Much has been written in recent years about the concept of digital twins. In this article, we will explore the benefits of this technology as well as the role that artificial intelligence (AI) can play in optimizing digital twin systems. Specifically we will explore the ways in which SparkCognition offerings like Maintenance Advisor and Visual AI Advisor leverage the latest in machine learning technology to ensure that digital twin programs deliver the performance levels organizations need to be successful.
Let’s start with an obvious question: What exactly is a digital twin? Here’s the official definition from a large manufacturer:
A digital twin is a virtual representation of an object or system that spans its lifecycle, is updated from real-time data, and uses simulation, machine learning, and reasoning to help decision-making.
The computerized digital twin is a concept that’s been around in one form or another for a couple of decades. But—as so often happens with newish technical jargon—it can mean different things to different people. In its simplest terms, a digital twin is a simulation, but with some important differences from the sorts of simulations that have been a part of technology in decades past.
The origination of the digital twin concept is a bit tough to pin down, but the ultimate credit probably goes to NASA. Since the 1960s America’s space agency has made a practice of emulating complete spacecraft back on Earth that operate in real-time in conjunction with their physical counterparts on missions orbiting the Earth or traveling farther afield. The concept—at least in the days of the Apollo program—was similar, though not quite the same. NASA was operating identical physical systems at home and in space rather than modeling the latter on computers, the capability for which did not exist at the time.
Whereas a traditional simulation tends to be about modeling one specific process, a digital twin is an identical but virtual reproduction of an entire physical system or device, one that runs in parallel with the analogous system or device. And so, rather than a computer program whose purpose is to forecast the performance of a process in advance, a digital twin is a tool for predicting and/or maintaining operational performance in real-time, working in concert with its physical counterpart.
How do digital twins work?
A digital twin can be created for any system or device that can be constructed in real life, be it a jet engine, wind turbine, or factory assembly line. Another important difference between a digital twin and a basic computerized simulation is that the digital twin operates using real-time data inputs from sensors on the actual system or device rather than predicted or OEM-provided values for important performance parameters. In this way, early stage failures or operational degradation can be identified before they occur in the real system.
In addition, alternative performance data can be input into the model to conduct scenario analyses in support of proposed maintenance actions. Finally, the digital twin model can make real-time recommendations about actions that can be taken proactively to avoid failures or performance compromises.
Digital twin applications
Digital twins have a number of important uses:
- Monitor real-world conditions and operating performance of a physical system.
- Model system performance prior to actual production or modification of a physical asset.
- Improve operation of physical systems in the field.
- Predict maintenance needs and prevent failures.
- Model proposed maintenance or physical system changes prior to implementation.
- Identify potential operational hazards.
Digital twins work well with machine learning capabilities like those employed by SparkCognition to add greater understanding to systems which over time develop new inputs that influence performance.
Josh Fox, SparkCognition’s director of product marketing, notes that there are differing viewpoints on what components are needed to create a full digital twin, with one factor being whether a 3D image or some other visual model is implemented. He added that machine learning approaches can often gain useful predictions and insights without the visual component coming into play.
“If you have a physics-based model, a means of visualizing the system, and a way to perform simulations, you can use that in a number of ways. It is, for example, useful to engineers who are building out a new production line or process to test how things are working. But more and more, customers want to use it for predictive analytics,” he said.
“Generally speaking, it involves developing some sort of model, often physics- or rules-based, to enable simulations to forecast performance in a way that can be used to compare against what’s actually occurring. A data-based approach can detect unknown unknowns and detect anomalies for an overall system whereas digital twin models tend to be more rigid and centered around individual assets.”
A sample of current digital twin applications provides a sense for the scale and scope of benefits the technology is delivering across the industrial world.
- Siemens Agent-based Turbine Operations & Maintenance (ATOM) system emulates the global maintenance, repair, and overhaul (MRO) of their fleet of aero-derivative gas turbines.
- Jet engine manufacturer GE develops digital twins of all its aerospace products. Built into each physical engine are hundreds of sensors that supply temperature, pressure, and other performance data to the model for each unique engine. Potential performance issues identified by the model are then shared with maintenance crews at the site where the physical engine is operating.
- CNH Industrial (parent company of CASE and other large manufacturing firms) uses digital twin technology to model its vehicle assembly lines and identify effective maintenance practices on the line. The system provides detailed information about the economic and production consequences of different maintenance policy configurations.
- The Port of Rotterdam uses digital twins to more effectively manage its shipping operations, including identifying optimal times for vessel mooring and departure.
It’s useful to think about digital twins as two-way streets when it comes to the flow of information: sensor performance data flowing from the physical system into the model, and performance improvement recommendations flowing from the model back into the physical system. This bidirectional flow serves at least two purposes: it allows the physical system to operate better over time, and the cyclical nature of the data flow allows the digital twin model to constantly improve its own analytical abilities, thus becoming an ever-improving source of value going forward.
The interplay of digital twins and artificial intelligence
This last point leads squarely into a discussion of how digital twins and artificial intelligence (AI) complement one another. To better understand the relationship between the two technologies, it’s important to acknowledge a couple of potential drawbacks to digital twin models.
First, because model outputs are typically based on well-defined, inflexible performance thresholds, the frequent recurrence of alarms can lead to alert fatigue, i.e., an overwhelming quantity of alerts, only a small percentage of which are severe enough to require human intervention. Second, digital twin models are based on OEM standards of performance for new systems and do not modify their alerting thresholds as the system ages and its performance evolves with time.
AI-based normal behavior modeling (NBM) is uniquely well-equipped to deal with both of these limitations. By continuously updating its understanding of “normal” system operation based on ongoing performance and frequent updates, a far smaller quantity of meaningful alerts are generated. And because NBM models are periodically updated to reflect the very latest operational performance of the system, the model evolves with time, keeping it relevant and accurate.
Finally, whereas digital twins are typically focused on a single piece of equipment or asset (e.g., a turbine), NBM models are effective at modeling entire complex systems comprising dozens or even hundreds of individual assets. These factors make AI-powered NBM predictive maintenance models excellent complements to digital twins.
The future of digital twins
Industry analysts expect the digital twin market to rise from $6.9B in 2022 to as much as $73.5B by 2027, representing a CAGR of 60.6%, with the healthcare and pharmaceutical sectors serving as the largest drivers of the market. And so, with the simultaneous growth of the digital twin and artificial intelligence marketplaces, it makes sense to regard the two as companion technologies, each ideally equipped to maximize the benefits provided by the other.
SparkCognition offerings like Maintenance Advisor and Visual AI Advisor use the latest in machine learning technology to ensure that your digital twins program delivers the performance upgrades your organization requires while eliminating false positives and the alert fatigue they can cause.
To learn more about how SparkCognition’s AI solutions can complement your digital twin program, check out our webinar.