Digitalized programmable physical world
Imagine that we create digital replicas, commonly referred to as digital twins, of everything we engage with in the real world. All physical things and places in the real world – buildings, roads, factories, farm fields, and pets – will become defined by software and powered by AI. These physical objects and places can then be programmed, later automated, or even self-learning, in the long run, enabled and empowered by the future network platform.
Where are we heading?
We see tremendous growth today in how things get connected for various purposes, like monitoring and controlling everyday objects in what we call the Internet of Things. This could be anything from simple household appliances to large utility infrastructures such as the Smart Grid. Discrete and process manufacturing is another example, which builds on cyber-physical systems. We see how IoT and cyber-physical systems blend, moving towards Industry 4.0 and later Industry 5.0. This is a development towards a physical reality where every imaginable thing is connected and digitally engaged in various activities and processes across industries and society at large.
Digital representations of the physical world will be enabled by billions of embedded sensors and actuators that provide the necessary data and make it possible to control physical actions. We are moving to a future where it will also be possible to create digital representations of very large entities – like whole cities with their intricate infrastructures and sprawling activities. These digital twins can be simultaneously accessed and controlled by large numbers of humans and applications alike, for detailed planning and execution of activities. They will be precise in both space and time as well as accurate enough to allow correct management and optimization of the city activities. In this digital representation, we can move backward and forward in time or observe and act here and now in real time. The network provides intelligence, ever-present connectivity, and the necessary processing capabilities to enable this cyber-physical continuum.
What is needed?
Sensors and actuators - the interface between the physical and digital worlds
Objects equipped with sensors, actuators, computing, and networking are the foundation for a digitalized reality. Today, we already have wearables. In the future, with new materials and sensor components, sensing can be embedded in clothing or even implanted into our bodies.
Tiny sensors for nonliving objects like buildings, cars, roads, robots, are getting smaller and smaller, cheaper and cheaper, and are shipping in billions today. New materials and new sensing technologies increase the number of sensing modalities – from temperature and heart rates to chemical compounds and various gases, to mention two. Printed and bio-degradable electronics can open up for new deployment scenarios and applications that are not possible today. Embedded computing, including support for Machine Learning based inferencing, has already made even tiny processors capable of extracting insights. The sensor, actuation, and computing evolutions not only serve the low-end of the scale of physical objects, but also the instrumentation of heavy and expensive machinery, like factory production cells and wind turbines.
Sensing and actuation can involve a wide variety of data types, reflecting the availability of various sensing and actuation modalities, like humidity, nutrition level, or turning on an insulin pump embedded in a person’s body. Data speeds and reliability requirements can range from single sensor readings once in a while, perhaps on a daily basis, to strict real time streaming of data and control commands used in advanced deployments, like in industrial processing sites. A factor to consider is that sensor and actuator deployments can be intended to work for up to 10-20 years.
Depending on the industry segment they are applied to, the sensor and actuator technologies will be many and different. Ensuring interoperability across all layers, from protocols to data syntax and semantics, will continue to be of paramount importance.
From programmability to automation and cognition
Programmability is today mainly achieved through low-level programming or manual setting and tuning of parameters used in various models. The use of AI, machine learning, and machine reasoning technologies is increasing rapidly, leading to digital twins becoming more self-adaptive and automated, able to deliver on desires and tasks expressed on a more abstract or higher level. This can be done by defining intents or objectives, a wanted behavior, or a certain outcome. For instance, we can program a machine to order service on parts before it fails and to do so with minimum impact on production yield or delivery lead times. With the right prediction, knowledge of production plans, and availability of service support, this can be worked out by the digital twin itself. This implies that all necessary resources are networked and interoperable, and the right information is constantly available.
Over time, digital twins will become more and more automated, adaptive, and eventually cognitive. They will start as infants, learn by interacting with others, and become wise adults, possibly without humans featuring anywhere in the loop.
Digital representation of the physical object – a digital twin
Creating a digital twin of a physical object implies that it accurately enough captures the properties as well as the behavior of the object, and at varying time scales from real time to relaxed, depending on the type of object and the processes it is involved in.
A digital twin can exist even before its physical counterpart does, meaning it can be a model and design of something that is yet to be manufactured or assembled. That way, the impact from the production process for each manufactured object can be captured in the digital twin, for example for quality assurance. Likewise, simulations by the digital twin can forecast and predict future behaviors such as a future malfunctioning machine for instance. Proper models are key here, based on the laws of physics, chemistry, fluid dynamics, etc., all depending on the object at hand. Adaptability is an important feature; the digital twin needs to keep observing the behavior of the physical object in order to fine-tune any preprogrammed models, parameter sets, and so on.
In one aspect, digital twins create a layer of indirection of data and information to deliver a more end-user-centric representation of knowledge that is intuitive and focuses on desired properties or behaviors at a more abstract and comprehensive level. AI, machine learning, and machine reasoning are key capabilities for deriving and capturing this knowledge accurately in representations, for example, through ontologies. These technologies are also used for inferencing new - a priori unknown or undiscovered - dependencies, behaviors, and relations.
As already alluded to, digital twins can be both collaborative and engage in aggregations hierarchically, as a system of systems. A typical example of the former is digital twins of AGVs collaborating on solving a particular task like keeping a construction site clear of litter and replenished with materials and tools. An example of digital twins in a more hierarchical setup is manufacturing machines that are dynamically aggregated and configured to manufacturing cells, factory floors, and entire factories.
Needless to say, whether simple or rich, the mutual interactivity of digital twins is key in any process where physical objects integrate into various physical processes. This interactivity is driven by a continuous exchange of data and knowledge, behavior, and intents. Again, proper semantics and automated processes to solve any heterogeneity will be needed in the future, along with appropriate contextual information.
Positioning and mapping
Navigating among objects in the physical world is obvious, and therefore, a key capability for digital representations is positioning and mapping. Position, and wider orientation and direction, is in itself one part of the information needed for the digital representation whether it is geolocation or locations inside buildings and sites, especially considering both absolute and relative positions. Mapping is about capturing and describing the structural view of the physical world. For both of these capabilities, the use of network and device information will be important, e.g. using radio propagation information or device reported information.
Cyber-physical security and safety
As these systems control physical objects and machines, ensuring trust, reliability, and safety is important. Undesired actions by a machine can cause harm to people and damage physical property. Reliable models and understanding of any uncertainties will be of increasing importance as more machines rely on digital representations and automation for their actions. The integrity of information to protect individuals is also an important aspect.
