Open Access
Subscription or Fee Access
Vision-based Crack Detection on Metallic Surfaces using Deep Convolutional Neural Network with Patch Clustering
Abstract
In the U.S., there were eight nuclear power plant accidents cost more than 140 million US dollars in property damage. Regular inspection of the components of nuclear power plants is important to improve their resilience. However, current inspection practices are time consuming, tedious, and subjective: they involve operators manually locating cracks by watching videos. Very few of vision-based crack detection algorithms for nuclear power plant components has been developed. Prevalent crack detection algorithms for other applications cannot detect those cracks because they are typically very small and have low contrast. Moreover, the existence of scratches, welds, and grind marks on the surfaces leads to a large number of false positives when prevalent crack detection algorithms are used. The previous study applied local binary patterns (LBP) and support vector machine (SVM) to detect cracks in inspection videos and outperformed the other prevalent algorithms. This paper utilizes a deep convolutional neural network (CNN) to detect cracks in videos. The performance is evaluated using several inspection videos and it achieves more than 94% hit rate with only 0.5 false positive-per-frame which is much better than the previous study
DOI
10.12783/shm2017/14240
10.12783/shm2017/14240
Full Text:
PDFRefbacks
- There are currently no refbacks.