Below are a handful of my favourite projects to date. Although there are many more that I would be happy to burn your ear off about if you ask me about them...
AI-accelerated physics-based simulations
Physics-based simulations of batteries for electrical vehicles (EV) are inherently slow due to their complexity. Together with a leading international EV-battery simulation and testing company, we are developing a product that uses our cutting-edge sampler in combination with machine learning to deliver an AI twin of the original simulation. This AI twin can then be used in place of the original simulation to facilitate previously impossible advanced analytics, including more expansive design searches and predictions. Our AI twin is able to do in a lunch break what the simulation used would take 11 weeks to do. In this project I am responsible for the day-to-day technical oversight and project management. This exciting project will revolutionalise how EV-battery designers work with physics-based simulations, improving their efficiencies and the quality of the designs they produce. It also will allow physics-based simulation to be used in battery-management systems for the first time, which will improve the safety of, and confidence in, EV batteries in the field. It also has application in second-life usage of EV batteries.
Physics-based simulations of batteries for electrical vehicles (EV) are inherently slow due to their complexity. Together with a leading international EV-battery simulation and testing company, we are developing a product that uses our cutting-edge sampler in combination with machine learning to deliver an AI twin of the original simulation. This AI twin can then be used in place of the original simulation to facilitate previously impossible advanced analytics, including more expansive design searches and predictions. Our AI twin is able to do in a lunch break what the simulation used would take 11 weeks to do. In this project I am responsible for the day-to-day technical oversight and project management. This exciting project will revolutionalise how EV-battery designers work with physics-based simulations, improving their efficiencies and the quality of the designs they produce. It also will allow physics-based simulation to be used in battery-management systems for the first time, which will improve the safety of, and confidence in, EV batteries in the field. It also has application in second-life usage of EV batteries.
Protein-folding proof of concept
My first project on joining PolyChord Ltd was to deliver a protein-folding proof-of-concept. I was solely responsible for the technical development and project management of this Amadeus-Capital-funded R&D project. I really enjoyed this project as it really stretched me. I went from knowing almost nothing about protein folding, to delivering a pipeline that exhibited PolyChord's core sampling technology's capabilities to minimise the energy function of a protein chain - folding it into a functional folded protein. The bottom right plot is an animation of a CASP protein folding through the sampling process.
My first project on joining PolyChord Ltd was to deliver a protein-folding proof-of-concept. I was solely responsible for the technical development and project management of this Amadeus-Capital-funded R&D project. I really enjoyed this project as it really stretched me. I went from knowing almost nothing about protein folding, to delivering a pipeline that exhibited PolyChord's core sampling technology's capabilities to minimise the energy function of a protein chain - folding it into a functional folded protein. The bottom right plot is an animation of a CASP protein folding through the sampling process.
A key element of this project was to demonstrate the ability of our methodology to capture dynamic motion of a folded protein, as would be exhibited in its natural form (information that mainstream methods based on machine learning, such as AlphaFold cannot provide). This vibrational movement of the predicted folded protein can be seen in the top right animation. Success in this project led to a follow-on project that was fully funded by a major drug discovery company.
Optimisation of a fusion reactor design
In order to be a realistic solution to removing our dependence on non-renewable energy sources, a fusion reactor must be competitive in price with energy sources such as gas, coal and nuclear-fission. In collaboration with a UK company that is pioneering R&D in projectile-driven fusion, we integrated their neutronics simulation (simulating the motion and interactions of neutrons in a CAD-designed fusion plant reactor) with an economic model for inertial fusion in order to optimise the plant design. Importantly, our solution provided knowledge of uncertainties in the models. This delivers designs that guarantee competitive energy generation using projectile fusion plants. I was solely responsible for the development of this optimisation tool, and also for project management.
In order to be a realistic solution to removing our dependence on non-renewable energy sources, a fusion reactor must be competitive in price with energy sources such as gas, coal and nuclear-fission. In collaboration with a UK company that is pioneering R&D in projectile-driven fusion, we integrated their neutronics simulation (simulating the motion and interactions of neutrons in a CAD-designed fusion plant reactor) with an economic model for inertial fusion in order to optimise the plant design. Importantly, our solution provided knowledge of uncertainties in the models. This delivers designs that guarantee competitive energy generation using projectile fusion plants. I was solely responsible for the development of this optimisation tool, and also for project management.
Planning transmitter locations for communications network
Alongside UK connected and automated-vehicle testbeds, we developed and verified a tool for deciding on where to place transmitters in a network in order to achieve optimal coverage, minimal power and infrastructure usage. The tool we developed can also guarantee network robustness in the event of a transmitter failure. I designed the technical plan, project managed and provided day-to-day technical guidance for this project. We successfully redesigned the existing network to remove a large dead spot in coverage. We also exhibited how the network provides flexibility as to the ultimate placement of transmitters. This information is key, since it is common that when installers come to fit a network, they discover a constraint on placement, e.g. unforeseen building restrictions, that means the original design can't be implemented. We also showed how our tool can deliver a more safe network by optimising with knowledge that a transmitter will likely fail at some point. This feature is critical since a connected vehicle can have an accident if it loses connectivity. You can see the tool in action here.
Alongside UK connected and automated-vehicle testbeds, we developed and verified a tool for deciding on where to place transmitters in a network in order to achieve optimal coverage, minimal power and infrastructure usage. The tool we developed can also guarantee network robustness in the event of a transmitter failure. I designed the technical plan, project managed and provided day-to-day technical guidance for this project. We successfully redesigned the existing network to remove a large dead spot in coverage. We also exhibited how the network provides flexibility as to the ultimate placement of transmitters. This information is key, since it is common that when installers come to fit a network, they discover a constraint on placement, e.g. unforeseen building restrictions, that means the original design can't be implemented. We also showed how our tool can deliver a more safe network by optimising with knowledge that a transmitter will likely fail at some point. This feature is critical since a connected vehicle can have an accident if it loses connectivity. You can see the tool in action here.
Exploitation of structure information to better understand the formation of the first stars and galaxies
Future radio telescopes such as the Square-Kilometre Array have the promise to revolutionise our understanding of the first stars and galaxies. This will be achieved by mapping with time the change of state of the hydrogen that pervades our universe from neutral to ionised, as the radiation from early generations of starts and galaxies ejects the electrons from hydrogen atoms. The evolving maps of the remaining neutral hydrogen will not be statistically smooth, instead there will be complex structure as the stars and galaxies create ionised bubbles that grow with time. During my career I was one of the leading researchers in understanding how we might use so called non-Gaussian statistics to detect this non-linear structure. My publication list summarises my contributions to this endeavour, and the publicly available codes I developed for the community (that are still in use) are available on BitBucket: https://bitbucket.org/caw11.
Future radio telescopes such as the Square-Kilometre Array have the promise to revolutionise our understanding of the first stars and galaxies. This will be achieved by mapping with time the change of state of the hydrogen that pervades our universe from neutral to ionised, as the radiation from early generations of starts and galaxies ejects the electrons from hydrogen atoms. The evolving maps of the remaining neutral hydrogen will not be statistically smooth, instead there will be complex structure as the stars and galaxies create ionised bubbles that grow with time. During my career I was one of the leading researchers in understanding how we might use so called non-Gaussian statistics to detect this non-linear structure. My publication list summarises my contributions to this endeavour, and the publicly available codes I developed for the community (that are still in use) are available on BitBucket: https://bitbucket.org/caw11.