Researchers Use Machine Learning to Detect Defects in Additive Manufacturing

Researchers at the University of Illinois at Urbana-Champaign have developed a new method to detect defects in additively manufactured parts.

One of the most important tasks in any factory is determining whether manufactured parts are free of defects. Finding defects can be especially difficult with additive manufacturing (3D printing) because it produces parts with complex three-dimensional shapes and critical internal features that are not easily observable.

This new technique uses deep learning to make it much easier to identify defects in additively manufactured parts. To build the model, the researchers used computer simulations to generate tens of thousands of artificial defects that exist only in the computer.

Each computer-generated defect varies in size, shape and location, allowing the deep learning model to train on a range of possible defects and recognize the difference between defective and non-defective components.

The algorithm was then tested on physical parts, both defective and non-defective, and was able to successfully identify hundreds of defects in the real physical parts that had not previously been detected by the deep learning model.

“This technology addresses one of the toughest challenges in additive manufacturing,” said William King, professor in the University of Illinois' Department of Mechanical Science and Engineering and project leader. “Using computer simulation, we can very quickly build machine learning models that identify defects with high accuracy. Deep learning allows us to accurately detect defects that were previously undetectable by computer.”

This study Intelligent Manufacturing Journal The paper, titled “Detection and Classification of Hidden Defects in Additively Manufactured Parts Using Deep Learning and X-ray Computed Tomography,” used X-ray computed tomography to inspect the inside of 3D parts that have invisible internal features and defects. While 3D parts can be easily produced with additive manufacturing, inspection is difficult when important features are not visible.

The authors are Miles Bimrose, Sameh Tawfick and William King of the University of Illinois at Urbana-Champaign, Davis McGregor of the University of Maryland, Chenhui Shao of the University of Michigan, Tianxiang Hu, Jiongxin Wang and Zuozhu Liu of Zhejiang University.