Manufacturing Perfection: How to Make Smart Machines Smarter

With each line of code and crunch of data, University of Virginia graduate engineering students Oliver Holzmond and Jamison Bartlett get a step closer to revolutionizing the way things are made.

Across several sprawling labs in the UVA School of Engineering’s departments of Mechanical and Aerospace Engineering and Materials Science and Engineering, Holzmond, Bartlett and their Ph.D. adviser, Professor Xiaodong “Chris” Li, are fine-tuning sensors and computer code that will analyze each layer of a 3-D-printed object, spot flaws, decide how to fix the flaws and direct the 3-D printer to make corrections.

Related Story

“Manufacturing is experiencing an evolutionary moment with the sophistication of 3-D printers,” said Li, Rolls-Royce Commonwealth Professor of mechanical and aerospace engineering. “Yet as good as they are, they inherently create products with flaws. As such, they can’t be relied on to create parts for critical devices.

“We are designing smart machines that can create flawless products.”

For about four years, Holzmond, Bartlett and Li have been working on their system to help 3-D printers make perfect products with fewer resources and virtually no waste, all in a fraction of the time it takes to make the same products today. In addition, the system will be able to learn from its mistakes, be resilient to hacking and train other machines.

“Humans use all of their senses to inform their decisions. We are bringing that same perspective to machines,” Li said. “We have created optical sensors, or eyes, capable of seeing down to the micron level, and infrared and auditory sensors that assess temperatures and can hear reverberations. And then, of course, we have created a ‘brain,’ a computer code that can process all this information in real time to make immediate decisions about how to address a problem.”

It is no overstatement to say that these are major breakthroughs in machine learning and advanced manufacturing. The technological advances Li and his students are making have the potential to pay huge dividends in increased efficiency and reduced waste in the $12 trillion manufacturing industry, and ultimately benefit society in almost every way.

“We have strengths in additive manufacturing, machine learning and artificial intelligence, and decades of experience in sensor technology here at UVA Engineering, and we are leveraging that knowledge in a way that will transform the manufacturing industry,” Li said.

Engineering Ph.D. students Oliver Holzmond and Jamison Bartlett calibrate a sensor that will use video and their computer code to analyze 3-D printed material for imperfections. (Photo by Christopher Tyree)
Engineering Ph.D. students Oliver Holzmond and Jamison Bartlett calibrate a sensor that will use video and their computer code to analyze 3-D printed material for imperfections. (Photo by Christopher Tyree)

Additive manufacturing, in which 3-D printers layer material together to make an object – as opposed to traditional manufacturing that involves cutting away or casting materials – has the potential to replace the way in which many goods are produced, and many leading corporations are betting on these devices for the future.

“Additive manufacturing offers the opportunity to make more complicated parts, parts with internal geometric features that you could not make with any other process,” Bartlett said. “You can make a part lighter, you can make it stronger, you can make it with less material – so lower costs, better performance.”

Manufacturing watchdog SmarTech Publishing released a report at the end of 2018 that showed the additive manufacturing market has more than doubled since 2014, to $9.3 billion in 2019; and SmarTech predicts the industry will top $41 billion by 2027.

To date, Li’s lab has received $400,000 in funding from the U.S. Department of Energy and the Commonwealth Center for Advanced Manufacturing, a consortium of industry, government and universities of which UVA is a founding member. Li and his team are looking for industry partnerships to test and refine their devices.

Ph.D. student Jamison Bartlett calibrates a sensor that will analyze 3-D printed materials for imperfections in Professor Xiaodong “Chris” Li’s lab. (Photo by Christopher Tyree)
Ph.D. student Jamison Bartlett calibrates a sensor that will analyze 3-D printed materials for imperfections in Professor Xiaodong “Chris” Li’s lab. (Photo by Christopher Tyree)

Corporations use additive machines to print with just about any type of material, from plastic to metal and even human tissue. The main drawback to additive printing as a manufacturing process is that the current technology produces defects due to variables like quality of the starting material or intense internal stresses resulting from the melting and solidifying of material. Every layer laid down has the potential for defects, and those problems can be magnified and multiplied as other layers are built on top.

For some goods, that might not be a huge issue; if, say, there is a millimeter-sized void in a 3-D-printed plastic gasket that isn’t critical for a machine’s normal functioning, the defect might not matter. But if you are talking about 3-D-printing aircraft turbine blades, engine gears, or even a bicycle wheel, a small flaw can have catastrophic consequences.

“If you get down to it, pretty much every part created by a 3-D printer has defects. I don’t think I’ve ever seen one you can consider 100% defect-free,” Holzmond said.

“They always have defects, even with what they call optimal, carefully studied parameter sets,” Bartlett agreed. “Often, they aren’t small problems, either, which can be a source for fatigue in that part. Certainly the aerospace industry wouldn’t want to use a defective part.”

Daily Report
Delivered Right to Your Inbox
Sign up for the Daily Report, your free, comprehensive source for all UVA News.

Today, manufacturers must make costly and time-consuming inspections of the pieces they create with 3-D printers. Some data suggest that nearly 20% to 30% of all additive manufactured parts are so defective they must be scrapped.

To combat the problems, Bartlett’s and Holzmond’s sensors use a pair of cameras to make 3-D digital images that reveal the full topography of a printed layer and focus down to the micron level in order to detect anomalies. Part of their study includes thermal mapping with an infrared sensor to monitor 3-D-printed metals. They also work with auditory sensors, which alert the machine to unexpected sounds that could signal a defect.

“All of these sensors are hard-wired to the 3-D printer’s computerized ‘brain,’” Li explained. “The brain takes all that data, analyzes it in real time, and then makes an informed decision on how to repair that layer before moving on to the next layer.” Repairing could mean the 3-D printer changes a variable, such as applying less or more heat, depending on what is needed.

As an added benefit, the sensor-equipped 3-D printers can be designed to work independently of a network, making them less vulnerable to hacking. “Essentially we’re trying to make an automated engineer that can just constantly watch the operations and sound off if it notices something out of the ordinary,” Holzmond said.

The aim is to eventually be able to produce a flawless product on the first try. Bartlett, Holzmond and Li believe they have at least a year of work left to do, especially on the computer code that will be used to make correct decisions for how to fix an issue, and then working with corporate partners to test the technology in 3-D printers used in industry.

“Our goal is to make smart machines smarter,” Li explained. “In this way we will save industry time and money, and society benefits by having precision instruments that last longer, cost less to produce and save precious natural resources.”