Hybrid Machine-Learning and Computational Fluid Dynamics Methods in the Energy Industry
An opportunity exists to undertake a PhD with the Matar Fluids Group (https://www.imperial.ac.uk/matar-fluids-group/) in the Department of Chemical Engineering at Imperial College London funded by BP. The successful student will contribute to the modelling and simulation of multiphase flows using hybrid methods that rely on a combination of machine-learning and computational fluid dynamics.
Engineering applications of turbulent multi-phase flows typically involve optimising hyper-parameters (related to flow, geometry etc.) to maximise a defined performance metric. In the energy industry, in spite of decades of research, there are a number of significant challenges; overcoming them will lead to a step-change in productivity, efficiency and reduction in emissions. For instance, three-phase flows comprising oil, water, and air, are exceedingly complex and feature poorly understood dynamics, phase formations and transitions. Characterising the three-phase mixture properties, e.g. rheology, is complex, and the prediction of the system behaviour is fraught with large uncertainties.
There is also a dearth of predictive models used by industry that can handle fluids that are either Newtonian but have viscosities that exceed those of water by orders of magnitude (‘heavy’ oils), or exhibit highly-complex rheological behaviour. The majority of current models do not provide accurate predictions in terms of flow pattern maps, phase holdup, and pressure gradient when dealing with such systems. Determining sensitivity of the predictions to the use of the numerous closures depending on the flow regime (a common feature of models in the energy industry) through any kind of statistical analysis is also, to a large extent, not part of the workflow in the modelling process. The issues associated with this lack of robustness can be propagated to higher level simulators (e.g. for reservoirs) with a profound impact on the design of production facilities that rely critically on the quality of the models. We either need a physics-based model to disentangle the individual effects of geometry, chemistry, temperature and pressure, and physico-chemical factors on the flow behaviour (difficult to achieve); or a predictive framework through a hybrid approach involving a combination, and tight integration, of data and mechanistic models for solutions with well-defined uncertainty.
We focus in this project on modelling fluid-fluid displacements during Enhanced Oil Recovery (EOR) and well bore clean-up, central to energy applications, cross-cutting a number of EPSRC research areas, e.g. energy efficiency, fluid dynamics and aerodynamics, and continuum mechanics. Automated, efficient, derivative free, surrogate model-based optimisation will be developed to replace manual hyper-parameter CFD tuning (current practice), to deal with the 3D flows, strongly-coupled variables, and complex geometries in our applications.
Informal enquiries about the post and the application process can be made to Prof. Omar Matar ([Email Address Removed]) by including a motivation letter and CV.
The PhD scholarship is available from October 1st 2019 and is open to all UK applicants. The scholarship covers both the tuition fees and an annual tax-free bursary, and its standard period is 42 months. There will also be an opportunity for a three-month placement at BP Sunbury office to gain industry experience. The successful applicant is expected to have obtained (or be heading for) a First Class Honours degree at Master’s level (or equivalent) in a branch of engineering or related science. The post is based in the Department of Chemical Engineering at Imperial College London (South Kensington Campus).
How good is research at Imperial College London in Aeronautical, Mechanical, Chemical and Manufacturing Engineering?
FTE Category A staff submitted: 172.80
Research output data provided by the Research Excellence Framework (REF)
Click here to see the results for all UK universities