Digital Twins Reveal Hidden Information

It can be difficult to determine what happens inside a composite product during the manufacturing process – where anomalies may have occurred or how closely the final product matches the original design. Digital twin technology can provide those interior views.

A digital twin is a virtual, data-based representation of the properties and behaviors of a physical system or process throughout its operational life. Using a digital twin, also known as a virtual twin, a manufacturer can optimize the performance of a composite product, reduce production errors and maintain a record of any changes made to the product.

Tracing Causes of Failure

Aligned Vision has been creating digital twins for aerospace customers for several years. The first step in that direction was the addition of cameras to its laser projection system, which resulted in an inspection system that captured, during manufacturing, calibrated images of the plies in layers of the composite component. These images, analyzed by machine vision and early machine learning models, enabled a manufacturer to determine if the fiber orientation in the actual composite component matched the orientation in the original CAD design. That real-time information allowed manufacturers to resolve some quality problems during production, reducing the number of parts that were not built to spec.

Now AI has made analyzing those images even more powerful and easier to do. “We can take these small, highly magnified, calibrated images and run them through deep learning classifiers to look for edges of the plies, fiber orientation and foreign objects and debris (FODs),” says Scott Blake, president of Aligned Vision.

The data from those images, along with other attributes like the part’s dimension and shape, are incorporated into an as-built digital twin to the originally designed component. Multiple component images are stitched together to create a digital twin of an entire part, such as a wind turbine blade or a helicopter rotor. By relying on this information rather than on a theoretical or design model, manufacturers gain a more accurate representation of what’s going on inside the part.

If the composite part fails while in use, the manufacturer can refer to its digital twin to try to determine the cause. A digital twin can also serve as a record of repair work that’s been done when a composite part is damaged, providing valuable information for technicians if the part later requires more work.

Digital twins could reduce part overdesign as well. By analyzing data from multiple digital twins of a particular component, a manufacturer could determine how much variation typically occurs in the manufacturing process. That could enable them to reduce the design allowables for the part while still meeting the design requirements.

“If you know how precisely you’re building what you’ve designed, you can more tightly optimize that design. This will lower the weight of the part and the energy used to produce it,” says Blake. “This is what industry 4.0 is all about, having this real-world data that can be compared with the nominal design data so you can make smart decisions.”

Increased Manufacturing Reliability

The complex manufacturing processes used to produce many composite parts today often require operators to closely monitor running equipment. Such close surveillance might not be necessary if a digital twin can trigger an alarm when there is a problem.

“The digital twin could help by putting the operator’s attention on a part when it’s needed, rather than having operators sit there all the time,” says Kris Villez, senior R&D staff at Oak Ridge National Laboratory (ORNL).

Some of Villez’s work at ORNL involves the use of statistical models to determine how quickly and accurately a digital twin system that uses a camera-based anomaly detection system can find errors during the additive manufacturing of a part. Using cameras to record data, researchers printed two 80-layer hexagonal patterns, with the first pattern printed correctly so it could serve as the normal image. In the second pattern, researchers varied the extruder speed in different trials, intentionally producing anomalies in the printed hexagon. Using machine learning to analyze and compare the normal statistical model with digital twins created at slower running speeds, Villez’s team detected large faults – anomalies representing a 30% reduction in speed – every time.

“Our results show that images are easily identified as anomalous for extruder speeds at or below 85% of the nominal speed, meaning that an anomalous reduction of the material deposition rate can be detected within seconds of its onset,” Villez says.

An alarm system tied to the digital twin could reduce manufacturers’ labor costs and/or allow them to assign operators to more important tasks, such as analyzing long-term reliability or data sets from multiple prints.

In another project, Villez and his team explored how similar the digital twin must be to the original normal image to trigger an alarm. The question becomes how much deviation a manufacturer can tolerate. There’s a tradeoff between having too many false alarms, when the operator is constantly being warned that something is wrong when it’s not, and missing any true anomaly or true fault. If the system generates too many false alarms, operators would be more likely to switch it off.

The researchers demonstrated that under ideal conditions it is possible to automatically set the right threshold for triggering alarms using statistical process control and principal component analysis (PCA) to analyze the data of the normal image versus its digital twin.

“One of the drivers of my research is to put the AI and statistical tools into the hands of people who are not necessarily experts,” Villez says. “If the threshold can be set automatically, then we don’t need a model expert or an AI expert to do this work.”

Exploring Production Options

Digital twin technology could assist the composites industry in meeting the large demand for parts for electric vertical takeoff and landing (eVTOL) vehicles and more. Autoclaving, the current method of producing those parts, is too slow; digital twins could speed the adoption of alternative production methods.

Boeing is developing a stamp forming production process to produce airframe structural parts in collaboration with Purdue University’s Composites Manufacturing & Simulation Center (CMSC) and Joby Aviation. In stamp forming, composite materials change shape in different directions as they are heated, shaped and cooled. To study these effects, CMSC developed a digital twin of a generic shape with a complex geometry of double curvature using FORM3D, which replicates a composite stamping process and analyzes the virtual product. Joby and the CMSC are working to validate the virtual twin to optimize this manufacturing method.

“We modeled all of the physics-based processes involved, the heat flow of the material, how it shrinks and all of the physics around how it changes shape and why,” says R. Byron Pipes, CMSC executive director. Determining how all these elements interact using empirical methods would take an enormous amount of time to conduct the required studies, but in a virtual twin everything can be integrated into one model, which makes it possible to understand and control the process.

With the virtual twin, researchers observed an unexpected change in form in the shape. The small radius of curvature decreased upon cooling, while the large radius of curvature increased. This unexpected behavior guided the redesign, which was only possible with the virtual twin.

“The virtual twin allows you to look at all of the design possibilities and evaluate them against each other,” says Pipes. Instead of running expensive and time-consuming real-world experiments changing one variable at a time, researchers could use the virtual twin to determine the guide design.

In another project, CMSC and Boeing used digital twin technology to develop a method of using composite prepreg molding to produce complex geometry components previously made by machining aluminum or titanium. When the discontinuous fiber prepreg platelet material was heated, it flowed into the geometries of the mold, but the fiber orientation changed during this process and impacted the properties of the material. With the digital twin that CMSC developed, researchers were able to simulate the flow of the platelet and predict the final fiber orientation.

“This met one of the important challenges of prepreg molded composites in aircraft structure, namely the ability of Boeing engineers to predict how part shape and strength were related,” says Pipes.

The Future of Digital Twins

Digital twin technology would be advantageous for composites manufacturers in industries like aerospace, but Blake sees several hurdles to overcome. He says that realization of digital twins’ benefits will likely be slowed by lack of cooperation between design, quality and manufacturing departments, and a lack of understanding of the technology’s benefits in C-suites.

While digital twin technology is too expensive for most companies to use today – it requires specialized computer expertise – Pipes envisions a time when manufacturers could access it via software providers that use AI. Manufacturers could run massive amounts of data through this software to build data that supports AI systems.

“The ultimate goal for a virtual twin in manufacturing is that it’s so good and runs so fast that it can tell the machine that’s doing the manufacturing what to do and how to adjust to any changes it sees in the process in real time,” Pipes says.

Mary Lou Jay is a freelance writer based in Timonium, Md. Email comments to mljay@comcast.net.

The post Digital Twins Reveal Hidden Information appeared first on American Composites Manufacturers Association.