Moog and University at Buffalo Developing Artificial Intelligence for Image Recognition in Metal 3D Printing
3D printing is used increasingly in projects that require some type of recognition – for instance, CENIT’s 3D software tool can analyze a topologically optimized component and allocate it to the correct bionic component in a CAD catalog, and a cyber security firm fooled the iPhone X facial recognition system with a 3D printed mask. Now, motion control technology provider Moog is teaming up with the University at Buffalo, part of the SUNY System and no stranger to 3D printing, to develop artificial intelligence (AI) for image recognition in metal 3D printing.
Metal 3D printing is undeniably on the rise these days, and is frequently among the major themes at industry events. The advanced manufacturing technology makes it possible to create metal parts with complex geometries at a faster rate, but it’s not foolproof, which is why we still often hear about ongoing research and development efforts centered around advancing and improving the technology.
Process development and control are important components in the process of 3D printing critical metal parts, and even when industrial statistical techniques, such as the Design of Experiments (DoE), are applied, the amount of experimental work to be done is vast, requiring lots of time in the laboratory inspecting sample parts under a microscope, or through X-ray CT scans. And as we know, human vision is not always 100% accurate.
While Moog has been spending time making process improvements in order to reduce this tiring experimental workload, Professor Rahul Rai, with the University at Buffalo’s Department of Mechanical and Aerospace Engineering, has been using AI to master image recognition in metal 3D printing.
The collaborative team of Professor Rai and engineers from Moog received funding from the UB New York State Center of Excellence in Materials Informatics (CMI) for their experimental work, and successfully applied convolutional neural networks to parts 3D printed out of metal. This resulted in a trained computer algorithm which is able to not only recognize metal 3D printed parts with high quality, but also reject ones with lower quality; you can see an illustration of this in the diagram below.
This diagram was reconstructed out of nearly 150 sub images, which were evaluated and colored individually by the new algorithm. Of 144 sub images, 136 were inspected and deemed “undermelt” by the AI algorithm, while six were classified as “overmelt.”
These terms – undermelt and overmelt – are non-optimal conditions that are not often used in the manufacturing of production parts. However, the undermelt in these cases was expected, as the process parameters which were employed to create the 136 images did not result in enough energy.
As a result of their collaboration, UB and Moog now have a computerized inspection tool that’s been trained to recognize high-quality metal 3D printed parts. Employing this tool will allow Moog to continue increasing the quality of metal 3D printed by running more experiments and closely inspecting the results.
Engineers at Moog are now making efforts to use the technology, which could have a major impact in the robotics field, in other areas, such as training the navigation system of an autonomous vehicle to recognize trees as obstacles from digital photographs, then employ that recognition in order to plan a smooth course around them.
Discuss this and other 3D printing topics at 3DPrintBoard.com or share your thoughts in the Facebook comments below.[Source/Images: Moog]
You May Also Like
Make Your Own 3D Print Design with Expert Guidance
3DPrint.com’s Design Essentials for 3D Printing online crash-course starts October 23 with two live presentations from industry leaders. Over the course of the one-week, learn how to create 3D print-ready...
3DEXPERIENCE Modeling & Simulation Conference: Finn Mast Project & Guest Speaker Dr. Byron Pipes
Recently, Dassault Systèmes, the 3DEXPERIENCE software company, held its 3DEXPERIENCE Modeling & Simulation Conference. I was lucky enough to be invited to attend the event in Novi, Michigan, and the first...
Coding for 3D Part 3: Tools of Choice
This is an in-depth article explaining the plan of attack for learning parametric design through code. I will be consulting various online resources as well as books to strengthen my knowledge base.
3DEXPERIENCE Modeling & Simulation Conference: Keynote by Skunk Works & Executive Roundtable
Last week, Dassault Systèmes, the 3DEXPERIENCE software company, held the 3DEXPERIENCE Modeling & Simulation Conference in Michigan. I was invited to attend the event, which was held from September 18-19...
View our broad assortment of in house and third party products.