Digital twin constructs—digital models of physical, real-world devices—are taking on increased importance as Internet of Things (IoT) technology spreads, especially in the industrial space. Digital twins go by a variety of names (e.g., Amazon Web Services (AWS) prefers “device shadow,” Microsoft uses “device twin” as part of its Azure IoT Hub) but in all cases the concept is roughly the same: digital representations of real-time configuration and state information for physical devices. This makes it easier for software applications to interact with remote devices, whether to query them for conditions or instruct them to perform certain actions. But while useful, digital twins containing only real-time state information are limited in their utility. The next step in digital twin evolution is to imbue them with behavioral information.
In order to move to more advanced use cases, such as adaptive diagnostics, condition-based maintenance or predictive failure, industrial IoT systems need to know more than simply what the current device state is. They need to know why. Knowing current device state only helps from a monitoring standpoint; important, yes, but really just the beginning of what we expect out of IoT systems. If we know why an asset exhibits a certain state we can determine what conditions lead to that state and take proactive steps to prevent future occurrences (assuming the state is undesirable).
We can “fast forward” the device and see what its condition might be tomorrow or next week or next month (obviously with diminishing accuracy the further out in time we peer). We can more precisely determine diagnostics steps that will minimize mean-time-to-repair (MTTR). Collectively, we can improve asset uptime and performance; reducing repair, maintenance and warranty costs.
As innovation in IoT continues at breakneck speed, expect that “behavioral digital twins” will continue to grow in complexity, which mean an increase in model accuracy. Achieving this may mean digital twins built using multiple discrete machine learning algorithms potentially spread across multiple IoT platforms, not simply relying on one. Further, we should expect that digital twins will interact with one another in virtual space. An engine, transmission, and braking system may all have discrete digital twins but will need to interact with each other every bit as much as the real engine, transmission, and braking system do in order to garner even deeper insight into overall system behavior.
The bottom line—digital twins are central to the vast majority of industrial IoT use cases, especially those that yield the highest ROI.