As we typically hear it referred to in industry today, a digital twin is a virtual representation of both the components and the real-time dynamics of a physical product throughout its life cycle. For example, an accurate digital twin of an aircraft engine not only matches all of the engine’s initial qualities (weight, composition, structure), but also continues to match the engine’s condition by accounting for every flight taken, every weather condition encountered, every maintenance or repair performed, type of fuel used, etc. This extensive data, gathered from IoT devices like smart sensors and cameras, should enable the extrapolation of the actual engine’s conditions, making it easier to predict failure and schedule preventative maintenance. Because the data used to maintain the digital twin is unique—no two engines will take identical flights or encounter identical conditions—the digital twins of two different units of the same aircraft engine should evolve differently, dependent on the different conditions each engine is subjected to.
In this article, we would like to extend the concept of digital twins beyond just the end product and into the process and resources used to create and maintain the product. When we refer here to a digital twin, we include the digital representation of both the components and dynamics of a product, and of the process, facilities, and equipment used to create it. Each element of this manufacturing cycle should have its own unique digital twin. Analysis of all of these moving parts of the manufacturing process could lead to more efficiency, faster detection of quality issues, and the identification and removal of bottlenecks, all with reduced cost of production and potentially higher quality.
However, creating these types of digital twins can be challenging for two key reasons. The first is that in order to create a true digital twin of a physical product, process, or piece of equipment, the number of features that need to be accounted for is staggering. Identifying millions of them and tracking changes in them in real time is not only difficult but also may not create a representation that satisfies every potential use case.
Secondly, the raw data from sensors on physical devices can lend themselves to multiple interpretations, depending on the business challenge being addressed or the specific AI solution being created. No matter how simple the data—for instance, a temperature reading—it can be interpreted differently depending on the stage of the manufacturing process or on the problem being solved. It may not be simple to decide which interpretation is correct, and digging deeper takes more time.
The solution to these challenges is to follow a new process for the creation of digital twins:
- Identify all of the individual business challenges that need to be addressed with AI and automation.
- For each potential problem to be solved, determine the elements in the manufacturing process that must be represented digitally to identify a solution. In other words, which elements are relevant to this specific problem?
- Represent each physical element as a separate digital twin, tailored specifically to the AI solution being built. In this way, the number of features that must be accounted for and tracked are reduced, and only the relevant ones remain. Plus, there can be only one correct interpretation of the data.
- There are now multiple digital twins for each physical element in the process: one for each AI solution that will be built.
This approach generates many more digital versions of each physical object involved, but since each one is tuned to a specific solution being sought, they generate more accurate results and therefore more trustworthy AI solutions that are simpler to manage.
One way to understand the efficacy of this approach is to imagine digital twins of yourself. If you could create separate versions of yourself carefully tuned to each aspect of your life, you might even achieve more success than you do now! Imagine a version of you that goes to work and is never distracted by worrying about mowing the lawn, thinking about what the children are doing at school, or deciding what to make for dinner that night. Meanwhile, the version of you that spends time at home never gets distracted by what happened at work that day or by the meeting scheduled for tomorrow. Having specifically tuned digital twins is similarly effective.
Any effort towards implementing accurate AI solutions should start with our four step process outlined above. This will create a foundational layer on top of existing control systems and sensors that are probably already embedded into the manufacturing process for most enterprises. Once these foundations are in place, enterprises can establish more intelligent and predictive manufacturing processes that require less human oversight and consistently perform more efficiently and more accurately than before.
Organizations like CrowdANALYTIX specialize in creating AI solutions that achieve automation and cost reduction goals for businesses across industries, including manufacturing. Our solutions are customized to organize business data of all kinds, and to maintain and deploy custom models that can accelerate digital transformation and help businesses reach their goals more efficiently and more profitably.