Inspection of steel components at nuclear power plants is essential because radiation exposure can render metals more brittle and causes them to swell.

Crack detection is typically performed by power station staff examining video footage from inspection cameras.

However, current inspection practices are time consuming and subjective because they involve an operator manually locating cracks in metallic surfaces.

Automatic crack detection algorithms under development often do not detect cracks in metallic surfaces because the cracks are usually small, have low contrast, and are difficult to distinguish from welds, scratches, and grind marks.

New technology from Purdue University of West Lafayette, IN, uses an advanced algorithm and a powerful machine learning technique to detect cracks based on the changing texture surrounding cracks on steel surfaces. Unlike systems designed for processing single images, the crack recognition and quantification (CRAQ) system provides more robust results by processing multiple video frames.

The researchers apply Bayesian decision theory to information extracted from multiple frames to determine the probability that an object is a crack and to assign it a confidence level. If the algorithm assigns a high confidence level to a crack, the box outline is red. The processing procedure takes about a minute (see video).