In the IoT world, digital twins are sometimes already included in the scope of delivery. With such a software extension as an accessory, every facet of a product, process or service can be mapped in terms of data and used in a variety of ways.
A digital twin is the virtual (digital) representation of a physical object or system over its life cycle using real-time data. Such a digital image can be used as a basis for traceability, knowledge transfer (training), simulation of possible events, optimization and predictive maintenance (condition monitoring, predictive maintenance), inference of AI models and many other tasks. Current studies are even discussing the possibility of using digital twins for intrusion detection in industrial plants.

An important sub-task of a digital twin is data integration. The respective data can originate from different sources and sometimes have completely different formats. They are usually stored in a central database over the entire product life cycle. There they can be linked, analyzed, prepared for visualization tasks or processed in other ways as required.

Driving forces

The idea of the digital twin is not new. It became more widely known at the beginning of the 21st century through a conceptual development by Michael Grieves and John Vickers in the USA. The first implementations were in the aerospace industry. In the past decade, the concept of the digital twin was first integrated into the Internet of Things and then linked to Industry 4.0 ideas. There is now an unmanageable variety of research contributions from various universities around the world and at least as many topic-related but marketing-oriented contributions from various companies.

From a technical perspective, three areas can be identified as "enablers" for digital twins: 1. sensor technology, 2. data processing and 3. visualization options. There is a high pace of innovation in these different technology segments, which will further accelerate the practical use of digital twins. At the moment, cost aspects are a real showstopper when it comes to the use of sensors: valuable applications with digital twins require a very high number of sensors in order to create high-quality real-time data images. However, suitable sensor technology is sometimes too expensive. In functional terms, just a few sensors are enough to create a digital twin as part of a condition monitoring application, for example. However, the user benefit is then often insufficient. The spread of MEMS semiconductor sensors has already resulted in significant cost reductions. But that is not enough. The next major innovation step with printed sensors (3D-printed sensor devices), low-power wireless interfaces and energy harvesting power supply will enable high-resolution IoT data images with a hopefully acceptable price/performance ratio.

In terms of the required data processing options, you can still rely on Moore's Law and the semiconductor manufacturers. Computing power and storage options are constantly growing, while the layer thicknesses in semiconductor manufacturing technology are moving in opposite directions. However, further progress in the field of AI hardware accelerators would be helpful for the real-time data analyses of a digital twin.

Virtual reality (VR) and augmented reality (AR) are likely to play a particularly important role with regard to the data and information visualization possibilities that arise through the use of digital twins. Using VR and the data of a digital twin, products and systems can be realistically simulated before they are built. AR allows, for example, context-related status data from the digital twin to be superimposed onto the live image of a physical object at which the camera sensor of a smartphone is pointed. Combined with a zoom-in effect, the object can first be viewed from a bird's eye view and then zoomed in on the part of interest based on the superimposed data. This makes it possible, for example, to create virtual IoT service assistants for machines and systems that enable an untrained user on site to carry out complex maintenance and service work.

Different twin variants

As part of the planning phase for the use of a digital twin (DZ), both the theoretical and the practical status quo should be examined. A suitable evaluation system is helpful here. The Arup report "Digital twin: towards a meaningful framework" provides a very useful guide. It classifies digital twins into different categories based on four different capabilities and a five-stage degree of realization (level). Here is an overview in Arup metrics:

Autonomy: This capability assesses the extent to which a technical system can act without human intervention. At level 1, the capability is completely absent (the DZ practically only stores data and information). However, only the user acts. In the second level, the DZ software already offers prompts and notifications about certain states of a physical object as user support. At autonomy level 3, a DZ is already capable of triggering an alarm regarding a specific system status and carrying out a switching action (e.g. stopping a system in the event of a fault and switching on a red light). A DZ with autonomy level 4 features can monitor certain conditions and perform critical switching actions with minimal human intervention. At level 5, the DZ can use the data and software functions to safely perform a specific task with the corresponding physical object or system without user intervention.

Intelligence: The extent of this capability depends very much on future AI development (e.g: What will OpenAI-ChatGPT be able to do in 5 to 10 years? How will the further development of cognitive systems proceed?) Level 1 means no intelligence. At level 2, the DZ already reacts to stimuli, but cannot use previously gained experience to influence current actions. A level 3 DZ is already capable of learning in order to improve its reaction. It can also learn from historical data in order to make decisions. At intelligence level 4, the DZ takes into account the requirements (needs) of other intelligent systems. At level 5, the future state of AI development may even enable the DC to develop something like virtual self-awareness.

Learning ability: This characteristic can be used to determine the machine learning capabilities of a DC. The level 1 degree of realization means that no machine learning is used in the software stack of the DZ. Level 2 indicates solutions that use at least one rule-based expert system to obtain and use information. In learning capability level 3, there is a task-based supervised machine learning model with which the digital twin can perform inference-based data analyses at any time, for example to classify sensor data images in order to recognize known patterns. In level 4, data is evaluated with the help of unsupervised machine learning algorithms, for example to detect an anomaly in the available sensor data (the main task in this level is the automated detection of data clusters). Level 5 means that the DZ uses reinforcement machine learning. This method of machine learning enables the DZ to independently learn a behavioural strategy with the help of a reward system and to maximize the result in terms of sensor data usage.

The author: is a member of the management team at SSV Software.

© SSV Software Systems

Data accuracy: The deviation of the data and information image of a DC from the actual state of the physical object or system can be assessed via the level of this capability. At level 1, the accuracy is so low that the DC corresponds at best to a conceptual model. Level 2 stands for a low to medium accuracy range; however, useful measurement data can be extracted from the data image of the DZ for further use. At level 3, the DZ offers a medium level of data accuracy for the state of the physical object or system. With a level 4 degree of realization, precise measurement data describes the actual physical object or system states (the information generated has a very low error rate). At data accuracy level 5, the high-precision data and information images are already suitable for the critical operating decisions of a fully autonomous vehicle.