Automatic real-time detection of structural damages on site is a
required technology to enable rapid, accurate, and on-site inspection.
This paper introduces an automated intelligent inspection system capable
of detecting structural problems, such as cracks, in real-time at the
edge of power plant components. Since no available dataset was suitable
for this case study, a real dataset was created by combining new and
existing. For inspection, this project customized a Deep Neural Network
(DNN) model to fit our application, including its quantization to enable
deployment at the edge. Real-time, on-site results from aerial and
hand-held setup images of the stack of an old power plant show that the
system is capable of identifying and localizing cracks within the field
of view (FOV) of the camera with a mean average precision (mAP) of
98.44% and ∼ at 2.5 frames per second (FPS) with real-time inference
for crack detection at the edge.