How AI is tracking RAAC

But a group of CIOB members at Loughborough University had known about the risks of RAAC for some time, and had already carried out extensive research into the material – also developing a machine learning tool to help mitigate the problems it can cause.

Hear more in our latest podcast

RAAC is an aerated lightweight form of concrete with no coarse aggregate, hence its material properties and structural behaviour are different from ‘traditional’ reinforced concrete. It has been used in the UK since the 1950s, but it is only recently that concerns over its tendency to ‘crumble’ – usually due to maintenance issues or water ingress – have emerged.

A report by the Standing Committee on Structural Safety, issued in May 2019, highlighted a significant risk of failure in RAAC panels.

“It is essential that those responsible for the management, maintenance or alteration of buildings know whether their buildings contain RAAC,” says Chris Goodier MCIOB, professor of construction engineering and materials at Loughborough University.

“If they do not know, they should seek appropriate expert advice. If not properly managed, RAAC structural building components have significant safety implications.”

In 2021, the Loughborough team were called in by the NHS to examine RAAC issues in seven hospitals.

“The NHS commissioned us to try and understand how RAAC behaves and look at their survey methodology and how data is captured,” explains Dr Karen Blay MCIOB, senior lecturer in digital construction and quantity surveying at Loughborough University.

“Also, co-develop a solution with the maintenance team for predictive maintenance – understanding what the risks are and what they need to do differently.”

The Loughborough team looked at how AI could be used to identify RAAC defects – but with very little data out there, the first task was to build up a library of photographs that the machine could ‘learn’ from.

“Initially we trained the AI tool to identify cracks, using 85,220 images of concrete, of which 1,800 were RAAC – taken from the estates of the NHS trusts we were working with, including timestamps of the pictures,” Blay explains.

Identifying cracks

Using that base data, when new RAAC images were scanned by the AI tool, the machine was able to identify cracks in RAAC with an accuracy level of 95.8%, she adds. The machine learning tool was created at Loughborough using the Python-based PyTorch tool.

“What we are now looking to do is integrate use of the tool within the survey processes of the NHS trusts estate teams,” Blay says. “Using this machine learning solution, we can generate insights and predict the behaviour of RAAC panels, which will help us make better decisions and put in place ‘fail-safes’.”

The AI tool could also help with the requirements of the Building Safety Act and the golden thread of information.

Chris Gorse MCIOB, professor of construction engineering and management at Loughborough University, and chair of the CIOB sustainability panel, says: “AI has been given a lot of bad press. But one of the things that it is able to do is recognise things relatively easily and sort through thousands of sets of data. Here, it has the ability to recognise changes in RAAC panels and that is going to be very useful.

“And this project aligns well with the Building Safety Act. We know about Dame Judith Hackitt’s idea for a golden thread of information. What became apparent with the Grenfell tragedy was that we just haven’t got a good enough knowledge of our buildings and the products used to construct them and how they perform.”

Looking ahead, Blay says that the Loughborough team is examining how to create a RAAC digital twin. “We are going to have to live with RAAC, so we need to be able to understand the changes that will occur in RAAC panels,” she says. “A digital twin will help facilitate that.”