Research into the feasibility of an additive-manufactured ultra high efficiency, high temperature micro gas turbine. The project aims to carry out fundamental research into a highly novel micro gas turbine by designing, manufacturing and testing a combustion system with industry support from HiETA Technologies utilising Additive Manufacturing to create high efficiency cooling systems. The objective is to prove the feasibility of running a system at very high gas temperatures to yield efficiency improvements. To start, research will be conducted on already existing combustor designs for similar micro-gas turbine applications, to gain an understanding of the already existing technology in the market and identify possible improvements that can be implemented with the use of additive manufacturing. This research will then feed into the initial proof of concept design that will then be analysed using CFD, manufactured by the project industrial partner HiETA and tested in the hot gas stand cell at Bath once it is fitted with a high temperature turbine. Further research on state of the art combustion cooling designs and CFD analysis on fuel delivery and combustion processes will follow, which will lead to multiple designs for a state of the art combustion system, which HiETA will assist in manufacturing. The designs will then be tested at high temperatures in the hot gas stand test cell at Bath again to validate the designs.

Fast and widespread uptake of low-carbon technologies (LCTs) - such as electric vehicles and heat pumps - is necessary for decarbonising the UK's energy consumption, but presents significant challenges for the country's power systems. Needed in advance of this are costly upgrades to the nation's electricity distribution networks and policies that decrease demand during periods of peak consumption. Determining where network infrastructure is most needed, and how to most effectively mitigate costly network activity, for accelerating the decarbonisation of the UK's energy consumption is necessary for reaching the country's 2035 and 2050 goals. However, there is significant uncertainty on the rate of uptake, nature, and location of the LCTs being installed - hindering our ability to estimate future energy consumption behaviour. To account for this uncertainty, a probabilistic approach is taken to model network changes and consumer behaviour in order to inform planning, pricing, and investment. AIMS, OBJECTIVES & BENEFITS In this research project, I plan to probabilistically model power systems - and individuals interacting with it - via a hierarchical Bayesian model. This approach seeks to use empirical data to train and improve the scientific models governing power system simulations. With this simulation platform, the effect of different technologies, consumer behaviours, and policies are to be modelled. The aim of this is to determine strategic investments for boosting the transition to LCTs. Additionally, we aim to identify the potential for strategic network pricing methodology, so that consumer behaviours can be influenced to optimally utilise renewable energy with existing electricity infrastructure. The benefits of these findings have the potential to bring cost savings to energy consumers, while accelerating the transition to renewables. RESEARCH COUNCIL RELEVANCE This project is being undertaken as part of the Accountable, Responsible and Transparent AI (ART-AI) - a UK Research and Innovation (UKRI) funded Centre for Doctoral Training (CDT). UKRI, with its strategic investment in artificial intelligence research, seeks to support the use of artificial intelligence advances for "application-driven research and innovation in discovery science and in areas such as health, the environment, agriculture, security, and government policy". In particular, according to the UKRI's "Transforming Our World With AI" report, they are looking to encourage the adoption of AI technologies to "manage smart energy networks, tackle climate change and deliver net zero CO2 targets". This project is strongly aligned with these aims, by seeking to answer pressing questions about the installation of LCTs and the design of smart energy networks that influence their usage. Furthermore, this project seeks to specifically conduct AI research in a responsible manner, while establishing interpretable machine learning techniques. As decisions must be made for power systems with technologies, behaviour, and phenomena that will not be representable from existing power system data, this project is using model-based methods so that scientifically-determined rules and human judgement are used (as well as data). This approach should therefore result in conclusions that properly account for the anticipated uncertainty associated with power systems, and avoid the overconfident and misguided predictions that could result from purely data-driven approaches. If successful, the methods used may be instructive for increasing the robustness of AI-based research to distributional shifts. Additionally, it is important that this research, as it looks to inform the design of consumer electricity pricing schemes, be transparent and auditable. By eschewing black-box techniques, and instead building models with interpretable variables, unjust correlations between different household attributes and network usage costs can be easily determined.

TBC 22/23

TBC 22/23

This project will investigate the use of machine learning and neural network methodologies to solve problems involving partial and stochastic differential equations. We will initially consider contaminant dispersal models as an exemplar; in this problem pollutant particles are modelled individually and we are interested in learning the distribution of a large number of such particles. The Fokker-Planck equation for this models is high dimensional, and currently only solvable using Monte Carlo methods. The first part of the project will focus on the efficient approximation of the solution to the forward problem using deep learning methods. A TensorFlow implementation of a deep learning high dimensional PDE solver will be created which incorporates suitable boundary conditions and background flow field. This approach will be analysed analytically where possible and compared to existing methods, such as the MLMC method developed by G. Katsiolides. Once implemented this method of solution will create avenues which can be used to approach the inverse problem of using data to parameterise the model by applying deep learning techniques and/or Bayesian methods; this part of the problem will be explored subsequently. There are a range of applications which could be considered in the later stages of the project, these include, but are not limited to, stochastic PDE models of particle movements, and stochastic optimal control.