Computer Vision / Video Analytics

Using AI To Detect Damage In Nuclear Reactors

Designing materials that can withstand the force of nuclear power is pivotal to maintaining the integrity of nuclear reactors. However, performing manual inspections of materials is time-consuming, prone to error, inconsistent, and does not scale well. To solve the problem, researchers from the University of Wisconsin-Madison and the Oak Ridge National Laboratory in Tennessee developed a deep learning-based system that can detect and analyze microscopic radiation damage to materials with superb accuracy and at a fraction of the time.
“Irradiation damage in materials for nuclear applications greatly affects the durability of existing nuclear reactor facilities and advanced reactor designs. Understanding the effects of irradiation on materials properties and performance is critical to safe and reliable nuclear reactor operation,” the researchers stated in their paper.
Using a single NVIDIA GeForce GTX 1070 GPU with the cuDNN-accelerated MATLAB deep learning framework, the team trained a convolutional neural network, as well as a cascade object detector, on over 60,000 images. The neural network identified and classified around 86 percent of dislocation errors in the materials. For comparison, the humans found 80 percent of the defects.

Schematic flow chart of the proposed automated detection approach. Input micrographic images go through the pipeline of module I —Cascade Object Detector, module II—CNN Screening, and module III—Local Image Analysis. After module I, the loop locations and bounding boxes are identified and then further refined to remove false positives using module II. Then module III determines the loop shape and size.

“Machine learning has great potential to transform the current, human-involved approach of image analysis in microscopy,” says Wei Li, the lead researcher who earned his master’s degree in materials science and engineering this year from UW–Madison, and is now a software engineer at Google.
The work was published this week in the journal npj Computational Materials.
Read more >

Discuss (0)

Tags