Applying deep convolutional neural network with 3D reality mesh model for water tank crack detection and evaluation

ABSTRACT

Water tanks have been built for decades and likely deteriorated over time. To ensure its integrity, Unmanned Aerial Vehicle (UAV) e.g. drones with cameras are used for inspecting the elevated tanks and the images are collected for detecting defects such as cracks. To automatically detect and segment the defect in images, Mask Regional Convolutional Neural Network (Mask-RCNN) the latest convolution neural network model is trained with the real-world infrastructure inspection images. The trained model has been applied to crack detection and segmentation for a water tower of 50 meters height and storage of 500 m³. The images are used to construct a 3D mesh model by photogrammetry technology. The 3D model with annotated cracks enables intuitive visualization and quantitative assessment of 1704 detected cracks, which are evaluated in range of 2 mm to 10 mm wide with total crack length of 77.6 m and total crack area of 0.28 m².

KEYWORDS:

• Tank inspection
• deep learning
• crack detection
• crack segmentation
• convolutional neural network
• 3D reality mesh model

Data availability

Some or all data, models, or code generated or used during the study are proprietary or confidential in nature and may only be provided with restrictions.

Disclosure statement

No potential conflict of interest was reported by the author(s).