What’s So New About Digital Twinning?

Digital twins are improving predictive maintenance capabilities, but more data inputs are needed to make the process more robust.

Printed headline: Digital Twins’ Evolution

When the Internet of Things became trendy several years ago, one way to grasp it was to realize it was basically what aircraft engine makers had been doing long before the internet came along—using sensor data from machines to detect possible problems—only doing all this better, faster, cheaper and much more broadly.

The same might be said of digital twinning, the name for making virtual models of aircraft assets that can support better decisions about asset production, operation, repair and eventual retirement. We are already well down this road, having learned important lessons. We have new tools to do it better and more broadly. And we are starting to see huge potential benefits in applying the twinning approach.

The future of using digital models of physical assets is expanding and changing across several dimensions.


1. Breadth Of Applications

Modeling and monitoring aircraft systems started with propulsion engines and then extended to their cousins, auxiliary power units. Now it is being extended to broader types of components, mechanical, electro-mechanical, hydraulic, pneumatics and possibly even structural and avionic parts.

GE and Infosys have developed a digital twin of landing gear as part of an integrated vehicle health management initiative, notes Infosys manufacturing and marketing head Akhil Srivastava. This twin applies to both nose and main landing gears and to the hydraulic system that drives the gear. It is used in a testbed to test gear architecture, analyze data and develop diagnostic and prognostic models. “A typical digital twin can be created in a few months provided the physical asset is mounted with sensors to capture its behavior,” he says.

GE already has used its engine experience to develop twins of components in several ATA chapters. “Even non-sensored parts can have predictive analytics using data-science techniques and correlated data sets, notes GE General Manager of Advanced Technology Darin DiTommaso.

Lufthansa Technik (LHT) now is working with digital twins of air-cycle machines and brake steering-control units, according to Gerko Wende, head of innovation for component services. Wende says twinning enables “data-centric production” in his repair shops. “The physical unit comes in with a digital footprint, and we can steer and supervise production.” But LHT needs component data from the entire life cycle, especially from operation and testing, and this data is so far only partially available.

2. Accuracy

As experience is gained and sensors added, virtual models become more accurate representations of their physical counterparts. The behavior of real assets in a wider variety of environments and circumstances becomes more precisely predictable and thus valuable in decision-making. Wende acknowledges his digital twins may never represent the physical part completely. “We’ll never have perfect digital twin, but we will get close,” he says.

3. Extension Over the Life Cycle

Digital twinning is starting earlier, in preliminary design phases and continuing longer, through all the repair events and eventual teardown and recycling actions. The ideal is an accurate digital representation of the asset from the time of its birth as an engineer’s concept to its death as a piece of scrap.

“Beginning with digital twinning during design, it’s logistically easier to link design and mechanical models with operational data,” says DiTommaso. Digital models can optimize designs for efficient production, especially for 3D printing.

4. Levels Of Representation

The principle of digital twinning can be applied to piece parts, which are then assembled and understood as digital components, which in turn cumulate to aircraft systems and then into whole airplanes. This is how real parts work.

This accumulation approach could be taken further. Virtual aircraft could be assembled into virtual fleets, whose aggregate behavior could be predicted by understanding probable behavior by tail number. Connected to other airline systems, scheduling, maintenance, revenue management and others, this might even someday enable a virtual airline to be roughly modeled.

GE is headed down that path. “Data analytics at aircraft level provides predictions of parts and systems driving delays and cancellations,” DiTommaso notes. “At the operations level, we can help airlines recover from fleet disruptions due to events like weather. Beyond that, we are creating applications that assist a variety of airline operations in cockpits and on the ground.”

5. Bigger And Different Data

All this is possible only because new aircraft are generating much more data per part, per minute and per flight and because other techniques are exploiting unstructured data, handwritten notes, pictures, voices and videos. It will take some time before all this bigger and different data is truly useful, but each step in understanding new data types makes the next step more feasible.

GE combines data from manufacturing, operations, full-flight cycles and other sources along with its physics-based understanding to predict engine behavior, says DiTommaso.

The biggest benefits of twinning will come from “clever fusions of different data sets,” says Wende. For example, LHT can combine aircraft position from ADS-B signals, add in weather updates, sensor data from aircraft and repair data from shops and vendors. “We will not predict degradation in theory, but will see it in a very robust data set under real-world conditions.” But one data set, text-mining from remote mechanic reports, is still far from complete.

6. Data Timeliness

New systems of broadband communications now move a tiny part of sensor data off aircraft in real time to inform day-of-operations maintenance decisions, while big chunks of less time-sensitive flight data are offloaded with Wi-Fi or cell networks on the ground. New services now transfer and format all this data for immediate use.

7. Who Is Digitizing

Early digital models of engines were built by the engineers who designed them, based mostly on physical principles and some testing. That is not enough anymore. Sensors generate data that adds to understanding of many assets, as do non-structured data. Specialized companies of data scientists have been acquired or contracted to sift these resources, using increasingly sophisticated statistical techniques. MROs and airlines collect data on repairs and operations that also can boost predictive precision. Since no one company owns all the data that can enable digital twinning, data-sharing needs have grown, adding new concerns about data ownership and confidentiality.

Lufthansa Technik has been digital-twinning engines for a decade, according to Mona Stuenckel, head of innovation at LHT Engine Services. The MRO begins with data collection on each engine as it is delivered, then develops individual twins of the engine, reflecting engine operator, age, usage and repairs.

LHT monitors the health of engines in operations, collecting data during various phases of each flight. “We started with smaller data sets,” Stuenckel explains. “Now we get full-flight data and snapshots of takeoff and climb.”

After 10 years of analyzing such data, LHT does not really need OEM digital twins. “We develop our own models and have most of the data we need,” -Stuenckel says. The MRO also sees engines in its shops during overhauls, noticing the wear and condition of parts and observing which parts must be replaced. “As long as we see data from operations and the shop side, we can do the models.” The LHT executive believes data analysis during operations provides a more valid basis for predicting engine behavior than OEM design data. She says the benefits are substantial, preventing aircraft-on-ground incidents, predicting airline maintenance costs, better fleet planning and smarter inventory volumes and locations.

8. Costs And Challenges

Realizing the true potential of digital twinning requires substantial resources and funding. For that reason and because of scale considerations, OEMs have tended to be most active in developing digital tools, MROs second and even major airlines third. Getting full-flight engine data from older aircraft remains a challenge. GE is working with airlines on downloading this data via modern connectivity systems and has set up a joint venture with Avionica to do this.

Handling huge amounts of data from the newest aircraft is still a challenge for digital twinning, Stuenckel admits: “There is so much in such a short time, computers have to struggle a little.”

Another twinning challenge is that “there are so many people with data in different systems,” says Wende. He argues twinning must focus on parts that justify expenditures; the minimum equipment lists no-fly defects. And planners must recognize the costs of sensors, possible faults in sensors and alternatives to sensors, such as images of aircraft and parts on the ground.

DiTomasso emphasizes progress is needed in data storage, computational horsepower and connectivity. 

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.