PhD Studentship: Multiscale Scientific Machine Learning Digital Twins for High-Torque Electric Propulsion under Extreme Duty Cycles

Electrifying heavy-duty maritime vessels, mining equipment and off-highway machinery is central to Net Zero and Clean Energy goals. Unlike conventional automotive traction systems, electric propulsion for these applications must often deliver sustained high torque at low rotational speed during demanding operations such as bollard pull, winching and heavy excavation. Under these extreme duty cycles, motor behaviour is governed by processes that evolve over very different timescales: fast electromagnetic dynamics determine torque response and saturation behaviour, electromechanical and load-history effects shape transient performance, and slow thermal dynamics govern the gradual build-up and dissipation of heat over extended operating periods.

These processes are strongly coupled, since changes in temperature alter electrical resistance, efficiency and torque capability, making the overall behaviour difficult to predict with physics-based models or conventional machine learning alone. Characterising these operating regimes experimentally is genuinely difficult. Testing high-power motors repeatedly to their thermal limits is energy-intensive, costly and potentially damaging, with risks including insulation degradation, excessive winding temperatures and permanent-magnet demagnetisation. Detailed physics-based simulations can be too slow to run for long, variable duty cycles, while conventional machine-learning models often generalise poorly beyond the conditions they were trained on, particularly for long-horizon thermal behaviour. This project will develop a multiscale Scientific Machine Learning (SciML) digital-twin framework predicting thermal trajectories, torque capability and efficiency variation in low speed, high-torque propulsion systems, investigating physics-informed hybrid architectures spanning fast, intermediate and slow system dynamics: Deep Neural Networks for nonlinear electromagnetic relationships and saturation effects, recurrent or neural state-space models for transient behaviour and load-history dependence, and Neural Ordinary Differential Equations for continuous-time thermal accumulation over multi-hour horizons.

The student will design safety-constrained dynamometer experiments at the University's Centre for Future Clean Mobility (CFCM), acquiring high-frequency electrical, mechanical and thermal measurements to train and validate the digital twin. Engineering and Physical Sciences Research Council Doctoral Landscape Award A central research challenge is determining how reliable operating-envelope predictions can be obtained from a small number of carefully chosen, non-destructive experiments rather than exhaustive testing to physical limits, requiring data-efficient experimental design, physics based constraints, and uncertainty quantification to assess confidence when the motor operates under conditions it hasn't directly been tested on. The resulting digital twin will provide fast, uncertainty-aware predictions of thermal loading, torque capability and efficiency over time, supporting safe evaluation of severe operating profiles that are difficult or risky to reproduce repeatedly on a physical testbed, and providing evidence to guide motor sizing and duty-cycle decisions for next-generation zero-emission propulsion systems.

This is a collaborative studentship with Ecomar Innovations, a marine power-systems company. The student will spend time on-site with Ecomar's engineering team, testing the framework against real industrial hardware and supporting improved thermal-management strategies for their systems, giving direct exposure to industry-scale equipment and a route from research to real-world deployment, alongside supervision from a team with an established track record in maritime electrification and control systems. The project welcomes applicants from electrical/mechanical engineering, control systems, applied mathematics or computer science, and offers interdisciplinary training in electric machines, SciML, thermal modelling and digital-twin validation.

Funded by: EPSRC Engineering and Physical Sciences Research Council