Researchers explore the possibility of A.I to catch defected metal AM parts
The Lawrence Livermore National Laboratory are developing a suite of convolutional neural networks to analyze data from 3D builds in real-time.0
A group of scientists and researchers from the Lawrence Livermore National Laboratory in California are developing a suite of convolutional neural networks (CNNs)—a popular algorithm used for process image and video—to analyze data from 3D builds in real-time. The idea being, that these machine learning algorithms will be able to know within milliseconds whether or not a product will meet proper build quality.
The Laboratory findings were published online by Advanced Materials Technologies, and speaks to the expensive nature of post-production work on 3D printed objects. The research team developed the CNN’s by collecting over 2,000 video clips of melted laser tracks under various speed and power conditions. The surfaces of parts were scanned with a tool that generates 3D height maps, using that information to train the algorithms to analyze sections of video frames. The process would be too difficult and time-consuming for a human to do manually, says principal investigator and LLNL researcher Brian Giera.
“This is a revolutionary way to look at the data that you can label video by video, or better yet, frame by frame,” says Giera. “The advantage is that you can collect video while you’re printing something and ultimately make conclusions as you’re printing it. A lot of people can collect this data, but they don’t know what to do with it on the fly, and this work is a step in that direction.”
According to the paper’s lead author, Bodi Yuan used the same algorithm that label the height maps of each build, to also predict the width of the build track and whether or not the track was broken. Bodi, a student at the University of California, was a part of the research team taking video of in-progress builds to determine if the part exhibited acceptable quality. Researchers reported that the neural networks were able to detect whether a part would be continuous with 93 percent accuracy, making other strong predictions on part width.
The neural networks described in the paper could theoretically be used in other 3-D printing systems and other researchers should be able to follow the same formula, creating parts under different conditions, collecting video and scanning them with a height map to generate a labeled video set that could be used with standard machine-learning techniques.
Giera said work still needs to be done to detect voids within parts that can’t be predicted with height map scans but could be measured using ex situ X-ray radiography.
Researchers also will be looking to create algorithms to incorporate multiple sensing modalities besides image and video.
“Right now, any type of detection is considered a huge win. If we can fix it on the fly, that is the greater end goal,” Giera said. “Given the volumes of data we’re collecting that machine learning algorithms are designed to handle, machine learning is going to play a central role in creating parts right the first time.”