In brief

Machine learning has been used to automatically detect structural defects on 3D printed parts.

© Xiling Yao

Finding flaws fast

26 Jul 2021

A machine learning method that finds defects or dimensional deviation on 3D-printed surfaces ‘on-the-fly’ is paving the way for smart, fully automated systems.

As the new kid on the block, 3D printing is challenging traditional ‘subtractive’ manufacturing methods which remove, rather than add, material to produce an object. Although versatile and effective for building complex structures, 3D printing methods can sometimes lack robustness and repeatability, especially when used to make large metal parts.

Various aspects of the 3D printing process—like thermal stress, localized overheating and inconsistent machine speed—make it prone to producing surface defects or dimensional deviation, which, if not detected early, can be detrimental to the quality of the final piece.

“Vision-based surface topography measurement methods use cameras and computer vision algorithms to digitally reconstruct the 3D surface,” explained Xiling Yao, a Research Scientist at A*STAR’s Singapore Institute of Manufacturing Technology (SIMTech). “The surface reconstruction process is usually computationally heavy and limited in terms of precision and resolution.”

Now, a research team including Yao, led by Guijun Bi, a Senior Scientist and Manager of the Joining & Machining Group at SIMTech, has developed a method that can detect defects ‘on-the-fly’ thanks to in-house designed software that simultaneously executes multiple point cloud processing functions and integrates the data into machine learning models.

In the new system, a rapid and accurate sensor scans the surface of a printed object to obtain height data. Wasting no time, the multimodal software automatically processes this 3D spatial data into simple statistics for use by machine learning models, which then isolate and identify potential surface defects.

The researchers developed their rapid defect identification method for a robot-based laser-aided additive manufacturing (LAAM) system, a unique metal 3D printing process. They trained classifier algorithms to identify three main classes of surface non-conformance—bulge, dent and wavy defects—as well as defect-free surfaces using 73 LAAM-printed samples of varying size and shape. Compared to a state-of-the-art surface monitoring technique, the new method rapidly detected and classified defects even when there were multiple defect regions present in the same layer.

© A*STAR Research

Despite being developed for LAAM, Yao said the technology can be used to monitor the surface conditions of other printed materials. Simply by developing a new set of training data, the framework can also easily be adapted for other processes.

Our ultimate goal is to make LAAM technology smart and fully automatic, with self-learning and self-rectification capability, hence enhancing the printing quality and maximizing its productivity with minimal human intervention,” Yao noted. “The next step, which we are working on now, is to develop a way for the machine to perform intelligent decision making and adaptive process planning and execution after errors are detected.”

The A*STAR-affiliated researchers contributing to this research are from the Singapore Institute of Manufacturing Technology (SIMTech).

Want to stay up to date with breakthroughs from A*STAR? Follow us on Twitter and LinkedIn!


Chen, L., Yao, X., Xu, P., Moon, SK., Bi, G. Rapid surface defect identification for additive manufacturing with in-situ point cloud processing and machine learning. Virtual and Physical Prototyping 16, 50–67 (2021) | article

About the Researcher

Xiling Yao is a Research Scientist at A*STAR’s Singapore Institute of Manufacturing Technology, Joining & Machining Group. His research interests include laser-aided additive manufacturing, hybrid manufacturing and computational method for manufacturing digitalization and intelligence.

This article was made for A*STAR Research by Wildtype Media Group