Abstract
Piping is a part of industrial structures that acts like human blood vessels. Since pipe leakages are a threat to the integrity of a structure, it is one of the major monitoring targets. If inspectors are unaware of leakages, access to pipes for inspection can cause serious injury to the human body. Therefore, it is necessary to operate a monitoring system that detects pipe leakage regions for the safety of not only facilities but also inspectors. In this study, a multi-kernel neural network was introduced to visualize the pipe leakage regions through deep learning of the characteristics of pixel- wise color variation in normal and leakage regions from camera footage. Furthermore, we present the results of properly adjusting the visualization properties through an analysis of precision and recall according to the threshold for leakage judgment based on the output of deep learning. The results show that leakage areas can be visualized in accordance with the leakage diagnosis environment and purpose by adjusting the threshold.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call