Imperial College London has secured £1.8m to develop artificial intelligence systems that help manufacturers assess whether complex components can be produced reliably, including identifying when automated predictions require further engineering verification.
The three-year project will be led by Dr Nan Li, Reader in Lightweight Design and Manufacturing at Imperial College London‘s Dyson School of Design Engineering.
Its focus is advanced material forming, a manufacturing process used to produce components including vehicle body panels, aircraft structures, and parts requiring demanding safety and quality standards.
The funding is part of a £39m UK Research and Innovation programme supporting 20 researchers developing applications of AI across scientific and industrial disciplines.
Li’s project, AIMForming, will combine information about component geometry, material behaviour, and manufacturing processes to help engineers evaluate whether proposed designs are practical to manufacture.
The system is intended to support decisions during product development, allowing engineers to identify manufacturing difficulties before committing to tooling, physical trials, or production.
A particular objective is to improve the reliability of AI predictions when operating conditions differ from those represented in the information used to train a model.
Manufacturers increasingly work with recycled materials and production methods intended to reduce emissions. These changes can introduce variations in material properties that complicate conventional forming assessments.
AI models trained on established production conditions may not predict performance reliably when component designs, material characteristics, or manufacturing processes change.
Identifying when AI predictions cannot be trusted —
AIMForming will be designed to recognise circumstances where its evidence is insufficient to support a confident prediction.
Rather than automatically approving a proposed component design, the system will indicate when additional simulation, testing, or engineering judgement is required.
That distinction is important in manufacturing applications where an incorrect prediction can lead to tooling changes, wasted materials, delays, or components that fail to meet production requirements.
Traditional assessments of complex formed components can involve repeated computer simulations and physical trials, particularly when designers introduce unfamiliar geometries or materials.
These processes require engineering time, specialist equipment, and testing resources. They can also limit the number of design alternatives that manufacturers can practically assess.
Li’s existing research group has developed deep-learning and graph-based simulation systems capable of predicting how components will form.
Some of those technologies have already been validated within commercial engineering software through two projects funded by Innovate UK.
The new fellowship builds on that work by focusing on the reliability and limitations of predictions rather than simply producing results more quickly.
The project is supported by 18 industrial partners, including component manufacturers, materials suppliers, engineering software developers, and organisations from the automotive and aerospace sectors.
Those partners have committed approximately £1.2m of support in kind, providing expertise and resources separately from the £1.8m research funding.
Their participation is intended to ensure the research addresses practical manufacturing conditions and can be evaluated against relevant industrial requirements.
Li’s previous work encompasses forming processes involving steel, aluminium, titanium, and composite materials, providing a basis for exploring the technology across different production environments.
The project will also establish an open benchmarking resource and a UK research community focused on AI applications in material forming.
These arrangements are intended to allow companies and research teams to test methods against shared evidence rather than developing entirely separate datasets and evaluation processes.
Smaller manufacturers and specialist engineering software providers could benefit from access to common testing resources, particularly where developing their own AI systems would require substantial computing capacity and technical expertise.
Potential reductions in development times, material waste, and manufacturing costs remain research objectives rather than independently demonstrated commercial outcomes.
The fellowship will run for three years, with Imperial and its industrial partners developing the system and evaluating its suitability for engineering decisions involving complex manufactured components.



