-Sketch, an elegant and versatile sketch that exploits multidimensional information, including item statistics and bucket stability, and adopts a stochastic approach to drive replacement decisions. We present a theoretical analysis of the error bounds of Stable-Sketch, and conduct extensive experiments to demonstrate that our solution achieves substantially higher accuracy and faster processing speeds than state-of-the-art sketches in a range of item detection tasks, even with tight memories.

Development of acoustic and airflow sensors inspired by the hair sensilla of insects and arachnids

Nature has always inspired humans in creating innovative tools. Arachnids and insects show exceptional and functional sensory receptors at a small scale. Air flow mechanoreceptors, commonly called trichobothria, are used in different shapes and sizes by several arachnid species. Moreover, some trichobothria appear to be sensitive to low-frequency near-field acoustic signals. The goal of this work is to develop flat hair-like sensors inspired by the adult Buthus occitanus scorpion, that can react to either airflow or acoustic narrow frequency bands. A sensor that responds to airflow has been developed and realized using multi-material additive manufacturing (AM), also known as 3D printing. Furthermore, using the same production technique, it is possible to create sensor structures that react to sound. They show different narrow frequency bands responses based on differences in shape and size of the artificial hair. These 3D printed structures were successfully coated with metal and the conversion of their mechanical movements into an electric signal was achieved. Nevertheless, further work is being conducted to proper condition their electric response, tailoring it for real world applications.

A novel technique for the calibration of the acoustoelastic constant

Residual stresses are present after most manufacturing processes, and can cause early, unexpected failure or strength deterioration within a structure. As such, measurement of residual stresses after manufacturing is an essential part of the diet of inspections a part is subjected to. There are several methods to measure residual stress, both destructively and non-destructively, and in my work I am studying the non-destructive ultrasonic technique, and developing a specific technique utilising Phased Array probes. This is all to give context for the methodology used, which is based on the principal of acoustoelasticity – that a material’s stress will affect the acoustic wave velocity within it. This varies based upon the acoustoelastic constant, which varies between materials, and even within batches of the same material. Therefore, before a residual stress investigation is carried out, the acoustoelastic constant must be calibrated. This poster will focus on the novel procedure for acoustoelastic calibration that I have developed and will discuss the methodology, the advantages that Phased Array probes bring to this essential calibration process. Finally, areas of potential improvement will be discussed, as well as the results of this investigation in the wider context of a residual stress investigation.

Real-time threat detection in IoT networks with machine learning and AI

The proposed research study revolves around improving threat identification in Internet of Things (IoT) networks with the help of machine learning and Artificial intelligence. The importance of IoT security increases as IoT devices spread across industries due to the threat that can occur due to new attack vectors the IoT devices can introduce. The purpose of this project is to design sophisticated ML and AI models for interception and neutralization of the threats in the IoT networks within the shortest amount of time possible.

It entails using large database and develop machine learning models that analyze network traffic activity and identify risks. They are used in real-time and are able to look into the data transmitted in a network to identify threats, and deal with them efficiently. Therefore, in an attempt to enhance the capability of threat detection systems, this research utilises techniques like anomaly detection; supervised learning; and deep learning.

Wireless sensor network for monitoring surface and ground water sources for pesticide pollution

This study is focused on the design of a comprehensive system for pesticide testing in regions where water sources are highly distributed, with a particular emphasis on addressing the pressing issue of water contamination by pesticides in Kenya. Between 2015 and 2018, pesticide imports in Kenya surged by 144%, with 76% of these pesticides classified as highly hazardous. Agriculture remains the cornerstone of Kenya’s economy, contributing over 30% of the nation’s gross domestic product (GDP). However, water scarcity poses a significant challenge, especially in rural areas where more than 70% of the population resides. Approximately 41.5% of these residents depend on shallow wells and surface water sources, such as lakes, streams, and ponds, for their daily water needs.

The study proposes the development of an advanced wireless sensor system that integrates RFID, Long Range (LoRa) protocol, and GSM to monitor the prevalence of pesticides in groundwater sources. This wireless architecture offers a point-of-care testing (POCT) solution through its integration with cost-effective, paper-based immunochromatographic biosensors. These biosensors leverage the specificity of antibodies for the detection of pesticides, enabling straightforward visual identification. The gathered information is disseminated effectively, with a focus on reaching marginalized groups, including those with limited literacy, the elderly, and the economically disadvantaged.

Towards rapid detection of microplastic particles in human blood samples

Small particles of plastic, called micro/nano plastics, have been found in all environments, including oceans, land, freshwaters and within human blood, lungs, and liver. The long-term human health impact of these micro/nano plastics is still unknown, but several reports suggest that the presence of plastic inside humans leads to inflammation, oxidative stress and potentially organ damage. With plastic pollution growing relentlessly, understanding the fate and impact of micro/nano plastic pollution on humans is of paramount importance, but the techniques used to detect the presence of these plastics is difficult and time consuming. In this project, our goal is to overcome these constraints by creating a user-friendly device designed for on-the-spot detection of micro/nano plastics in blood, eliminating the necessity for costly laboratory skills or equipment. The concept involves utilizing a biological recognition molecule to enzymatically identify microplastic particles. By introducing a colorimetric substrate specific to the enzyme, we can identify the existence of the peptide bound to PS, and consequently, the presence of plastic. To support low cost detection we describe an electrochemical detection approach.

Sustainable sensors manufacturing

The goal is to develop materials for fully compostable sensors, including substrates, encapsulation, conductive tracks, sensing layers, and electronic layers. The project emphasizes the use of everyday materials and biodegradable polymers, such as cellulose, polylactic acid (PLA), and starch-based materials, alongside metal nanoparticles, to achieve good electrical conductivity while ensuring the entire device can break down naturally and enrich the soil after its lifetime.

Organic optoelectronic devices for neuromorphic engineering

In recent years, there has been a great interest in optoelectronic devices for use in artificial intelligence vision systems (AIVs). The aim of this project is to design such devices which are capable of both light sensing and synaptic functionality just like our retina and visual cortex, using organic/non-toxic materials. The human visual system is based on a neural network formed by synaptic connections between neurons, through which a vast number of photoreceptor neurons convert incident light to neuro-electrical signals. Similarly, the operation of the aforementioned devices can realize event driven responsivity to optical stimuli, real time processing, as well as data storage (memory), effectively bio-mimicking the operation of neuronal synapses. Such devices have been implemented in Artificial Neural networks (ANNs) in an effort to solve complex tasks such as image processing, speech imagery data recognition, spatiotemporal pattern recognition and visual signal classification. Regarding AIVs, their key functionality is the ability to modify the operation of the synaptic devices to distinguish and remember different wavelengths and intensities. Moreover, the use of organic/non-toxic materials simplifies the fabrication process, minimizes fabrication costs and ultimately reduces the impact on the environment.

Metamaterials applied to biomedical ultrasound: novel backing layers for dynamic control

Investigating the optimisation of 3D printed acoustic metamaterials as probe backing layers for biomedical ultrasound imaging. Aiming to demonstrate an investigative method to aid acoustic designers in optimising their device design for high-frequency manipulation, as well as to choose the most suitable material for a wide range of applications.

Enhancing cyber-security in smart seaports with TwinPot: a digital twin-assisted honeypot system

My research focuses on the development and implementation of TwinPot, a Digital Twin (DT)-assisted honeypot system designed to enhance the cyber-security of smart seaports. With the increasing digitization of seaports driven by technologies such as IoT and AI, the need for robust security mechanisms has become paramount. Traditional security measures often fall short in handling sophisticated cyber-attacks. TwinPot addresses this gap by integrating DT technology with honeypots to create high-fidelity, realistic traps that attract and analyze cyber-attacks, providing a more sophisticated defence mechanism against cyber-attacks. TwinPot duplicates seaport assets into a high-fidelity virtual environment, attracting attackers and enabling detailed behavioural analysis. This system detects attacks through simulated entities and employs an intelligent attack detection mechanism. By leveraging advanced classification methods, TwinPot can dynamically adapt to various attack types, ensuring comprehensive protection. It offers a scalable, effective solution for safeguarding the critical infrastructure of smart seaports, contributing significantly to maritime cyber-security.