Artificial intelligence accelerates the process design of 3D printing of metal alloys

In order to successfully 3D print metal parts to meet the strict specifications required by many industries, it is necessary to optimize process parameters, including printing speed, laser power, and layer thickness of deposited materials.

However, in order to develop additive manufacturing process diagrams that ensure these optimal results, researchers have to rely on traditional methods, such as using off-site material characterization to test laboratory experiments on parts printed with various parameters. Testing so many parameter combinations to develop the best process may be time-consuming and expensive, especially considering the various metals and alloys available for additive manufacturing.

David Guirguis, Jack Beuth, and Conrad Tucker from the Department of Mechanical Engineering at Carnegie Mellon University have developed a system that utilizes ultra high speed in situ imaging and visual transformers. This system not only optimizes these process parameters, but also has scalability, making it applicable to various metal alloys.

Their research findings are published in the journal Nature Communications.
Visual converter is a form of machine learning that applies neural network architectures originally developed for natural language processing tasks to computer vision tasks, such as image classification. The video visual converter goes further by using video sequences instead of still images to capture spatial and temporal relationships, enabling the model to learn complex patterns and dependencies in video data.

The self attention mechanism allows natural language processing models to balance the importance of different words in a sequence, and allows models created by Guirguis to balance the importance of different parts of the input sequence to predict the occurrence of defects.

"We need to automate this process, but it cannot be achieved solely through computer programming," explained Guirguis, a postdoctoral researcher in mechanical engineering. To capture these patterns, we need to apply machine learning.

"We are pleased to have developed an artificial intelligence method that utilizes the temporal characteristics of additive manufacturing imaging data to detect different types of defects. This demonstrates the groundbreaking generalizability of AI methods using different AM metals and reveals that the same trained AI model can be used without the need for expensive retraining with additional data," commented Professor Tucker of Mechanical Engineering.

Guirguis said he is fortunate to have received such powerful machine learning training at Carnegie Mellon University because mechanical engineers know how to apply experimental and computational solutions to the problems they solve, which is more important than ever before.

In this case, Guirguis attempts to overcome the main limitations of in-situ imaging in laser powder bed melt additive manufacturing processes. This technology uses high-power laser as an energy source to melt and melt powder at specific locations to form certain shapes, then a new layer of powder is spread out by a recoating machine, and the process is repeated until a 3D object is formed.

However, during the printing process, the molten metal seen by the camera is saturated, so its physical characteristics cannot be seen, which can identify defects that may reduce mechanical performance and fatigue life of printed parts.

Source: Laser Net