US DoE detects 3D printing defects in real time

According to the US Department of Energy (US DoE), researchers have used diagnostic tools and machine learning to develop a new method for detecting and predicting defects in 3D printed materials. The method uses various imaging and machine learning techniques to detect the generation of pores in real time, with near-perfect accuracy. The researchers will soon develop sensing technologies that can detect other types of defects that occur during the additive manufacturing process, with the goal of creating a system that not only detects defects but enables repairs during AM.

Many industries rely on metal additive manufacturing to rapidly build complex parts and components – everything from rocket engine nozzles, to pistons for high-performance cars, to custom-made orthopedic implants. New advanced diagnostic tools for detecting and potentially repairing defects will expand the use of AM in aerospace and other industries that rely on high-performance metal parts.

One of the major defects in laser powder bed fusion is the formation of keyhole pores. These pores or structural defects can compromise the performance of the printed parts. Many 3D printing machines have thermal imaging sensors that monitor what’s being built, but can still miss the formation of pores. The only way to directly detect pores inside dense metal is by X-ray imaging, using highly intense beams such as those at the Advanced Photon Source, a DoE Office of Science user facility.

Here, researchers correlated the X-ray images of the sample interior and the thermal images of the melt pool and discovered that the formation of a keyhole pore creates a distinct signal at the material’s surface that can be detected by thermal cameras. First, they trained a machine learning model with X-ray images to predict the formation of pores using only thermal images. Then, they tested the model’s ability to decipher the complex thermal signals and predict pore generation in unlabeled samples. The researchers found that the approach could detect the exact moment when a pore formed during the printing process on timescales of less than a millisecond.