[2B1] Automated defect classification for X-ray backscatter (XBS) inspection of tenanted arches
N Pietrow, M Begg, D Russel, H Murtaza, J Elliott and M Rahman
The Manufacturing Technology Centre (MTC), UK
Railway viaduct structures are highly common across the UK and Europe and many of the arch structures contained within them are converted into useable spaces for business tenants. These so-called ‘tenanted arches’ have strict legal requirements to undergo regular structural inspections and maintenance, which requires direct access to the arch structure that is often cladded with either plastic or metal sheeting. Traditional examination techniques that are used during such inspections require extensive planning and manual inspection processes at a high cost, as well as causing significant disruption to tenants. This research paper aims to investigate novel methodologies for automated classification of defects detected via X-ray backscatter (XBS) inspections of tenanted arch structures. An image data processing pipeline and artificial intelligence (AI) models have been deployed to automatically classify common defects observed within the brickwork, including spalling and mortar degradation, without the need to remove the cladding or undergo extensive manual image data analysis. Four supervised classification models and three deep learning models were compared, with the aim to down select one for future development. The AI-driven approach significantly reduces the manual analysis time of XBS data. This approach is key to enable more efficient data-driven maintenance and decision-making on the structural condition of the tenanted arches, to ensure the safety of these key infrastructure assets at a reduced operational cost.
