A new approach for identifying defects in additively manufactured components has been developed by a team of researchers at the University of Illinois Urbana-Champaign.

While typically difficult to determine if an additively manufactured product is free of defects due to the components having complex three-dimensional shapes and internal features that are not easily observed, the new approach was devised to make defect identification easier using deep machine learning.

Longitudinal (top) and axial (middle) images of X-Ray CT data of parts with 6 internal defects: a spherical clog, a stellated shaped clog, a cone shaped void, a blob shaped void, an elliptical warp of the inner channel, and a nonconcentric center nozzle. Source: The Grainger College of Engineering at the University of Illinois Urbana-ChampaignLongitudinal (top) and axial (middle) images of X-Ray CT data of parts with 6 internal defects: a spherical clog, a stellated shaped clog, a cone shaped void, a blob shaped void, an elliptical warp of the inner channel, and a nonconcentric center nozzle. Source: The Grainger College of Engineering at the University of Illinois Urbana-Champaign

To accomplish this, the researchers built their model using computer simulations to produce tens of thousands of synthetic defects that exist only in the computer.

According to the researchers, each of these computer-generated defects featured a different size, shape and location, thereby enabling the deep learning model to train on a range of potential defects. This also enabled the model to recognize components that are defective versus those that are not.

The team also tested the algorithm on physical parts featuring an assortment of both defective and defect free parts. During these tests, the algorithm reportedly identified hundreds of defects in those real parts that had not been previously detected by the computer.

An article detailing the approach, "Detecting and classifying hidden defects in additively manufactured parts using deep learning and X-ray computed tomography," appears in the Journal of Intelligent Manufacturing.

To contact the author of this article, email mdonlon@globalspec.com