The era of smart, connected physical assets is upon us, thanks to ubiquitous connectivity and the availability of inexpensive sensors. Insights derived from sensor data are being employed to optimize asset operations, and additionally are being fed back into the product design cycle for future product improvements. Sensor data coupled with advances in data science and machine learning have resulted in exciting new paradigms focused on the intersection of the physical and digital worlds, with the ultimate goal being to drive business value for companies. In the Industrial Internet of Things (IIoT) space, one such paradigm that is gaining increasing prominence is the concept of the ‘digital twin.’
What is a digital twin?
A digital twin is a virtual representation of a physical asset that enables an asset operator to derive actionable insights on both the performance and health of the asset. These insights can result in reduced costs, new revenue opportunities, and improved overall business operations. Sensors provide data on the asset’s operating conditions and key performance parameters that describe the asset’s real-world behavior. When both real-time and historical operational data are combined with physics based scientific insights from asset design, a unique digital representation of each asset emerges: the digital twin.
The scope of the digital twin can encompass all subcomponents of a physical asset. For instance, the digital twin of a car can include critical parts such as the engine, water pump, alternator, brakes, fuel pump, and so on. When these parts are equipped with sensors, they can measure the thermal or mechanical loads that these parts experience when the car is in service. When the car manufacturer combines design insights with this sensor data, it can predict when a critical part in the automobile will fail and notify the car owner well in advance of the projected failure. This practice allows for critical parts to be preemptively replaced before they break or wear out. Sensor data allows the manufacturer to study the part’s performance under real-world conditions and suggest improvements to future designs.
Every digital twin has a unique one-to-one correspondence with a physical asset in the field: if there were 1000 cars in the real world, they would correspond to 1000 unique digital twins.
The digital twin is a confluence of physics, sensors, and data
Designing a complex asset involves an elaborate process spanning several stages, starting from preliminary design all the way to detailed design, ultimately resulting in the final production-ready design. At each stage, various studies and analyses, such as fluid flow, mechanical, thermal, and control systems, are employed to harden the design elements in an effort to predict the performance of the asset in the field. When combined with testing in controlled environments, such analyses confirm the design objectives of the asset at the part level, system level, and system-of-systems level.
These studies result in physics-based ‘mathematical’ models that can predict the behavior of a part in service for a given set of environmental conditions and operational parameters. Expert opinion is an important ingredient in building such models, although data and analysis-driven approaches are now being employed to limit human intervention in building physics-based models. These models, along with the insights derived from them, constitute the first building block of a digital twin.
Once the asset is deployed in service, its sensors release a massive and varied amount of data that necessitate the use of data-driven techniques to obtain operational insights. Advanced techniques in data science, such as anomaly and outlier detection and Bayesian methods, are employed to obtain statistical insights from the field sensor data. Sensor enabled data and associated insights constitute the second building block of a digital twin. Figure 1 illustrates these building blocks for the case of a digital twin of a car engine.