Purdue University researchers are developing an artificial intelligence system to detect cracks captured in videos of a nuclear reactor.
The researchers want to create a deep-learning framework called a naive Bayes-convolutional neural network to analyze individual video frames for crack detection.
The researchers developed a data fusion scheme that aggregates the information extracted from each video frame to enhance the overall performance and robustness of the system.
Purdue professor Mohammad R. Jahanshahi says the new system detects cracks in overlapping "patches" in each frame while the data fusion algorithm scheme can track the crack from one frame to the next, resulting in a 98.3% success rate that is markedly higher than other state-of-the-art approaches.
The researchers used graphical processing units to train the neural network in how to detect cracks with a dataset consisting of about 300,000 crack and non-crack patches.
From Purdue University News
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA