Cognitive Digital Twins
A Digital Twin, as explained in the previous posts, is the sum of three data sets:
- one representing the model of the physical entity
- one representing the current status of the physical entity
- one recording the history of the physical entity, including the relations it had (like what robots built it) and the sequence of states (timeline of the shadows)
All together these data set represent “what” is the physical entity.
It is possible to create a Digital Twin to represent a process, rather than a product. As an example an enterprise can have several digital twins, one for each of its processes (supply, manufacturing, resource management, sales, maintenance,… ). Here again the Digital Twin will keep a data set modelling the process, another mirroring the current status of the process and the digital thread recording the evolution over time of the process. When looking at the Digital Twin of a process one can understand “how” enterprise activities are executed, including, as an example, how a product is built/operated/maintained.
This information can, as an example, be used by an assembly line of robots to “teach” them how to build a specific product. In a sense this can be seen as the representation of a given space of knowledge.
IBM was probably the first company to investigate this type of Digital Twins and they called them “Cognitive Digital Twins” (GE was probably the first to use the first type of Digital Twins at an industry level).
This type of Digital Twins is very important for companies in general since a company is defined by its processes (people can come and go –and they do- the company stays and its processes are its trademark) and their efficiency is what make it competitive on the market.
Modelling processes using bit is not particularly tricky, most companies already have a computerised management flow that supports the various activities and the interplay of resources (including human workers). Hence it is usually straightforward to create a formal digital model of an industrial process (this can be very detailed, like what piece of papers have to be filled, who is authorising what, where are spare parts located, what to do to manage emergencies…).
The shadowing is usually more complex (companies are using various methodologies, more and more computerised to monitor “what is going on” and the results of what is going on). The crucial aspects related to shadowing are:
- the velocity, i.e. how often is the status of the process updated
- the resolution, i.e. at what granularity is the process monitored
- the exception handling, i.e. what should be done if the monitored process diverges from the expected path.
The shadowing is crucial to the smooth running of the enterprise. The more frequently a process is monitored and the finer the granularity the better from the point of view of detecting divergences and activate appropriate measures when needed. On the other hand, this increase in velocity and resolution calls for bigger cost, so a trade-off is needed.
However, the Digital Transformation can result in a very high velocity and resolution with no cost increase!
The key is to leverage on IoTs, pervasive in the whole value chain and to use Cognitive Digital Twins to aggregate and pre-process the data. This is what platforms like Mindsphere and FIWare make possible. The use of robots in many activities provides the sensing as well as the actuation points needed.
The communication infrastructure is obviously crucial and the performance provided by 5G becomes interesting. This is why, as an example, Siemens has acquired 5G spectrum for use in the factory plan.
Cognitive Digital Twins closely mirror factory processes and either directly, by embedding intelligence or indirectly by interacting with intelligent applications can replicate in the cyberspace the factory processes and steer the physical processes in case something unexpected/unwanted happen.
There is actually much more in store for Cognitive Digital Twins….