SELECTED WORK
From physical models
to practical intelligence.
Three connected research themes across energy infrastructure, structural integrity, and advanced manufacturing.
01 / Power grid assets
Physics-informed digital twins
Fast thermal insight for bushings and fluid-immersed transformers under changing load and weather.
Conceptual illustration · not a simulation result
The challenge
High-fidelity thermal simulations are valuable for understanding critical assets, but expensive to repeat across operating conditions.
Approach
Neural operator models combine operating conditions with spatial representations of engineering geometry to predict temperatures and hot spots.
Research directions
Multi-domain thermal modelling, solid–fluid interfaces, physics-informed training and parameterised operating conditions.
THE GOAL
Enable near-real-time thermal prediction while retaining the physical structure of the engineering problem.
02 / Structural integrity
Neural ODEs for fatigue crack growth
Continuous crack-growth models that connect loading, material physics, and uncertainty.
Conceptual illustration · not a simulation result
The challenge
Crack evolution depends on material behaviour, geometry and loading. Probabilistic assessment requires many repeated predictions.
Approach
Physics-guided dynamical models incorporate NASGRO material degradation equations and propagate uncertainty in initial states and loading.
Research directions
Controlled and latent Neural ODEs, monotonic dynamics, variable-amplitude loading, load-sequence effects and long-horizon prediction.
THE GOAL
Build fast surrogate models for probabilistic structural-integrity assessment.
03 / Advanced manufacturing
Scientific ML for manufacturing
Sensor-informed dynamics for composite processing, casting, and manufacturing quality.
Conceptual illustration · not a simulation result
The challenge
Manufacturing combines thermal and fluid dynamics with time-varying process controls, sensor measurements and final quality outcomes.
Approach
Neural ODE models track process dynamics in composite molds to predict the final post-curing repair index. Related work includes data-driven robotic control for casting.
Research directions
Latent-state modelling, sensor-informed state estimation, sensitivity analysis, inverse optimisation and generative engineering design.
THE GOAL
Move toward predictive process optimisation and more informed quality decisions.
RELATED WORK
Generative design of turbine blades
Diffusion-based generation of turbine-blade geometries in the spectral domain, conditioned on conjugate heat-transfer performance metrics.
THE NEXT CONVERSATION