Get our free email newsletter

Deep Learning Technology Used to Detect Cracks in Nuclear Reactors


Scientists at Purdue University are working on a system that will use artificial intelligence to identify cracks in nuclear reactors. This technology could be instrumental in reducing accidents and maintenance costs, as well as saving lives.

Current methods of inspecting nuclear reactors are problematic to say the least. Part of the reason for this is purely structural: reactors are submerged in water to keep them cool. This means that direct manual inspection is pretty much impossible for safety reasons. The solution so far has been for technicians to watch remotely recorded videos of the underwater reactors, and identify any cracks that way. Obviously, this method leaves a lot to be desired.

“Regular inspection of nuclear power plant components is important to guarantee safe operations. However, current practice is time-consuming, tedious, and subjective and involves human technicians reviewing inspection videos to identify cracks on reactors.”

- Partner Content -

A Dash of Maxwell’s: A Maxwell’s Equations Primer – Part One

Solving Maxwell’s Equations for real-life situations, like predicting the RF emissions from a cell tower, requires more mathematical horsepower than any individual mind can muster. These equations don’t give the scientist or engineer just insight, they are literally the answer to everything RF.
Mohammad R. Jahanshahi, assistant professor in Purdue’s Lyles School of Civil Engineering

Scientists believe the solution to this problem is a deep learning framework. Called a naive Bayes-convolutional neural network, this system would be able to analyze video footage for any cracks or irregularities in nuclear reactors. The information would be aggregated and extracted from the video frames by a ‘data fusion scheme.’

The system would be a convolutional neural network, otherwise known as a form of artificial intelligence called ‘deep learning.’ Such systems have been used in a variety of applications before, and scientists believe deep learning would be ideal for keeping tabs on nuclear reactors.

Cracks are detected on overlapping ‘patches’ over each video frame. From there, a data fusion algorithm scheme tracks the progression of the crack from frame to frame. This system has a 98.3% rate of success.

Nuclear power incidents are extremely expensive; from 1952 to 2010 there were 99 major nuclear incidents worldwide. The total bill for those incidents added up to more than $20 billion. But with 20% of the United State’s total electric energy coming from nuclear reactors, it doesn’t look as though our dependency on nuclear power is ending anytime soon.

This new system promises to catch issues before they become crises, keep people safe, and our nuclear reactors crack-free. Research is ongoing to streamline the technology and to determine if it would be viable in other applications.

Related Articles

Digital Sponsors

Become a Sponsor

Discover new products, review technical whitepapers, read the latest compliance news, trending engineering news, and weekly recall alerts.

Get our email updates

What's New

- From Our Sponsors -

Sign up for the In Compliance Email Newsletter

Discover new products, review technical whitepapers, read the latest compliance news, trending engineering news, and weekly recall alerts.