PhD Studentship: Physics-Informed Machine Learning for Control of Dynamic Systems University College London in United Kingdom

This four-year PhD studentship at University College London, within the Department of Chemical Engineering and the Sargent Centre for Process Systems Engineering, offers an exciting opportunity to develop a physics-informed machine learning framework for the control of nonlinear dynamic systems governed by partial differential equations (PDEs). The project is motivated by the need for advanced control strategies in industrial processes critical to the green energy transition, such as carbon capture in adsorption systems and hydrogen storage in cryogenic tanks. These processes are often subject to significant uncertainty due to unmeasured disturbances, model mismatch, and sensor noise, making conventional control strategies insufficient under non-ideal conditions.

The research will focus on embedding mechanistic priors and error quantification into machine learning models using Bayesian optimisation, and will build upon recent advances in data-driven stochastic model predictive control (MPC) and Gaussian Process (GP) inference for PDE systems. The aim is to generalise GP-based approaches to nonlinear systems, integrating physical constraints and uncertainty quantification directly into the model structure for data-efficient learning and reliable extrapolation. The developed physics-informed GP surrogate will be embedded into a stochastic MPC framework, enabling robust, adaptive, and uncertainty-aware control of dynamic systems. Additionally, a Bayesian optimisation module will be created for adaptive controller tuning, with hyperparameters updated online under physics-informed priors to ensure physically plausible operation.

Validation will be conducted through case studies of increasing complexity, including fixed-bed adsorption systems, CO2 capture via vacuum pressure swing adsorption (VPSA), and cryogenic tanks for liquid hydrogen storage, all governed by nonlinear heat and mass transfer PDEs. The framework is designed to be broadly applicable across chemical and process engineering domains. The successful candidate will be supervised by Dr Paulina Quintanilla and Professor Federico Galvanin, gaining expertise in physics-informed machine learning, Bayesian optimisation, Gaussian process modelling, stochastic MPC, PDE modelling, numerical analysis, and programming (Python, MATLAB, or Julia).

The project also offers the possibility of international collaboration with experts in cryogenic storage modelling at Pontificia Universidad Católica de Chile, potentially including a short research placement. Applicants should have a strong background in chemical engineering, control engineering, applied mathematics, or computational science, with excellent analytical and programming skills and an interest in sustainable process optimisation.

Funding covers a stipend of approximately £23,466 per annum plus UK-equivalent tuition fees for four years; overseas students may apply but must cover the difference in fees. Applications should be submitted via the UCL portal, nominating Dr Paulina Quintanilla as supervisor and including a statement of interest. Informal enquiries can be directed to Dr Paulina Quintanilla at [email protected].