Deep learning for masonry lined tunnel condition assessment

The condition assessment of masonry lined railway tunnels typically involves manually identifying lining defects from photographic and lidar surveys taken of the tunnel intrados. This process is time-consuming and subjective to the assessor’s judgement. However, recent developments in machine learning achieve the quality metrics required to automate the detection of defects from noisy and irregular tunnel data, offering the potential to reduce tunnel assessment and maintenance costs. This paper proposes a deep learning workflow for defect segmentation. The method is evaluated on the task of masonry block segmentation from lidar data. Acceptable performance is achieved on a sample tunnel section, suggesting that similar methods are applicable to other masonry lined tunnel defect segmentation tasks.