Researchers have developed a deep learning alternative to laser powder bed fusion monitoring

Many problems can arise during additive manufacturing (AM) metal fabrication, and without on-site process monitoring, defects can be detected and identified after the product is manufactured. Typically, manufacturers will use high-speed cameras to keep a close eye on the geometry of the melt pool and its changes during the short duration of the laser powder bed fusion (LPBF) process.

It requires expensive equipment, large amounts of memory storage (ie, 200,000 to 30,000 high-resolution images are stored per second), and laborious labor to collect and sort the data. These ultimately increase the cost of online visual tracking and process analysis.

To enable automated, cost-effective in-situ visual monitoring of metal additive manufacturing processes, researchers at Carnegie Mellon University’s School of Engineering have developed a deep learning method that provides a method to capture and characterize only airborne acoustics or thermally molten pools. In LBPF. Discharge.

The team’s approach was recently published Additive Manufacturing MagazineEnables manufacturers to obtain basic melt pool geometry and predict transient melt pool changes almost instantaneously.

“Using the underlying physical principles and data-driven advantages of multimodal process signals, AI enables our pipeline engineers to reconstruct critical melt pool characteristics using highly affordable and easy-to-use sensors, such as microphones or photodiodes,” said Haolin Liu, doctoral candidate in mechanical engineering.

An obvious advantage of this new method is its ability to detect spatially correlated lack of fusion (LOF) defects in LPBF. As one of the most common process anomalies, LOF occurs when there is insufficient melt pool overlap when the laser passes through the powder layer.

The resulting molten powder leaves large unalloyed gaps and residual porous parts that can severely compromise the durability and other mechanical properties of the final product. Therefore, capturing these local defects as well as changing the melt pool in real time is essential to produce durable products.

The team conducted a series of LPBF experiments to explore various printing parameters of the titanium alloy Ti-6Al-4V (Ti-64). Airborne acoustic, thermal and high-speed imaging data are collected for each respective process condition and synchronized from the pre-designed as-built structure to successfully reconstruct the precise melt pool geometry. The team even tracked the oscillating behavior of the molten pool within milliseconds. The method also shows good ability to effectively detect local LOF defects between two adjacent laser scanning lines.

“This method allows monitoring of the melt pool using low-cost sensors that can be installed on any laser powder bed additive manufacturing machine. Creating synthetic videos of high-speed melt pools based on acoustic and photodiode sensor data is unique in the field of additive manufacturing. “said Jack Beuth, professor of mechanical engineering and co-director of the Next Manufacturing Center.

Additionally, the team’s research takes an important step toward better understanding the physical relationships between multimodal in situ process signals.

“The interrelationships between these signals have yet to be fully explored by the scientific community,” Liu said.

“While our research focuses on deep sensing data-driven pipelines, we found some fundamental connections between acoustic signatures, thermal emissions and melt pool morphology, whose physics and dynamics require further scientific exploration and experimental investigation. “

“Although many experts are aware of the correlation between acoustic emissions, thermal emissions and melt pool dynamics resulting from laser printing,” said Levant Burak Kara, professor of mechanical engineering, the precise relationship is largely unknown.

“In this work, we develop and demonstrate a data-driven predictive model that links these three phenomena in a very accurate and physically meaningful way.”

According to Anthony Rowlett, professor of materials science engineering and co-director of the Next Manufacturing Center, acoustic behavior requires fundamental physical interactions between lasers and materials.

“To our surprise, it revealed more than we expected, and it proved to be very useful for reporting process-related quantities that could affect production quality.”

Going forward, the team plans to explore more real-time monitoring applications driven by acoustic and thermal emission data from materials other than Ti-64, as well as across different platforms and additive manufacturing processes.

“By providing a deeper interpretation of acoustic and thermal emission potentials, we hope to better understand their relationship to pool changes, keyhole oscillations, and other spatially relevant process features,” said Liu.

“One day we may be able to build fully functional digital twins of advanced surrogate models and other process characterization equipment such as synchrotron x-ray machines and entire additive manufacturing processes!”