Research

My research develops AI methods for modelling, reasoning about, and designing complex physical systems. It connects high-fidelity simulation and scientific data with foundation models and agentic engineering workflows that remain verifiable, traceable, and accountable to human engineering judgement.

Current research

Research programme

Agentic computational engineering

AI systems that formulate engineering tasks; select and orchestrate numerical, analysis, and optimisation tools; verify results for physical consistency; preserve provenance and traceability; handle uncertainty and recover from failures; and retain human oversight for consequential decisions.

NASA Ames talk

Open scientific data

CAE ML Datasets

I lead and collaborate on large-scale, high-fidelity CFD datasets for reproducible machine-learning research. AhmedML, WindsorML, DrivAerML, and HiLiftAeroML cover automotive and aerospace configurations of increasing geometric and physical complexity.

Project website

Perspective paper

Fluid Intelligence

An analysis of scaling laws, data-generation cost, and the technical requirements for foundation models in computational fluid dynamics.

Paper

Community workshops

I founded AutoCFD and serve on the HLPW organising committee, where I lead the AI/ML Technical Focus Group. These programmes establish shared test cases, evaluation methods, and community research priorities for computational aerodynamics.

AIAA CFD High-Lift Prediction Workshop (HLPW)

Organising committee and AI/ML Technical Focus Group lead

A collaborative workshop series advancing computational methods and validation practices for complex high-lift aircraft configurations.

Open engineering datasets

AhmedML

Scale-resolving CFD for 500 geometric variants of the Ahmed body.

WindsorML

GPU-native wall-modelled large-eddy simulations for 355 Windsor-body variants.

DrivAerML

Surface and volume flow data for 500 realistic road-car variants.

HiLiftAeroML

High-fidelity simulations across geometry variants and angles of attack for high-lift aircraft aerodynamics.

Research questions

Physical-system representations

How can models learn from high-fidelity simulations and generalise across geometries, operating conditions, and engineering tasks?

Tool-using engineering agents

How can agents select and orchestrate simulation, analysis, and optimisation tools while verifying physical consistency, preserving provenance, recovering from failures, representing uncertainty, and retaining appropriate human control?

Trustworthy evaluation

Which datasets and benchmarks demonstrate physical fidelity, engineering utility, computational efficiency, and out-of-distribution performance?

Research correspondence

I welcome correspondence from researchers working on related problems in industry and academia.

contact@neilashton.co.uk