Civil engineering in the age of artificial intelligence
I work on AI for infrastructure: how ageing assets are assessed, how failure is anticipated, and what has to be true before an engineering judgement can be trusted. The work runs from mechanics and sensing through to the standards and assurance that decide whether any of it can be relied upon.
Four layers, each harder to replicate than the one above it. The fourth is the one that decides whether the first three ever reach an asset.
Layer 1 and 2
Hazard-structure coupling, failure probability and residual capacity, topology optimisation, data governance. Above that sit the things that travel: digital twin platform architecture, a data-asset meta-model and access standard for ageing assets, inspection procedures and draft evaluation standards that do not depend on the original data.
Layer 3
Replaceable steel connections, modular emergency tower design, climbing inspection and assembly robotics, and before-and-after strengthening trials on decommissioned structures. Most comparable work stops at proof of concept and papers.
Layer 4
Research groups join with their own teams and plug into a shared framework of directions, targets, funding and facilities. The projects they win sign through the platform, while a small in-house team runs finance, delivery, market development and commercialisation. Mature results can spin out as companies with the platform holding equity and continuing support. Overseas centres run the same model, with the platform paying the partner side.
How claims are handled when the work is technical, the client is accountable, and part of the delivery is AI-assisted.
What gets written down
Where a benchmark was not met, it goes into the stage report next to the tests that passed, with the calibration work that follows. Where a target is contractual rather than achieved, we say so: an acceptance target is not a realised outcome until it has been checked against real cases.
AI-assisted delivery
On AI-assisted deliverables, each statement traces back to material the client provided. The model does not get to find its own facts. Assurance is the product, not the packaging.
Where the research has been examined, taught, standardised and put to work.
Publications
High-impact work on digital twin-enabled predictive maintenance for building operations, contributing to asset management and infrastructure data research.
Standards
Algorithms are replaced continuously and are dated by the time they reach the field. Complete, consistent, standardised data is the shared input every future generation of algorithm will need, which is where the durable asset sits. I work on this from inside the standards process: committee member of ISO/TC 59/SC 13 JWG 14, the joint working group on GIS and BIM, and deputy chair of the standards digitalisation committee of the China Association for Engineering Construction Standardization.
Teaching
Former OsloMet Associate Professor experience, postgraduate teaching in BIM and digital engineering, Loughborough guest teaching, and supervision of doctoral and master's research.
Knowledge transfer
Published methods written into industry inspection procedures, commissioned resilience research for a national grid operator, and a research platform built with district and municipal support in Chongqing, reported independently by national and regional media.