Construction researchers at Loughborough University have created a digital, machine-learning tool that can analyse thousands of pictures of building interiors to detect the presence of distressed RAAC structural elements and predict how they’ll behave.
The work, led by Prof Chris Gorse MCIOB and Dr Karen Blay MCIOB, is explored in depth in the latest episode of the CIOB’s 21CC podcast.
RAAC – short for reinforced autoclaved aerated concrete – is a lightweight form of concrete with no coarse aggregate.
Tens of thousands of RAAC structural panels exist across a broad range of buildings, many constructed in the 1950s, 60s, and 70s.
Many are showing signs of deterioration. The vast majority form the roof of the structure, usually flat, and hence are difficult to access, survey, maintain, and replace.
The Loughborough team were commissioned by the UK’s National Health Service (NHS) in August 2021 to help it understand how RAAC behaves in its buildings and hospitals, and to help develop a predictive maintenance strategy.
They set out to “train” their AI system first with more than 18,000 photographs they took of cracks in normal concrete in a sample of NHS buildings.
Then they finessed that training with around 1,850 images they took of cracked RAAC elements in the NHS buildings.
RAAC deterioration
Using software developed at Loughborough, the system can detect RAAC-specific deterioration to an accuracy of around 95%, Dr. Karen Blay told us.
Using the time stamp of the photographs and the cracks’ location, they are now developing digital twins of the RAAC elements to help plan maintenance.
“We’re creating a digital twin of these RAACs so that we can understand how changes happen. So it’s all about capturing change,” said Blay.
“We are going to live with RAAC, so it’s about how do we capture the data after you put in in your failsafes, moving forward? How can we extend their lifespan? How can we proactively predict the behaviour of RAAC using digital solutions?”
Find out more by listening to the 21CC podcast.
