Digital twins, virtual representations of physical assets, can bring benefits to a shopfloor such as reducing time and costs. Mapping the real asset against the digital world can provide optimisation, error analyses and simulations.
Digital twins often store geometry data, as well as kinematics and behaviour, but feeding them with data from the shop floor requires a more sophisticated use of sensors and computing, such as edge and cloud devices. Protocols such as OPC UA (open platform communications unified architecture) can be used to connect the sensors and actuators via ethernet to the digital twin, allowing live monitoring of the production and subsequent optimisations.
To use the digital twin on SCADA, the digital twin must meet industrial requirements such as real-time capability. Most systems at this level have hard real-time requirements to ensure predictable behaviour. Current digital twin solutions do not support these requirements and can only be used for a retrospective analysis of the process, not for optimisation during the runtime. However, those optimisations have potential to reduce rejects and optimise quality. This raises the question: can the digital twin also be used down to the field level? How can these systems benefit from a digital twin? What would be tangible use cases? And could they be implemented using software frameworks? What are the specifics and challenges of a digital twin in a real-time context?
The goal of this ICNAP study is to answer these questions.
ISO 23247-1 defines a digital twin as ‘a fit for purpose digital representation of an observable manufacturing element with synchronisation between the element and its digital representation.’ The observable manufacturing elements can include processes and documents. Synchronisation allows users to update the digital twin and the observable manufacturing element with the respective values.
The standard also presents a digital twin framework for manufacturing. It consists of a user domain, a digital twin domain and a device communication domain that collects data from the observable manufacturing elements and actuates them. The user domain analyses the digital twin for humans, for example via a human machine interface. Alternatively, the user domain also provides access to the digital twin for enterprise resource planning or manufacturing execution systems. The digital twin domain holds the digital twin, takes care of updating it with the data from the device communications, and manages access to the digital twin by the user domain.
Real-time computation and communication
Important processes within automation and production technology are subject to real-time requirements. There is a difference between hard and soft real-time, in terms of how long after the deadline the value of a computation or communication is valid. In hard real-time, the data loses its value directly after the deadline. If the system misses the deadline, the quality of the product will be affected, or someone could be in danger. In soft real-time, the data is also valid if the deadline is missed occasionally. If the system misses the deadline, the operation degrades.
Most digital twins only apply soft real-time. Real-time computation systems must do the respective data processing according to the requirements of the connected outside process. Therefore, they have to be synchronised with the external events and react directly. The availability of the result of the processing must then meet the deadline predetermined by the outside process.
For real-time communication, special physical Fieldbus systems are used for time-critical processes in automation and production technology. These connect control and field devices and enable cyclic and deterministic communication. However, it is difficult or impossible to combine them with networks of devices that use other communication standards such as cloud or edge Systems relying on IEEE 802.3 ethernet.
The first application for digital twins in manufacturing listed by ISO 23247-1 is real-time control of manufacturing process, which could be categorised as a hard real-time process. A digital twin in a real-time context has hard or soft real-time timing requirements placed on it.
Use cases
In a whitepaper from the TransContinuum Initiative, the term ‘real-time digital twin’ refers to the use ‘for online prediction and optimisation of highly dynamic industrial assets and processes.’ This is posing soft and hard real-time requirements on digital twins. One proposed use case is in a milling process, when accuracy depends on the stiffness of the milling machine. In this use case, the vibration that occurs during the milling process is predicted during the process. The results are then used to compensate for the vibrations occurring in a robot during the milling process (see also article, pp12-13).
Another use case is increasing motor uptime. The temperature inside the electrical drive is a limiting factor for the operation of the motors. In this use case, the digital twin is used as virtual sensor to sense this temperature. With this information, the availability of the motor can be increased because the system can quickly detect when the temperature has dropped enough for the motor to be used again.
The authors devised a testing framework consisting of the device communication entity, the digital twin entity, the user entity and a cross system entity to serve as the base for the software evaluation. The observable manufacturing elements communicate with the digital twin via the device communication entity, which has sub-entities for data collection and device control. The information is then processed by the digital twin entity, which is the virtual representation of the observable manufacturing elements. The evaluation focuses on functional entities that would be included into a closed loop setup (depicted in diagram at left).
The authors list a number of potential software frameworks that could be used with real-time digital twins which were evaluated using the testing framework.
First is the Asset Administration Shell (AAS), which, in combination with an asset, is called an Industry 4.0 component. Assets can be machines or material, but also documents or contracts – everything else that needs to be integrated into the virtual world for an Industry 4.0 solution.
Second is the OPC UA standard provides a base information model, which is extended by more detailed specifications by the OPC Foundation. Companion specifications are built on top by industry partners agreeing on a common information model for a particular type of asset. For example, German industrial standards body VDMA has working groups developing companion specs for robotics or drive technology.
Third is Eclipse Ditto, an open-source software framework that provides an Internet of Things middleware that connects and abstracts the devices to and from the applications. It provides access to the digital twin via an application programming interface and provides authentication.
Fourth is the Robot Operating System ROS2, an open-source meta operating system for robotics. ROS can be divided into middeware, algorithms, and developer tools. These developer tools contain tools for configuration, visualisation and simulation. The algorithms are suited for robotic applications.
Each of the presented software frameworks has advantages and disadvantages. For example, the AAS provides standardised information modelling, but lacks maturity and easy deployment. Meanwhile, Eclipse Ditto offers APIs easy communication with the digital twin.
Conclusions
It is not possible to cover every function of the digital twin within a single framework. Building a digital twin requires building architecture spanning the whole framework of communications with the assets and building connections to further applications or other digital twins. The presented frameworks can play a part in building such an architecture, but it is important to keep standardisation and interoperability in mind.
The digital twin functions that can be fulfilled in real-time are limited. For soft real-time systems, all frameworks presented might be sufficient, but this should be evaluated on a use-case basis before the implementation.
None of the software frameworks can currently be used to implement a digital twin off the shelf for hard real-time use cases.
The most promising solutions are ROS2 and OPC UAFX. ROS2 offers a real-time framework intended for robotics with the potential for implementing digital twins, although this possibility should be explored further. The real-time communication that is already possible via OPC UAFX could be advanced.
In addition, the current model creation and deployment are done manually, which is not scalable. New service-orientated architectures and solutions for the exchange of data and simulation models are missing. Also, the effort of implementing a digital twin needs to be reduced. DevOps, the conjunction of development and operation used in software development, should be applied to digital twins. Additionally, the usability of the digital twin has to be improved. Here, mixed reality techniques as well as low code approaches are promising to make interacting with a digital twin easier.
Finally, it is important to be able to trust the predictions of digital twins that are used in a safety or quality critical process. Trust in data can partly be enhanced using metadata by including a timestamp, the last calibration or type of sensor used. This requires an agreed upon standard of metadata and a specification on what data is needed for which digital twin. Trust in the model can be increased by using uncertainty quantification and a validation strategy that encompasses chains of interdependent models.
Real-time digital twins have a lot of potential for industrial production. While more research is needed, some implementations are already possible using specific models and carefully selected software.
This article is an edited summary of a chapter of the 2022 ICNAP study report (www.is.gd/qurope), which in addition to digital twins also covered pricing models for industrial data, digital infrastructures for sustainable production, cybersecurity and data spaces for data-driven production. It is reprinted with permission from ICNAP, www.is.gd/afatibe, an initiative comprising the German state of North Rhine-Westphalia and the national R&D institute Fraunhofer-Gesellschaft.