November/December AMP_Digital

1 0 A D V A N C E D M A T E R I A L S & P R O C E S S E S | N O V E M B E R / D E C E M B E R 2 0 1 8 METHOD INTERPRETS MATERIAL SPECTRA Modern spectroscopy methods en- able rapid generation of huge amounts of material spectra, but it must be correctly interpreted to gather useful information about the material being studied. Use of big data analysis is attracting attention in materials sci- ence and researchers at The University of Tokyo Institute of Industrial Science realized that such techniques could be used to interpret much larger numbers of spectra than traditional approaches. “We developed a data-driven approach based on machine learning techniques using a combination of the layer clus- tering and decision tree methods,” says researcher Teruyasu Mizoguchi. The team used theoretical calcu- lations to construct a spectral database TESTING | CHARACTERIZATION MACHINE LEARNING ANALYZES 3D PART BUILDS Researchers at Lawrence Liver- more National Laboratory (LLNL), Calif., are using machine learning to process data obtained during 3D part builds in real time, detecting within milliseconds whether a build will be of satisfactory quality. The team reports developing convolutional neural networks (CNNs), a type of algorithm primarily used to process images and videos, to predict whether a part will be good by looking at as little as 10 milliseconds of video. “This is a revolutionary way to look at the data that you can label video by video, or better yet, frame by frame,” says principal investigator Brian Giera. “The advantage is that you can collect video while you’re printing something and ultimately make conclusions as you’re printing it.” Researchers devel- oped the neural networks using about 2000 video clips of melted laser tracks under varying conditions, such as speed or power. They scanned the part surfaces with a tool that generated 3D height maps, using that information to train the algorithms to analyze sections of video frames. Team member Bodi Yuan devel- oped algorithms that automatically label the height maps of each build and employed the same model to predict the width of the build track, whether the track was broken, and the standard width deviation. Using the algorithms, the group is able to take video of in-progress builds and determine if the part exhibits acceptable quality. They report that the neural networks are able to detect whether a part would be con- tinuous with 93% accuracy. llnl.gov. Drawing from the spectrum, trees exchange information andmake the “interpretation” (apples) bloom. Courtesy of T. Mizoguchi/University of Tokyo. BRIEF Greene Tweed’s nondestructive test (NDT) lab, Kulpsville, Pa., recently completed Nadcap accreditation for NDT Testing Facility Digital Radiography of composite materials, using a digital detector array. The accreditation enables in-house NDT of production composite hardware for commercial aircraft. gtweed.com. Researchers developedmachine learning algorithms capable of detecting within milliseconds whether a 3D part will be satisfactory. Courtesy of LLNL.